diff --git a/.gitignore b/.gitignore
index 3f7b860..09707c6 100644
--- a/.gitignore
+++ b/.gitignore
@@ -4,11 +4,31 @@
 *.a
 *.ci
 *.csv
+*.t.*
+*.b.*
+*.gcno
+*.gcda
+*.perf
+lfs
+liblfs.a
 
 # Testing things
-blocks/
-lfs
-test.c
-tests/*.toml.*
-scripts/__pycache__
+runners/test_runner
+runners/bench_runner
+lfs.code.csv
+lfs.data.csv
+lfs.stack.csv
+lfs.structs.csv
+lfs.cov.csv
+lfs.perf.csv
+lfs.perfbd.csv
+lfs.test.csv
+lfs.bench.csv
+
+# Misc
+tags
 .gdb_history
+scripts/__pycache__
+
+# Historical, probably should remove at some point
+tests/*.toml.*
diff --git a/Makefile b/Makefile
index 7cc59f8..24865e5 100644
--- a/Makefile
+++ b/Makefile
@@ -1,173 +1,582 @@
-ifdef BUILDDIR
-# make sure BUILDDIR ends with a slash
-override BUILDDIR := $(BUILDDIR)/
-# bit of a hack, but we want to make sure BUILDDIR directory structure
-# is correct before any commands
-$(if $(findstring n,$(MAKEFLAGS)),, $(shell mkdir -p \
-	$(BUILDDIR) \
-	$(BUILDDIR)bd \
-	$(BUILDDIR)tests))
-endif
-
+# overrideable build dir, default is in-place
+BUILDDIR ?= .
 # overridable target/src/tools/flags/etc
 ifneq ($(wildcard test.c main.c),)
-TARGET ?= $(BUILDDIR)lfs
+TARGET ?= $(BUILDDIR)/lfs
 else
-TARGET ?= $(BUILDDIR)lfs.a
+TARGET ?= $(BUILDDIR)/liblfs.a
 endif
 
 
-CC      ?= gcc
-AR      ?= ar
-SIZE    ?= size
-CTAGS   ?= ctags
-NM      ?= nm
-OBJDUMP ?= objdump
-LCOV    ?= lcov
+CC       ?= gcc
+AR       ?= ar
+SIZE     ?= size
+CTAGS    ?= ctags
+NM       ?= nm
+OBJDUMP  ?= objdump
+VALGRIND ?= valgrind
+GDB		 ?= gdb
+PERF	 ?= perf
 
-SRC ?= $(wildcard *.c)
-OBJ := $(SRC:%.c=$(BUILDDIR)%.o)
-DEP := $(SRC:%.c=$(BUILDDIR)%.d)
-ASM := $(SRC:%.c=$(BUILDDIR)%.s)
-CGI := $(SRC:%.c=$(BUILDDIR)%.ci)
+SRC  ?= $(filter-out $(wildcard *.t.* *.b.*),$(wildcard *.c))
+OBJ  := $(SRC:%.c=$(BUILDDIR)/%.o)
+DEP  := $(SRC:%.c=$(BUILDDIR)/%.d)
+ASM  := $(SRC:%.c=$(BUILDDIR)/%.s)
+CI   := $(SRC:%.c=$(BUILDDIR)/%.ci)
+GCDA := $(SRC:%.c=$(BUILDDIR)/%.t.gcda)
 
+TESTS ?= $(wildcard tests/*.toml)
+TEST_SRC ?= $(SRC) \
+		$(filter-out $(wildcard bd/*.t.* bd/*.b.*),$(wildcard bd/*.c)) \
+		runners/test_runner.c
+TEST_RUNNER ?= $(BUILDDIR)/runners/test_runner
+TEST_A     := $(TESTS:%.toml=$(BUILDDIR)/%.t.a.c) \
+		$(TEST_SRC:%.c=$(BUILDDIR)/%.t.a.c)
+TEST_C     := $(TEST_A:%.t.a.c=%.t.c)
+TEST_OBJ   := $(TEST_C:%.t.c=%.t.o)
+TEST_DEP   := $(TEST_C:%.t.c=%.t.d)
+TEST_CI	   := $(TEST_C:%.t.c=%.t.ci)
+TEST_GCNO  := $(TEST_C:%.t.c=%.t.gcno)
+TEST_GCDA  := $(TEST_C:%.t.c=%.t.gcda)
+TEST_PERF  := $(TEST_RUNNER:%=%.perf)
+TEST_TRACE := $(TEST_RUNNER:%=%.trace)
+TEST_CSV   := $(TEST_RUNNER:%=%.csv)
+
+BENCHES ?= $(wildcard benches/*.toml)
+BENCH_SRC ?= $(SRC) \
+		$(filter-out $(wildcard bd/*.t.* bd/*.b.*),$(wildcard bd/*.c)) \
+		runners/bench_runner.c
+BENCH_RUNNER ?= $(BUILDDIR)/runners/bench_runner
+BENCH_A     := $(BENCHES:%.toml=$(BUILDDIR)/%.b.a.c) \
+		$(BENCH_SRC:%.c=$(BUILDDIR)/%.b.a.c)
+BENCH_C     := $(BENCH_A:%.b.a.c=%.b.c)
+BENCH_OBJ   := $(BENCH_C:%.b.c=%.b.o)
+BENCH_DEP   := $(BENCH_C:%.b.c=%.b.d)
+BENCH_CI    := $(BENCH_C:%.b.c=%.b.ci)
+BENCH_GCNO  := $(BENCH_C:%.b.c=%.b.gcno)
+BENCH_GCDA  := $(BENCH_C:%.b.c=%.b.gcda)
+BENCH_PERF  := $(BENCH_RUNNER:%=%.perf)
+BENCH_TRACE := $(BENCH_RUNNER:%=%.trace)
+BENCH_CSV   := $(BENCH_RUNNER:%=%.csv)
+
+CFLAGS += -fcallgraph-info=su
+CFLAGS += -g3
+CFLAGS += -I.
+CFLAGS += -std=c99 -Wall -Wextra -pedantic
+CFLAGS += -ftrack-macro-expansion=0
 ifdef DEBUG
-override CFLAGS += -O0
+CFLAGS += -O0
 else
-override CFLAGS += -Os
+CFLAGS += -Os
 endif
 ifdef TRACE
-override CFLAGS += -DLFS_YES_TRACE
+CFLAGS += -DLFS_YES_TRACE
+endif
+ifdef YES_COV
+CFLAGS += --coverage
+endif
+ifdef YES_PERF
+CFLAGS += -fno-omit-frame-pointer
+endif
+ifdef YES_PERFBD
+CFLAGS += -fno-omit-frame-pointer
 endif
-override CFLAGS += -g3
-override CFLAGS += -I.
-override CFLAGS += -std=c99 -Wall -pedantic
-override CFLAGS += -Wextra -Wshadow -Wjump-misses-init -Wundef
 
 ifdef VERBOSE
-override TESTFLAGS     += -v
-override CALLSFLAGS    += -v
-override CODEFLAGS     += -v
-override DATAFLAGS     += -v
-override STACKFLAGS    += -v
-override STRUCTSFLAGS  += -v
-override COVERAGEFLAGS += -v
-endif
-ifdef EXEC
-override TESTFLAGS += --exec="$(EXEC)"
-endif
-ifdef COVERAGE
-override TESTFLAGS += --coverage
-endif
-ifdef BUILDDIR
-override TESTFLAGS     += --build-dir="$(BUILDDIR:/=)"
-override CALLSFLAGS    += --build-dir="$(BUILDDIR:/=)"
-override CODEFLAGS     += --build-dir="$(BUILDDIR:/=)"
-override DATAFLAGS     += --build-dir="$(BUILDDIR:/=)"
-override STACKFLAGS    += --build-dir="$(BUILDDIR:/=)"
-override STRUCTSFLAGS  += --build-dir="$(BUILDDIR:/=)"
-override COVERAGEFLAGS += --build-dir="$(BUILDDIR:/=)"
+CODEFLAGS    += -v
+DATAFLAGS    += -v
+STACKFLAGS   += -v
+STRUCTSFLAGS += -v
+COVFLAGS     += -v
+PERFFLAGS    += -v
+PERFBDFLAGS  += -v
 endif
+# forward -j flag
+PERFFLAGS   += $(filter -j%,$(MAKEFLAGS))
+PERFBDFLAGS += $(filter -j%,$(MAKEFLAGS))
 ifneq ($(NM),nm)
-override CODEFLAGS += --nm-tool="$(NM)"
-override DATAFLAGS += --nm-tool="$(NM)"
+CODEFLAGS += --nm-path="$(NM)"
+DATAFLAGS += --nm-path="$(NM)"
 endif
 ifneq ($(OBJDUMP),objdump)
-override STRUCTSFLAGS += --objdump-tool="$(OBJDUMP)"
+CODEFLAGS    += --objdump-path="$(OBJDUMP)"
+DATAFLAGS    += --objdump-path="$(OBJDUMP)"
+STRUCTSFLAGS += --objdump-path="$(OBJDUMP)"
+PERFFLAGS    += --objdump-path="$(OBJDUMP)"
+PERFBDFLAGS  += --objdump-path="$(OBJDUMP)"
+endif
+ifneq ($(PERF),perf)
+PERFFLAGS += --perf-path="$(PERF)"
+endif
+
+TESTFLAGS  += -b
+BENCHFLAGS += -b
+# forward -j flag
+TESTFLAGS  += $(filter -j%,$(MAKEFLAGS))
+BENCHFLAGS += $(filter -j%,$(MAKEFLAGS))
+ifdef YES_PERF
+TESTFLAGS  += -p $(TEST_PERF)
+BENCHFLAGS += -p $(BENCH_PERF)
+endif
+ifdef YES_PERFBD
+TESTFLAGS += -t $(TEST_TRACE) --trace-backtrace --trace-freq=100
+endif
+ifndef NO_PERFBD
+BENCHFLAGS += -t $(BENCH_TRACE) --trace-backtrace --trace-freq=100
+endif
+ifdef YES_TESTMARKS
+TESTFLAGS += -o $(TEST_CSV)
+endif
+ifndef NO_BENCHMARKS
+BENCHFLAGS += -o $(BENCH_CSV)
+endif
+ifdef VERBOSE
+TESTFLAGS   += -v
+TESTCFLAGS  += -v
+BENCHFLAGS  += -v
+BENCHCFLAGS += -v
+endif
+ifdef EXEC
+TESTFLAGS  += --exec="$(EXEC)"
+BENCHFLAGS += --exec="$(EXEC)"
+endif
+ifneq ($(GDB),gdb)
+TESTFLAGS  += --gdb-path="$(GDB)"
+BENCHFLAGS += --gdb-path="$(GDB)"
+endif
+ifneq ($(VALGRIND),valgrind)
+TESTFLAGS  += --valgrind-path="$(VALGRIND)"
+BENCHFLAGS += --valgrind-path="$(VALGRIND)"
+endif
+ifneq ($(PERF),perf)
+TESTFLAGS  += --perf-path="$(PERF)"
+BENCHFLAGS += --perf-path="$(PERF)"
+endif
+
+# this is a bit of a hack, but we want to make sure the BUILDDIR
+# directory structure is correct before we run any commands
+ifneq ($(BUILDDIR),.)
+$(if $(findstring n,$(MAKEFLAGS)),, $(shell mkdir -p \
+	$(addprefix $(BUILDDIR)/,$(dir \
+		$(SRC) \
+		$(TESTS) \
+		$(TEST_SRC) \
+		$(BENCHES) \
+		$(BENCH_SRC)))))
 endif
 
 
 # commands
+
+## Build littlefs
 .PHONY: all build
 all build: $(TARGET)
 
+## Build assembly files
 .PHONY: asm
 asm: $(ASM)
 
+## Find the total size
 .PHONY: size
 size: $(OBJ)
 	$(SIZE) -t $^
 
+## Generate a ctags file
 .PHONY: tags
 tags:
 	$(CTAGS) --totals --c-types=+p $(shell find -H -name '*.h') $(SRC)
 
-.PHONY: calls
-calls: $(CGI)
-	./scripts/calls.py $^ $(CALLSFLAGS)
-
-.PHONY: test
-test:
-	./scripts/test.py $(TESTFLAGS)
-.SECONDEXPANSION:
-test%: tests/test$$(firstword $$(subst \#, ,%)).toml
-	./scripts/test.py $@ $(TESTFLAGS)
+## Show this help text
+.PHONY: help
+help:
+	@$(strip awk '/^## / { \
+			sub(/^## /,""); \
+			getline rule; \
+			while (rule ~ /^(#|\.PHONY|ifdef|ifndef)/) getline rule; \
+			gsub(/:.*/, "", rule); \
+			printf " "" %-25s %s\n", rule, $$0 \
+		}' $(MAKEFILE_LIST))
 
+## Find the per-function code size
 .PHONY: code
-code: $(OBJ)
-	./scripts/code.py $^ -S $(CODEFLAGS)
+code: CODEFLAGS+=-S
+code: $(OBJ) $(BUILDDIR)/lfs.code.csv
+	./scripts/code.py $(OBJ) $(CODEFLAGS)
 
+## Compare per-function code size
+.PHONY: code-diff
+code-diff: $(OBJ)
+	./scripts/code.py $^ $(CODEFLAGS) -d $(BUILDDIR)/lfs.code.csv
+
+## Find the per-function data size
 .PHONY: data
-data: $(OBJ)
-	./scripts/data.py $^ -S $(DATAFLAGS)
+data: DATAFLAGS+=-S
+data: $(OBJ) $(BUILDDIR)/lfs.data.csv
+	./scripts/data.py $(OBJ) $(DATAFLAGS)
 
+## Compare per-function data size
+.PHONY: data-diff
+data-diff: $(OBJ)
+	./scripts/data.py $^ $(DATAFLAGS) -d $(BUILDDIR)/lfs.data.csv
+
+## Find the per-function stack usage
 .PHONY: stack
-stack: $(CGI)
-	./scripts/stack.py $^ -S $(STACKFLAGS)
+stack: STACKFLAGS+=-S
+stack: $(CI) $(BUILDDIR)/lfs.stack.csv
+	./scripts/stack.py $(CI) $(STACKFLAGS)
 
+## Compare per-function stack usage
+.PHONY: stack-diff
+stack-diff: $(CI)
+	./scripts/stack.py $^ $(STACKFLAGS) -d $(BUILDDIR)/lfs.stack.csv
+
+## Find function sizes
+.PHONY: funcs
+funcs: SUMMARYFLAGS+=-S
+funcs: \
+		$(BUILDDIR)/lfs.code.csv \
+		$(BUILDDIR)/lfs.data.csv \
+		$(BUILDDIR)/lfs.stack.csv
+	$(strip ./scripts/summary.py $^ \
+		-bfunction \
+		-fcode=code_size \
+		-fdata=data_size \
+		-fstack=stack_limit --max=stack \
+		$(SUMMARYFLAGS))
+
+## Compare function sizes
+.PHONY: funcs-diff
+funcs-diff: SHELL=/bin/bash
+funcs-diff: $(OBJ) $(CI)
+	$(strip ./scripts/summary.py \
+		<(./scripts/code.py $(OBJ) -q $(CODEFLAGS) -o-) \
+		<(./scripts/data.py $(OBJ) -q $(DATAFLAGS) -o-) \
+		<(./scripts/stack.py $(CI) -q $(STACKFLAGS) -o-) \
+		-bfunction \
+		-fcode=code_size \
+		-fdata=data_size \
+		-fstack=stack_limit --max=stack \
+		$(SUMMARYFLAGS) -d <(./scripts/summary.py \
+			$(BUILDDIR)/lfs.code.csv \
+			$(BUILDDIR)/lfs.data.csv \
+			$(BUILDDIR)/lfs.stack.csv \
+			-q $(SUMMARYFLAGS) -o-))
+
+## Find struct sizes
 .PHONY: structs
-structs: $(OBJ)
-	./scripts/structs.py $^ -S $(STRUCTSFLAGS)
+structs: STRUCTSFLAGS+=-S
+structs: $(OBJ) $(BUILDDIR)/lfs.structs.csv
+	./scripts/structs.py $(OBJ) $(STRUCTSFLAGS)
 
-.PHONY: coverage
-coverage:
-	./scripts/coverage.py $(BUILDDIR)tests/*.toml.info -s $(COVERAGEFLAGS)
+## Compare struct sizes
+.PHONY: structs-diff
+structs-diff: $(OBJ)
+	./scripts/structs.py $^ $(STRUCTSFLAGS) -d $(BUILDDIR)/lfs.structs.csv
+
+## Find the line/branch coverage after a test run
+.PHONY: cov
+cov: COVFLAGS+=-s
+cov: $(GCDA) $(BUILDDIR)/lfs.cov.csv
+	$(strip ./scripts/cov.py $(GCDA) \
+		$(patsubst %,-F%,$(SRC)) \
+		$(COVFLAGS))
+
+## Compare line/branch coverage
+.PHONY: cov-diff
+cov-diff: $(GCDA)
+	$(strip ./scripts/cov.py $^ \
+		$(patsubst %,-F%,$(SRC)) \
+		$(COVFLAGS) -d $(BUILDDIR)/lfs.cov.csv)
+
+## Find the perf results after bench run with YES_PERF
+.PHONY: perf
+perf: PERFFLAGS+=-S
+perf: $(BENCH_PERF) $(BUILDDIR)/lfs.perf.csv
+	$(strip ./scripts/perf.py $(BENCH_PERF) \
+		$(patsubst %,-F%,$(SRC)) \
+		$(PERFFLAGS))
+
+## Compare perf results
+.PHONY: perf-diff
+perf-diff: $(BENCH_PERF)
+	$(strip ./scripts/perf.py $^ \
+		$(patsubst %,-F%,$(SRC)) \
+		$(PERFFLAGS) -d $(BUILDDIR)/lfs.perf.csv)
+
+## Find the perfbd results after a bench run
+.PHONY: perfbd
+perfbd: PERFBDFLAGS+=-S
+perfbd: $(BENCH_TRACE) $(BUILDDIR)/lfs.perfbd.csv
+	$(strip ./scripts/perfbd.py $(BENCH_RUNNER) $(BENCH_TRACE) \
+		$(patsubst %,-F%,$(SRC)) \
+		$(PERFBDFLAGS))
+
+## Compare perfbd results
+.PHONY: perfbd-diff
+perfbd-diff: $(BENCH_TRACE)
+	$(strip ./scripts/perfbd.py $(BENCH_RUNNER) $^ \
+		$(patsubst %,-F%,$(SRC)) \
+		$(PERFBDFLAGS) -d $(BUILDDIR)/lfs.perfbd.csv)
+
+## Find a summary of compile-time sizes
+.PHONY: summary sizes
+summary sizes: \
+		$(BUILDDIR)/lfs.code.csv \
+		$(BUILDDIR)/lfs.data.csv \
+		$(BUILDDIR)/lfs.stack.csv \
+		$(BUILDDIR)/lfs.structs.csv
+	$(strip ./scripts/summary.py $^ \
+		-fcode=code_size \
+		-fdata=data_size \
+		-fstack=stack_limit --max=stack \
+		-fstructs=struct_size \
+		-Y $(SUMMARYFLAGS))
+
+## Compare compile-time sizes
+.PHONY: summary-diff sizes-diff
+summary-diff sizes-diff: SHELL=/bin/bash
+summary-diff sizes-diff: $(OBJ) $(CI)
+	$(strip ./scripts/summary.py \
+		<(./scripts/code.py $(OBJ) -q $(CODEFLAGS) -o-) \
+		<(./scripts/data.py $(OBJ) -q $(DATAFLAGS) -o-) \
+		<(./scripts/stack.py $(CI) -q $(STACKFLAGS) -o-) \
+		<(./scripts/structs.py $(OBJ) -q $(STRUCTSFLAGS) -o-) \
+		-fcode=code_size \
+		-fdata=data_size \
+		-fstack=stack_limit --max=stack \
+		-fstructs=struct_size \
+		-Y $(SUMMARYFLAGS) -d <(./scripts/summary.py \
+			$(BUILDDIR)/lfs.code.csv \
+			$(BUILDDIR)/lfs.data.csv \
+			$(BUILDDIR)/lfs.stack.csv \
+			$(BUILDDIR)/lfs.structs.csv \
+			-q $(SUMMARYFLAGS) -o-))
+
+## Build the test-runner
+.PHONY: test-runner build-test
+ifndef NO_COV
+test-runner build-test: CFLAGS+=--coverage
+endif
+ifdef YES_PERF
+test-runner build-test: CFLAGS+=-fno-omit-frame-pointer
+endif
+ifdef YES_PERFBD
+test-runner build-test: CFLAGS+=-fno-omit-frame-pointer
+endif
+# note we remove some binary dependent files during compilation,
+# otherwise it's way to easy to end up with outdated results
+test-runner build-test: $(TEST_RUNNER)
+ifndef NO_COV
+	rm -f $(TEST_GCDA)
+endif
+ifdef YES_PERF
+	rm -f $(TEST_PERF)
+endif
+ifdef YES_PERFBD
+	rm -f $(TEST_TRACE)
+endif
+
+## Run the tests, -j enables parallel tests
+.PHONY: test
+test: test-runner
+	./scripts/test.py $(TEST_RUNNER) $(TESTFLAGS)
+
+## List the tests
+.PHONY: test-list
+test-list: test-runner
+	./scripts/test.py $(TEST_RUNNER) $(TESTFLAGS) -l
+
+## Summarize the testmarks
+.PHONY: testmarks
+testmarks: SUMMARYFLAGS+=-spassed
+testmarks: $(TEST_CSV) $(BUILDDIR)/lfs.test.csv
+	$(strip ./scripts/summary.py $(TEST_CSV) \
+		-bsuite \
+		-fpassed=test_passed \
+		$(SUMMARYFLAGS))
+
+## Compare testmarks against a previous run
+.PHONY: testmarks-diff
+testmarks-diff: $(TEST_CSV)
+	$(strip ./scripts/summary.py $^ \
+		-bsuite \
+		-fpassed=test_passed \
+		$(SUMMARYFLAGS) -d $(BUILDDIR)/lfs.test.csv)
+
+## Build the bench-runner
+.PHONY: bench-runner build-bench
+ifdef YES_COV
+bench-runner build-bench: CFLAGS+=--coverage
+endif
+ifdef YES_PERF
+bench-runner build-bench: CFLAGS+=-fno-omit-frame-pointer
+endif
+ifndef NO_PERFBD
+bench-runner build-bench: CFLAGS+=-fno-omit-frame-pointer
+endif
+# note we remove some binary dependent files during compilation,
+# otherwise it's way to easy to end up with outdated results
+bench-runner build-bench: $(BENCH_RUNNER)
+ifdef YES_COV 
+	rm -f $(BENCH_GCDA)
+endif
+ifdef YES_PERF
+	rm -f $(BENCH_PERF)
+endif
+ifndef NO_PERFBD
+	rm -f $(BENCH_TRACE)
+endif
+
+## Run the benchmarks, -j enables parallel benchmarks
+.PHONY: bench
+bench: bench-runner
+	./scripts/bench.py $(BENCH_RUNNER) $(BENCHFLAGS)
+
+## List the benchmarks
+.PHONY: bench-list
+bench-list: bench-runner
+	./scripts/bench.py $(BENCH_RUNNER) $(BENCHFLAGS) -l
+
+## Summarize the benchmarks
+.PHONY: benchmarks
+benchmarks: SUMMARYFLAGS+=-Serased -Sproged -Sreaded
+benchmarks: $(BENCH_CSV) $(BUILDDIR)/lfs.bench.csv
+	$(strip ./scripts/summary.py $(BENCH_CSV) \
+		-bsuite \
+		-freaded=bench_readed \
+		-fproged=bench_proged \
+		-ferased=bench_erased \
+		$(SUMMARYFLAGS))
+
+## Compare benchmarks against a previous run
+.PHONY: benchmarks-diff
+benchmarks-diff: $(BENCH_CSV)
+	$(strip ./scripts/summary.py $^ \
+		-bsuite \
+		-freaded=bench_readed \
+		-fproged=bench_proged \
+		-ferased=bench_erased \
+		$(SUMMARYFLAGS) -d $(BUILDDIR)/lfs.bench.csv)
 
-.PHONY: summary
-summary: $(BUILDDIR)lfs.csv
-	./scripts/summary.py -Y $^ $(SUMMARYFLAGS)
 
 
 # rules
 -include $(DEP)
+-include $(TEST_DEP)
 .SUFFIXES:
+.SECONDARY:
 
-$(BUILDDIR)lfs: $(OBJ)
+$(BUILDDIR)/lfs: $(OBJ)
 	$(CC) $(CFLAGS) $^ $(LFLAGS) -o $@
 
-$(BUILDDIR)lfs.a: $(OBJ)
+$(BUILDDIR)/liblfs.a: $(OBJ)
 	$(AR) rcs $@ $^
 
-$(BUILDDIR)lfs.csv: $(OBJ) $(CGI)
-	./scripts/code.py $(OBJ) -q $(CODEFLAGS) -o $@
-	./scripts/data.py $(OBJ) -q -m $@ $(DATAFLAGS) -o $@
-	./scripts/stack.py $(CGI) -q -m $@ $(STACKFLAGS) -o $@
-	./scripts/structs.py $(OBJ) -q -m $@ $(STRUCTSFLAGS) -o $@
-	$(if $(COVERAGE),\
-		./scripts/coverage.py $(BUILDDIR)tests/*.toml.info \
-			-q -m $@ $(COVERAGEFLAGS) -o $@)
+$(BUILDDIR)/lfs.code.csv: $(OBJ)
+	./scripts/code.py $^ -q $(CODEFLAGS) -o $@
 
-$(BUILDDIR)%.o: %.c
-	$(CC) -c -MMD $(CFLAGS) $< -o $@
+$(BUILDDIR)/lfs.data.csv: $(OBJ)
+	./scripts/data.py $^ -q $(DATAFLAGS) -o $@
 
-$(BUILDDIR)%.s: %.c
+$(BUILDDIR)/lfs.stack.csv: $(CI)
+	./scripts/stack.py $^ -q $(STACKFLAGS) -o $@
+
+$(BUILDDIR)/lfs.structs.csv: $(OBJ)
+	./scripts/structs.py $^ -q $(STRUCTSFLAGS) -o $@
+
+$(BUILDDIR)/lfs.cov.csv: $(GCDA)
+	$(strip ./scripts/cov.py $^ \
+		$(patsubst %,-F%,$(SRC)) \
+		-q $(COVFLAGS) -o $@)
+
+$(BUILDDIR)/lfs.perf.csv: $(BENCH_PERF)
+	$(strip ./scripts/perf.py $^ \
+		$(patsubst %,-F%,$(SRC)) \
+		-q $(PERFFLAGS) -o $@)
+
+$(BUILDDIR)/lfs.perfbd.csv: $(BENCH_TRACE)
+	$(strip ./scripts/perfbd.py $(BENCH_RUNNER) $^ \
+		$(patsubst %,-F%,$(SRC)) \
+		-q $(PERFBDFLAGS) -o $@)
+
+$(BUILDDIR)/lfs.test.csv: $(TEST_CSV)
+	cp $^ $@
+
+$(BUILDDIR)/lfs.bench.csv: $(BENCH_CSV)
+	cp $^ $@
+
+$(BUILDDIR)/runners/test_runner: $(TEST_OBJ)
+	$(CC) $(CFLAGS) $^ $(LFLAGS) -o $@
+
+$(BUILDDIR)/runners/bench_runner: $(BENCH_OBJ)
+	$(CC) $(CFLAGS) $^ $(LFLAGS) -o $@
+
+# our main build rule generates .o, .d, and .ci files, the latter
+# used for stack analysis
+$(BUILDDIR)/%.o $(BUILDDIR)/%.ci: %.c
+	$(CC) -c -MMD $(CFLAGS) $< -o $(BUILDDIR)/$*.o
+
+$(BUILDDIR)/%.o $(BUILDDIR)/%.ci: $(BUILDDIR)/%.c
+	$(CC) -c -MMD $(CFLAGS) $< -o $(BUILDDIR)/$*.o
+
+$(BUILDDIR)/%.s: %.c
 	$(CC) -S $(CFLAGS) $< -o $@
 
-# gcc depends on the output file for intermediate file names, so
-# we can't omit to .o output. We also need to serialize with the
-# normal .o rule because otherwise we can end up with multiprocess
-# problems with two instances of gcc modifying the same .o
-$(BUILDDIR)%.ci: %.c | $(BUILDDIR)%.o
-	$(CC) -c -MMD -fcallgraph-info=su $(CFLAGS) $< -o $|
+$(BUILDDIR)/%.c: %.a.c
+	./scripts/prettyasserts.py -p LFS_ASSERT $< -o $@
 
-# clean everything
+$(BUILDDIR)/%.c: $(BUILDDIR)/%.a.c
+	./scripts/prettyasserts.py -p LFS_ASSERT $< -o $@
+
+$(BUILDDIR)/%.t.a.c: %.toml
+	./scripts/test.py -c $< $(TESTCFLAGS) -o $@
+
+$(BUILDDIR)/%.t.a.c: %.c $(TESTS)
+	./scripts/test.py -c $(TESTS) -s $< $(TESTCFLAGS) -o $@
+
+$(BUILDDIR)/%.b.a.c: %.toml
+	./scripts/bench.py -c $< $(BENCHCFLAGS) -o $@
+
+$(BUILDDIR)/%.b.a.c: %.c $(BENCHES)
+	./scripts/bench.py -c $(BENCHES) -s $< $(BENCHCFLAGS) -o $@
+
+## Clean everything
 .PHONY: clean
 clean:
-	rm -f $(BUILDDIR)lfs
-	rm -f $(BUILDDIR)lfs.a
-	rm -f $(BUILDDIR)lfs.csv
+	rm -f $(BUILDDIR)/lfs
+	rm -f $(BUILDDIR)/liblfs.a
+	rm -f $(BUILDDIR)/lfs.code.csv
+	rm -f $(BUILDDIR)/lfs.data.csv
+	rm -f $(BUILDDIR)/lfs.stack.csv
+	rm -f $(BUILDDIR)/lfs.structs.csv
+	rm -f $(BUILDDIR)/lfs.cov.csv
+	rm -f $(BUILDDIR)/lfs.perf.csv
+	rm -f $(BUILDDIR)/lfs.perfbd.csv
+	rm -f $(BUILDDIR)/lfs.test.csv
+	rm -f $(BUILDDIR)/lfs.bench.csv
 	rm -f $(OBJ)
-	rm -f $(CGI)
 	rm -f $(DEP)
 	rm -f $(ASM)
-	rm -f $(BUILDDIR)tests/*.toml.*
+	rm -f $(CI)
+	rm -f $(TEST_RUNNER)
+	rm -f $(TEST_A)
+	rm -f $(TEST_C)
+	rm -f $(TEST_OBJ)
+	rm -f $(TEST_DEP)
+	rm -f $(TEST_CI)
+	rm -f $(TEST_GCNO)
+	rm -f $(TEST_GCDA)
+	rm -f $(TEST_PERF)
+	rm -f $(TEST_TRACE)
+	rm -f $(TEST_CSV)
+	rm -f $(BENCH_RUNNER)
+	rm -f $(BENCH_A)
+	rm -f $(BENCH_C)
+	rm -f $(BENCH_OBJ)
+	rm -f $(BENCH_DEP)
+	rm -f $(BENCH_CI)
+	rm -f $(BENCH_GCNO)
+	rm -f $(BENCH_GCDA)
+	rm -f $(BENCH_PERF)
+	rm -f $(BENCH_TRACE)
+	rm -f $(BENCH_CSV)
diff --git a/README.OpenSource b/README.OpenSource
index b99c3b4..7a4496a 100644
--- a/README.OpenSource
+++ b/README.OpenSource
@@ -3,7 +3,7 @@
         "Name"                  : "littlefs",
         "License"               : "BSD 3-Clause License",
         "License File"          : "LICENSE.md",
-        "Version Number"        : "V2.5",
+        "Version Number"        : "V2.8",
         "Owner"                 : "wangmihu@huawei.com",
         "Upstream URL"          : "https://github.com/littlefs-project/littlefs",
         "Description"           : "A little fail-safe filesystem designed for microcontrollers."
diff --git a/README.md b/README.md
index 584ada3..df7ee00 100644
--- a/README.md
+++ b/README.md
@@ -226,6 +226,13 @@ License Identifiers that are here available: http://spdx.org/licenses/
   to create images of the filesystem on your PC. Check if littlefs will fit
   your needs, create images for a later download to the target memory or
   inspect the content of a binary image of the target memory.
+  
+- [littlefs2-rust] - A Rust wrapper for littlefs. This project allows you
+  to use littlefs in a Rust-friendly API, reaping the benefits of Rust's memory
+  safety and other guarantees.
+
+- [littlefs-disk-img-viewer] - A memory-efficient web application for viewing
+  littlefs disk images in your web browser.
 
 - [mklfs] - A command line tool built by the [Lua RTOS] guys for making
   littlefs images from a host PC. Supports Windows, Mac OS, and Linux.
@@ -243,8 +250,16 @@ License Identifiers that are here available: http://spdx.org/licenses/
   MCUs. It offers static wear-leveling and power-resilience with only a fixed
   _O(|address|)_ pointer structure stored on each block and in RAM.
 
+- [ChaN's FatFs] - A lightweight reimplementation of the infamous FAT filesystem
+  for microcontroller-scale devices. Due to limitations of FAT it can't provide
+  power-loss resilience, but it does allow easy interop with PCs.
+
+- [chamelon] - A pure-OCaml implementation of (most of) littlefs, designed for
+  use with the MirageOS library operating system project. It is interoperable
+  with the reference implementation, with some caveats.
 
 [BSD-3-Clause]: https://spdx.org/licenses/BSD-3-Clause.html
+[littlefs-disk-img-viewer]: https://github.com/tniessen/littlefs-disk-img-viewer
 [littlefs-fuse]: https://github.com/geky/littlefs-fuse
 [FUSE]: https://github.com/libfuse/libfuse
 [littlefs-js]: https://github.com/geky/littlefs-js
@@ -252,7 +267,10 @@ License Identifiers that are here available: http://spdx.org/licenses/
 [mklfs]: https://github.com/whitecatboard/Lua-RTOS-ESP32/tree/master/components/mklfs/src
 [Lua RTOS]: https://github.com/whitecatboard/Lua-RTOS-ESP32
 [Mbed OS]: https://github.com/armmbed/mbed-os
-[LittleFileSystem]: https://os.mbed.com/docs/mbed-os/v5.12/apis/littlefilesystem.html
+[LittleFileSystem]: https://os.mbed.com/docs/mbed-os/latest/apis/littlefilesystem.html
 [SPIFFS]: https://github.com/pellepl/spiffs
 [Dhara]: https://github.com/dlbeer/dhara
+[ChaN's FatFs]: http://elm-chan.org/fsw/ff/00index_e.html
 [littlefs-python]: https://pypi.org/project/littlefs-python/
+[littlefs2-rust]: https://crates.io/crates/littlefs2
+[chamelon]: https://github.com/yomimono/chamelon
diff --git a/SPEC.md b/SPEC.md
index 3663ea5..2370ea6 100644
--- a/SPEC.md
+++ b/SPEC.md
@@ -1,10 +1,10 @@
 ## littlefs technical specification
 
-This is the technical specification of the little filesystem. This document
-covers the technical details of how the littlefs is stored on disk for
-introspection and tooling. This document assumes you are familiar with the
-design of the littlefs, for more info on how littlefs works check
-out [DESIGN.md](DESIGN.md).
+This is the technical specification of the little filesystem with on-disk
+version lfs2.1. This document covers the technical details of how the littlefs
+is stored on disk for introspection and tooling. This document assumes you are
+familiar with the design of the littlefs, for more info on how littlefs works
+check out [DESIGN.md](DESIGN.md).
 
 ```
    | | |     .---._____
@@ -133,12 +133,6 @@ tags XORed together, starting with `0xffffffff`.
 '-------------------'                    '-------------------'
 ```
 
-One last thing to note before we get into the details around tag encoding. Each
-tag contains a valid bit used to indicate if the tag and containing commit is
-valid. This valid bit is the first bit found in the tag and the commit and can
-be used to tell if we've attempted to write to the remaining space in the
-block.
-
 Here's a more complete example of metadata block containing 4 entries:
 
 ```
@@ -191,6 +185,53 @@ Here's a more complete example of metadata block containing 4 entries:
                                              '---- most recent D
 ```
 
+Two things to note before we get into the details around tag encoding:
+
+1. Each tag contains a valid bit used to indicate if the tag and containing
+   commit is valid. After XORing, this bit should always be zero.
+
+   At the end of each commit, the valid bit of the previous tag is XORed
+   with the lowest bit in the type field of the CRC tag. This allows
+   the CRC tag to force the next commit to fail the valid bit test if it
+   has not yet been written to.
+
+2. The valid bit alone is not enough info to know if the next commit has been
+   erased. We don't know the order bits will be programmed in a program block,
+   so it's possible that the next commit had an attempted program that left the
+   valid bit unchanged.
+
+   To ensure we only ever program erased bytes, each commit can contain an
+   optional forward-CRC (FCRC). An FCRC contains a checksum of some amount of
+   bytes in the next commit at the time it was erased.
+
+   ```
+   .-------------------. \      \
+   |  revision count   | |      |
+   |-------------------| |      |
+   |     metadata      | |      |
+   |                   | +---.  +-- current commit
+   |                   | |   |  |
+   |-------------------| |   |  |
+   |       FCRC       ---|-. |  |
+   |-------------------| / | |  |
+   |        CRC       -----|-'  /
+   |-------------------|   |
+   |      padding      |   |        padding (does't need CRC)
+   |                   |   |
+   |-------------------| \ |    \
+   |      erased?      | +-'    |
+   |         |         | |      +-- next commit
+   |         v         | /      |
+   |                   |        /
+   |                   |
+   '-------------------'
+   ```
+
+   If the FCRC is missing or the checksum does not match, we must assume a
+   commit was attempted but failed due to power-loss.
+
+   Note that end-of-block commits do not need an FCRC.
+
 ## Metadata tags
 
 So in littlefs, 32-bit tags describe every type of metadata. And this means
@@ -785,3 +826,41 @@ CRC fields:
    are made about the contents.
 
 ---
+#### `0x5ff` LFS_TYPE_FCRC
+
+Added in lfs2.1, the optional FCRC tag contains a checksum of some amount of
+bytes in the next commit at the time it was erased. This allows us to ensure
+that we only ever program erased bytes, even if a previous commit failed due
+to power-loss.
+
+When programming a commit, the FCRC size must be at least as large as the
+program block size. However, the program block is not saved on disk, and can
+change between mounts, so the FCRC size on disk may be different than the
+current program block size.
+
+If the FCRC is missing or the checksum does not match, we must assume a
+commit was attempted but failed due to power-loss.
+
+Layout of the FCRC tag:
+
+```
+        tag                          data
+[--      32      --][--      32      --|--      32      --]
+[1|- 11 -| 10 | 10 ][--      32      --|--      32      --]
+ ^    ^     ^    ^            ^- fcrc size       ^- fcrc
+ |    |     |    '- size (8)
+ |    |     '------ id (0x3ff)
+ |    '------------ type (0x5ff)
+ '----------------- valid bit
+```
+
+FCRC fields:
+
+1. **FCRC size (32-bits)** - Number of bytes after this commit's CRC tag's
+   padding to include in the FCRC.
+
+2. **FCRC (32-bits)** - CRC of the bytes after this commit's CRC tag's padding
+   when erased. Like the CRC tag, this uses a CRC-32 with a polynomial of
+   `0x04c11db7` initialized with `0xffffffff`.
+
+---
diff --git a/bd/lfs_emubd.c b/bd/lfs_emubd.c
new file mode 100644
index 0000000..c27ae30
--- /dev/null
+++ b/bd/lfs_emubd.c
@@ -0,0 +1,645 @@
+/*
+ * Emulating block device, wraps filebd and rambd while providing a bunch
+ * of hooks for testing littlefs in various conditions.
+ *
+ * Copyright (c) 2022, The littlefs authors.
+ * Copyright (c) 2017, Arm Limited. All rights reserved.
+ * SPDX-License-Identifier: BSD-3-Clause
+ */
+
+#ifndef _POSIX_C_SOURCE
+#define _POSIX_C_SOURCE 199309L
+#endif
+
+#include "bd/lfs_emubd.h"
+
+#include <stdlib.h>
+#include <fcntl.h>
+#include <unistd.h>
+#include <errno.h>
+#include <time.h>
+
+#ifdef _WIN32
+#include <windows.h>
+#endif
+
+
+// access to lazily-allocated/copy-on-write blocks
+//
+// Note we can only modify a block if we have exclusive access to it (rc == 1)
+//
+
+static lfs_emubd_block_t *lfs_emubd_incblock(lfs_emubd_block_t *block) {
+    if (block) {
+        block->rc += 1;
+    }
+    return block;
+}
+
+static void lfs_emubd_decblock(lfs_emubd_block_t *block) {
+    if (block) {
+        block->rc -= 1;
+        if (block->rc == 0) {
+            free(block);
+        }
+    }
+}
+
+static lfs_emubd_block_t *lfs_emubd_mutblock(
+        const struct lfs_config *cfg,
+        lfs_emubd_block_t **block) {
+    lfs_emubd_t *bd = cfg->context;
+    lfs_emubd_block_t *block_ = *block;
+    if (block_ && block_->rc == 1) {
+        // rc == 1? can modify
+        return block_;
+
+    } else if (block_) {
+        // rc > 1? need to create a copy
+        lfs_emubd_block_t *nblock = malloc(
+                sizeof(lfs_emubd_block_t) + bd->cfg->erase_size);
+        if (!nblock) {
+            return NULL;
+        }
+
+        memcpy(nblock, block_,
+                sizeof(lfs_emubd_block_t) + bd->cfg->erase_size);
+        nblock->rc = 1;
+
+        lfs_emubd_decblock(block_);
+        *block = nblock;
+        return nblock;
+
+    } else {
+        // no block? need to allocate
+        lfs_emubd_block_t *nblock = malloc(
+                sizeof(lfs_emubd_block_t) + bd->cfg->erase_size);
+        if (!nblock) {
+            return NULL;
+        }
+
+        nblock->rc = 1;
+        nblock->wear = 0;
+
+        // zero for consistency
+        memset(nblock->data,
+                (bd->cfg->erase_value != -1) ? bd->cfg->erase_value : 0,
+                bd->cfg->erase_size);
+
+        *block = nblock;
+        return nblock;
+    }
+}
+
+
+// emubd create/destroy
+
+int lfs_emubd_create(const struct lfs_config *cfg,
+        const struct lfs_emubd_config *bdcfg) {
+    LFS_EMUBD_TRACE("lfs_emubd_create(%p {.context=%p, "
+                ".read=%p, .prog=%p, .erase=%p, .sync=%p}, "
+                "%p {.read_size=%"PRIu32", .prog_size=%"PRIu32", "
+                ".erase_size=%"PRIu32", .erase_count=%"PRIu32", "
+                ".erase_value=%"PRId32", .erase_cycles=%"PRIu32", "
+                ".badblock_behavior=%"PRIu8", .power_cycles=%"PRIu32", "
+                ".powerloss_behavior=%"PRIu8", .powerloss_cb=%p, "
+                ".powerloss_data=%p, .track_branches=%d})",
+            (void*)cfg, cfg->context,
+            (void*)(uintptr_t)cfg->read, (void*)(uintptr_t)cfg->prog,
+            (void*)(uintptr_t)cfg->erase, (void*)(uintptr_t)cfg->sync,
+            (void*)bdcfg,
+            bdcfg->read_size, bdcfg->prog_size, bdcfg->erase_size,
+            bdcfg->erase_count, bdcfg->erase_value, bdcfg->erase_cycles,
+            bdcfg->badblock_behavior, bdcfg->power_cycles,
+            bdcfg->powerloss_behavior, (void*)(uintptr_t)bdcfg->powerloss_cb,
+            bdcfg->powerloss_data, bdcfg->track_branches);
+    lfs_emubd_t *bd = cfg->context;
+    bd->cfg = bdcfg;
+
+    // allocate our block array, all blocks start as uninitialized
+    bd->blocks = malloc(bd->cfg->erase_count * sizeof(lfs_emubd_block_t*));
+    if (!bd->blocks) {
+        LFS_EMUBD_TRACE("lfs_emubd_create -> %d", LFS_ERR_NOMEM);
+        return LFS_ERR_NOMEM;
+    }
+    memset(bd->blocks, 0, bd->cfg->erase_count * sizeof(lfs_emubd_block_t*));
+
+    // setup testing things
+    bd->readed = 0;
+    bd->proged = 0;
+    bd->erased = 0;
+    bd->power_cycles = bd->cfg->power_cycles;
+    bd->disk = NULL;
+
+    if (bd->cfg->disk_path) {
+        bd->disk = malloc(sizeof(lfs_emubd_disk_t));
+        if (!bd->disk) {
+            LFS_EMUBD_TRACE("lfs_emubd_create -> %d", LFS_ERR_NOMEM);
+            return LFS_ERR_NOMEM;
+        }
+        bd->disk->rc = 1;
+        bd->disk->scratch = NULL;
+
+        #ifdef _WIN32
+        bd->disk->fd = open(bd->cfg->disk_path,
+                O_RDWR | O_CREAT | O_BINARY, 0666);
+        #else
+        bd->disk->fd = open(bd->cfg->disk_path,
+                O_RDWR | O_CREAT, 0666);
+        #endif
+        if (bd->disk->fd < 0) {
+            int err = -errno;
+            LFS_EMUBD_TRACE("lfs_emubd_create -> %d", err);
+            return err;
+        }
+
+        // if we're emulating erase values, we can keep a block around in
+        // memory of just the erase state to speed up emulated erases
+        if (bd->cfg->erase_value != -1) {
+            bd->disk->scratch = malloc(bd->cfg->erase_size);
+            if (!bd->disk->scratch) {
+                LFS_EMUBD_TRACE("lfs_emubd_create -> %d", LFS_ERR_NOMEM);
+                return LFS_ERR_NOMEM;
+            }
+            memset(bd->disk->scratch,
+                    bd->cfg->erase_value,
+                    bd->cfg->erase_size);
+
+            // go ahead and erase all of the disk, otherwise the file will not
+            // match our internal representation
+            for (size_t i = 0; i < bd->cfg->erase_count; i++) {
+                ssize_t res = write(bd->disk->fd,
+                        bd->disk->scratch,
+                        bd->cfg->erase_size);
+                if (res < 0) {
+                    int err = -errno;
+                    LFS_EMUBD_TRACE("lfs_emubd_create -> %d", err);
+                    return err;
+                }
+            }
+        }
+    }
+
+    LFS_EMUBD_TRACE("lfs_emubd_create -> %d", 0);
+    return 0;
+}
+
+int lfs_emubd_destroy(const struct lfs_config *cfg) {
+    LFS_EMUBD_TRACE("lfs_emubd_destroy(%p)", (void*)cfg);
+    lfs_emubd_t *bd = cfg->context;
+
+    // decrement reference counts
+    for (lfs_block_t i = 0; i < bd->cfg->erase_count; i++) {
+        lfs_emubd_decblock(bd->blocks[i]);
+    }
+    free(bd->blocks);
+
+    // clean up other resources 
+    if (bd->disk) {
+        bd->disk->rc -= 1;
+        if (bd->disk->rc == 0) {
+            close(bd->disk->fd);
+            free(bd->disk->scratch);
+            free(bd->disk);
+        }
+    }
+
+    LFS_EMUBD_TRACE("lfs_emubd_destroy -> %d", 0);
+    return 0;
+}
+
+
+
+// block device API
+
+int lfs_emubd_read(const struct lfs_config *cfg, lfs_block_t block,
+        lfs_off_t off, void *buffer, lfs_size_t size) {
+    LFS_EMUBD_TRACE("lfs_emubd_read(%p, "
+                "0x%"PRIx32", %"PRIu32", %p, %"PRIu32")",
+            (void*)cfg, block, off, buffer, size);
+    lfs_emubd_t *bd = cfg->context;
+
+    // check if read is valid
+    LFS_ASSERT(block < bd->cfg->erase_count);
+    LFS_ASSERT(off  % bd->cfg->read_size == 0);
+    LFS_ASSERT(size % bd->cfg->read_size == 0);
+    LFS_ASSERT(off+size <= bd->cfg->erase_size);
+
+    // get the block
+    const lfs_emubd_block_t *b = bd->blocks[block];
+    if (b) {
+        // block bad?
+        if (bd->cfg->erase_cycles && b->wear >= bd->cfg->erase_cycles &&
+                bd->cfg->badblock_behavior == LFS_EMUBD_BADBLOCK_READERROR) {
+            LFS_EMUBD_TRACE("lfs_emubd_read -> %d", LFS_ERR_CORRUPT);
+            return LFS_ERR_CORRUPT;
+        }
+
+        // read data
+        memcpy(buffer, &b->data[off], size);
+    } else {
+        // zero for consistency
+        memset(buffer,
+                (bd->cfg->erase_value != -1) ? bd->cfg->erase_value : 0,
+                size);
+    }   
+
+    // track reads
+    bd->readed += size;
+    if (bd->cfg->read_sleep) {
+        int err = nanosleep(&(struct timespec){
+                .tv_sec=bd->cfg->read_sleep/1000000000,
+                .tv_nsec=bd->cfg->read_sleep%1000000000},
+            NULL);
+        if (err) {
+            err = -errno;
+            LFS_EMUBD_TRACE("lfs_emubd_read -> %d", err);
+            return err;
+        }
+    }
+
+    LFS_EMUBD_TRACE("lfs_emubd_read -> %d", 0);
+    return 0;
+}
+
+int lfs_emubd_prog(const struct lfs_config *cfg, lfs_block_t block,
+        lfs_off_t off, const void *buffer, lfs_size_t size) {
+    LFS_EMUBD_TRACE("lfs_emubd_prog(%p, "
+                "0x%"PRIx32", %"PRIu32", %p, %"PRIu32")",
+            (void*)cfg, block, off, buffer, size);
+    lfs_emubd_t *bd = cfg->context;
+
+    // check if write is valid
+    LFS_ASSERT(block < bd->cfg->erase_count);
+    LFS_ASSERT(off  % bd->cfg->prog_size == 0);
+    LFS_ASSERT(size % bd->cfg->prog_size == 0);
+    LFS_ASSERT(off+size <= bd->cfg->erase_size);
+
+    // get the block
+    lfs_emubd_block_t *b = lfs_emubd_mutblock(cfg, &bd->blocks[block]);
+    if (!b) {
+        LFS_EMUBD_TRACE("lfs_emubd_prog -> %d", LFS_ERR_NOMEM);
+        return LFS_ERR_NOMEM;
+    }
+
+    // block bad?
+    if (bd->cfg->erase_cycles && b->wear >= bd->cfg->erase_cycles) {
+        if (bd->cfg->badblock_behavior ==
+                LFS_EMUBD_BADBLOCK_PROGERROR) {
+            LFS_EMUBD_TRACE("lfs_emubd_prog -> %d", LFS_ERR_CORRUPT);
+            return LFS_ERR_CORRUPT;
+        } else if (bd->cfg->badblock_behavior ==
+                LFS_EMUBD_BADBLOCK_PROGNOOP ||
+                bd->cfg->badblock_behavior ==
+                LFS_EMUBD_BADBLOCK_ERASENOOP) {
+            LFS_EMUBD_TRACE("lfs_emubd_prog -> %d", 0);
+            return 0;
+        }
+    }
+
+    // were we erased properly?
+    if (bd->cfg->erase_value != -1) {
+        for (lfs_off_t i = 0; i < size; i++) {
+            LFS_ASSERT(b->data[off+i] == bd->cfg->erase_value);
+        }
+    }
+
+    // prog data
+    memcpy(&b->data[off], buffer, size);
+
+    // mirror to disk file?
+    if (bd->disk) {
+        off_t res1 = lseek(bd->disk->fd,
+                (off_t)block*bd->cfg->erase_size + (off_t)off,
+                SEEK_SET);
+        if (res1 < 0) {
+            int err = -errno;
+            LFS_EMUBD_TRACE("lfs_emubd_prog -> %d", err);
+            return err;
+        }
+
+        ssize_t res2 = write(bd->disk->fd, buffer, size);
+        if (res2 < 0) {
+            int err = -errno;
+            LFS_EMUBD_TRACE("lfs_emubd_prog -> %d", err);
+            return err;
+        }
+    }
+
+    // track progs
+    bd->proged += size;
+    if (bd->cfg->prog_sleep) {
+        int err = nanosleep(&(struct timespec){
+                .tv_sec=bd->cfg->prog_sleep/1000000000,
+                .tv_nsec=bd->cfg->prog_sleep%1000000000},
+            NULL);
+        if (err) {
+            err = -errno;
+            LFS_EMUBD_TRACE("lfs_emubd_prog -> %d", err);
+            return err;
+        }
+    }
+
+    // lose power?
+    if (bd->power_cycles > 0) {
+        bd->power_cycles -= 1;
+        if (bd->power_cycles == 0) {
+            // simulate power loss
+            bd->cfg->powerloss_cb(bd->cfg->powerloss_data);
+        }
+    }
+
+    LFS_EMUBD_TRACE("lfs_emubd_prog -> %d", 0);
+    return 0;
+}
+
+int lfs_emubd_erase(const struct lfs_config *cfg, lfs_block_t block) {
+    LFS_EMUBD_TRACE("lfs_emubd_erase(%p, 0x%"PRIx32" (%"PRIu32"))",
+            (void*)cfg, block, ((lfs_emubd_t*)cfg->context)->cfg->erase_size);
+    lfs_emubd_t *bd = cfg->context;
+
+    // check if erase is valid
+    LFS_ASSERT(block < bd->cfg->erase_count);
+
+    // get the block
+    lfs_emubd_block_t *b = lfs_emubd_mutblock(cfg, &bd->blocks[block]);
+    if (!b) {
+        LFS_EMUBD_TRACE("lfs_emubd_prog -> %d", LFS_ERR_NOMEM);
+        return LFS_ERR_NOMEM;
+    }
+
+    // block bad?
+    if (bd->cfg->erase_cycles) {
+        if (b->wear >= bd->cfg->erase_cycles) {
+            if (bd->cfg->badblock_behavior ==
+                    LFS_EMUBD_BADBLOCK_ERASEERROR) {
+                LFS_EMUBD_TRACE("lfs_emubd_erase -> %d", LFS_ERR_CORRUPT);
+                return LFS_ERR_CORRUPT;
+            } else if (bd->cfg->badblock_behavior ==
+                    LFS_EMUBD_BADBLOCK_ERASENOOP) {
+                LFS_EMUBD_TRACE("lfs_emubd_erase -> %d", 0);
+                return 0;
+            }
+        } else {
+            // mark wear
+            b->wear += 1;
+        }
+    }
+
+    // emulate an erase value?
+    if (bd->cfg->erase_value != -1) {
+        memset(b->data, bd->cfg->erase_value, bd->cfg->erase_size);
+
+        // mirror to disk file?
+        if (bd->disk) {
+            off_t res1 = lseek(bd->disk->fd,
+                    (off_t)block*bd->cfg->erase_size,
+                    SEEK_SET);
+            if (res1 < 0) {
+                int err = -errno;
+                LFS_EMUBD_TRACE("lfs_emubd_erase -> %d", err);
+                return err;
+            }
+
+            ssize_t res2 = write(bd->disk->fd,
+                    bd->disk->scratch,
+                    bd->cfg->erase_size);
+            if (res2 < 0) {
+                int err = -errno;
+                LFS_EMUBD_TRACE("lfs_emubd_erase -> %d", err);
+                return err;
+            }
+        }
+    }
+
+    // track erases
+    bd->erased += bd->cfg->erase_size;
+    if (bd->cfg->erase_sleep) {
+        int err = nanosleep(&(struct timespec){
+                .tv_sec=bd->cfg->erase_sleep/1000000000,
+                .tv_nsec=bd->cfg->erase_sleep%1000000000},
+            NULL);
+        if (err) {
+            err = -errno;
+            LFS_EMUBD_TRACE("lfs_emubd_erase -> %d", err);
+            return err;
+        }
+    }
+
+    // lose power?
+    if (bd->power_cycles > 0) {
+        bd->power_cycles -= 1;
+        if (bd->power_cycles == 0) {
+            // simulate power loss
+            bd->cfg->powerloss_cb(bd->cfg->powerloss_data);
+        }
+    }
+
+    LFS_EMUBD_TRACE("lfs_emubd_erase -> %d", 0);
+    return 0;
+}
+
+int lfs_emubd_sync(const struct lfs_config *cfg) {
+    LFS_EMUBD_TRACE("lfs_emubd_sync(%p)", (void*)cfg);
+
+    // do nothing
+    (void)cfg;
+
+    LFS_EMUBD_TRACE("lfs_emubd_sync -> %d", 0);
+    return 0;
+}
+
+/// Additional extended API for driving test features ///
+
+static int lfs_emubd_rawcrc(const struct lfs_config *cfg,
+        lfs_block_t block, uint32_t *crc) {
+    lfs_emubd_t *bd = cfg->context;
+
+    // check if crc is valid
+    LFS_ASSERT(block < cfg->block_count);
+
+    // crc the block
+    uint32_t crc_ = 0xffffffff;
+    const lfs_emubd_block_t *b = bd->blocks[block];
+    if (b) {
+        crc_ = lfs_crc(crc_, b->data, cfg->block_size);
+    } else {
+        uint8_t erase_value = (bd->cfg->erase_value != -1)
+                ? bd->cfg->erase_value
+                : 0;
+        for (lfs_size_t i = 0; i < cfg->block_size; i++) {
+            crc_ = lfs_crc(crc_, &erase_value, 1);
+        }
+    }
+    *crc = 0xffffffff ^ crc_;
+
+    return 0;
+}
+
+int lfs_emubd_crc(const struct lfs_config *cfg,
+        lfs_block_t block, uint32_t *crc) {
+    LFS_EMUBD_TRACE("lfs_emubd_crc(%p, %"PRIu32", %p)",
+            (void*)cfg, block, crc);
+    int err = lfs_emubd_rawcrc(cfg, block, crc);
+    LFS_EMUBD_TRACE("lfs_emubd_crc -> %d", err);
+    return err;
+}
+
+int lfs_emubd_bdcrc(const struct lfs_config *cfg, uint32_t *crc) {
+    LFS_EMUBD_TRACE("lfs_emubd_bdcrc(%p, %p)", (void*)cfg, crc);
+
+    uint32_t crc_ = 0xffffffff;
+    for (lfs_block_t i = 0; i < cfg->block_count; i++) {
+        uint32_t i_crc;
+        int err = lfs_emubd_rawcrc(cfg, i, &i_crc);
+        if (err) {
+            LFS_EMUBD_TRACE("lfs_emubd_bdcrc -> %d", err);
+            return err;
+        }
+
+        crc_ = lfs_crc(crc_, &i_crc, sizeof(uint32_t));
+    }
+    *crc = 0xffffffff ^ crc_;
+
+    LFS_EMUBD_TRACE("lfs_emubd_bdcrc -> %d", 0);
+    return 0;
+}
+
+lfs_emubd_sio_t lfs_emubd_readed(const struct lfs_config *cfg) {
+    LFS_EMUBD_TRACE("lfs_emubd_readed(%p)", (void*)cfg);
+    lfs_emubd_t *bd = cfg->context;
+    LFS_EMUBD_TRACE("lfs_emubd_readed -> %"PRIu64, bd->readed);
+    return bd->readed;
+}
+
+lfs_emubd_sio_t lfs_emubd_proged(const struct lfs_config *cfg) {
+    LFS_EMUBD_TRACE("lfs_emubd_proged(%p)", (void*)cfg);
+    lfs_emubd_t *bd = cfg->context;
+    LFS_EMUBD_TRACE("lfs_emubd_proged -> %"PRIu64, bd->proged);
+    return bd->proged;
+}
+
+lfs_emubd_sio_t lfs_emubd_erased(const struct lfs_config *cfg) {
+    LFS_EMUBD_TRACE("lfs_emubd_erased(%p)", (void*)cfg);
+    lfs_emubd_t *bd = cfg->context;
+    LFS_EMUBD_TRACE("lfs_emubd_erased -> %"PRIu64, bd->erased);
+    return bd->erased;
+}
+
+int lfs_emubd_setreaded(const struct lfs_config *cfg, lfs_emubd_io_t readed) {
+    LFS_EMUBD_TRACE("lfs_emubd_setreaded(%p, %"PRIu64")", (void*)cfg, readed);
+    lfs_emubd_t *bd = cfg->context;
+    bd->readed = readed;
+    LFS_EMUBD_TRACE("lfs_emubd_setreaded -> %d", 0);
+    return 0;
+}
+
+int lfs_emubd_setproged(const struct lfs_config *cfg, lfs_emubd_io_t proged) {
+    LFS_EMUBD_TRACE("lfs_emubd_setproged(%p, %"PRIu64")", (void*)cfg, proged);
+    lfs_emubd_t *bd = cfg->context;
+    bd->proged = proged;
+    LFS_EMUBD_TRACE("lfs_emubd_setproged -> %d", 0);
+    return 0;
+}
+
+int lfs_emubd_seterased(const struct lfs_config *cfg, lfs_emubd_io_t erased) {
+    LFS_EMUBD_TRACE("lfs_emubd_seterased(%p, %"PRIu64")", (void*)cfg, erased);
+    lfs_emubd_t *bd = cfg->context;
+    bd->erased = erased;
+    LFS_EMUBD_TRACE("lfs_emubd_seterased -> %d", 0);
+    return 0;
+}
+
+lfs_emubd_swear_t lfs_emubd_wear(const struct lfs_config *cfg,
+        lfs_block_t block) {
+    LFS_EMUBD_TRACE("lfs_emubd_wear(%p, %"PRIu32")", (void*)cfg, block);
+    lfs_emubd_t *bd = cfg->context;
+
+    // check if block is valid
+    LFS_ASSERT(block < bd->cfg->erase_count);
+
+    // get the wear
+    lfs_emubd_wear_t wear;
+    const lfs_emubd_block_t *b = bd->blocks[block];
+    if (b) {
+        wear = b->wear;
+    } else {
+        wear = 0;
+    }
+
+    LFS_EMUBD_TRACE("lfs_emubd_wear -> %"PRIi32, wear);
+    return wear;
+}
+
+int lfs_emubd_setwear(const struct lfs_config *cfg,
+        lfs_block_t block, lfs_emubd_wear_t wear) {
+    LFS_EMUBD_TRACE("lfs_emubd_setwear(%p, %"PRIu32", %"PRIi32")",
+            (void*)cfg, block, wear);
+    lfs_emubd_t *bd = cfg->context;
+
+    // check if block is valid
+    LFS_ASSERT(block < bd->cfg->erase_count);
+
+    // set the wear
+    lfs_emubd_block_t *b = lfs_emubd_mutblock(cfg, &bd->blocks[block]);
+    if (!b) {
+        LFS_EMUBD_TRACE("lfs_emubd_setwear -> %d", LFS_ERR_NOMEM);
+        return LFS_ERR_NOMEM;
+    }
+    b->wear = wear;
+
+    LFS_EMUBD_TRACE("lfs_emubd_setwear -> %d", 0);
+    return 0;
+}
+
+lfs_emubd_spowercycles_t lfs_emubd_powercycles(
+        const struct lfs_config *cfg) {
+    LFS_EMUBD_TRACE("lfs_emubd_powercycles(%p)", (void*)cfg);
+    lfs_emubd_t *bd = cfg->context;
+
+    LFS_EMUBD_TRACE("lfs_emubd_powercycles -> %"PRIi32, bd->power_cycles);
+    return bd->power_cycles;
+}
+
+int lfs_emubd_setpowercycles(const struct lfs_config *cfg,
+        lfs_emubd_powercycles_t power_cycles) {
+    LFS_EMUBD_TRACE("lfs_emubd_setpowercycles(%p, %"PRIi32")",
+            (void*)cfg, power_cycles);
+    lfs_emubd_t *bd = cfg->context;
+
+    bd->power_cycles = power_cycles;
+
+    LFS_EMUBD_TRACE("lfs_emubd_powercycles -> %d", 0);
+    return 0;
+}
+
+int lfs_emubd_copy(const struct lfs_config *cfg, lfs_emubd_t *copy) {
+    LFS_EMUBD_TRACE("lfs_emubd_copy(%p, %p)", (void*)cfg, (void*)copy);
+    lfs_emubd_t *bd = cfg->context;
+
+    // lazily copy over our block array
+    copy->blocks = malloc(bd->cfg->erase_count * sizeof(lfs_emubd_block_t*));
+    if (!copy->blocks) {
+        LFS_EMUBD_TRACE("lfs_emubd_copy -> %d", LFS_ERR_NOMEM);
+        return LFS_ERR_NOMEM;
+    }
+
+    for (size_t i = 0; i < bd->cfg->erase_count; i++) {
+        copy->blocks[i] = lfs_emubd_incblock(bd->blocks[i]);
+    }
+
+    // other state
+    copy->readed = bd->readed;
+    copy->proged = bd->proged;
+    copy->erased = bd->erased;
+    copy->power_cycles = bd->power_cycles;
+    copy->disk = bd->disk;
+    if (copy->disk) {
+        copy->disk->rc += 1;
+    }
+    copy->cfg = bd->cfg;
+
+    LFS_EMUBD_TRACE("lfs_emubd_copy -> %d", 0);
+    return 0;
+}
+
diff --git a/bd/lfs_emubd.h b/bd/lfs_emubd.h
new file mode 100644
index 0000000..9049649
--- /dev/null
+++ b/bd/lfs_emubd.h
@@ -0,0 +1,241 @@
+/*
+ * Emulating block device, wraps filebd and rambd while providing a bunch
+ * of hooks for testing littlefs in various conditions.
+ *
+ * Copyright (c) 2022, The littlefs authors.
+ * Copyright (c) 2017, Arm Limited. All rights reserved.
+ * SPDX-License-Identifier: BSD-3-Clause
+ */
+#ifndef LFS_EMUBD_H
+#define LFS_EMUBD_H
+
+#include "lfs.h"
+#include "lfs_util.h"
+#include "bd/lfs_rambd.h"
+#include "bd/lfs_filebd.h"
+
+#ifdef __cplusplus
+extern "C"
+{
+#endif
+
+
+// Block device specific tracing
+#ifndef LFS_EMUBD_TRACE
+#ifdef LFS_EMUBD_YES_TRACE
+#define LFS_EMUBD_TRACE(...) LFS_TRACE(__VA_ARGS__)
+#else
+#define LFS_EMUBD_TRACE(...)
+#endif
+#endif
+
+// Mode determining how "bad-blocks" behave during testing. This simulates
+// some real-world circumstances such as progs not sticking (prog-noop),
+// a readonly disk (erase-noop), and ECC failures (read-error).
+//
+// Not that read-noop is not allowed. Read _must_ return a consistent (but
+// may be arbitrary) value on every read.
+typedef enum lfs_emubd_badblock_behavior {
+    LFS_EMUBD_BADBLOCK_PROGERROR,
+    LFS_EMUBD_BADBLOCK_ERASEERROR,
+    LFS_EMUBD_BADBLOCK_READERROR,
+    LFS_EMUBD_BADBLOCK_PROGNOOP,
+    LFS_EMUBD_BADBLOCK_ERASENOOP,
+} lfs_emubd_badblock_behavior_t;
+
+// Mode determining how power-loss behaves during testing. For now this
+// only supports a noop behavior, leaving the data on-disk untouched.
+typedef enum lfs_emubd_powerloss_behavior {
+    LFS_EMUBD_POWERLOSS_NOOP,
+} lfs_emubd_powerloss_behavior_t;
+
+// Type for measuring read/program/erase operations
+typedef uint64_t lfs_emubd_io_t;
+typedef int64_t lfs_emubd_sio_t;
+
+// Type for measuring wear
+typedef uint32_t lfs_emubd_wear_t;
+typedef int32_t lfs_emubd_swear_t;
+
+// Type for tracking power-cycles
+typedef uint32_t lfs_emubd_powercycles_t;
+typedef int32_t lfs_emubd_spowercycles_t;
+
+// Type for delays in nanoseconds
+typedef uint64_t lfs_emubd_sleep_t;
+typedef int64_t lfs_emubd_ssleep_t;
+
+// emubd config, this is required for testing
+struct lfs_emubd_config {
+    // Minimum size of a read operation in bytes.
+    lfs_size_t read_size;
+
+    // Minimum size of a program operation in bytes.
+    lfs_size_t prog_size;
+
+    // Size of an erase operation in bytes.
+    lfs_size_t erase_size;
+
+    // Number of erase blocks on the device.
+    lfs_size_t erase_count;
+
+    // 8-bit erase value to use for simulating erases. -1 does not simulate
+    // erases, which can speed up testing by avoiding the extra block-device
+    // operations to store the erase value.
+    int32_t erase_value;
+
+    // Number of erase cycles before a block becomes "bad". The exact behavior
+    // of bad blocks is controlled by badblock_behavior.
+    uint32_t erase_cycles;
+
+    // The mode determining how bad-blocks fail
+    lfs_emubd_badblock_behavior_t badblock_behavior;
+
+    // Number of write operations (erase/prog) before triggering a power-loss.
+    // power_cycles=0 disables this. The exact behavior of power-loss is
+    // controlled by a combination of powerloss_behavior and powerloss_cb.
+    lfs_emubd_powercycles_t power_cycles;
+
+    // The mode determining how power-loss affects disk
+    lfs_emubd_powerloss_behavior_t powerloss_behavior;
+
+    // Function to call to emulate power-loss. The exact behavior of power-loss
+    // is up to the runner to provide.
+    void (*powerloss_cb)(void*);
+
+    // Data for power-loss callback
+    void *powerloss_data;
+
+    // True to track when power-loss could have occured. Note this involves 
+    // heavy memory usage!
+    bool track_branches;
+
+    // Path to file to use as a mirror of the disk. This provides a way to view
+    // the current state of the block device.
+    const char *disk_path;
+
+    // Artificial delay in nanoseconds, there is no purpose for this other
+    // than slowing down the simulation.
+    lfs_emubd_sleep_t read_sleep;
+
+    // Artificial delay in nanoseconds, there is no purpose for this other
+    // than slowing down the simulation.
+    lfs_emubd_sleep_t prog_sleep;
+
+    // Artificial delay in nanoseconds, there is no purpose for this other
+    // than slowing down the simulation.
+    lfs_emubd_sleep_t erase_sleep;
+};
+
+// A reference counted block
+typedef struct lfs_emubd_block {
+    uint32_t rc;
+    lfs_emubd_wear_t wear;
+
+    uint8_t data[];
+} lfs_emubd_block_t;
+
+// Disk mirror
+typedef struct lfs_emubd_disk {
+    uint32_t rc;
+    int fd;
+    uint8_t *scratch;
+} lfs_emubd_disk_t;
+
+// emubd state
+typedef struct lfs_emubd {
+    // array of copy-on-write blocks
+    lfs_emubd_block_t **blocks;
+
+    // some other test state
+    lfs_emubd_io_t readed;
+    lfs_emubd_io_t proged;
+    lfs_emubd_io_t erased;
+    lfs_emubd_powercycles_t power_cycles;
+    lfs_emubd_disk_t *disk;
+
+    const struct lfs_emubd_config *cfg;
+} lfs_emubd_t;
+
+
+/// Block device API ///
+
+// Create an emulating block device using the geometry in lfs_config
+int lfs_emubd_create(const struct lfs_config *cfg,
+        const struct lfs_emubd_config *bdcfg);
+
+// Clean up memory associated with block device
+int lfs_emubd_destroy(const struct lfs_config *cfg);
+
+// Read a block
+int lfs_emubd_read(const struct lfs_config *cfg, lfs_block_t block,
+        lfs_off_t off, void *buffer, lfs_size_t size);
+
+// Program a block
+//
+// The block must have previously been erased.
+int lfs_emubd_prog(const struct lfs_config *cfg, lfs_block_t block,
+        lfs_off_t off, const void *buffer, lfs_size_t size);
+
+// Erase a block
+//
+// A block must be erased before being programmed. The
+// state of an erased block is undefined.
+int lfs_emubd_erase(const struct lfs_config *cfg, lfs_block_t block);
+
+// Sync the block device
+int lfs_emubd_sync(const struct lfs_config *cfg);
+
+
+/// Additional extended API for driving test features ///
+
+// A CRC of a block for debugging purposes
+int lfs_emubd_crc(const struct lfs_config *cfg,
+        lfs_block_t block, uint32_t *crc);
+
+// A CRC of the entire block device for debugging purposes
+int lfs_emubd_bdcrc(const struct lfs_config *cfg, uint32_t *crc);
+
+// Get total amount of bytes read
+lfs_emubd_sio_t lfs_emubd_readed(const struct lfs_config *cfg);
+
+// Get total amount of bytes programmed
+lfs_emubd_sio_t lfs_emubd_proged(const struct lfs_config *cfg);
+
+// Get total amount of bytes erased
+lfs_emubd_sio_t lfs_emubd_erased(const struct lfs_config *cfg);
+
+// Manually set amount of bytes read
+int lfs_emubd_setreaded(const struct lfs_config *cfg, lfs_emubd_io_t readed);
+
+// Manually set amount of bytes programmed
+int lfs_emubd_setproged(const struct lfs_config *cfg, lfs_emubd_io_t proged);
+
+// Manually set amount of bytes erased
+int lfs_emubd_seterased(const struct lfs_config *cfg, lfs_emubd_io_t erased);
+
+// Get simulated wear on a given block
+lfs_emubd_swear_t lfs_emubd_wear(const struct lfs_config *cfg,
+        lfs_block_t block);
+
+// Manually set simulated wear on a given block
+int lfs_emubd_setwear(const struct lfs_config *cfg,
+        lfs_block_t block, lfs_emubd_wear_t wear);
+
+// Get the remaining power-cycles
+lfs_emubd_spowercycles_t lfs_emubd_powercycles(
+        const struct lfs_config *cfg);
+
+// Manually set the remaining power-cycles
+int lfs_emubd_setpowercycles(const struct lfs_config *cfg,
+        lfs_emubd_powercycles_t power_cycles);
+
+// Create a copy-on-write copy of the state of this block device
+int lfs_emubd_copy(const struct lfs_config *cfg, lfs_emubd_t *copy);
+
+
+#ifdef __cplusplus
+} /* extern "C" */
+#endif
+
+#endif
diff --git a/bd/lfs_filebd.c b/bd/lfs_filebd.c
index 98e5abc..4ff25d4 100644
--- a/bd/lfs_filebd.c
+++ b/bd/lfs_filebd.c
@@ -15,19 +15,20 @@
 #include <windows.h>
 #endif
 
-int lfs_filebd_createcfg(const struct lfs_config *cfg, const char *path,
+int lfs_filebd_create(const struct lfs_config *cfg, const char *path,
         const struct lfs_filebd_config *bdcfg) {
-    LFS_FILEBD_TRACE("lfs_filebd_createcfg(%p {.context=%p, "
-                ".read=%p, .prog=%p, .erase=%p, .sync=%p, "
-                ".read_size=%"PRIu32", .prog_size=%"PRIu32", "
-                ".block_size=%"PRIu32", .block_count=%"PRIu32"}, "
+    LFS_FILEBD_TRACE("lfs_filebd_create(%p {.context=%p, "
+                ".read=%p, .prog=%p, .erase=%p, .sync=%p}, "
                 "\"%s\", "
-                "%p {.erase_value=%"PRId32"})",
+                "%p {.read_size=%"PRIu32", .prog_size=%"PRIu32", "
+                ".erase_size=%"PRIu32", .erase_count=%"PRIu32"})",
             (void*)cfg, cfg->context,
             (void*)(uintptr_t)cfg->read, (void*)(uintptr_t)cfg->prog,
             (void*)(uintptr_t)cfg->erase, (void*)(uintptr_t)cfg->sync,
-            cfg->read_size, cfg->prog_size, cfg->block_size, cfg->block_count,
-            path, (void*)bdcfg, bdcfg->erase_value);
+            path,
+            (void*)bdcfg,
+            bdcfg->read_size, bdcfg->prog_size, bdcfg->erase_size,
+            bdcfg->erase_count);
     lfs_filebd_t *bd = cfg->context;
     bd->cfg = bdcfg;
 
@@ -40,31 +41,14 @@ int lfs_filebd_createcfg(const struct lfs_config *cfg, const char *path,
 
     if (bd->fd < 0) {
         int err = -errno;
-        LFS_FILEBD_TRACE("lfs_filebd_createcfg -> %d", err);
+        LFS_FILEBD_TRACE("lfs_filebd_create -> %d", err);
         return err;
     }
 
-    LFS_FILEBD_TRACE("lfs_filebd_createcfg -> %d", 0);
+    LFS_FILEBD_TRACE("lfs_filebd_create -> %d", 0);
     return 0;
 }
 
-int lfs_filebd_create(const struct lfs_config *cfg, const char *path) {
-    LFS_FILEBD_TRACE("lfs_filebd_create(%p {.context=%p, "
-                ".read=%p, .prog=%p, .erase=%p, .sync=%p, "
-                ".read_size=%"PRIu32", .prog_size=%"PRIu32", "
-                ".block_size=%"PRIu32", .block_count=%"PRIu32"}, "
-                "\"%s\")",
-            (void*)cfg, cfg->context,
-            (void*)(uintptr_t)cfg->read, (void*)(uintptr_t)cfg->prog,
-            (void*)(uintptr_t)cfg->erase, (void*)(uintptr_t)cfg->sync,
-            cfg->read_size, cfg->prog_size, cfg->block_size, cfg->block_count,
-            path);
-    static const struct lfs_filebd_config defaults = {.erase_value=-1};
-    int err = lfs_filebd_createcfg(cfg, path, &defaults);
-    LFS_FILEBD_TRACE("lfs_filebd_create -> %d", err);
-    return err;
-}
-
 int lfs_filebd_destroy(const struct lfs_config *cfg) {
     LFS_FILEBD_TRACE("lfs_filebd_destroy(%p)", (void*)cfg);
     lfs_filebd_t *bd = cfg->context;
@@ -86,18 +70,17 @@ int lfs_filebd_read(const struct lfs_config *cfg, lfs_block_t block,
     lfs_filebd_t *bd = cfg->context;
 
     // check if read is valid
-    LFS_ASSERT(off  % cfg->read_size == 0);
-    LFS_ASSERT(size % cfg->read_size == 0);
-    LFS_ASSERT(block < cfg->block_count);
+    LFS_ASSERT(block < bd->cfg->erase_count);
+    LFS_ASSERT(off  % bd->cfg->read_size == 0);
+    LFS_ASSERT(size % bd->cfg->read_size == 0);
+    LFS_ASSERT(off+size <= bd->cfg->erase_size);
 
     // zero for reproducibility (in case file is truncated)
-    if (bd->cfg->erase_value != -1) {
-        memset(buffer, bd->cfg->erase_value, size);
-    }
+    memset(buffer, 0, size);
 
     // read
     off_t res1 = lseek(bd->fd,
-            (off_t)block*cfg->block_size + (off_t)off, SEEK_SET);
+            (off_t)block*bd->cfg->erase_size + (off_t)off, SEEK_SET);
     if (res1 < 0) {
         int err = -errno;
         LFS_FILEBD_TRACE("lfs_filebd_read -> %d", err);
@@ -117,41 +100,20 @@ int lfs_filebd_read(const struct lfs_config *cfg, lfs_block_t block,
 
 int lfs_filebd_prog(const struct lfs_config *cfg, lfs_block_t block,
         lfs_off_t off, const void *buffer, lfs_size_t size) {
-    LFS_FILEBD_TRACE("lfs_filebd_prog(%p, 0x%"PRIx32", %"PRIu32", %p, %"PRIu32")",
+    LFS_FILEBD_TRACE("lfs_filebd_prog(%p, "
+                "0x%"PRIx32", %"PRIu32", %p, %"PRIu32")",
             (void*)cfg, block, off, buffer, size);
     lfs_filebd_t *bd = cfg->context;
 
     // check if write is valid
-    LFS_ASSERT(off  % cfg->prog_size == 0);
-    LFS_ASSERT(size % cfg->prog_size == 0);
-    LFS_ASSERT(block < cfg->block_count);
-
-    // check that data was erased? only needed for testing
-    if (bd->cfg->erase_value != -1) {
-        off_t res1 = lseek(bd->fd,
-                (off_t)block*cfg->block_size + (off_t)off, SEEK_SET);
-        if (res1 < 0) {
-            int err = -errno;
-            LFS_FILEBD_TRACE("lfs_filebd_prog -> %d", err);
-            return err;
-        }
-
-        for (lfs_off_t i = 0; i < size; i++) {
-            uint8_t c;
-            ssize_t res2 = read(bd->fd, &c, 1);
-            if (res2 < 0) {
-                int err = -errno;
-                LFS_FILEBD_TRACE("lfs_filebd_prog -> %d", err);
-                return err;
-            }
-
-            LFS_ASSERT(c == bd->cfg->erase_value);
-        }
-    }
+    LFS_ASSERT(block < bd->cfg->erase_count);
+    LFS_ASSERT(off  % bd->cfg->prog_size == 0);
+    LFS_ASSERT(size % bd->cfg->prog_size == 0);
+    LFS_ASSERT(off+size <= bd->cfg->erase_size);
 
     // program data
     off_t res1 = lseek(bd->fd,
-            (off_t)block*cfg->block_size + (off_t)off, SEEK_SET);
+            (off_t)block*bd->cfg->erase_size + (off_t)off, SEEK_SET);
     if (res1 < 0) {
         int err = -errno;
         LFS_FILEBD_TRACE("lfs_filebd_prog -> %d", err);
@@ -170,30 +132,15 @@ int lfs_filebd_prog(const struct lfs_config *cfg, lfs_block_t block,
 }
 
 int lfs_filebd_erase(const struct lfs_config *cfg, lfs_block_t block) {
-    LFS_FILEBD_TRACE("lfs_filebd_erase(%p, 0x%"PRIx32")", (void*)cfg, block);
+    LFS_FILEBD_TRACE("lfs_filebd_erase(%p, 0x%"PRIx32" (%"PRIu32"))",
+            (void*)cfg, block, ((lfs_file_t*)cfg->context)->cfg->erase_size);
     lfs_filebd_t *bd = cfg->context;
 
     // check if erase is valid
-    LFS_ASSERT(block < cfg->block_count);
+    LFS_ASSERT(block < bd->cfg->erase_count);
 
-    // erase, only needed for testing
-    if (bd->cfg->erase_value != -1) {
-        off_t res1 = lseek(bd->fd, (off_t)block*cfg->block_size, SEEK_SET);
-        if (res1 < 0) {
-            int err = -errno;
-            LFS_FILEBD_TRACE("lfs_filebd_erase -> %d", err);
-            return err;
-        }
-
-        for (lfs_off_t i = 0; i < cfg->block_size; i++) {
-            ssize_t res2 = write(bd->fd, &(uint8_t){bd->cfg->erase_value}, 1);
-            if (res2 < 0) {
-                int err = -errno;
-                LFS_FILEBD_TRACE("lfs_filebd_erase -> %d", err);
-                return err;
-            }
-        }
-    }
+    // erase is a noop
+    (void)block;
 
     LFS_FILEBD_TRACE("lfs_filebd_erase -> %d", 0);
     return 0;
@@ -201,10 +148,11 @@ int lfs_filebd_erase(const struct lfs_config *cfg, lfs_block_t block) {
 
 int lfs_filebd_sync(const struct lfs_config *cfg) {
     LFS_FILEBD_TRACE("lfs_filebd_sync(%p)", (void*)cfg);
+
     // file sync
     lfs_filebd_t *bd = cfg->context;
     #ifdef _WIN32
-    int err = FlushFileBuffers((HANDLE) _get_osfhandle(fd)) ? 0 : -1;
+    int err = FlushFileBuffers((HANDLE) _get_osfhandle(bd->fd)) ? 0 : -1;
     #else
     int err = fsync(bd->fd);
     #endif
diff --git a/bd/lfs_filebd.h b/bd/lfs_filebd.h
index 1a9456c..d7d2fd9 100644
--- a/bd/lfs_filebd.h
+++ b/bd/lfs_filebd.h
@@ -18,18 +18,27 @@ extern "C"
 
 
 // Block device specific tracing
+#ifndef LFS_FILEBD_TRACE
 #ifdef LFS_FILEBD_YES_TRACE
 #define LFS_FILEBD_TRACE(...) LFS_TRACE(__VA_ARGS__)
 #else
 #define LFS_FILEBD_TRACE(...)
 #endif
+#endif
 
-// filebd config (optional)
+// filebd config
 struct lfs_filebd_config {
-    // 8-bit erase value to use for simulating erases. -1 does not simulate
-    // erases, which can speed up testing by avoiding all the extra block-device
-    // operations to store the erase value.
-    int32_t erase_value;
+    // Minimum size of a read operation in bytes.
+    lfs_size_t read_size;
+
+    // Minimum size of a program operation in bytes.
+    lfs_size_t prog_size;
+
+    // Size of an erase operation in bytes.
+    lfs_size_t erase_size;
+
+    // Number of erase blocks on the device.
+    lfs_size_t erase_count;
 };
 
 // filebd state
@@ -39,9 +48,8 @@ typedef struct lfs_filebd {
 } lfs_filebd_t;
 
 
-// Create a file block device using the geometry in lfs_config
-int lfs_filebd_create(const struct lfs_config *cfg, const char *path);
-int lfs_filebd_createcfg(const struct lfs_config *cfg, const char *path,
+// Create a file block device
+int lfs_filebd_create(const struct lfs_config *cfg, const char *path,
         const struct lfs_filebd_config *bdcfg);
 
 // Clean up memory associated with block device
diff --git a/bd/lfs_rambd.c b/bd/lfs_rambd.c
index 39bb815..a6a0572 100644
--- a/bd/lfs_rambd.c
+++ b/bd/lfs_rambd.c
@@ -7,18 +7,19 @@
  */
 #include "bd/lfs_rambd.h"
 
-int lfs_rambd_createcfg(const struct lfs_config *cfg,
+int lfs_rambd_create(const struct lfs_config *cfg,
         const struct lfs_rambd_config *bdcfg) {
-    LFS_RAMBD_TRACE("lfs_rambd_createcfg(%p {.context=%p, "
-                ".read=%p, .prog=%p, .erase=%p, .sync=%p, "
-                ".read_size=%"PRIu32", .prog_size=%"PRIu32", "
-                ".block_size=%"PRIu32", .block_count=%"PRIu32"}, "
-                "%p {.erase_value=%"PRId32", .buffer=%p})",
+    LFS_RAMBD_TRACE("lfs_rambd_create(%p {.context=%p, "
+                ".read=%p, .prog=%p, .erase=%p, .sync=%p}, "
+                "%p {.read_size=%"PRIu32", .prog_size=%"PRIu32", "
+                ".erase_size=%"PRIu32", .erase_count=%"PRIu32", "
+                ".buffer=%p})",
             (void*)cfg, cfg->context,
             (void*)(uintptr_t)cfg->read, (void*)(uintptr_t)cfg->prog,
             (void*)(uintptr_t)cfg->erase, (void*)(uintptr_t)cfg->sync,
-            cfg->read_size, cfg->prog_size, cfg->block_size, cfg->block_count,
-            (void*)bdcfg, bdcfg->erase_value, bdcfg->buffer);
+            (void*)bdcfg,
+            bdcfg->read_size, bdcfg->prog_size, bdcfg->erase_size,
+            bdcfg->erase_count, bdcfg->buffer);
     lfs_rambd_t *bd = cfg->context;
     bd->cfg = bdcfg;
 
@@ -26,40 +27,20 @@ int lfs_rambd_createcfg(const struct lfs_config *cfg,
     if (bd->cfg->buffer) {
         bd->buffer = bd->cfg->buffer;
     } else {
-        bd->buffer = lfs_malloc(cfg->block_size * cfg->block_count);
+        bd->buffer = lfs_malloc(bd->cfg->erase_size * bd->cfg->erase_count);
         if (!bd->buffer) {
-            LFS_RAMBD_TRACE("lfs_rambd_createcfg -> %d", LFS_ERR_NOMEM);
+            LFS_RAMBD_TRACE("lfs_rambd_create -> %d", LFS_ERR_NOMEM);
             return LFS_ERR_NOMEM;
         }
     }
 
-    // zero for reproducibility?
-    if (bd->cfg->erase_value != -1) {
-        memset(bd->buffer, bd->cfg->erase_value,
-                cfg->block_size * cfg->block_count);
-    } else {
-        memset(bd->buffer, 0, cfg->block_size * cfg->block_count);
-    }
+    // zero for reproducibility
+    memset(bd->buffer, 0, bd->cfg->erase_size * bd->cfg->erase_count);
 
-    LFS_RAMBD_TRACE("lfs_rambd_createcfg -> %d", 0);
+    LFS_RAMBD_TRACE("lfs_rambd_create -> %d", 0);
     return 0;
 }
 
-int lfs_rambd_create(const struct lfs_config *cfg) {
-    LFS_RAMBD_TRACE("lfs_rambd_create(%p {.context=%p, "
-                ".read=%p, .prog=%p, .erase=%p, .sync=%p, "
-                ".read_size=%"PRIu32", .prog_size=%"PRIu32", "
-                ".block_size=%"PRIu32", .block_count=%"PRIu32"})",
-            (void*)cfg, cfg->context,
-            (void*)(uintptr_t)cfg->read, (void*)(uintptr_t)cfg->prog,
-            (void*)(uintptr_t)cfg->erase, (void*)(uintptr_t)cfg->sync,
-            cfg->read_size, cfg->prog_size, cfg->block_size, cfg->block_count);
-    static const struct lfs_rambd_config defaults = {.erase_value=-1};
-    int err = lfs_rambd_createcfg(cfg, &defaults);
-    LFS_RAMBD_TRACE("lfs_rambd_create -> %d", err);
-    return err;
-}
-
 int lfs_rambd_destroy(const struct lfs_config *cfg) {
     LFS_RAMBD_TRACE("lfs_rambd_destroy(%p)", (void*)cfg);
     // clean up memory
@@ -79,12 +60,13 @@ int lfs_rambd_read(const struct lfs_config *cfg, lfs_block_t block,
     lfs_rambd_t *bd = cfg->context;
 
     // check if read is valid
-    LFS_ASSERT(off  % cfg->read_size == 0);
-    LFS_ASSERT(size % cfg->read_size == 0);
-    LFS_ASSERT(block < cfg->block_count);
+    LFS_ASSERT(block < bd->cfg->erase_count);
+    LFS_ASSERT(off  % bd->cfg->read_size == 0);
+    LFS_ASSERT(size % bd->cfg->read_size == 0);
+    LFS_ASSERT(off+size <= bd->cfg->erase_size);
 
     // read data
-    memcpy(buffer, &bd->buffer[block*cfg->block_size + off], size);
+    memcpy(buffer, &bd->buffer[block*bd->cfg->erase_size + off], size);
 
     LFS_RAMBD_TRACE("lfs_rambd_read -> %d", 0);
     return 0;
@@ -98,37 +80,28 @@ int lfs_rambd_prog(const struct lfs_config *cfg, lfs_block_t block,
     lfs_rambd_t *bd = cfg->context;
 
     // check if write is valid
-    LFS_ASSERT(off  % cfg->prog_size == 0);
-    LFS_ASSERT(size % cfg->prog_size == 0);
-    LFS_ASSERT(block < cfg->block_count);
-
-    // check that data was erased? only needed for testing
-    if (bd->cfg->erase_value != -1) {
-        for (lfs_off_t i = 0; i < size; i++) {
-            LFS_ASSERT(bd->buffer[block*cfg->block_size + off + i] ==
-                    bd->cfg->erase_value);
-        }
-    }
+    LFS_ASSERT(block < bd->cfg->erase_count);
+    LFS_ASSERT(off  % bd->cfg->prog_size == 0);
+    LFS_ASSERT(size % bd->cfg->prog_size == 0);
+    LFS_ASSERT(off+size <= bd->cfg->erase_size);
 
     // program data
-    memcpy(&bd->buffer[block*cfg->block_size + off], buffer, size);
+    memcpy(&bd->buffer[block*bd->cfg->erase_size + off], buffer, size);
 
     LFS_RAMBD_TRACE("lfs_rambd_prog -> %d", 0);
     return 0;
 }
 
 int lfs_rambd_erase(const struct lfs_config *cfg, lfs_block_t block) {
-    LFS_RAMBD_TRACE("lfs_rambd_erase(%p, 0x%"PRIx32")", (void*)cfg, block);
+    LFS_RAMBD_TRACE("lfs_rambd_erase(%p, 0x%"PRIx32" (%"PRIu32"))",
+            (void*)cfg, block, ((lfs_rambd_t*)cfg->context)->cfg->erase_size);
     lfs_rambd_t *bd = cfg->context;
 
     // check if erase is valid
-    LFS_ASSERT(block < cfg->block_count);
+    LFS_ASSERT(block < bd->cfg->erase_count);
 
-    // erase, only needed for testing
-    if (bd->cfg->erase_value != -1) {
-        memset(&bd->buffer[block*cfg->block_size],
-                bd->cfg->erase_value, cfg->block_size);
-    }
+    // erase is a noop
+    (void)block;
 
     LFS_RAMBD_TRACE("lfs_rambd_erase -> %d", 0);
     return 0;
@@ -136,8 +109,10 @@ int lfs_rambd_erase(const struct lfs_config *cfg, lfs_block_t block) {
 
 int lfs_rambd_sync(const struct lfs_config *cfg) {
     LFS_RAMBD_TRACE("lfs_rambd_sync(%p)", (void*)cfg);
-    // sync does nothing because we aren't backed by anything real
+
+    // sync is a noop
     (void)cfg;
+
     LFS_RAMBD_TRACE("lfs_rambd_sync -> %d", 0);
     return 0;
 }
diff --git a/bd/lfs_rambd.h b/bd/lfs_rambd.h
index 3a70bc6..8663702 100644
--- a/bd/lfs_rambd.h
+++ b/bd/lfs_rambd.h
@@ -18,17 +18,27 @@ extern "C"
 
 
 // Block device specific tracing
+#ifndef LFS_RAMBD_TRACE
 #ifdef LFS_RAMBD_YES_TRACE
 #define LFS_RAMBD_TRACE(...) LFS_TRACE(__VA_ARGS__)
 #else
 #define LFS_RAMBD_TRACE(...)
 #endif
+#endif
 
-// rambd config (optional)
+// rambd config
 struct lfs_rambd_config {
-    // 8-bit erase value to simulate erasing with. -1 indicates no erase
-    // occurs, which is still a valid block device
-    int32_t erase_value;
+    // Minimum size of a read operation in bytes.
+    lfs_size_t read_size;
+
+    // Minimum size of a program operation in bytes.
+    lfs_size_t prog_size;
+
+    // Size of an erase operation in bytes.
+    lfs_size_t erase_size;
+
+    // Number of erase blocks on the device.
+    lfs_size_t erase_count;
 
     // Optional statically allocated buffer for the block device.
     void *buffer;
@@ -41,9 +51,8 @@ typedef struct lfs_rambd {
 } lfs_rambd_t;
 
 
-// Create a RAM block device using the geometry in lfs_config
-int lfs_rambd_create(const struct lfs_config *cfg);
-int lfs_rambd_createcfg(const struct lfs_config *cfg,
+// Create a RAM block device
+int lfs_rambd_create(const struct lfs_config *cfg,
         const struct lfs_rambd_config *bdcfg);
 
 // Clean up memory associated with block device
diff --git a/benches/bench_dir.toml b/benches/bench_dir.toml
new file mode 100644
index 0000000..5f8cb49
--- /dev/null
+++ b/benches/bench_dir.toml
@@ -0,0 +1,270 @@
+[cases.bench_dir_open]
+# 0 = in-order
+# 1 = reversed-order
+# 2 = random-order
+defines.ORDER = [0, 1, 2]
+defines.N = 1024
+defines.FILE_SIZE = 8
+defines.CHUNK_SIZE = 8
+code = '''
+    lfs_t lfs;
+    lfs_format(&lfs, cfg) => 0;
+    lfs_mount(&lfs, cfg) => 0;
+
+    // first create the files
+    char name[256];
+    uint8_t buffer[CHUNK_SIZE];
+    for (lfs_size_t i = 0; i < N; i++) {
+        sprintf(name, "file%08x", i);
+        lfs_file_t file;
+        lfs_file_open(&lfs, &file, name,
+                LFS_O_WRONLY | LFS_O_CREAT | LFS_O_EXCL) => 0;
+
+        uint32_t file_prng = i;
+        for (lfs_size_t j = 0; j < FILE_SIZE; j += CHUNK_SIZE) {
+            for (lfs_size_t k = 0; k < CHUNK_SIZE; k++) {
+                buffer[k] = BENCH_PRNG(&file_prng);
+            }
+            lfs_file_write(&lfs, &file, buffer, CHUNK_SIZE) => CHUNK_SIZE;
+        }
+
+        lfs_file_close(&lfs, &file) => 0;
+    }
+
+    // then read the files
+    BENCH_START();
+    uint32_t prng = 42;
+    for (lfs_size_t i = 0; i < N; i++) {
+        lfs_off_t i_
+            = (ORDER == 0) ? i
+            : (ORDER == 1) ? (N-1-i)
+            : BENCH_PRNG(&prng) % N;
+        sprintf(name, "file%08x", i_);
+        lfs_file_t file;
+        lfs_file_open(&lfs, &file, name, LFS_O_RDONLY) => 0;
+
+        uint32_t file_prng = i_;
+        for (lfs_size_t j = 0; j < FILE_SIZE; j += CHUNK_SIZE) {
+            lfs_file_read(&lfs, &file, buffer, CHUNK_SIZE) => CHUNK_SIZE;
+            for (lfs_size_t k = 0; k < CHUNK_SIZE; k++) {
+                assert(buffer[k] == BENCH_PRNG(&file_prng));
+            }
+        }
+
+        lfs_file_close(&lfs, &file) => 0;
+    }
+    BENCH_STOP();
+
+    lfs_unmount(&lfs) => 0;
+'''
+
+[cases.bench_dir_creat]
+# 0 = in-order
+# 1 = reversed-order
+# 2 = random-order
+defines.ORDER = [0, 1, 2]
+defines.N = 1024
+defines.FILE_SIZE = 8
+defines.CHUNK_SIZE = 8
+code = '''
+    lfs_t lfs;
+    lfs_format(&lfs, cfg) => 0;
+    lfs_mount(&lfs, cfg) => 0;
+
+    BENCH_START();
+    uint32_t prng = 42;
+    char name[256];
+    uint8_t buffer[CHUNK_SIZE];
+    for (lfs_size_t i = 0; i < N; i++) {
+        lfs_off_t i_
+            = (ORDER == 0) ? i
+            : (ORDER == 1) ? (N-1-i)
+            : BENCH_PRNG(&prng) % N;
+        sprintf(name, "file%08x", i_);
+        lfs_file_t file;
+        lfs_file_open(&lfs, &file, name,
+                LFS_O_WRONLY | LFS_O_CREAT | LFS_O_TRUNC) => 0;
+
+        uint32_t file_prng = i_;
+        for (lfs_size_t j = 0; j < FILE_SIZE; j += CHUNK_SIZE) {
+            for (lfs_size_t k = 0; k < CHUNK_SIZE; k++) {
+                buffer[k] = BENCH_PRNG(&file_prng);
+            }
+            lfs_file_write(&lfs, &file, buffer, CHUNK_SIZE) => CHUNK_SIZE;
+        }
+
+        lfs_file_close(&lfs, &file) => 0;
+    }
+    BENCH_STOP();
+
+    lfs_unmount(&lfs) => 0;
+'''
+
+[cases.bench_dir_remove]
+# 0 = in-order
+# 1 = reversed-order
+# 2 = random-order
+defines.ORDER = [0, 1, 2]
+defines.N = 1024
+defines.FILE_SIZE = 8
+defines.CHUNK_SIZE = 8
+code = '''
+    lfs_t lfs;
+    lfs_format(&lfs, cfg) => 0;
+    lfs_mount(&lfs, cfg) => 0;
+
+    // first create the files
+    char name[256];
+    uint8_t buffer[CHUNK_SIZE];
+    for (lfs_size_t i = 0; i < N; i++) {
+        sprintf(name, "file%08x", i);
+        lfs_file_t file;
+        lfs_file_open(&lfs, &file, name,
+                LFS_O_WRONLY | LFS_O_CREAT | LFS_O_EXCL) => 0;
+
+        uint32_t file_prng = i;
+        for (lfs_size_t j = 0; j < FILE_SIZE; j += CHUNK_SIZE) {
+            for (lfs_size_t k = 0; k < CHUNK_SIZE; k++) {
+                buffer[k] = BENCH_PRNG(&file_prng);
+            }
+            lfs_file_write(&lfs, &file, buffer, CHUNK_SIZE) => CHUNK_SIZE;
+        }
+
+        lfs_file_close(&lfs, &file) => 0;
+    }
+
+    // then remove the files
+    BENCH_START();
+    uint32_t prng = 42;
+    for (lfs_size_t i = 0; i < N; i++) {
+        lfs_off_t i_
+            = (ORDER == 0) ? i
+            : (ORDER == 1) ? (N-1-i)
+            : BENCH_PRNG(&prng) % N;
+        sprintf(name, "file%08x", i_);
+        int err = lfs_remove(&lfs, name);
+        assert(!err || err == LFS_ERR_NOENT);
+    }
+    BENCH_STOP();
+
+    lfs_unmount(&lfs) => 0;
+'''
+
+[cases.bench_dir_read]
+defines.N = 1024
+defines.FILE_SIZE = 8
+defines.CHUNK_SIZE = 8
+code = '''
+    lfs_t lfs;
+    lfs_format(&lfs, cfg) => 0;
+    lfs_mount(&lfs, cfg) => 0;
+
+    // first create the files
+    char name[256];
+    uint8_t buffer[CHUNK_SIZE];
+    for (lfs_size_t i = 0; i < N; i++) {
+        sprintf(name, "file%08x", i);
+        lfs_file_t file;
+        lfs_file_open(&lfs, &file, name,
+                LFS_O_WRONLY | LFS_O_CREAT | LFS_O_EXCL) => 0;
+
+        uint32_t file_prng = i;
+        for (lfs_size_t j = 0; j < FILE_SIZE; j += CHUNK_SIZE) {
+            for (lfs_size_t k = 0; k < CHUNK_SIZE; k++) {
+                buffer[k] = BENCH_PRNG(&file_prng);
+            }
+            lfs_file_write(&lfs, &file, buffer, CHUNK_SIZE) => CHUNK_SIZE;
+        }
+
+        lfs_file_close(&lfs, &file) => 0;
+    }
+
+    // then read the directory
+    BENCH_START();
+    lfs_dir_t dir;
+    lfs_dir_open(&lfs, &dir, "/") => 0;
+    struct lfs_info info;
+    lfs_dir_read(&lfs, &dir, &info) => 1;
+    assert(info.type == LFS_TYPE_DIR);
+    assert(strcmp(info.name, ".") == 0);
+    lfs_dir_read(&lfs, &dir, &info) => 1;
+    assert(info.type == LFS_TYPE_DIR);
+    assert(strcmp(info.name, "..") == 0);
+    for (int i = 0; i < N; i++) {
+        sprintf(name, "file%08x", i);
+        lfs_dir_read(&lfs, &dir, &info) => 1;
+        assert(info.type == LFS_TYPE_REG);
+        assert(strcmp(info.name, name) == 0);
+    }
+    lfs_dir_read(&lfs, &dir, &info) => 0;
+    lfs_dir_close(&lfs, &dir) => 0;
+    BENCH_STOP();
+
+    lfs_unmount(&lfs) => 0;
+'''
+
+[cases.bench_dir_mkdir]
+# 0 = in-order
+# 1 = reversed-order
+# 2 = random-order
+defines.ORDER = [0, 1, 2]
+defines.N = 8
+code = '''
+    lfs_t lfs;
+    lfs_format(&lfs, cfg) => 0;
+    lfs_mount(&lfs, cfg) => 0;
+
+    BENCH_START();
+    uint32_t prng = 42;
+    char name[256];
+    for (lfs_size_t i = 0; i < N; i++) {
+        lfs_off_t i_
+            = (ORDER == 0) ? i
+            : (ORDER == 1) ? (N-1-i)
+            : BENCH_PRNG(&prng) % N;
+        printf("hm %d\n", i);
+        sprintf(name, "dir%08x", i_);
+        int err = lfs_mkdir(&lfs, name);
+        assert(!err || err == LFS_ERR_EXIST);
+    }
+    BENCH_STOP();
+
+    lfs_unmount(&lfs) => 0;
+'''
+
+[cases.bench_dir_rmdir]
+# 0 = in-order
+# 1 = reversed-order
+# 2 = random-order
+defines.ORDER = [0, 1, 2]
+defines.N = 8
+code = '''
+    lfs_t lfs;
+    lfs_format(&lfs, cfg) => 0;
+    lfs_mount(&lfs, cfg) => 0;
+
+    // first create the dirs
+    char name[256];
+    for (lfs_size_t i = 0; i < N; i++) {
+        sprintf(name, "dir%08x", i);
+        lfs_mkdir(&lfs, name) => 0;
+    }
+
+    // then remove the dirs
+    BENCH_START();
+    uint32_t prng = 42;
+    for (lfs_size_t i = 0; i < N; i++) {
+        lfs_off_t i_
+            = (ORDER == 0) ? i
+            : (ORDER == 1) ? (N-1-i)
+            : BENCH_PRNG(&prng) % N;
+        sprintf(name, "dir%08x", i_);
+        int err = lfs_remove(&lfs, name);
+        assert(!err || err == LFS_ERR_NOENT);
+    }
+    BENCH_STOP();
+
+    lfs_unmount(&lfs) => 0;
+'''
+
+
diff --git a/benches/bench_file.toml b/benches/bench_file.toml
new file mode 100644
index 0000000..168eaad
--- /dev/null
+++ b/benches/bench_file.toml
@@ -0,0 +1,95 @@
+[cases.bench_file_read]
+# 0 = in-order
+# 1 = reversed-order
+# 2 = random-order
+defines.ORDER = [0, 1, 2]
+defines.SIZE = '128*1024'
+defines.CHUNK_SIZE = 64
+code = '''
+    lfs_t lfs;
+    lfs_format(&lfs, cfg) => 0;
+    lfs_mount(&lfs, cfg) => 0;
+    lfs_size_t chunks = (SIZE+CHUNK_SIZE-1)/CHUNK_SIZE;
+
+    // first write the file
+    lfs_file_t file;
+    uint8_t buffer[CHUNK_SIZE];
+    lfs_file_open(&lfs, &file, "file",
+            LFS_O_WRONLY | LFS_O_CREAT | LFS_O_EXCL) => 0;
+    for (lfs_size_t i = 0; i < chunks; i++) {
+        uint32_t chunk_prng = i;
+        for (lfs_size_t j = 0; j < CHUNK_SIZE; j++) {
+            buffer[j] = BENCH_PRNG(&chunk_prng);
+        }
+
+        lfs_file_write(&lfs, &file, buffer, CHUNK_SIZE) => CHUNK_SIZE;
+    }
+    lfs_file_write(&lfs, &file, buffer, CHUNK_SIZE) => CHUNK_SIZE;
+    lfs_file_close(&lfs, &file) => 0;
+
+    // then read the file
+    BENCH_START();
+    lfs_file_open(&lfs, &file, "file", LFS_O_RDONLY) => 0;
+
+    uint32_t prng = 42;
+    for (lfs_size_t i = 0; i < chunks; i++) {
+        lfs_off_t i_
+            = (ORDER == 0) ? i
+            : (ORDER == 1) ? (chunks-1-i)
+            : BENCH_PRNG(&prng) % chunks;
+        lfs_file_seek(&lfs, &file, i_*CHUNK_SIZE, LFS_SEEK_SET)
+                => i_*CHUNK_SIZE;
+        lfs_file_read(&lfs, &file, buffer, CHUNK_SIZE) => CHUNK_SIZE;
+
+        uint32_t chunk_prng = i_;
+        for (lfs_size_t j = 0; j < CHUNK_SIZE; j++) {
+            assert(buffer[j] == BENCH_PRNG(&chunk_prng));
+        }
+    }
+
+    lfs_file_close(&lfs, &file) => 0;
+    BENCH_STOP();
+
+    lfs_unmount(&lfs) => 0;
+'''
+
+[cases.bench_file_write]
+# 0 = in-order
+# 1 = reversed-order
+# 2 = random-order
+defines.ORDER = [0, 1, 2]
+defines.SIZE = '128*1024'
+defines.CHUNK_SIZE = 64
+code = '''
+    lfs_t lfs;
+    lfs_format(&lfs, cfg) => 0;
+    lfs_mount(&lfs, cfg) => 0;
+    lfs_size_t chunks = (SIZE+CHUNK_SIZE-1)/CHUNK_SIZE;
+
+    BENCH_START();
+    lfs_file_t file;
+    lfs_file_open(&lfs, &file, "file",
+            LFS_O_WRONLY | LFS_O_CREAT | LFS_O_EXCL) => 0;
+
+    uint8_t buffer[CHUNK_SIZE];
+    uint32_t prng = 42;
+    for (lfs_size_t i = 0; i < chunks; i++) {
+        lfs_off_t i_
+            = (ORDER == 0) ? i
+            : (ORDER == 1) ? (chunks-1-i)
+            : BENCH_PRNG(&prng) % chunks;
+        uint32_t chunk_prng = i_;
+        for (lfs_size_t j = 0; j < CHUNK_SIZE; j++) {
+            buffer[j] = BENCH_PRNG(&chunk_prng);
+        }
+
+        lfs_file_seek(&lfs, &file, i_*CHUNK_SIZE, LFS_SEEK_SET)
+                => i_*CHUNK_SIZE;
+        lfs_file_write(&lfs, &file, buffer, CHUNK_SIZE) => CHUNK_SIZE;
+    }
+
+    lfs_file_close(&lfs, &file) => 0;
+    BENCH_STOP();
+
+    lfs_unmount(&lfs) => 0;
+'''
diff --git a/benches/bench_superblock.toml b/benches/bench_superblock.toml
new file mode 100644
index 0000000..37659d4
--- /dev/null
+++ b/benches/bench_superblock.toml
@@ -0,0 +1,56 @@
+[cases.bench_superblocks_found]
+# support benchmarking with files
+defines.N = [0, 1024]
+defines.FILE_SIZE = 8
+defines.CHUNK_SIZE = 8
+code = '''
+    lfs_t lfs;
+    lfs_format(&lfs, cfg) => 0;
+
+    // create files?
+    lfs_mount(&lfs, cfg) => 0;
+    char name[256];
+    uint8_t buffer[CHUNK_SIZE];
+    for (lfs_size_t i = 0; i < N; i++) {
+        sprintf(name, "file%08x", i);
+        lfs_file_t file;
+        lfs_file_open(&lfs, &file, name,
+                LFS_O_WRONLY | LFS_O_CREAT | LFS_O_EXCL) => 0;
+
+        for (lfs_size_t j = 0; j < FILE_SIZE; j += CHUNK_SIZE) {
+            for (lfs_size_t k = 0; k < CHUNK_SIZE; k++) {
+                buffer[k] = i+j+k;
+            }
+            lfs_file_write(&lfs, &file, buffer, CHUNK_SIZE) => CHUNK_SIZE;
+        }
+
+        lfs_file_close(&lfs, &file) => 0;
+    }
+    lfs_unmount(&lfs) => 0;
+
+    BENCH_START();
+    lfs_mount(&lfs, cfg) => 0;
+    BENCH_STOP();
+
+    lfs_unmount(&lfs) => 0;
+'''
+
+[cases.bench_superblocks_missing]
+code = '''
+    lfs_t lfs;
+
+    BENCH_START();
+    int err = lfs_mount(&lfs, cfg);
+    assert(err != 0);
+    BENCH_STOP();
+'''
+
+[cases.bench_superblocks_format]
+code = '''
+    lfs_t lfs;
+
+    BENCH_START();
+    lfs_format(&lfs, cfg) => 0;
+    BENCH_STOP();
+'''
+
diff --git a/lfs.c b/lfs.c
index f1d0310..0827331 100644
--- a/lfs.c
+++ b/lfs.c
@@ -5,10 +5,10 @@
  * Copyright (c) 2017, Arm Limited. All rights reserved.
  * SPDX-License-Identifier: BSD-3-Clause
  */
-#define LFS_NO_ASSERT
 #include "lfs.h"
 #include "lfs_util.h"
 
+
 // some constants used throughout the code
 #define LFS_BLOCK_NULL ((lfs_block_t)-1)
 #define LFS_BLOCK_INLINE ((lfs_block_t)-2)
@@ -46,8 +46,8 @@ static int lfs_bd_read(lfs_t *lfs,
         lfs_block_t block, lfs_off_t off,
         void *buffer, lfs_size_t size) {
     uint8_t *data = buffer;
-    if (block >= lfs->cfg->block_count ||
-            off+size > lfs->cfg->block_size) {
+    if (off+size > lfs->cfg->block_size
+            || (lfs->block_count && block >= lfs->block_count)) {
         return LFS_ERR_CORRUPT;
     }
 
@@ -104,7 +104,7 @@ static int lfs_bd_read(lfs_t *lfs,
         }
 
         // load to cache, first condition can no longer fail
-        LFS_ASSERT(block < lfs->cfg->block_count);
+        LFS_ASSERT(!lfs->block_count || block < lfs->block_count);
         rcache->block = block;
         rcache->off = lfs_aligndown(off, lfs->cfg->read_size);
         rcache->size = lfs_min(
@@ -135,14 +135,14 @@ static int lfs_bd_cmp(lfs_t *lfs,
         uint8_t dat[8];
 
         diff = lfs_min(size-i, sizeof(dat));
-        int res = lfs_bd_read(lfs,
+        int err = lfs_bd_read(lfs,
                 pcache, rcache, hint-i,
                 block, off+i, &dat, diff);
-        if (res) {
-            return res;
+        if (err) {
+            return err;
         }
 
-        res = memcmp(dat, data + i, diff);
+        int res = memcmp(dat, data + i, diff);
         if (res) {
             return res < 0 ? LFS_CMP_LT : LFS_CMP_GT;
         }
@@ -151,11 +151,32 @@ static int lfs_bd_cmp(lfs_t *lfs,
     return LFS_CMP_EQ;
 }
 
+static int lfs_bd_crc(lfs_t *lfs,
+        const lfs_cache_t *pcache, lfs_cache_t *rcache, lfs_size_t hint,
+        lfs_block_t block, lfs_off_t off, lfs_size_t size, uint32_t *crc) {
+    lfs_size_t diff = 0;
+
+    for (lfs_off_t i = 0; i < size; i += diff) {
+        uint8_t dat[8];
+        diff = lfs_min(size-i, sizeof(dat));
+        int err = lfs_bd_read(lfs,
+                pcache, rcache, hint-i,
+                block, off+i, &dat, diff);
+        if (err) {
+            return err;
+        }
+
+        *crc = lfs_crc(*crc, &dat, diff);
+    }
+
+    return 0;
+}
+
 #ifndef LFS_READONLY
 static int lfs_bd_flush(lfs_t *lfs,
         lfs_cache_t *pcache, lfs_cache_t *rcache, bool validate) {
     if (pcache->block != LFS_BLOCK_NULL && pcache->block != LFS_BLOCK_INLINE) {
-        LFS_ASSERT(pcache->block < lfs->cfg->block_count);
+        LFS_ASSERT(pcache->block < lfs->block_count);
         lfs_size_t diff = lfs_alignup(pcache->size, lfs->cfg->prog_size);
         int err = lfs->cfg->prog(lfs->cfg, pcache->block,
                 pcache->off, pcache->buffer, diff);
@@ -208,7 +229,7 @@ static int lfs_bd_prog(lfs_t *lfs,
         lfs_block_t block, lfs_off_t off,
         const void *buffer, lfs_size_t size) {
     const uint8_t *data = buffer;
-    LFS_ASSERT(block == LFS_BLOCK_INLINE || block < lfs->cfg->block_count);
+    LFS_ASSERT(block == LFS_BLOCK_INLINE || block < lfs->block_count);
     LFS_ASSERT(off + size <= lfs->cfg->block_size);
 
     while (size > 0) {
@@ -252,7 +273,7 @@ static int lfs_bd_prog(lfs_t *lfs,
 
 #ifndef LFS_READONLY
 static int lfs_bd_erase(lfs_t *lfs, lfs_block_t block) {
-    LFS_ASSERT(block < lfs->cfg->block_count);
+    LFS_ASSERT(block < lfs->block_count);
     int err = lfs->cfg->erase(lfs->cfg, block);
     LFS_ASSERT(err <= 0);
     return err;
@@ -279,14 +300,12 @@ static inline int lfs_pair_cmp(
              paira[0] == pairb[1] || paira[1] == pairb[0]);
 }
 
-#ifndef LFS_READONLY
-static inline bool lfs_pair_sync(
+static inline bool lfs_pair_issync(
         const lfs_block_t paira[2],
         const lfs_block_t pairb[2]) {
     return (paira[0] == pairb[0] && paira[1] == pairb[1]) ||
            (paira[0] == pairb[1] && paira[1] == pairb[0]);
 }
-#endif
 
 static inline void lfs_pair_fromle32(lfs_block_t pair[2]) {
     pair[0] = lfs_fromle32(pair[0]);
@@ -325,6 +344,10 @@ static inline uint16_t lfs_tag_type1(lfs_tag_t tag) {
     return (tag & 0x70000000) >> 20;
 }
 
+static inline uint16_t lfs_tag_type2(lfs_tag_t tag) {
+    return (tag & 0x78000000) >> 20;
+}
+
 static inline uint16_t lfs_tag_type3(lfs_tag_t tag) {
     return (tag & 0x7ff00000) >> 20;
 }
@@ -386,7 +409,7 @@ static inline bool lfs_gstate_hasorphans(const lfs_gstate_t *a) {
 }
 
 static inline uint8_t lfs_gstate_getorphans(const lfs_gstate_t *a) {
-    return lfs_tag_size(a->tag);
+    return lfs_tag_size(a->tag) & 0x1ff;
 }
 
 static inline bool lfs_gstate_hasmove(const lfs_gstate_t *a) {
@@ -394,6 +417,10 @@ static inline bool lfs_gstate_hasmove(const lfs_gstate_t *a) {
 }
 #endif
 
+static inline bool lfs_gstate_needssuperblock(const lfs_gstate_t *a) {
+    return lfs_tag_size(a->tag) >> 9;
+}
+
 static inline bool lfs_gstate_hasmovehere(const lfs_gstate_t *a,
         const lfs_block_t *pair) {
     return lfs_tag_type1(a->tag) && lfs_pair_cmp(a->pair, pair) == 0;
@@ -413,6 +440,24 @@ static inline void lfs_gstate_tole32(lfs_gstate_t *a) {
 }
 #endif
 
+// operations on forward-CRCs used to track erased state
+struct lfs_fcrc {
+    lfs_size_t size;
+    uint32_t crc;
+};
+
+static void lfs_fcrc_fromle32(struct lfs_fcrc *fcrc) {
+    fcrc->size = lfs_fromle32(fcrc->size);
+    fcrc->crc = lfs_fromle32(fcrc->crc);
+}
+
+#ifndef LFS_READONLY
+static void lfs_fcrc_tole32(struct lfs_fcrc *fcrc) {
+    fcrc->size = lfs_tole32(fcrc->size);
+    fcrc->crc = lfs_tole32(fcrc->crc);
+}
+#endif
+
 // other endianness operations
 static void lfs_ctz_fromle32(struct lfs_ctz *ctz) {
     ctz->head = lfs_fromle32(ctz->head);
@@ -473,6 +518,28 @@ static void lfs_mlist_append(lfs_t *lfs, struct lfs_mlist *mlist) {
     lfs->mlist = mlist;
 }
 
+// some other filesystem operations
+static uint32_t lfs_fs_disk_version(lfs_t *lfs) {
+    (void)lfs;
+#ifdef LFS_MULTIVERSION
+    if (lfs->cfg->disk_version) {
+        return lfs->cfg->disk_version;
+    } else
+#endif
+    {
+        return LFS_DISK_VERSION;
+    }
+}
+
+static uint16_t lfs_fs_disk_version_major(lfs_t *lfs) {
+    return 0xffff & (lfs_fs_disk_version(lfs) >> 16);
+
+}
+
+static uint16_t lfs_fs_disk_version_minor(lfs_t *lfs) {
+    return 0xffff & (lfs_fs_disk_version(lfs) >> 0);
+}
+
 
 /// Internal operations predeclared here ///
 #ifndef LFS_READONLY
@@ -500,6 +567,8 @@ static lfs_stag_t lfs_fs_parent(lfs_t *lfs, const lfs_block_t dir[2],
 static int lfs_fs_forceconsistency(lfs_t *lfs);
 #endif
 
+static void lfs_fs_prepsuperblock(lfs_t *lfs, bool needssuperblock);
+
 #ifdef LFS_MIGRATE
 static int lfs1_traverse(lfs_t *lfs,
         int (*cb)(void*, lfs_block_t), void *data);
@@ -528,7 +597,7 @@ static int lfs_rawunmount(lfs_t *lfs);
 static int lfs_alloc_lookahead(void *p, lfs_block_t block) {
     lfs_t *lfs = (lfs_t*)p;
     lfs_block_t off = ((block - lfs->free.off)
-            + lfs->cfg->block_count) % lfs->cfg->block_count;
+            + lfs->block_count) % lfs->block_count;
 
     if (off < lfs->free.size) {
         lfs->free.buffer[off / 32] |= 1U << (off % 32);
@@ -542,7 +611,7 @@ static int lfs_alloc_lookahead(void *p, lfs_block_t block) {
 // is to prevent blocks from being garbage collected in the middle of a
 // commit operation
 static void lfs_alloc_ack(lfs_t *lfs) {
-    lfs->free.ack = lfs->cfg->block_count;
+    lfs->free.ack = lfs->block_count;
 }
 
 // drop the lookahead buffer, this is done during mounting and failed
@@ -553,6 +622,26 @@ static void lfs_alloc_drop(lfs_t *lfs) {
     lfs_alloc_ack(lfs);
 }
 
+#ifndef LFS_READONLY
+static int lfs_fs_rawgc(lfs_t *lfs) {
+    // Move free offset at the first unused block (lfs->free.i)
+    // lfs->free.i is equal lfs->free.size when all blocks are used
+    lfs->free.off = (lfs->free.off + lfs->free.i) % lfs->block_count;
+    lfs->free.size = lfs_min(8*lfs->cfg->lookahead_size, lfs->free.ack);
+    lfs->free.i = 0;
+
+    // find mask of free blocks from tree
+    memset(lfs->free.buffer, 0, lfs->cfg->lookahead_size);
+    int err = lfs_fs_rawtraverse(lfs, lfs_alloc_lookahead, lfs, true);
+    if (err) {
+        lfs_alloc_drop(lfs);
+        return err;
+    }
+
+    return 0;
+}
+#endif
+
 #ifndef LFS_READONLY
 static int lfs_alloc(lfs_t *lfs, lfs_block_t *block) {
     while (true) {
@@ -563,7 +652,7 @@ static int lfs_alloc(lfs_t *lfs, lfs_block_t *block) {
 
             if (!(lfs->free.buffer[off / 32] & (1U << (off % 32)))) {
                 // found a free block
-                *block = (lfs->free.off + off) % lfs->cfg->block_count;
+                *block = (lfs->free.off + off) % lfs->block_count;
 
                 // eagerly find next off so an alloc ack can
                 // discredit old lookahead blocks
@@ -585,16 +674,8 @@ static int lfs_alloc(lfs_t *lfs, lfs_block_t *block) {
             return LFS_ERR_NOSPC;
         }
 
-        lfs->free.off = (lfs->free.off + lfs->free.size)
-                % lfs->cfg->block_count;
-        lfs->free.size = lfs_min(8*lfs->cfg->lookahead_size, lfs->free.ack);
-        lfs->free.i = 0;
-
-        // find mask of free blocks from tree
-        memset(lfs->free.buffer, 0, lfs->cfg->lookahead_size);
-        int err = lfs_fs_rawtraverse(lfs, lfs_alloc_lookahead, lfs, true);
-        if (err) {
-            lfs_alloc_drop(lfs);
+        int err = lfs_fs_rawgc(lfs);
+        if(err) {
             return err;
         }
     }
@@ -808,7 +889,7 @@ static int lfs_dir_traverse(lfs_t *lfs,
     // iterate over directory and attrs
     lfs_tag_t tag;
     const void *buffer;
-    struct lfs_diskoff disk;
+    struct lfs_diskoff disk = {0};
     while (true) {
         {
             if (off+lfs_tag_dsize(ptag) < dir->off) {
@@ -865,11 +946,6 @@ static int lfs_dir_traverse(lfs_t *lfs,
                 };
                 sp += 1;
 
-                dir = dir;
-                off = off;
-                ptag = ptag;
-                attrs = attrs;
-                attrcount = attrcount;
                 tmask = 0;
                 ttag = 0;
                 begin = 0;
@@ -1003,7 +1079,8 @@ static lfs_stag_t lfs_dir_fetchmatch(lfs_t *lfs,
 
     // if either block address is invalid we return LFS_ERR_CORRUPT here,
     // otherwise later writes to the pair could fail
-    if (pair[0] >= lfs->cfg->block_count || pair[1] >= lfs->cfg->block_count) {
+    if (lfs->block_count 
+            && (pair[0] >= lfs->block_count || pair[1] >= lfs->block_count)) {
         return LFS_ERR_CORRUPT;
     }
 
@@ -1040,6 +1117,11 @@ static lfs_stag_t lfs_dir_fetchmatch(lfs_t *lfs,
         bool tempsplit = false;
         lfs_stag_t tempbesttag = besttag;
 
+        // assume not erased until proven otherwise
+        bool maybeerased = false;
+        bool hasfcrc = false;
+        struct lfs_fcrc fcrc;
+
         dir->rev = lfs_tole32(dir->rev);
         uint32_t crc = lfs_crc(0xffffffff, &dir->rev, sizeof(dir->rev));
         dir->rev = lfs_fromle32(dir->rev);
@@ -1054,7 +1136,6 @@ static lfs_stag_t lfs_dir_fetchmatch(lfs_t *lfs,
             if (err) {
                 if (err == LFS_ERR_CORRUPT) {
                     // can't continue?
-                    dir->erased = false;
                     break;
                 }
                 return err;
@@ -1063,19 +1144,19 @@ static lfs_stag_t lfs_dir_fetchmatch(lfs_t *lfs,
             crc = lfs_crc(crc, &tag, sizeof(tag));
             tag = lfs_frombe32(tag) ^ ptag;
 
-            // next commit not yet programmed or we're not in valid range
+            // next commit not yet programmed?
             if (!lfs_tag_isvalid(tag)) {
-                dir->erased = (lfs_tag_type1(ptag) == LFS_TYPE_CRC &&
-                        dir->off % lfs->cfg->prog_size == 0);
+                // we only might be erased if the last tag was a crc
+                maybeerased = (lfs_tag_type2(ptag) == LFS_TYPE_CCRC);
                 break;
+            // out of range?
             } else if (off + lfs_tag_dsize(tag) > lfs->cfg->block_size) {
-                dir->erased = false;
                 break;
             }
 
             ptag = tag;
 
-            if (lfs_tag_type1(tag) == LFS_TYPE_CRC) {
+            if (lfs_tag_type2(tag) == LFS_TYPE_CCRC) {
                 // check the crc attr
                 uint32_t dcrc;
                 err = lfs_bd_read(lfs,
@@ -1083,7 +1164,6 @@ static lfs_stag_t lfs_dir_fetchmatch(lfs_t *lfs,
                         dir->pair[0], off+sizeof(tag), &dcrc, sizeof(dcrc));
                 if (err) {
                     if (err == LFS_ERR_CORRUPT) {
-                        dir->erased = false;
                         break;
                     }
                     return err;
@@ -1091,7 +1171,6 @@ static lfs_stag_t lfs_dir_fetchmatch(lfs_t *lfs,
                 dcrc = lfs_fromle32(dcrc);
 
                 if (crc != dcrc) {
-                    dir->erased = false;
                     break;
                 }
 
@@ -1113,26 +1192,21 @@ static lfs_stag_t lfs_dir_fetchmatch(lfs_t *lfs,
                 dir->tail[1] = temptail[1];
                 dir->split = tempsplit;
 
-                // reset crc
+                // reset crc, hasfcrc
                 crc = 0xffffffff;
                 continue;
             }
 
             // crc the entry first, hopefully leaving it in the cache
-            for (lfs_off_t j = sizeof(tag); j < lfs_tag_dsize(tag); j++) {
-                uint8_t dat;
-                err = lfs_bd_read(lfs,
-                        NULL, &lfs->rcache, lfs->cfg->block_size,
-                        dir->pair[0], off+j, &dat, 1);
-                if (err) {
-                    if (err == LFS_ERR_CORRUPT) {
-                        dir->erased = false;
-                        break;
-                    }
-                    return err;
+            err = lfs_bd_crc(lfs,
+                    NULL, &lfs->rcache, lfs->cfg->block_size,
+                    dir->pair[0], off+sizeof(tag),
+                    lfs_tag_dsize(tag)-sizeof(tag), &crc);
+            if (err) {
+                if (err == LFS_ERR_CORRUPT) {
+                    break;
                 }
-
-                crc = lfs_crc(crc, &dat, 1);
+                return err;
             }
 
             // directory modification tags?
@@ -1159,11 +1233,24 @@ static lfs_stag_t lfs_dir_fetchmatch(lfs_t *lfs,
                         dir->pair[0], off+sizeof(tag), &temptail, 8);
                 if (err) {
                     if (err == LFS_ERR_CORRUPT) {
-                        dir->erased = false;
+                        break;
+                    }
+                    return err;
+                }
+                lfs_pair_fromle32(temptail);
+            } else if (lfs_tag_type3(tag) == LFS_TYPE_FCRC) {
+                err = lfs_bd_read(lfs,
+                        NULL, &lfs->rcache, lfs->cfg->block_size,
+                        dir->pair[0], off+sizeof(tag),
+                        &fcrc, sizeof(fcrc));
+                if (err) {
+                    if (err == LFS_ERR_CORRUPT) {
                         break;
                     }
                 }
-                lfs_pair_fromle32(temptail);
+
+                lfs_fcrc_fromle32(&fcrc);
+                hasfcrc = true;
             }
 
             // found a match for our fetcher?
@@ -1172,7 +1259,6 @@ static lfs_stag_t lfs_dir_fetchmatch(lfs_t *lfs,
                         dir->pair[0], off+sizeof(tag)});
                 if (res < 0) {
                     if (res == LFS_ERR_CORRUPT) {
-                        dir->erased = false;
                         break;
                     }
                     return res;
@@ -1194,35 +1280,67 @@ static lfs_stag_t lfs_dir_fetchmatch(lfs_t *lfs,
             }
         }
 
-        // consider what we have good enough
-        if (dir->off > 0) {
-            // synthetic move
-            if (lfs_gstate_hasmovehere(&lfs->gdisk, dir->pair)) {
-                if (lfs_tag_id(lfs->gdisk.tag) == lfs_tag_id(besttag)) {
-                    besttag |= 0x80000000;
-                } else if (besttag != -1 &&
-                        lfs_tag_id(lfs->gdisk.tag) < lfs_tag_id(besttag)) {
-                    besttag -= LFS_MKTAG(0, 1, 0);
+        // found no valid commits?
+        if (dir->off == 0) {
+            // try the other block?
+            lfs_pair_swap(dir->pair);
+            dir->rev = revs[(r+1)%2];
+            continue;
+        }
+
+        // did we end on a valid commit? we may have an erased block
+        dir->erased = false;
+        if (maybeerased && dir->off % lfs->cfg->prog_size == 0) {
+        #ifdef LFS_MULTIVERSION
+            // note versions < lfs2.1 did not have fcrc tags, if
+            // we're < lfs2.1 treat missing fcrc as erased data
+            //
+            // we don't strictly need to do this, but otherwise writing
+            // to lfs2.0 disks becomes very inefficient
+            if (lfs_fs_disk_version(lfs) < 0x00020001) {
+                dir->erased = true;
+
+            } else
+        #endif
+            if (hasfcrc) {
+                // check for an fcrc matching the next prog's erased state, if
+                // this failed most likely a previous prog was interrupted, we
+                // need a new erase
+                uint32_t fcrc_ = 0xffffffff;
+                int err = lfs_bd_crc(lfs,
+                        NULL, &lfs->rcache, lfs->cfg->block_size,
+                        dir->pair[0], dir->off, fcrc.size, &fcrc_);
+                if (err && err != LFS_ERR_CORRUPT) {
+                    return err;
                 }
-            }
 
-            // found tag? or found best id?
-            if (id) {
-                *id = lfs_min(lfs_tag_id(besttag), dir->count);
-            }
-
-            if (lfs_tag_isvalid(besttag)) {
-                return besttag;
-            } else if (lfs_tag_id(besttag) < dir->count) {
-                return LFS_ERR_NOENT;
-            } else {
-                return 0;
+                // found beginning of erased part?
+                dir->erased = (fcrc_ == fcrc.crc);
             }
         }
 
-        // failed, try the other block?
-        lfs_pair_swap(dir->pair);
-        dir->rev = revs[(r+1)%2];
+        // synthetic move
+        if (lfs_gstate_hasmovehere(&lfs->gdisk, dir->pair)) {
+            if (lfs_tag_id(lfs->gdisk.tag) == lfs_tag_id(besttag)) {
+                besttag |= 0x80000000;
+            } else if (besttag != -1 &&
+                    lfs_tag_id(lfs->gdisk.tag) < lfs_tag_id(besttag)) {
+                besttag -= LFS_MKTAG(0, 1, 0);
+            }
+        }
+
+        // found tag? or found best id?
+        if (id) {
+            *id = lfs_min(lfs_tag_id(besttag), dir->count);
+        }
+
+        if (lfs_tag_isvalid(besttag)) {
+            return besttag;
+        } else if (lfs_tag_id(besttag) < dir->count) {
+            return LFS_ERR_NOENT;
+        } else {
+            return 0;
+        }
     }
 
     LFS_ERROR("Corrupted dir pair at {0x%"PRIx32", 0x%"PRIx32"}",
@@ -1496,9 +1614,15 @@ static int lfs_dir_commitattr(lfs_t *lfs, struct lfs_commit *commit,
 #endif
 
 #ifndef LFS_READONLY
+
 static int lfs_dir_commitcrc(lfs_t *lfs, struct lfs_commit *commit) {
     // align to program units
-    const lfs_off_t end = lfs_alignup(commit->off + 2*sizeof(uint32_t),
+    //
+    // this gets a bit complex as we have two types of crcs:
+    // - 5-word crc with fcrc to check following prog (middle of block)
+    // - 2-word crc with no following prog (end of block)
+    const lfs_off_t end = lfs_alignup(
+            lfs_min(commit->off + 5*sizeof(uint32_t), lfs->cfg->block_size),
             lfs->cfg->prog_size);
 
     lfs_off_t off1 = 0;
@@ -1508,89 +1632,128 @@ static int lfs_dir_commitcrc(lfs_t *lfs, struct lfs_commit *commit) {
     // padding is not crced, which lets fetches skip padding but
     // makes committing a bit more complicated
     while (commit->off < end) {
-        lfs_off_t off = commit->off + sizeof(lfs_tag_t);
-        lfs_off_t noff = lfs_min(end - off, 0x3fe) + off;
+        lfs_off_t noff = (
+                lfs_min(end - (commit->off+sizeof(lfs_tag_t)), 0x3fe)
+                + (commit->off+sizeof(lfs_tag_t)));
+        // too large for crc tag? need padding commits
         if (noff < end) {
-            noff = lfs_min(noff, end - 2*sizeof(uint32_t));
+            noff = lfs_min(noff, end - 5*sizeof(uint32_t));
         }
 
-        // read erased state from next program unit
-        lfs_tag_t tag = 0xffffffff;
-        int err = lfs_bd_read(lfs,
-                NULL, &lfs->rcache, sizeof(tag),
-                commit->block, noff, &tag, sizeof(tag));
-        if (err && err != LFS_ERR_CORRUPT) {
-            return err;
+        // space for fcrc?
+        uint8_t eperturb = (uint8_t)-1;
+        if (noff >= end && noff <= lfs->cfg->block_size - lfs->cfg->prog_size) {
+            // first read the leading byte, this always contains a bit
+            // we can perturb to avoid writes that don't change the fcrc
+            int err = lfs_bd_read(lfs,
+                    NULL, &lfs->rcache, lfs->cfg->prog_size,
+                    commit->block, noff, &eperturb, 1);
+            if (err && err != LFS_ERR_CORRUPT) {
+                return err;
+            }
+
+        #ifdef LFS_MULTIVERSION
+            // unfortunately fcrcs break mdir fetching < lfs2.1, so only write
+            // these if we're a >= lfs2.1 filesystem
+            if (lfs_fs_disk_version(lfs) <= 0x00020000) {
+                // don't write fcrc
+            } else
+        #endif
+            {
+                // find the expected fcrc, don't bother avoiding a reread
+                // of the eperturb, it should still be in our cache
+                struct lfs_fcrc fcrc = {
+                    .size = lfs->cfg->prog_size,
+                    .crc = 0xffffffff
+                };
+                err = lfs_bd_crc(lfs,
+                        NULL, &lfs->rcache, lfs->cfg->prog_size,
+                        commit->block, noff, fcrc.size, &fcrc.crc);
+                if (err && err != LFS_ERR_CORRUPT) {
+                    return err;
+                }
+
+                lfs_fcrc_tole32(&fcrc);
+                err = lfs_dir_commitattr(lfs, commit,
+                        LFS_MKTAG(LFS_TYPE_FCRC, 0x3ff, sizeof(struct lfs_fcrc)),
+                        &fcrc);
+                if (err) {
+                    return err;
+                }
+            }
         }
 
-        // build crc tag
-        bool reset = ~lfs_frombe32(tag) >> 31;
-        tag = LFS_MKTAG(LFS_TYPE_CRC + reset, 0x3ff, noff - off);
+        // build commit crc
+        struct {
+            lfs_tag_t tag;
+            uint32_t crc;
+        } ccrc;
+        lfs_tag_t ntag = LFS_MKTAG(
+                LFS_TYPE_CCRC + (((uint8_t)~eperturb) >> 7), 0x3ff,
+                noff - (commit->off+sizeof(lfs_tag_t)));
+        ccrc.tag = lfs_tobe32(ntag ^ commit->ptag);
+        commit->crc = lfs_crc(commit->crc, &ccrc.tag, sizeof(lfs_tag_t));
+        ccrc.crc = lfs_tole32(commit->crc);
 
-        // write out crc
-        uint32_t footer[2];
-        footer[0] = lfs_tobe32(tag ^ commit->ptag);
-        commit->crc = lfs_crc(commit->crc, &footer[0], sizeof(footer[0]));
-        footer[1] = lfs_tole32(commit->crc);
-        err = lfs_bd_prog(lfs,
+        int err = lfs_bd_prog(lfs,
                 &lfs->pcache, &lfs->rcache, false,
-                commit->block, commit->off, &footer, sizeof(footer));
+                commit->block, commit->off, &ccrc, sizeof(ccrc));
         if (err) {
             return err;
         }
 
         // keep track of non-padding checksum to verify
         if (off1 == 0) {
-            off1 = commit->off + sizeof(uint32_t);
+            off1 = commit->off + sizeof(lfs_tag_t);
             crc1 = commit->crc;
         }
 
-        commit->off += sizeof(tag)+lfs_tag_size(tag);
-        commit->ptag = tag ^ ((lfs_tag_t)reset << 31);
-        commit->crc = 0xffffffff; // reset crc for next "commit"
+        commit->off = noff;
+        // perturb valid bit?
+        commit->ptag = ntag ^ ((0x80UL & ~eperturb) << 24);
+        // reset crc for next commit
+        commit->crc = 0xffffffff;
+
+        // manually flush here since we don't prog the padding, this confuses
+        // the caching layer
+        if (noff >= end || noff >= lfs->pcache.off + lfs->cfg->cache_size) {
+            // flush buffers
+            int err = lfs_bd_sync(lfs, &lfs->pcache, &lfs->rcache, false);
+            if (err) {
+                return err;
+            }
+        }
     }
 
-    // flush buffers
-    int err = lfs_bd_sync(lfs, &lfs->pcache, &lfs->rcache, false);
+    // successful commit, check checksums to make sure
+    //
+    // note that we don't need to check padding commits, worst
+    // case if they are corrupted we would have had to compact anyways
+    lfs_off_t off = commit->begin;
+    uint32_t crc = 0xffffffff;
+    int err = lfs_bd_crc(lfs,
+            NULL, &lfs->rcache, off1+sizeof(uint32_t),
+            commit->block, off, off1-off, &crc);
     if (err) {
         return err;
     }
 
-    // successful commit, check checksums to make sure
-    lfs_off_t off = commit->begin;
-    lfs_off_t noff = off1;
-    while (off < end) {
-        uint32_t crc = 0xffffffff;
-        for (lfs_off_t i = off; i < noff+sizeof(uint32_t); i++) {
-            // check against written crc, may catch blocks that
-            // become readonly and match our commit size exactly
-            if (i == off1 && crc != crc1) {
-                return LFS_ERR_CORRUPT;
-            }
+    // check non-padding commits against known crc
+    if (crc != crc1) {
+        return LFS_ERR_CORRUPT;
+    }
 
-            // leave it up to caching to make this efficient
-            uint8_t dat;
-            err = lfs_bd_read(lfs,
-                    NULL, &lfs->rcache, noff+sizeof(uint32_t)-i,
-                    commit->block, i, &dat, 1);
-            if (err) {
-                return err;
-            }
+    // make sure to check crc in case we happen to pick
+    // up an unrelated crc (frozen block?)
+    err = lfs_bd_crc(lfs,
+            NULL, &lfs->rcache, sizeof(uint32_t),
+            commit->block, off1, sizeof(uint32_t), &crc);
+    if (err) {
+        return err;
+    }
 
-            crc = lfs_crc(crc, &dat, 1);
-        }
-
-        // detected write error?
-        if (crc != 0) {
-            return LFS_ERR_CORRUPT;
-        }
-
-        // skip padding
-        off = lfs_min(end - noff, 0x3fe) + noff;
-        if (off < end) {
-            off = lfs_min(off, end - 2*sizeof(uint32_t));
-        }
-        noff = off + sizeof(uint32_t);
+    if (crc != 0) {
+        return LFS_ERR_CORRUPT;
     }
 
     return 0;
@@ -1931,11 +2094,20 @@ static int lfs_dir_splittingcompact(lfs_t *lfs, lfs_mdir_t *dir,
                 return err;
             }
 
-            // space is complicated, we need room for tail, crc, gstate,
-            // cleanup delete, and we cap at half a block to give room
-            // for metadata updates.
+            // space is complicated, we need room for:
+            //
+            // - tail:         4+2*4 = 12 bytes
+            // - gstate:       4+3*4 = 16 bytes
+            // - move delete:  4     = 4 bytes
+            // - crc:          4+4   = 8 bytes
+            //                 total = 40 bytes
+            //
+            // And we cap at half a block to avoid degenerate cases with
+            // nearly-full metadata blocks.
+            //
             if (end - split < 0xff
-                    && size <= lfs_min(lfs->cfg->block_size - 36,
+                    && size <= lfs_min(
+                        lfs->cfg->block_size - 40,
                         lfs_alignup(
                             (lfs->cfg->metadata_max
                                 ? lfs->cfg->metadata_max
@@ -1981,7 +2153,7 @@ static int lfs_dir_splittingcompact(lfs_t *lfs, lfs_mdir_t *dir,
 
         // do we have extra space? littlefs can't reclaim this space
         // by itself, so expand cautiously
-        if ((lfs_size_t)size < lfs->cfg->block_count/2) {
+        if ((lfs_size_t)size < lfs->block_count/2) {
             LFS_DEBUG("Expanding superblock at rev %"PRIu32, dir->rev);
             int err = lfs_dir_split(lfs, dir, attrs, attrcount,
                     source, begin, end);
@@ -2599,11 +2771,6 @@ static int lfs_dir_rawseek(lfs_t *lfs, lfs_dir_t *dir, lfs_off_t off) {
     dir->id = (off > 0 && lfs_pair_cmp(dir->head, lfs->root) == 0);
 
     while (off > 0) {
-        int diff = lfs_min(dir->m.count - dir->id, off);
-        dir->id += diff;
-        dir->pos += diff;
-        off -= diff;
-
         if (dir->id == dir->m.count) {
             if (!dir->m.split) {
                 return LFS_ERR_INVAL;
@@ -2616,6 +2783,11 @@ static int lfs_dir_rawseek(lfs_t *lfs, lfs_dir_t *dir, lfs_off_t off) {
 
             dir->id = 0;
         }
+
+        int diff = lfs_min(dir->m.count - dir->id, off);
+        dir->id += diff;
+        dir->pos += diff;
+        off -= diff;
     }
 
     return 0;
@@ -2998,12 +3170,14 @@ cleanup:
     return err;
 }
 
+#ifndef LFS_NO_MALLOC
 static int lfs_file_rawopen(lfs_t *lfs, lfs_file_t *file,
         const char *path, int flags) {
     static const struct lfs_file_config defaults = {0};
     int err = lfs_file_rawopencfg(lfs, file, path, flags, &defaults);
     return err;
 }
+#endif
 
 static int lfs_file_rawclose(lfs_t *lfs, lfs_file_t *file) {
 #ifndef LFS_READONLY
@@ -3350,7 +3524,7 @@ static lfs_ssize_t lfs_file_flushedwrite(lfs_t *lfs, lfs_file_t *file,
                     // find out which block we're extending from
                     int err = lfs_ctz_find(lfs, NULL, &file->cache,
                             file->ctz.head, file->ctz.size,
-                            file->pos-1, &file->block, &file->off);
+                            file->pos-1, &file->block, &(lfs_off_t){0});
                     if (err) {
                         file->flags |= LFS_F_ERRED;
                         return err;
@@ -3528,26 +3702,55 @@ static int lfs_file_rawtruncate(lfs_t *lfs, lfs_file_t *file, lfs_off_t size) {
     lfs_off_t pos = file->pos;
     lfs_off_t oldsize = lfs_file_rawsize(lfs, file);
     if (size < oldsize) {
-        // need to flush since directly changing metadata
-        int err = lfs_file_flush(lfs, file);
-        if (err) {
-            return err;
-        }
+        // revert to inline file?
+        if (size <= lfs_min(0x3fe, lfs_min(
+                lfs->cfg->cache_size,
+                (lfs->cfg->metadata_max ?
+                    lfs->cfg->metadata_max : lfs->cfg->block_size) / 8))) {
+            // flush+seek to head
+            lfs_soff_t res = lfs_file_rawseek(lfs, file, 0, LFS_SEEK_SET);
+            if (res < 0) {
+                return (int)res;
+            }
 
-        // lookup new head in ctz skip list
-        err = lfs_ctz_find(lfs, NULL, &file->cache,
-                file->ctz.head, file->ctz.size,
-                size, &file->block, &file->off);
-        if (err) {
-            return err;
-        }
+            // read our data into rcache temporarily
+            lfs_cache_drop(lfs, &lfs->rcache);
+            res = lfs_file_flushedread(lfs, file,
+                    lfs->rcache.buffer, size);
+            if (res < 0) {
+                return (int)res;
+            }
 
-        // need to set pos/block/off consistently so seeking back to
-        // the old position does not get confused
-        file->pos = size;
-        file->ctz.head = file->block;
-        file->ctz.size = size;
-        file->flags |= LFS_F_DIRTY | LFS_F_READING;
+            file->ctz.head = LFS_BLOCK_INLINE;
+            file->ctz.size = size;
+            file->flags |= LFS_F_DIRTY | LFS_F_READING | LFS_F_INLINE;
+            file->cache.block = file->ctz.head;
+            file->cache.off = 0;
+            file->cache.size = lfs->cfg->cache_size;
+            memcpy(file->cache.buffer, lfs->rcache.buffer, size);
+
+        } else {
+            // need to flush since directly changing metadata
+            int err = lfs_file_flush(lfs, file);
+            if (err) {
+                return err;
+            }
+
+            // lookup new head in ctz skip list
+            err = lfs_ctz_find(lfs, NULL, &file->cache,
+                    file->ctz.head, file->ctz.size,
+                    size-1, &file->block, &(lfs_off_t){0});
+            if (err) {
+                return err;
+            }
+
+            // need to set pos/block/off consistently so seeking back to
+            // the old position does not get confused
+            file->pos = size;
+            file->ctz.head = file->block;
+            file->ctz.size = size;
+            file->flags |= LFS_F_DIRTY | LFS_F_READING;
+        }
     } else if (size > oldsize) {
         // flush+seek if not already at end
         lfs_soff_t res = lfs_file_rawseek(lfs, file, 0, LFS_SEEK_END);
@@ -3905,8 +4108,24 @@ static int lfs_rawremoveattr(lfs_t *lfs, const char *path, uint8_t type) {
 /// Filesystem operations ///
 static int lfs_init(lfs_t *lfs, const struct lfs_config *cfg) {
     lfs->cfg = cfg;
+    lfs->block_count = cfg->block_count;  // May be 0
     int err = 0;
 
+#ifdef LFS_MULTIVERSION
+    // this driver only supports minor version < current minor version
+    LFS_ASSERT(!lfs->cfg->disk_version || (
+            (0xffff & (lfs->cfg->disk_version >> 16))
+                    == LFS_DISK_VERSION_MAJOR
+                && (0xffff & (lfs->cfg->disk_version >> 0))
+                    <= LFS_DISK_VERSION_MINOR));
+#endif
+
+    // check that bool is a truthy-preserving type
+    //
+    // note the most common reason for this failure is a before-c99 compiler,
+    // which littlefs currently does not support
+    LFS_ASSERT((bool)0x80000000);
+
     // validate that the lfs-cfg sizes were initiated properly before
     // performing any arithmetic logics with them
     LFS_ASSERT(lfs->cfg->read_size != 0);
@@ -3919,7 +4138,10 @@ static int lfs_init(lfs_t *lfs, const struct lfs_config *cfg) {
     LFS_ASSERT(lfs->cfg->cache_size % lfs->cfg->prog_size == 0);
     LFS_ASSERT(lfs->cfg->block_size % lfs->cfg->cache_size == 0);
 
-    // check that the block size is large enough to fit ctz pointers
+    // check that the block size is large enough to fit all ctz pointers
+    LFS_ASSERT(lfs->cfg->block_size >= 128);
+    // this is the exact calculation for all ctz pointers, if this fails
+    // and the simpler assert above does not, math must be broken
     LFS_ASSERT(4*lfs_npw2(0xffffffff / (lfs->cfg->block_size-2*4))
             <= lfs->cfg->block_size);
 
@@ -4029,6 +4251,8 @@ static int lfs_deinit(lfs_t *lfs) {
     return 0;
 }
 
+
+
 #ifndef LFS_READONLY
 static int lfs_rawformat(lfs_t *lfs, const struct lfs_config *cfg) {
     int err = 0;
@@ -4038,11 +4262,13 @@ static int lfs_rawformat(lfs_t *lfs, const struct lfs_config *cfg) {
             return err;
         }
 
+        LFS_ASSERT(cfg->block_count != 0);
+
         // create free lookahead
         memset(lfs->free.buffer, 0, lfs->cfg->lookahead_size);
         lfs->free.off = 0;
         lfs->free.size = lfs_min(8*lfs->cfg->lookahead_size,
-                lfs->cfg->block_count);
+                lfs->block_count);
         lfs->free.i = 0;
         lfs_alloc_ack(lfs);
 
@@ -4055,9 +4281,9 @@ static int lfs_rawformat(lfs_t *lfs, const struct lfs_config *cfg) {
 
         // write one superblock
         lfs_superblock_t superblock = {
-            .version     = LFS_DISK_VERSION,
+            .version     = lfs_fs_disk_version(lfs),
             .block_size  = lfs->cfg->block_size,
-            .block_count = lfs->cfg->block_count,
+            .block_count = lfs->block_count,
             .name_max    = lfs->name_max,
             .file_max    = lfs->file_max,
             .attr_max    = lfs->attr_max,
@@ -4103,14 +4329,23 @@ static int lfs_rawmount(lfs_t *lfs, const struct lfs_config *cfg) {
 
     // scan directory blocks for superblock and any global updates
     lfs_mdir_t dir = {.tail = {0, 1}};
-    lfs_block_t cycle = 0;
+    lfs_block_t tortoise[2] = {LFS_BLOCK_NULL, LFS_BLOCK_NULL};
+    lfs_size_t tortoise_i = 1;
+    lfs_size_t tortoise_period = 1;
     while (!lfs_pair_isnull(dir.tail)) {
-        if (cycle >= lfs->cfg->block_count/2) {
-            // loop detected
+        // detect cycles with Brent's algorithm
+        if (lfs_pair_issync(dir.tail, tortoise)) {
+            LFS_WARN("Cycle detected in tail list");
             err = LFS_ERR_CORRUPT;
             goto cleanup;
         }
-        cycle += 1;
+        if (tortoise_i == tortoise_period) {
+            tortoise[0] = dir.tail[0];
+            tortoise[1] = dir.tail[1];
+            tortoise_i = 0;
+            tortoise_period *= 2;
+        }
+        tortoise_i += 1;
 
         // fetch next block in tail list
         lfs_stag_t tag = lfs_dir_fetchmatch(lfs, &dir, dir.tail,
@@ -4144,14 +4379,33 @@ static int lfs_rawmount(lfs_t *lfs, const struct lfs_config *cfg) {
             // check version
             uint16_t major_version = (0xffff & (superblock.version >> 16));
             uint16_t minor_version = (0xffff & (superblock.version >>  0));
-            if ((major_version != LFS_DISK_VERSION_MAJOR ||
-                 minor_version > LFS_DISK_VERSION_MINOR)) {
-                LFS_ERROR("Invalid version v%"PRIu16".%"PRIu16,
-                        major_version, minor_version);
+            if (major_version != lfs_fs_disk_version_major(lfs)
+                    || minor_version > lfs_fs_disk_version_minor(lfs)) {
+                LFS_ERROR("Invalid version "
+                        "v%"PRIu16".%"PRIu16" != v%"PRIu16".%"PRIu16,
+                        major_version,
+                        minor_version,
+                        lfs_fs_disk_version_major(lfs),
+                        lfs_fs_disk_version_minor(lfs));
                 err = LFS_ERR_INVAL;
                 goto cleanup;
             }
 
+            // found older minor version? set an in-device only bit in the
+            // gstate so we know we need to rewrite the superblock before
+            // the first write
+            if (minor_version < lfs_fs_disk_version_minor(lfs)) {
+                LFS_DEBUG("Found older minor version "
+                        "v%"PRIu16".%"PRIu16" < v%"PRIu16".%"PRIu16,
+                        major_version,
+                        minor_version,
+                        lfs_fs_disk_version_major(lfs),
+                        lfs_fs_disk_version_minor(lfs));
+                // note this bit is reserved on disk, so fetching more gstate
+                // will not interfere here
+                lfs_fs_prepsuperblock(lfs, true);
+            }
+
             // check superblock configuration
             if (superblock.name_max) {
                 if (superblock.name_max > lfs->name_max) {
@@ -4186,16 +4440,20 @@ static int lfs_rawmount(lfs_t *lfs, const struct lfs_config *cfg) {
                 lfs->attr_max = superblock.attr_max;
             }
 
-            if (superblock.block_count != lfs->cfg->block_count) {
+            // this is where we get the block_count from disk if block_count=0
+            if (lfs->cfg->block_count
+                    && superblock.block_count != lfs->cfg->block_count) {
                 LFS_ERROR("Invalid block count (%"PRIu32" != %"PRIu32")",
                         superblock.block_count, lfs->cfg->block_count);
                 err = LFS_ERR_INVAL;
                 goto cleanup;
             }
 
+            lfs->block_count = superblock.block_count;
+
             if (superblock.block_size != lfs->cfg->block_size) {
                 LFS_ERROR("Invalid block size (%"PRIu32" != %"PRIu32")",
-                        superblock.block_count, lfs->cfg->block_count);
+                        superblock.block_size, lfs->cfg->block_size);
                 err = LFS_ERR_INVAL;
                 goto cleanup;
             }
@@ -4208,12 +4466,6 @@ static int lfs_rawmount(lfs_t *lfs, const struct lfs_config *cfg) {
         }
     }
 
-    // found superblock?
-    if (lfs_pair_isnull(lfs->root)) {
-        err = LFS_ERR_INVAL;
-        goto cleanup;
-    }
-
     // update littlefs with gstate
     if (!lfs_gstate_iszero(&lfs->gstate)) {
         LFS_DEBUG("Found pending gstate 0x%08"PRIx32"%08"PRIx32"%08"PRIx32,
@@ -4226,7 +4478,7 @@ static int lfs_rawmount(lfs_t *lfs, const struct lfs_config *cfg) {
 
     // setup free lookahead, to distribute allocations uniformly across
     // boots, we start the allocator at a random location
-    lfs->free.off = lfs->seed % lfs->cfg->block_count;
+    lfs->free.off = lfs->seed % lfs->block_count;
     lfs_alloc_drop(lfs);
 
     return 0;
@@ -4242,6 +4494,46 @@ static int lfs_rawunmount(lfs_t *lfs) {
 
 
 /// Filesystem filesystem operations ///
+static int lfs_fs_rawstat(lfs_t *lfs, struct lfs_fsinfo *fsinfo) {
+    // if the superblock is up-to-date, we must be on the most recent
+    // minor version of littlefs
+    if (!lfs_gstate_needssuperblock(&lfs->gstate)) {
+        fsinfo->disk_version = lfs_fs_disk_version(lfs);
+
+    // otherwise we need to read the minor version on disk
+    } else {
+        // fetch the superblock
+        lfs_mdir_t dir;
+        int err = lfs_dir_fetch(lfs, &dir, lfs->root);
+        if (err) {
+            return err;
+        }
+
+        lfs_superblock_t superblock;
+        lfs_stag_t tag = lfs_dir_get(lfs, &dir, LFS_MKTAG(0x7ff, 0x3ff, 0),
+                LFS_MKTAG(LFS_TYPE_INLINESTRUCT, 0, sizeof(superblock)),
+                &superblock);
+        if (tag < 0) {
+            return tag;
+        }
+        lfs_superblock_fromle32(&superblock);
+
+        // read the on-disk version
+        fsinfo->disk_version = superblock.version;
+    }
+
+    // filesystem geometry
+    fsinfo->block_size = lfs->cfg->block_size;
+    fsinfo->block_count = lfs->block_count;
+
+    // other on-disk configuration, we cache all of these for internal use
+    fsinfo->name_max = lfs->name_max;
+    fsinfo->file_max = lfs->file_max;
+    fsinfo->attr_max = lfs->attr_max;
+
+    return 0;
+}
+
 int lfs_fs_rawtraverse(lfs_t *lfs,
         int (*cb)(void *data, lfs_block_t block), void *data,
         bool includeorphans) {
@@ -4261,13 +4553,22 @@ int lfs_fs_rawtraverse(lfs_t *lfs,
     }
 #endif
 
-    lfs_block_t cycle = 0;
+    lfs_block_t tortoise[2] = {LFS_BLOCK_NULL, LFS_BLOCK_NULL};
+    lfs_size_t tortoise_i = 1;
+    lfs_size_t tortoise_period = 1;
     while (!lfs_pair_isnull(dir.tail)) {
-        if (cycle >= lfs->cfg->block_count/2) {
-            // loop detected
+        // detect cycles with Brent's algorithm
+        if (lfs_pair_issync(dir.tail, tortoise)) {
+            LFS_WARN("Cycle detected in tail list");
             return LFS_ERR_CORRUPT;
         }
-        cycle += 1;
+        if (tortoise_i == tortoise_period) {
+            tortoise[0] = dir.tail[0];
+            tortoise[1] = dir.tail[1];
+            tortoise_i = 0;
+            tortoise_period *= 2;
+        }
+        tortoise_i += 1;
 
         for (int i = 0; i < 2; i++) {
             int err = cb(data, dir.tail[i]);
@@ -4346,13 +4647,22 @@ static int lfs_fs_pred(lfs_t *lfs,
     // iterate over all directory directory entries
     pdir->tail[0] = 0;
     pdir->tail[1] = 1;
-    lfs_block_t cycle = 0;
+    lfs_block_t tortoise[2] = {LFS_BLOCK_NULL, LFS_BLOCK_NULL};
+    lfs_size_t tortoise_i = 1;
+    lfs_size_t tortoise_period = 1;
     while (!lfs_pair_isnull(pdir->tail)) {
-        if (cycle >= lfs->cfg->block_count/2) {
-            // loop detected
+        // detect cycles with Brent's algorithm
+        if (lfs_pair_issync(pdir->tail, tortoise)) {
+            LFS_WARN("Cycle detected in tail list");
             return LFS_ERR_CORRUPT;
         }
-        cycle += 1;
+        if (tortoise_i == tortoise_period) {
+            tortoise[0] = pdir->tail[0];
+            tortoise[1] = pdir->tail[1];
+            tortoise_i = 0;
+            tortoise_period *= 2;
+        }
+        tortoise_i += 1;
 
         if (lfs_pair_cmp(pdir->tail, pair) == 0) {
             return 0;
@@ -4402,13 +4712,22 @@ static lfs_stag_t lfs_fs_parent(lfs_t *lfs, const lfs_block_t pair[2],
     // use fetchmatch with callback to find pairs
     parent->tail[0] = 0;
     parent->tail[1] = 1;
-    lfs_block_t cycle = 0;
+    lfs_block_t tortoise[2] = {LFS_BLOCK_NULL, LFS_BLOCK_NULL};
+    lfs_size_t tortoise_i = 1;
+    lfs_size_t tortoise_period = 1;
     while (!lfs_pair_isnull(parent->tail)) {
-        if (cycle >= lfs->cfg->block_count/2) {
-            // loop detected
+        // detect cycles with Brent's algorithm
+        if (lfs_pair_issync(parent->tail, tortoise)) {
+            LFS_WARN("Cycle detected in tail list");
             return LFS_ERR_CORRUPT;
         }
-        cycle += 1;
+        if (tortoise_i == tortoise_period) {
+            tortoise[0] = parent->tail[0];
+            tortoise[1] = parent->tail[1];
+            tortoise_i = 0;
+            tortoise_period *= 2;
+        }
+        tortoise_i += 1;
 
         lfs_stag_t tag = lfs_dir_fetchmatch(lfs, parent, parent->tail,
                 LFS_MKTAG(0x7ff, 0, 0x3ff),
@@ -4425,9 +4744,15 @@ static lfs_stag_t lfs_fs_parent(lfs_t *lfs, const lfs_block_t pair[2],
 }
 #endif
 
+static void lfs_fs_prepsuperblock(lfs_t *lfs, bool needssuperblock) {
+    lfs->gstate.tag = (lfs->gstate.tag & ~LFS_MKTAG(0, 0, 0x200))
+            | (uint32_t)needssuperblock << 9;
+}
+
 #ifndef LFS_READONLY
 static int lfs_fs_preporphans(lfs_t *lfs, int8_t orphans) {
-    LFS_ASSERT(lfs_tag_size(lfs->gstate.tag) > 0 || orphans >= 0);
+    LFS_ASSERT(lfs_tag_size(lfs->gstate.tag) > 0x000 || orphans >= 0);
+    LFS_ASSERT(lfs_tag_size(lfs->gstate.tag) < 0x1ff || orphans <= 0);
     lfs->gstate.tag += orphans;
     lfs->gstate.tag = ((lfs->gstate.tag & ~LFS_MKTAG(0x800, 0, 0)) |
             ((uint32_t)lfs_gstate_hasorphans(&lfs->gstate) << 31));
@@ -4446,6 +4771,45 @@ static void lfs_fs_prepmove(lfs_t *lfs,
 }
 #endif
 
+#ifndef LFS_READONLY
+static int lfs_fs_desuperblock(lfs_t *lfs) {
+    if (!lfs_gstate_needssuperblock(&lfs->gstate)) {
+        return 0;
+    }
+
+    LFS_DEBUG("Rewriting superblock {0x%"PRIx32", 0x%"PRIx32"}",
+            lfs->root[0],
+            lfs->root[1]);
+
+    lfs_mdir_t root;
+    int err = lfs_dir_fetch(lfs, &root, lfs->root);
+    if (err) {
+        return err;
+    }
+
+    // write a new superblock
+    lfs_superblock_t superblock = {
+        .version     = lfs_fs_disk_version(lfs),
+        .block_size  = lfs->cfg->block_size,
+        .block_count = lfs->block_count,
+        .name_max    = lfs->name_max,
+        .file_max    = lfs->file_max,
+        .attr_max    = lfs->attr_max,
+    };
+
+    lfs_superblock_tole32(&superblock);
+    err = lfs_dir_commit(lfs, &root, LFS_MKATTRS(
+            {LFS_MKTAG(LFS_TYPE_INLINESTRUCT, 0, sizeof(superblock)),
+                &superblock}));
+    if (err) {
+        return err;
+    }
+
+    lfs_fs_prepsuperblock(lfs, false);
+    return 0;
+}
+#endif
+
 #ifndef LFS_READONLY
 static int lfs_fs_demove(lfs_t *lfs) {
     if (!lfs_gstate_hasmove(&lfs->gdisk)) {
@@ -4458,6 +4822,10 @@ static int lfs_fs_demove(lfs_t *lfs) {
             lfs->gdisk.pair[1],
             lfs_tag_id(lfs->gdisk.tag));
 
+    // no other gstate is supported at this time, so if we found something else
+    // something most likely went wrong in gstate calculation
+    LFS_ASSERT(lfs_tag_type3(lfs->gdisk.tag) == LFS_TYPE_DELETE);
+
     // fetch and delete the moved entry
     lfs_mdir_t movedir;
     int err = lfs_dir_fetch(lfs, &movedir, lfs->gdisk.pair);
@@ -4484,12 +4852,20 @@ static int lfs_fs_deorphan(lfs_t *lfs, bool powerloss) {
         return 0;
     }
 
-    int8_t found = 0;
-restart:
-    {
+    // Check for orphans in two separate passes:
+    // - 1 for half-orphans (relocations)
+    // - 2 for full-orphans (removes/renames)
+    //
+    // Two separate passes are needed as half-orphans can contain outdated
+    // references to full-orphans, effectively hiding them from the deorphan
+    // search.
+    //
+    int pass = 0;
+    while (pass < 2) {
         // Fix any orphans
         lfs_mdir_t pdir = {.split = true, .tail = {0, 1}};
         lfs_mdir_t dir;
+        bool moreorphans = false;
 
         // iterate over all directory directory entries
         while (!lfs_pair_isnull(pdir.tail)) {
@@ -4507,42 +4883,7 @@ restart:
                     return tag;
                 }
 
-                // note we only check for full orphans if we may have had a
-                // power-loss, otherwise orphans are created intentionally
-                // during operations such as lfs_mkdir
-                if (tag == LFS_ERR_NOENT && powerloss) {
-                    // we are an orphan
-                    LFS_DEBUG("Fixing orphan {0x%"PRIx32", 0x%"PRIx32"}",
-                            pdir.tail[0], pdir.tail[1]);
-
-                    // steal state
-                    err = lfs_dir_getgstate(lfs, &dir, &lfs->gdelta);
-                    if (err) {
-                        return err;
-                    }
-
-                    // steal tail
-                    lfs_pair_tole32(dir.tail);
-                    int state = lfs_dir_orphaningcommit(lfs, &pdir, LFS_MKATTRS(
-                            {LFS_MKTAG(LFS_TYPE_TAIL + dir.split, 0x3ff, 8),
-                                dir.tail}));
-                    lfs_pair_fromle32(dir.tail);
-                    if (state < 0) {
-                        return state;
-                    }
-
-                    found += 1;
-
-                    // did our commit create more orphans?
-                    if (state == LFS_OK_ORPHANED) {
-                        goto restart;
-                    }
-
-                    // refetch tail
-                    continue;
-                }
-
-                if (tag != LFS_ERR_NOENT) {
+                if (pass == 0 && tag != LFS_ERR_NOENT) {
                     lfs_block_t pair[2];
                     lfs_stag_t state = lfs_dir_get(lfs, &parent,
                             LFS_MKTAG(0x7ff, 0x3ff, 0), tag, pair);
@@ -4551,7 +4892,7 @@ restart:
                     }
                     lfs_pair_fromle32(pair);
 
-                    if (!lfs_pair_sync(pair, pdir.tail)) {
+                    if (!lfs_pair_issync(pair, pdir.tail)) {
                         // we have desynced
                         LFS_DEBUG("Fixing half-orphan "
                                 "{0x%"PRIx32", 0x%"PRIx32"} "
@@ -4581,33 +4922,69 @@ restart:
                             return state;
                         }
 
-                        found += 1;
-
                         // did our commit create more orphans?
                         if (state == LFS_OK_ORPHANED) {
-                            goto restart;
+                            moreorphans = true;
                         }
 
                         // refetch tail
                         continue;
                     }
                 }
+
+                // note we only check for full orphans if we may have had a
+                // power-loss, otherwise orphans are created intentionally
+                // during operations such as lfs_mkdir
+                if (pass == 1 && tag == LFS_ERR_NOENT && powerloss) {
+                    // we are an orphan
+                    LFS_DEBUG("Fixing orphan {0x%"PRIx32", 0x%"PRIx32"}",
+                            pdir.tail[0], pdir.tail[1]);
+
+                    // steal state
+                    err = lfs_dir_getgstate(lfs, &dir, &lfs->gdelta);
+                    if (err) {
+                        return err;
+                    }
+
+                    // steal tail
+                    lfs_pair_tole32(dir.tail);
+                    int state = lfs_dir_orphaningcommit(lfs, &pdir, LFS_MKATTRS(
+                            {LFS_MKTAG(LFS_TYPE_TAIL + dir.split, 0x3ff, 8),
+                                dir.tail}));
+                    lfs_pair_fromle32(dir.tail);
+                    if (state < 0) {
+                        return state;
+                    }
+
+                    // did our commit create more orphans?
+                    if (state == LFS_OK_ORPHANED) {
+                        moreorphans = true;
+                    }
+
+                    // refetch tail
+                    continue;
+                }
             }
 
             pdir = dir;
         }
+
+        pass = moreorphans ? 0 : pass+1;
     }
 
     // mark orphans as fixed
-    return lfs_fs_preporphans(lfs, -lfs_min(
-            lfs_gstate_getorphans(&lfs->gstate),
-            found));
+    return lfs_fs_preporphans(lfs, -lfs_gstate_getorphans(&lfs->gstate));
 }
 #endif
 
 #ifndef LFS_READONLY
 static int lfs_fs_forceconsistency(lfs_t *lfs) {
-    int err = lfs_fs_demove(lfs);
+    int err = lfs_fs_desuperblock(lfs);
+    if (err) {
+        return err;
+    }
+
+    err = lfs_fs_demove(lfs);
     if (err) {
         return err;
     }
@@ -4621,6 +4998,36 @@ static int lfs_fs_forceconsistency(lfs_t *lfs) {
 }
 #endif
 
+#ifndef LFS_READONLY
+int lfs_fs_rawmkconsistent(lfs_t *lfs) {
+    // lfs_fs_forceconsistency does most of the work here
+    int err = lfs_fs_forceconsistency(lfs);
+    if (err) {
+        return err;
+    }
+
+    // do we have any pending gstate?
+    lfs_gstate_t delta = {0};
+    lfs_gstate_xor(&delta, &lfs->gdisk);
+    lfs_gstate_xor(&delta, &lfs->gstate);
+    if (!lfs_gstate_iszero(&delta)) {
+        // lfs_dir_commit will implicitly write out any pending gstate
+        lfs_mdir_t root;
+        err = lfs_dir_fetch(lfs, &root, lfs->root);
+        if (err) {
+            return err;
+        }
+
+        err = lfs_dir_commit(lfs, &root, NULL, 0);
+        if (err) {
+            return err;
+        }
+    }
+
+    return 0;
+}
+#endif
+
 static int lfs_fs_size_count(void *p, lfs_block_t block) {
     (void)block;
     lfs_size_t *size = p;
@@ -4638,6 +5045,45 @@ static lfs_ssize_t lfs_fs_rawsize(lfs_t *lfs) {
     return size;
 }
 
+#ifndef LFS_READONLY
+int lfs_fs_rawgrow(lfs_t *lfs, lfs_size_t block_count) {
+    // shrinking is not supported
+    LFS_ASSERT(block_count >= lfs->block_count);
+
+    if (block_count > lfs->block_count) {
+        lfs->block_count = block_count;
+
+        // fetch the root
+        lfs_mdir_t root;
+        int err = lfs_dir_fetch(lfs, &root, lfs->root);
+        if (err) {
+            return err;
+        }
+
+        // update the superblock
+        lfs_superblock_t superblock;
+        lfs_stag_t tag = lfs_dir_get(lfs, &root, LFS_MKTAG(0x7ff, 0x3ff, 0),
+                LFS_MKTAG(LFS_TYPE_INLINESTRUCT, 0, sizeof(superblock)),
+                &superblock);
+        if (tag < 0) {
+            return tag;
+        }
+        lfs_superblock_fromle32(&superblock);
+
+        superblock.block_count = lfs->block_count;
+
+        lfs_superblock_tole32(&superblock);
+        err = lfs_dir_commit(lfs, &root, LFS_MKATTRS(
+                {tag, &superblock}));
+        if (err) {
+            return err;
+        }
+    }
+
+    return 0;
+}
+#endif
+
 #ifdef LFS_MIGRATE
 ////// Migration from littelfs v1 below this //////
 
@@ -5061,6 +5507,10 @@ static int lfs1_unmount(lfs_t *lfs) {
 /// v1 migration ///
 static int lfs_rawmigrate(lfs_t *lfs, const struct lfs_config *cfg) {
     struct lfs1 lfs1;
+
+    // Indeterminate filesystem size not allowed for migration.
+    LFS_ASSERT(cfg->block_count != 0);
+
     int err = lfs1_mount(lfs, &lfs1, cfg);
     if (err) {
         return err;
@@ -5757,6 +6207,20 @@ int lfs_dir_rewind(lfs_t *lfs, lfs_dir_t *dir) {
     return err;
 }
 
+int lfs_fs_stat(lfs_t *lfs, struct lfs_fsinfo *fsinfo) {
+    int err = LFS_LOCK(lfs->cfg);
+    if (err) {
+        return err;
+    }
+    LFS_TRACE("lfs_fs_stat(%p, %p)", (void*)lfs, (void*)fsinfo);
+
+    err = lfs_fs_rawstat(lfs, fsinfo);
+
+    LFS_TRACE("lfs_fs_stat -> %d", err);
+    LFS_UNLOCK(lfs->cfg);
+    return err;
+}
+
 lfs_ssize_t lfs_fs_size(lfs_t *lfs) {
     int err = LFS_LOCK(lfs->cfg);
     if (err) {
@@ -5786,6 +6250,54 @@ int lfs_fs_traverse(lfs_t *lfs, int (*cb)(void *, lfs_block_t), void *data) {
     return err;
 }
 
+#ifndef LFS_READONLY
+int lfs_fs_gc(lfs_t *lfs) {
+    int err = LFS_LOCK(lfs->cfg);
+    if (err) {
+        return err;
+    }
+    LFS_TRACE("lfs_fs_gc(%p)", (void*)lfs);
+
+    err = lfs_fs_rawgc(lfs);
+
+    LFS_TRACE("lfs_fs_gc -> %d", err);
+    LFS_UNLOCK(lfs->cfg);
+    return err;
+}
+#endif
+
+#ifndef LFS_READONLY
+int lfs_fs_mkconsistent(lfs_t *lfs) {
+    int err = LFS_LOCK(lfs->cfg);
+    if (err) {
+        return err;
+    }
+    LFS_TRACE("lfs_fs_mkconsistent(%p)", (void*)lfs);
+
+    err = lfs_fs_rawmkconsistent(lfs);
+
+    LFS_TRACE("lfs_fs_mkconsistent -> %d", err);
+    LFS_UNLOCK(lfs->cfg);
+    return err;
+}
+#endif
+
+#ifndef LFS_READONLY
+int lfs_fs_grow(lfs_t *lfs, lfs_size_t block_count) {
+    int err = LFS_LOCK(lfs->cfg);
+    if (err) {
+        return err;
+    }
+    LFS_TRACE("lfs_fs_grow(%p, %"PRIu32")", (void*)lfs, block_count);
+
+    err = lfs_fs_rawgrow(lfs, block_count);
+
+    LFS_TRACE("lfs_fs_grow -> %d", err);
+    LFS_UNLOCK(lfs->cfg);
+    return err;
+}
+#endif
+
 #ifdef LFS_MIGRATE
 int lfs_migrate(lfs_t *lfs, const struct lfs_config *cfg) {
     int err = LFS_LOCK(cfg);
diff --git a/lfs.h b/lfs.h
index 3fc1e98..9eeab23 100644
--- a/lfs.h
+++ b/lfs.h
@@ -8,8 +8,6 @@
 #ifndef LFS_H
 #define LFS_H
 
-#include <stdint.h>
-#include <stdbool.h>
 #include "lfs_util.h"
 
 #ifdef __cplusplus
@@ -23,14 +21,14 @@ extern "C"
 // Software library version
 // Major (top-nibble), incremented on backwards incompatible changes
 // Minor (bottom-nibble), incremented on feature additions
-#define LFS_VERSION 0x00020005
+#define LFS_VERSION 0x00020008
 #define LFS_VERSION_MAJOR (0xffff & (LFS_VERSION >> 16))
 #define LFS_VERSION_MINOR (0xffff & (LFS_VERSION >>  0))
 
 // Version of On-disk data structures
 // Major (top-nibble), incremented on backwards incompatible changes
 // Minor (bottom-nibble), incremented on feature additions
-#define LFS_DISK_VERSION 0x00020000
+#define LFS_DISK_VERSION 0x00020001
 #define LFS_DISK_VERSION_MAJOR (0xffff & (LFS_DISK_VERSION >> 16))
 #define LFS_DISK_VERSION_MINOR (0xffff & (LFS_DISK_VERSION >>  0))
 
@@ -114,6 +112,8 @@ enum lfs_type {
     LFS_TYPE_SOFTTAIL       = 0x600,
     LFS_TYPE_HARDTAIL       = 0x601,
     LFS_TYPE_MOVESTATE      = 0x7ff,
+    LFS_TYPE_CCRC           = 0x500,
+    LFS_TYPE_FCRC           = 0x5ff,
 
     // internal chip sources
     LFS_FROM_NOOP           = 0x000,
@@ -263,6 +263,14 @@ struct lfs_config {
     // can help bound the metadata compaction time. Must be <= block_size.
     // Defaults to block_size when zero.
     lfs_size_t metadata_max;
+
+#ifdef LFS_MULTIVERSION
+    // On-disk version to use when writing in the form of 16-bit major version
+    // + 16-bit minor version. This limiting metadata to what is supported by
+    // older minor versions. Note that some features will be lost. Defaults to 
+    // to the most recent minor version when zero.
+    uint32_t disk_version;
+#endif
 };
 
 // File info structure
@@ -280,6 +288,27 @@ struct lfs_info {
     char name[LFS_NAME_MAX+1];
 };
 
+// Filesystem info structure
+struct lfs_fsinfo {
+    // On-disk version.
+    uint32_t disk_version;
+
+    // Size of a logical block in bytes.
+    lfs_size_t block_size;
+
+    // Number of logical blocks in filesystem.
+    lfs_size_t block_count;
+
+    // Upper limit on the length of file names in bytes.
+    lfs_size_t name_max;
+
+    // Upper limit on the size of files in bytes.
+    lfs_size_t file_max;
+
+    // Upper limit on the size of custom attributes in bytes.
+    lfs_size_t attr_max;
+};
+
 // Custom attribute structure, used to describe custom attributes
 // committed atomically during file writes.
 struct lfs_attr {
@@ -410,6 +439,7 @@ typedef struct lfs {
     } free;
 
     const struct lfs_config *cfg;
+    lfs_size_t block_count;
     lfs_size_t name_max;
     lfs_size_t file_max;
     lfs_size_t attr_max;
@@ -534,8 +564,8 @@ int lfs_file_open(lfs_t *lfs, lfs_file_t *file,
 // are values from the enum lfs_open_flags that are bitwise-ored together.
 //
 // The config struct provides additional config options per file as described
-// above. The config struct must be allocated while the file is open, and the
-// config struct must be zeroed for defaults and backwards compatibility.
+// above. The config struct must remain allocated while the file is open, and
+// the config struct must be zeroed for defaults and backwards compatibility.
 //
 // Returns a negative error code on failure.
 int lfs_file_opencfg(lfs_t *lfs, lfs_file_t *file,
@@ -659,6 +689,12 @@ int lfs_dir_rewind(lfs_t *lfs, lfs_dir_t *dir);
 
 /// Filesystem-level filesystem operations
 
+// Find on-disk info about the filesystem
+//
+// Fills out the fsinfo structure based on the filesystem found on-disk.
+// Returns a negative error code on failure.
+int lfs_fs_stat(lfs_t *lfs, struct lfs_fsinfo *fsinfo);
+
 // Finds the current size of the filesystem
 //
 // Note: Result is best effort. If files share COW structures, the returned
@@ -676,6 +712,40 @@ lfs_ssize_t lfs_fs_size(lfs_t *lfs);
 // Returns a negative error code on failure.
 int lfs_fs_traverse(lfs_t *lfs, int (*cb)(void*, lfs_block_t), void *data);
 
+// Attempt to proactively find free blocks
+//
+// Calling this function is not required, but may allowing the offloading of
+// the expensive block allocation scan to a less time-critical code path.
+//
+// Note: littlefs currently does not persist any found free blocks to disk.
+// This may change in the future.
+//
+// Returns a negative error code on failure. Finding no free blocks is
+// not an error.
+int lfs_fs_gc(lfs_t *lfs);
+
+#ifndef LFS_READONLY
+// Attempt to make the filesystem consistent and ready for writing
+//
+// Calling this function is not required, consistency will be implicitly
+// enforced on the first operation that writes to the filesystem, but this
+// function allows the work to be performed earlier and without other
+// filesystem changes.
+//
+// Returns a negative error code on failure.
+int lfs_fs_mkconsistent(lfs_t *lfs);
+#endif
+
+#ifndef LFS_READONLY
+// Grows the filesystem to a new size, updating the superblock with the new
+// block count.
+//
+// Note: This is irreversible.
+//
+// Returns a negative error code on failure.
+int lfs_fs_grow(lfs_t *lfs, lfs_size_t block_count);
+#endif
+
 #ifndef LFS_READONLY
 #ifdef LFS_MIGRATE
 // Attempts to migrate a previous version of littlefs
diff --git a/lfs_util.h b/lfs_util.h
index 0cbc2a3..13e9396 100644
--- a/lfs_util.h
+++ b/lfs_util.h
@@ -167,10 +167,9 @@ static inline int lfs_scmp(uint32_t a, uint32_t b) {
 
 // Convert between 32-bit little-endian and native order
 static inline uint32_t lfs_fromle32(uint32_t a) {
-#if !defined(LFS_NO_INTRINSICS) && ( \
-    (defined(  BYTE_ORDER  ) && defined(  ORDER_LITTLE_ENDIAN  ) &&   BYTE_ORDER   ==   ORDER_LITTLE_ENDIAN  ) || \
+#if (defined(  BYTE_ORDER  ) && defined(  ORDER_LITTLE_ENDIAN  ) &&   BYTE_ORDER   ==   ORDER_LITTLE_ENDIAN  ) || \
     (defined(__BYTE_ORDER  ) && defined(__ORDER_LITTLE_ENDIAN  ) && __BYTE_ORDER   == __ORDER_LITTLE_ENDIAN  ) || \
-    (defined(__BYTE_ORDER__) && defined(__ORDER_LITTLE_ENDIAN__) && __BYTE_ORDER__ == __ORDER_LITTLE_ENDIAN__))
+    (defined(__BYTE_ORDER__) && defined(__ORDER_LITTLE_ENDIAN__) && __BYTE_ORDER__ == __ORDER_LITTLE_ENDIAN__)
     return a;
 #elif !defined(LFS_NO_INTRINSICS) && ( \
     (defined(  BYTE_ORDER  ) && defined(  ORDER_BIG_ENDIAN  ) &&   BYTE_ORDER   ==   ORDER_BIG_ENDIAN  ) || \
@@ -196,10 +195,9 @@ static inline uint32_t lfs_frombe32(uint32_t a) {
     (defined(__BYTE_ORDER  ) && defined(__ORDER_LITTLE_ENDIAN  ) && __BYTE_ORDER   == __ORDER_LITTLE_ENDIAN  ) || \
     (defined(__BYTE_ORDER__) && defined(__ORDER_LITTLE_ENDIAN__) && __BYTE_ORDER__ == __ORDER_LITTLE_ENDIAN__))
     return __builtin_bswap32(a);
-#elif !defined(LFS_NO_INTRINSICS) && ( \
-    (defined(  BYTE_ORDER  ) && defined(  ORDER_BIG_ENDIAN  ) &&   BYTE_ORDER   ==   ORDER_BIG_ENDIAN  ) || \
+#elif (defined(  BYTE_ORDER  ) && defined(  ORDER_BIG_ENDIAN  ) &&   BYTE_ORDER   ==   ORDER_BIG_ENDIAN  ) || \
     (defined(__BYTE_ORDER  ) && defined(__ORDER_BIG_ENDIAN  ) && __BYTE_ORDER   == __ORDER_BIG_ENDIAN  ) || \
-    (defined(__BYTE_ORDER__) && defined(__ORDER_BIG_ENDIAN__) && __BYTE_ORDER__ == __ORDER_BIG_ENDIAN__))
+    (defined(__BYTE_ORDER__) && defined(__ORDER_BIG_ENDIAN__) && __BYTE_ORDER__ == __ORDER_BIG_ENDIAN__)
     return a;
 #else
     return (((uint8_t*)&a)[0] << 24) |
diff --git a/runners/bench_runner.c b/runners/bench_runner.c
new file mode 100644
index 0000000..f4dce22
--- /dev/null
+++ b/runners/bench_runner.c
@@ -0,0 +1,2055 @@
+/*
+ * Runner for littlefs benchmarks
+ *
+ * Copyright (c) 2022, The littlefs authors.
+ * SPDX-License-Identifier: BSD-3-Clause
+ */
+#ifndef _POSIX_C_SOURCE
+#define _POSIX_C_SOURCE 199309L
+#endif
+
+#include "runners/bench_runner.h"
+#include "bd/lfs_emubd.h"
+
+#include <getopt.h>
+#include <sys/types.h>
+#include <errno.h>
+#include <setjmp.h>
+#include <fcntl.h>
+#include <stdarg.h>
+#include <stdio.h>
+#include <unistd.h>
+#include <execinfo.h>
+#include <time.h>
+
+
+// some helpers
+
+// append to an array with amortized doubling
+void *mappend(void **p,
+        size_t size,
+        size_t *count,
+        size_t *capacity) {
+    uint8_t *p_ = *p;
+    size_t count_ = *count;
+    size_t capacity_ = *capacity;
+
+    count_ += 1;
+    if (count_ > capacity_) {
+        capacity_ = (2*capacity_ < 4) ? 4 : 2*capacity_;
+
+        p_ = realloc(p_, capacity_*size);
+        if (!p_) {
+            return NULL;
+        }
+    }
+
+    *p = p_;
+    *count = count_;
+    *capacity = capacity_;
+    return &p_[(count_-1)*size];
+}
+
+// a quick self-terminating text-safe varint scheme
+static void leb16_print(uintmax_t x) {
+    // allow 'w' to indicate negative numbers
+    if ((intmax_t)x < 0) {
+        printf("w");
+        x = -x;
+    }
+
+    while (true) {
+        char nibble = (x & 0xf) | (x > 0xf ? 0x10 : 0);
+        printf("%c", (nibble < 10) ? '0'+nibble : 'a'+nibble-10);
+        if (x <= 0xf) {
+            break;
+        }
+        x >>= 4;
+    }
+}
+
+static uintmax_t leb16_parse(const char *s, char **tail) {
+    bool neg = false;
+    uintmax_t x = 0;
+    if (tail) {
+        *tail = (char*)s;
+    }
+
+    if (s[0] == 'w') {
+        neg = true;
+        s = s+1;
+    }
+
+    size_t i = 0;
+    while (true) {
+        uintmax_t nibble = s[i];
+        if (nibble >= '0' && nibble <= '9') {
+            nibble = nibble - '0';
+        } else if (nibble >= 'a' && nibble <= 'v') {
+            nibble = nibble - 'a' + 10;
+        } else {
+            // invalid?
+            return 0;
+        }
+
+        x |= (nibble & 0xf) << (4*i);
+        i += 1;
+        if (!(nibble & 0x10)) {
+            s = s + i;
+            break;
+        }
+    }
+
+    if (tail) {
+        *tail = (char*)s;
+    }
+    return neg ? -x : x;
+}
+
+
+
+// bench_runner types
+
+typedef struct bench_geometry {
+    const char *name;
+    bench_define_t defines[BENCH_GEOMETRY_DEFINE_COUNT];
+} bench_geometry_t;
+
+typedef struct bench_id {
+    const char *name;
+    const bench_define_t *defines;
+    size_t define_count;
+} bench_id_t;
+
+
+// bench suites are linked into a custom ld section
+extern struct bench_suite __start__bench_suites;
+extern struct bench_suite __stop__bench_suites;
+
+const struct bench_suite *bench_suites = &__start__bench_suites;
+#define BENCH_SUITE_COUNT \
+    ((size_t)(&__stop__bench_suites - &__start__bench_suites))
+
+
+// bench define management
+typedef struct bench_define_map {
+    const bench_define_t *defines;
+    size_t count;
+} bench_define_map_t;
+
+typedef struct bench_define_names {
+    const char *const *names;
+    size_t count;
+} bench_define_names_t;
+
+intmax_t bench_define_lit(void *data) {
+    return (intptr_t)data;
+}
+
+#define BENCH_CONST(x) {bench_define_lit, (void*)(uintptr_t)(x)}
+#define BENCH_LIT(x) ((bench_define_t)BENCH_CONST(x))
+
+
+#define BENCH_DEF(k, v) \
+    intmax_t bench_define_##k(void *data) { \
+        (void)data; \
+        return v; \
+    }
+
+    BENCH_IMPLICIT_DEFINES
+#undef BENCH_DEF
+
+#define BENCH_DEFINE_MAP_OVERRIDE    0
+#define BENCH_DEFINE_MAP_EXPLICIT    1
+#define BENCH_DEFINE_MAP_PERMUTATION 2
+#define BENCH_DEFINE_MAP_GEOMETRY    3
+#define BENCH_DEFINE_MAP_IMPLICIT    4
+#define BENCH_DEFINE_MAP_COUNT       5
+
+bench_define_map_t bench_define_maps[BENCH_DEFINE_MAP_COUNT] = {
+    [BENCH_DEFINE_MAP_IMPLICIT] = {
+        (const bench_define_t[BENCH_IMPLICIT_DEFINE_COUNT]) {
+            #define BENCH_DEF(k, v) \
+                [k##_i] = {bench_define_##k, NULL},
+
+                BENCH_IMPLICIT_DEFINES
+            #undef BENCH_DEF
+        },
+        BENCH_IMPLICIT_DEFINE_COUNT,
+    },
+};
+
+#define BENCH_DEFINE_NAMES_SUITE    0
+#define BENCH_DEFINE_NAMES_IMPLICIT 1
+#define BENCH_DEFINE_NAMES_COUNT    2
+
+bench_define_names_t bench_define_names[BENCH_DEFINE_NAMES_COUNT] = {
+    [BENCH_DEFINE_NAMES_IMPLICIT] = {
+        (const char *const[BENCH_IMPLICIT_DEFINE_COUNT]){
+            #define BENCH_DEF(k, v) \
+                [k##_i] = #k,
+
+                BENCH_IMPLICIT_DEFINES
+            #undef BENCH_DEF
+        },
+        BENCH_IMPLICIT_DEFINE_COUNT,
+    },
+};
+
+intmax_t *bench_define_cache;
+size_t bench_define_cache_count;
+unsigned *bench_define_cache_mask;
+
+const char *bench_define_name(size_t define) {
+    // lookup in our bench names
+    for (size_t i = 0; i < BENCH_DEFINE_NAMES_COUNT; i++) {
+        if (define < bench_define_names[i].count
+                && bench_define_names[i].names
+                && bench_define_names[i].names[define]) {
+            return bench_define_names[i].names[define];
+        }
+    }
+
+    return NULL;
+}
+
+bool bench_define_ispermutation(size_t define) {
+    // is this define specific to the permutation?
+    for (size_t i = 0; i < BENCH_DEFINE_MAP_IMPLICIT; i++) {
+        if (define < bench_define_maps[i].count
+                && bench_define_maps[i].defines[define].cb) {
+            return true;
+        }
+    }
+
+    return false;
+}
+
+intmax_t bench_define(size_t define) {
+    // is the define in our cache?
+    if (define < bench_define_cache_count
+            && (bench_define_cache_mask[define/(8*sizeof(unsigned))]
+                & (1 << (define%(8*sizeof(unsigned)))))) {
+        return bench_define_cache[define];
+    }
+
+    // lookup in our bench defines
+    for (size_t i = 0; i < BENCH_DEFINE_MAP_COUNT; i++) {
+        if (define < bench_define_maps[i].count
+                && bench_define_maps[i].defines[define].cb) {
+            intmax_t v = bench_define_maps[i].defines[define].cb(
+                    bench_define_maps[i].defines[define].data);
+
+            // insert into cache!
+            bench_define_cache[define] = v;
+            bench_define_cache_mask[define / (8*sizeof(unsigned))]
+                    |= 1 << (define%(8*sizeof(unsigned)));
+
+            return v;
+        }
+    }
+
+    return 0;
+
+    // not found?
+    const char *name = bench_define_name(define);
+    fprintf(stderr, "error: undefined define %s (%zd)\n",
+            name ? name : "(unknown)",
+            define);
+    assert(false);
+    exit(-1);
+}
+
+void bench_define_flush(void) {
+    // clear cache between permutations
+    memset(bench_define_cache_mask, 0,
+            sizeof(unsigned)*(
+                (bench_define_cache_count+(8*sizeof(unsigned))-1)
+                / (8*sizeof(unsigned))));
+}
+
+// geometry updates
+const bench_geometry_t *bench_geometry = NULL;
+
+void bench_define_geometry(const bench_geometry_t *geometry) {
+    bench_define_maps[BENCH_DEFINE_MAP_GEOMETRY] = (bench_define_map_t){
+            geometry->defines, BENCH_GEOMETRY_DEFINE_COUNT};
+}
+
+// override updates
+typedef struct bench_override {
+    const char *name;
+    const intmax_t *defines;
+    size_t permutations;
+} bench_override_t;
+
+const bench_override_t *bench_overrides = NULL;
+size_t bench_override_count = 0;
+
+bench_define_t *bench_override_defines = NULL;
+size_t bench_override_define_count = 0;
+size_t bench_override_define_permutations = 1;
+size_t bench_override_define_capacity = 0;
+
+// suite/perm updates
+void bench_define_suite(const struct bench_suite *suite) {
+    bench_define_names[BENCH_DEFINE_NAMES_SUITE] = (bench_define_names_t){
+            suite->define_names, suite->define_count};
+
+    // make sure our cache is large enough
+    if (lfs_max(suite->define_count, BENCH_IMPLICIT_DEFINE_COUNT)
+            > bench_define_cache_count) {
+        // align to power of two to avoid any superlinear growth
+        size_t ncount = 1 << lfs_npw2(
+                lfs_max(suite->define_count, BENCH_IMPLICIT_DEFINE_COUNT));
+        bench_define_cache = realloc(bench_define_cache, ncount*sizeof(intmax_t));
+        bench_define_cache_mask = realloc(bench_define_cache_mask,
+                sizeof(unsigned)*(
+                    (ncount+(8*sizeof(unsigned))-1)
+                    / (8*sizeof(unsigned))));
+        bench_define_cache_count = ncount;
+    }
+
+    // map any overrides
+    if (bench_override_count > 0) {
+        // first figure out the total size of override permutations
+        size_t count = 0;
+        size_t permutations = 1;
+        for (size_t i = 0; i < bench_override_count; i++) {
+            for (size_t d = 0;
+                    d < lfs_max(
+                        suite->define_count,
+                        BENCH_IMPLICIT_DEFINE_COUNT);
+                    d++) {
+                // define name match?
+                const char *name = bench_define_name(d);
+                if (name && strcmp(name, bench_overrides[i].name) == 0) {
+                    count = lfs_max(count, d+1);
+                    permutations *= bench_overrides[i].permutations;
+                    break;
+                }
+            }
+        }
+        bench_override_define_count = count;
+        bench_override_define_permutations = permutations;
+
+        // make sure our override arrays are big enough
+        if (count * permutations > bench_override_define_capacity) {
+            // align to power of two to avoid any superlinear growth
+            size_t ncapacity = 1 << lfs_npw2(count * permutations);
+            bench_override_defines = realloc(
+                    bench_override_defines,
+                    sizeof(bench_define_t)*ncapacity);
+            bench_override_define_capacity = ncapacity;
+        }
+
+        // zero unoverridden defines
+        memset(bench_override_defines, 0,
+                sizeof(bench_define_t) * count * permutations);
+
+        // compute permutations
+        size_t p = 1;
+        for (size_t i = 0; i < bench_override_count; i++) {
+            for (size_t d = 0;
+                    d < lfs_max(
+                        suite->define_count,
+                        BENCH_IMPLICIT_DEFINE_COUNT);
+                    d++) {
+                // define name match?
+                const char *name = bench_define_name(d);
+                if (name && strcmp(name, bench_overrides[i].name) == 0) {
+                    // scatter the define permutations based on already
+                    // seen permutations
+                    for (size_t j = 0; j < permutations; j++) {
+                        bench_override_defines[j*count + d] = BENCH_LIT(
+                                bench_overrides[i].defines[(j/p)
+                                    % bench_overrides[i].permutations]);
+                    }
+
+                    // keep track of how many permutations we've seen so far
+                    p *= bench_overrides[i].permutations;
+                    break;
+                }
+            }
+        }
+    }
+}
+
+void bench_define_perm(
+        const struct bench_suite *suite,
+        const struct bench_case *case_,
+        size_t perm) {
+    if (case_->defines) {
+        bench_define_maps[BENCH_DEFINE_MAP_PERMUTATION] = (bench_define_map_t){
+                case_->defines + perm*suite->define_count,
+                suite->define_count};
+    } else {
+        bench_define_maps[BENCH_DEFINE_MAP_PERMUTATION] = (bench_define_map_t){
+                NULL, 0};
+    }
+}
+
+void bench_define_override(size_t perm) {
+    bench_define_maps[BENCH_DEFINE_MAP_OVERRIDE] = (bench_define_map_t){
+            bench_override_defines + perm*bench_override_define_count,
+            bench_override_define_count};
+}
+
+void bench_define_explicit(
+        const bench_define_t *defines,
+        size_t define_count) {
+    bench_define_maps[BENCH_DEFINE_MAP_EXPLICIT] = (bench_define_map_t){
+            defines, define_count};
+}
+
+void bench_define_cleanup(void) {
+    // bench define management can allocate a few things
+    free(bench_define_cache);
+    free(bench_define_cache_mask);
+    free(bench_override_defines);
+}
+
+
+
+// bench state
+extern const bench_geometry_t *bench_geometries;
+extern size_t bench_geometry_count;
+
+const bench_id_t *bench_ids = (const bench_id_t[]) {
+    {NULL, NULL, 0},
+};
+size_t bench_id_count = 1;
+
+size_t bench_step_start = 0;
+size_t bench_step_stop = -1;
+size_t bench_step_step = 1;
+
+const char *bench_disk_path = NULL;
+const char *bench_trace_path = NULL;
+bool bench_trace_backtrace = false;
+uint32_t bench_trace_period = 0;
+uint32_t bench_trace_freq = 0;
+FILE *bench_trace_file = NULL;
+uint32_t bench_trace_cycles = 0;
+uint64_t bench_trace_time = 0;
+uint64_t bench_trace_open_time = 0;
+lfs_emubd_sleep_t bench_read_sleep = 0.0;
+lfs_emubd_sleep_t bench_prog_sleep = 0.0;
+lfs_emubd_sleep_t bench_erase_sleep = 0.0;
+
+// this determines both the backtrace buffer and the trace printf buffer, if
+// trace ends up interleaved or truncated this may need to be increased
+#ifndef BENCH_TRACE_BACKTRACE_BUFFER_SIZE
+#define BENCH_TRACE_BACKTRACE_BUFFER_SIZE 8192
+#endif
+void *bench_trace_backtrace_buffer[
+    BENCH_TRACE_BACKTRACE_BUFFER_SIZE / sizeof(void*)];
+
+// trace printing
+void bench_trace(const char *fmt, ...) {
+    if (bench_trace_path) {
+        // sample at a specific period?
+        if (bench_trace_period) {
+            if (bench_trace_cycles % bench_trace_period != 0) {
+                bench_trace_cycles += 1;
+                return;
+            }
+            bench_trace_cycles += 1;
+        }
+
+        // sample at a specific frequency?
+        if (bench_trace_freq) {
+            struct timespec t;
+            clock_gettime(CLOCK_MONOTONIC, &t);
+            uint64_t now = (uint64_t)t.tv_sec*1000*1000*1000
+                    + (uint64_t)t.tv_nsec;
+            if (now - bench_trace_time < (1000*1000*1000) / bench_trace_freq) {
+                return;
+            }
+            bench_trace_time = now;
+        }
+
+        if (!bench_trace_file) {
+            // Tracing output is heavy and trying to open every trace
+            // call is slow, so we only try to open the trace file every
+            // so often. Note this doesn't affect successfully opened files
+            struct timespec t;
+            clock_gettime(CLOCK_MONOTONIC, &t);
+            uint64_t now = (uint64_t)t.tv_sec*1000*1000*1000
+                    + (uint64_t)t.tv_nsec;
+            if (now - bench_trace_open_time < 100*1000*1000) {
+                return;
+            }
+            bench_trace_open_time = now;
+
+            // try to open the trace file
+            int fd;
+            if (strcmp(bench_trace_path, "-") == 0) {
+                fd = dup(1);
+                if (fd < 0) {
+                    return;
+                }
+            } else {
+                fd = open(
+                        bench_trace_path,
+                        O_WRONLY | O_CREAT | O_APPEND | O_NONBLOCK,
+                        0666);
+                if (fd < 0) {
+                    return;
+                }
+                int err = fcntl(fd, F_SETFL, O_WRONLY | O_CREAT | O_APPEND);
+                assert(!err);
+            }
+
+            FILE *f = fdopen(fd, "a");
+            assert(f);
+            int err = setvbuf(f, NULL, _IOFBF,
+                    BENCH_TRACE_BACKTRACE_BUFFER_SIZE);
+            assert(!err);
+            bench_trace_file = f;
+        }
+
+        // print trace
+        va_list va;
+        va_start(va, fmt);
+        int res = vfprintf(bench_trace_file, fmt, va);
+        va_end(va);
+        if (res < 0) {
+            fclose(bench_trace_file);
+            bench_trace_file = NULL;
+            return;
+        }
+
+        if (bench_trace_backtrace) {
+            // print backtrace
+            size_t count = backtrace(
+                    bench_trace_backtrace_buffer,
+                    BENCH_TRACE_BACKTRACE_BUFFER_SIZE);
+            // note we skip our own stack frame
+            for (size_t i = 1; i < count; i++) {
+                res = fprintf(bench_trace_file, "\tat %p\n",
+                        bench_trace_backtrace_buffer[i]);
+                if (res < 0) {
+                    fclose(bench_trace_file);
+                    bench_trace_file = NULL;
+                    return;
+                }
+            }
+        }
+
+        // flush immediately
+        fflush(bench_trace_file);
+    }
+}
+
+
+// bench prng
+uint32_t bench_prng(uint32_t *state) {
+    // A simple xorshift32 generator, easily reproducible. Keep in mind
+    // determinism is much more important than actual randomness here.
+    uint32_t x = *state;
+    x ^= x << 13;
+    x ^= x >> 17;
+    x ^= x << 5;
+    *state = x;
+    return x;
+}
+
+
+// bench recording state
+static struct lfs_config *bench_cfg = NULL;
+static lfs_emubd_io_t bench_last_readed = 0;
+static lfs_emubd_io_t bench_last_proged = 0;
+static lfs_emubd_io_t bench_last_erased = 0;
+lfs_emubd_io_t bench_readed = 0;
+lfs_emubd_io_t bench_proged = 0;
+lfs_emubd_io_t bench_erased = 0;
+
+void bench_reset(void) {
+    bench_readed = 0;
+    bench_proged = 0;
+    bench_erased = 0;
+    bench_last_readed = 0;
+    bench_last_proged = 0;
+    bench_last_erased = 0;
+}
+
+void bench_start(void) {
+    assert(bench_cfg);
+    lfs_emubd_sio_t readed = lfs_emubd_readed(bench_cfg);
+    assert(readed >= 0);
+    lfs_emubd_sio_t proged = lfs_emubd_proged(bench_cfg);
+    assert(proged >= 0);
+    lfs_emubd_sio_t erased = lfs_emubd_erased(bench_cfg);
+    assert(erased >= 0);
+
+    bench_last_readed = readed;
+    bench_last_proged = proged;
+    bench_last_erased = erased;
+}
+
+void bench_stop(void) {
+    assert(bench_cfg);
+    lfs_emubd_sio_t readed = lfs_emubd_readed(bench_cfg);
+    assert(readed >= 0);
+    lfs_emubd_sio_t proged = lfs_emubd_proged(bench_cfg);
+    assert(proged >= 0);
+    lfs_emubd_sio_t erased = lfs_emubd_erased(bench_cfg);
+    assert(erased >= 0);
+
+    bench_readed += readed - bench_last_readed;
+    bench_proged += proged - bench_last_proged;
+    bench_erased += erased - bench_last_erased;
+}
+
+
+// encode our permutation into a reusable id
+static void perm_printid(
+        const struct bench_suite *suite,
+        const struct bench_case *case_) {
+    (void)suite;
+    // case[:permutation]
+    printf("%s:", case_->name);
+    for (size_t d = 0;
+            d < lfs_max(
+                suite->define_count,
+                BENCH_IMPLICIT_DEFINE_COUNT);
+            d++) {
+        if (bench_define_ispermutation(d)) {
+            leb16_print(d);
+            leb16_print(BENCH_DEFINE(d));
+        }
+    }
+}
+
+// a quick trie for keeping track of permutations we've seen
+typedef struct bench_seen {
+    struct bench_seen_branch *branches;
+    size_t branch_count;
+    size_t branch_capacity;
+} bench_seen_t;
+
+struct bench_seen_branch {
+    intmax_t define;
+    struct bench_seen branch;
+};
+
+bool bench_seen_insert(
+        bench_seen_t *seen,
+        const struct bench_suite *suite,
+        const struct bench_case *case_) {
+    (void)case_;
+    bool was_seen = true;
+
+    // use the currently set defines
+    for (size_t d = 0;
+            d < lfs_max(
+                suite->define_count,
+                BENCH_IMPLICIT_DEFINE_COUNT);
+            d++) {
+        // treat unpermuted defines the same as 0
+        intmax_t define = bench_define_ispermutation(d) ? BENCH_DEFINE(d) : 0;
+
+        // already seen?
+        struct bench_seen_branch *branch = NULL;
+        for (size_t i = 0; i < seen->branch_count; i++) {
+            if (seen->branches[i].define == define) {
+                branch = &seen->branches[i];
+                break;
+            }
+        }
+
+        // need to create a new node
+        if (!branch) {
+            was_seen = false;
+            branch = mappend(
+                    (void**)&seen->branches,
+                    sizeof(struct bench_seen_branch),
+                    &seen->branch_count,
+                    &seen->branch_capacity);
+            branch->define = define;
+            branch->branch = (bench_seen_t){NULL, 0, 0};
+        }
+
+        seen = &branch->branch;
+    }
+
+    return was_seen;
+}
+
+void bench_seen_cleanup(bench_seen_t *seen) {
+    for (size_t i = 0; i < seen->branch_count; i++) {
+        bench_seen_cleanup(&seen->branches[i].branch);
+    }
+    free(seen->branches);
+}
+
+// iterate through permutations in a bench case
+static void case_forperm(
+        const struct bench_suite *suite,
+        const struct bench_case *case_,
+        const bench_define_t *defines,
+        size_t define_count,
+        void (*cb)(
+            void *data,
+            const struct bench_suite *suite,
+            const struct bench_case *case_),
+        void *data) {
+    // explicit permutation?
+    if (defines) {
+        bench_define_explicit(defines, define_count);
+
+        for (size_t v = 0; v < bench_override_define_permutations; v++) {
+            // define override permutation
+            bench_define_override(v);
+            bench_define_flush();
+
+            cb(data, suite, case_);
+        }
+
+        return;
+    }
+
+    bench_seen_t seen = {NULL, 0, 0};
+
+    for (size_t k = 0; k < case_->permutations; k++) {
+        // define permutation
+        bench_define_perm(suite, case_, k);
+
+        for (size_t v = 0; v < bench_override_define_permutations; v++) {
+            // define override permutation
+            bench_define_override(v);
+
+            for (size_t g = 0; g < bench_geometry_count; g++) {
+                // define geometry
+                bench_define_geometry(&bench_geometries[g]);
+                bench_define_flush();
+
+                // have we seen this permutation before?
+                bool was_seen = bench_seen_insert(&seen, suite, case_);
+                if (!(k == 0 && v == 0 && g == 0) && was_seen) {
+                    continue;
+                }
+
+                cb(data, suite, case_);
+            }
+        }
+    }
+
+    bench_seen_cleanup(&seen);
+}
+
+
+// how many permutations are there actually in a bench case
+struct perm_count_state {
+    size_t total;
+    size_t filtered;
+};
+
+void perm_count(
+        void *data,
+        const struct bench_suite *suite,
+        const struct bench_case *case_) {
+    struct perm_count_state *state = data;
+    (void)suite;
+    (void)case_;
+
+    state->total += 1;
+
+    if (case_->filter && !case_->filter()) {
+        return;
+    }
+
+    state->filtered += 1;
+}
+
+
+// operations we can do
+static void summary(void) {
+    printf("%-23s  %7s %7s %7s %11s\n",
+            "", "flags", "suites", "cases", "perms");
+    size_t suites = 0;
+    size_t cases = 0;
+    bench_flags_t flags = 0;
+    struct perm_count_state perms = {0, 0};
+
+    for (size_t t = 0; t < bench_id_count; t++) {
+        for (size_t i = 0; i < BENCH_SUITE_COUNT; i++) {
+            bench_define_suite(&bench_suites[i]);
+
+            for (size_t j = 0; j < bench_suites[i].case_count; j++) {
+                // does neither suite nor case name match?
+                if (bench_ids[t].name && !(
+                        strcmp(bench_ids[t].name,
+                            bench_suites[i].name) == 0
+                        || strcmp(bench_ids[t].name,
+                            bench_suites[i].cases[j].name) == 0)) {
+                    continue;
+                }
+
+                cases += 1;
+                case_forperm(
+                        &bench_suites[i],
+                        &bench_suites[i].cases[j],
+                        bench_ids[t].defines,
+                        bench_ids[t].define_count,
+                        perm_count,
+                        &perms);
+            }
+
+            suites += 1;
+            flags |= bench_suites[i].flags;
+        }
+    }
+
+    char perm_buf[64];
+    sprintf(perm_buf, "%zu/%zu", perms.filtered, perms.total);
+    char flag_buf[64];
+    sprintf(flag_buf, "%s%s",
+            (flags & BENCH_REENTRANT) ? "r" : "",
+            (!flags) ? "-" : "");
+    printf("%-23s  %7s %7zu %7zu %11s\n",
+            "TOTAL",
+            flag_buf,
+            suites,
+            cases,
+            perm_buf);
+}
+
+static void list_suites(void) {
+    // at least size so that names fit
+    unsigned name_width = 23;
+    for (size_t i = 0; i < BENCH_SUITE_COUNT; i++) {
+        size_t len = strlen(bench_suites[i].name);
+        if (len > name_width) {
+            name_width = len;
+        }
+    }
+    name_width = 4*((name_width+1+4-1)/4)-1;
+
+    printf("%-*s  %7s %7s %11s\n",
+            name_width, "suite", "flags", "cases", "perms");
+    for (size_t t = 0; t < bench_id_count; t++) {
+        for (size_t i = 0; i < BENCH_SUITE_COUNT; i++) {
+            bench_define_suite(&bench_suites[i]);
+
+            size_t cases = 0;
+            struct perm_count_state perms = {0, 0};
+
+            for (size_t j = 0; j < bench_suites[i].case_count; j++) {
+                // does neither suite nor case name match?
+                if (bench_ids[t].name && !(
+                        strcmp(bench_ids[t].name,
+                            bench_suites[i].name) == 0
+                        || strcmp(bench_ids[t].name,
+                            bench_suites[i].cases[j].name) == 0)) {
+                    continue;
+                }
+
+                cases += 1;
+                case_forperm(
+                        &bench_suites[i],
+                        &bench_suites[i].cases[j],
+                        bench_ids[t].defines,
+                        bench_ids[t].define_count,
+                        perm_count,
+                        &perms);
+            }
+
+            // no benches found?
+            if (!cases) {
+                continue;
+            }
+
+            char perm_buf[64];
+            sprintf(perm_buf, "%zu/%zu", perms.filtered, perms.total);
+            char flag_buf[64];
+            sprintf(flag_buf, "%s%s",
+                    (bench_suites[i].flags & BENCH_REENTRANT) ? "r" : "",
+                    (!bench_suites[i].flags) ? "-" : "");
+            printf("%-*s  %7s %7zu %11s\n",
+                    name_width,
+                    bench_suites[i].name,
+                    flag_buf,
+                    cases,
+                    perm_buf);
+        }
+    }
+}
+
+static void list_cases(void) {
+    // at least size so that names fit
+    unsigned name_width = 23;
+    for (size_t i = 0; i < BENCH_SUITE_COUNT; i++) {
+        for (size_t j = 0; j < bench_suites[i].case_count; j++) {
+            size_t len = strlen(bench_suites[i].cases[j].name);
+            if (len > name_width) {
+                name_width = len;
+            }
+        }
+    }
+    name_width = 4*((name_width+1+4-1)/4)-1;
+
+    printf("%-*s  %7s %11s\n", name_width, "case", "flags", "perms");
+    for (size_t t = 0; t < bench_id_count; t++) {
+        for (size_t i = 0; i < BENCH_SUITE_COUNT; i++) {
+            bench_define_suite(&bench_suites[i]);
+
+            for (size_t j = 0; j < bench_suites[i].case_count; j++) {
+                // does neither suite nor case name match?
+                if (bench_ids[t].name && !(
+                        strcmp(bench_ids[t].name,
+                            bench_suites[i].name) == 0
+                        || strcmp(bench_ids[t].name,
+                            bench_suites[i].cases[j].name) == 0)) {
+                    continue;
+                }
+
+                struct perm_count_state perms = {0, 0};
+                case_forperm(
+                        &bench_suites[i],
+                        &bench_suites[i].cases[j],
+                        bench_ids[t].defines,
+                        bench_ids[t].define_count,
+                        perm_count,
+                        &perms);
+
+                char perm_buf[64];
+                sprintf(perm_buf, "%zu/%zu", perms.filtered, perms.total);
+                char flag_buf[64];
+                sprintf(flag_buf, "%s%s",
+                        (bench_suites[i].cases[j].flags & BENCH_REENTRANT)
+                            ? "r" : "",
+                        (!bench_suites[i].cases[j].flags)
+                            ? "-" : "");
+                printf("%-*s  %7s %11s\n",
+                        name_width,
+                        bench_suites[i].cases[j].name,
+                        flag_buf,
+                        perm_buf);
+            }
+        }
+    }
+}
+
+static void list_suite_paths(void) {
+    // at least size so that names fit
+    unsigned name_width = 23;
+    for (size_t i = 0; i < BENCH_SUITE_COUNT; i++) {
+        size_t len = strlen(bench_suites[i].name);
+        if (len > name_width) {
+            name_width = len;
+        }
+    }
+    name_width = 4*((name_width+1+4-1)/4)-1;
+
+    printf("%-*s  %s\n", name_width, "suite", "path");
+    for (size_t t = 0; t < bench_id_count; t++) {
+        for (size_t i = 0; i < BENCH_SUITE_COUNT; i++) {
+            size_t cases = 0;
+
+            for (size_t j = 0; j < bench_suites[i].case_count; j++) {
+                // does neither suite nor case name match?
+                if (bench_ids[t].name && !(
+                        strcmp(bench_ids[t].name,
+                            bench_suites[i].name) == 0
+                        || strcmp(bench_ids[t].name,
+                            bench_suites[i].cases[j].name) == 0)) {
+                    continue;
+
+                    cases += 1;
+                }
+            }
+
+            // no benches found?
+            if (!cases) {
+                continue;
+            }
+
+            printf("%-*s  %s\n",
+                    name_width,
+                    bench_suites[i].name,
+                    bench_suites[i].path);
+        }
+    }
+}
+
+static void list_case_paths(void) {
+    // at least size so that names fit
+    unsigned name_width = 23;
+    for (size_t i = 0; i < BENCH_SUITE_COUNT; i++) {
+        for (size_t j = 0; j < bench_suites[i].case_count; j++) {
+            size_t len = strlen(bench_suites[i].cases[j].name);
+            if (len > name_width) {
+                name_width = len;
+            }
+        }
+    }
+    name_width = 4*((name_width+1+4-1)/4)-1;
+
+    printf("%-*s  %s\n", name_width, "case", "path");
+    for (size_t t = 0; t < bench_id_count; t++) {
+        for (size_t i = 0; i < BENCH_SUITE_COUNT; i++) {
+            for (size_t j = 0; j < bench_suites[i].case_count; j++) {
+                // does neither suite nor case name match?
+                if (bench_ids[t].name && !(
+                        strcmp(bench_ids[t].name,
+                            bench_suites[i].name) == 0
+                        || strcmp(bench_ids[t].name,
+                            bench_suites[i].cases[j].name) == 0)) {
+                    continue;
+                }
+
+                printf("%-*s  %s\n",
+                        name_width,
+                        bench_suites[i].cases[j].name,
+                        bench_suites[i].cases[j].path);
+            }
+        }
+    }
+}
+
+struct list_defines_define {
+    const char *name;
+    intmax_t *values;
+    size_t value_count;
+    size_t value_capacity;
+};
+
+struct list_defines_defines {
+    struct list_defines_define *defines;
+    size_t define_count;
+    size_t define_capacity;
+};
+
+static void list_defines_add(
+        struct list_defines_defines *defines,
+        size_t d) {
+    const char *name = bench_define_name(d);
+    intmax_t value = BENCH_DEFINE(d);
+
+    // define already in defines?
+    for (size_t i = 0; i < defines->define_count; i++) {
+        if (strcmp(defines->defines[i].name, name) == 0) {
+            // value already in values?
+            for (size_t j = 0; j < defines->defines[i].value_count; j++) {
+                if (defines->defines[i].values[j] == value) {
+                    return;
+                }
+            }
+
+            *(intmax_t*)mappend(
+                (void**)&defines->defines[i].values,
+                sizeof(intmax_t),
+                &defines->defines[i].value_count,
+                &defines->defines[i].value_capacity) = value;
+
+            return;
+        }
+    }
+
+    // new define?
+    struct list_defines_define *define = mappend(
+            (void**)&defines->defines,
+            sizeof(struct list_defines_define),
+            &defines->define_count,
+            &defines->define_capacity);
+    define->name = name;
+    define->values = malloc(sizeof(intmax_t));
+    define->values[0] = value;
+    define->value_count = 1;
+    define->value_capacity = 1;
+}
+
+void perm_list_defines(
+        void *data,
+        const struct bench_suite *suite,
+        const struct bench_case *case_) {
+    struct list_defines_defines *defines = data;
+    (void)suite;
+    (void)case_;
+
+    // collect defines
+    for (size_t d = 0;
+            d < lfs_max(suite->define_count,
+                BENCH_IMPLICIT_DEFINE_COUNT);
+            d++) {
+        if (d < BENCH_IMPLICIT_DEFINE_COUNT
+                || bench_define_ispermutation(d)) {
+            list_defines_add(defines, d);
+        }
+    }
+}
+
+void perm_list_permutation_defines(
+        void *data,
+        const struct bench_suite *suite,
+        const struct bench_case *case_) {
+    struct list_defines_defines *defines = data;
+    (void)suite;
+    (void)case_;
+
+    // collect permutation_defines
+    for (size_t d = 0;
+            d < lfs_max(suite->define_count,
+                BENCH_IMPLICIT_DEFINE_COUNT);
+            d++) {
+        if (bench_define_ispermutation(d)) {
+            list_defines_add(defines, d);
+        }
+    }
+}
+
+extern const bench_geometry_t builtin_geometries[];
+
+static void list_defines(void) {
+    struct list_defines_defines defines = {NULL, 0, 0};
+
+    // add defines
+    for (size_t t = 0; t < bench_id_count; t++) {
+        for (size_t i = 0; i < BENCH_SUITE_COUNT; i++) {
+            bench_define_suite(&bench_suites[i]);
+
+            for (size_t j = 0; j < bench_suites[i].case_count; j++) {
+                // does neither suite nor case name match?
+                if (bench_ids[t].name && !(
+                        strcmp(bench_ids[t].name,
+                            bench_suites[i].name) == 0
+                        || strcmp(bench_ids[t].name,
+                            bench_suites[i].cases[j].name) == 0)) {
+                    continue;
+                }
+
+                case_forperm(
+                        &bench_suites[i],
+                        &bench_suites[i].cases[j],
+                        bench_ids[t].defines,
+                        bench_ids[t].define_count,
+                        perm_list_defines,
+                        &defines);
+            }
+        }
+    }
+
+    for (size_t i = 0; i < defines.define_count; i++) {
+        printf("%s=", defines.defines[i].name);
+        for (size_t j = 0; j < defines.defines[i].value_count; j++) {
+            printf("%jd", defines.defines[i].values[j]);
+            if (j != defines.defines[i].value_count-1) {
+                printf(",");
+            }
+        }
+        printf("\n");
+    }
+
+    for (size_t i = 0; i < defines.define_count; i++) {
+        free(defines.defines[i].values);
+    }
+    free(defines.defines);
+}
+
+static void list_permutation_defines(void) {
+    struct list_defines_defines defines = {NULL, 0, 0};
+
+    // add permutation defines
+    for (size_t t = 0; t < bench_id_count; t++) {
+        for (size_t i = 0; i < BENCH_SUITE_COUNT; i++) {
+            bench_define_suite(&bench_suites[i]);
+
+            for (size_t j = 0; j < bench_suites[i].case_count; j++) {
+                // does neither suite nor case name match?
+                if (bench_ids[t].name && !(
+                        strcmp(bench_ids[t].name,
+                            bench_suites[i].name) == 0
+                        || strcmp(bench_ids[t].name,
+                            bench_suites[i].cases[j].name) == 0)) {
+                    continue;
+                }
+
+                case_forperm(
+                        &bench_suites[i],
+                        &bench_suites[i].cases[j],
+                        bench_ids[t].defines,
+                        bench_ids[t].define_count,
+                        perm_list_permutation_defines,
+                        &defines);
+            }
+        }
+    }
+
+    for (size_t i = 0; i < defines.define_count; i++) {
+        printf("%s=", defines.defines[i].name);
+        for (size_t j = 0; j < defines.defines[i].value_count; j++) {
+            printf("%jd", defines.defines[i].values[j]);
+            if (j != defines.defines[i].value_count-1) {
+                printf(",");
+            }
+        }
+        printf("\n");
+    }
+
+    for (size_t i = 0; i < defines.define_count; i++) {
+        free(defines.defines[i].values);
+    }
+    free(defines.defines);
+}
+
+static void list_implicit_defines(void) {
+    struct list_defines_defines defines = {NULL, 0, 0};
+
+    // yes we do need to define a suite, this does a bit of bookeeping
+    // such as setting up the define cache
+    bench_define_suite(&(const struct bench_suite){0});
+
+    // make sure to include builtin geometries here
+    extern const bench_geometry_t builtin_geometries[];
+    for (size_t g = 0; builtin_geometries[g].name; g++) {
+        bench_define_geometry(&builtin_geometries[g]);
+        bench_define_flush();
+
+        // add implicit defines
+        for (size_t d = 0; d < BENCH_IMPLICIT_DEFINE_COUNT; d++) {
+            list_defines_add(&defines, d);
+        }
+    }
+
+    for (size_t i = 0; i < defines.define_count; i++) {
+        printf("%s=", defines.defines[i].name);
+        for (size_t j = 0; j < defines.defines[i].value_count; j++) {
+            printf("%jd", defines.defines[i].values[j]);
+            if (j != defines.defines[i].value_count-1) {
+                printf(",");
+            }
+        }
+        printf("\n");
+    }
+
+    for (size_t i = 0; i < defines.define_count; i++) {
+        free(defines.defines[i].values);
+    }
+    free(defines.defines);
+}
+
+
+
+// geometries to bench
+
+const bench_geometry_t builtin_geometries[] = {
+    {"default", {{0}, BENCH_CONST(16),   BENCH_CONST(512),   {0}}},
+    {"eeprom",  {{0}, BENCH_CONST(1),    BENCH_CONST(512),   {0}}},
+    {"emmc",    {{0}, {0},               BENCH_CONST(512),   {0}}},
+    {"nor",     {{0}, BENCH_CONST(1),    BENCH_CONST(4096),  {0}}},
+    {"nand",    {{0}, BENCH_CONST(4096), BENCH_CONST(32768), {0}}},
+    {NULL, {{0}, {0}, {0}, {0}}},
+};
+
+const bench_geometry_t *bench_geometries = builtin_geometries;
+size_t bench_geometry_count = 5;
+
+static void list_geometries(void) {
+    // at least size so that names fit
+    unsigned name_width = 23;
+    for (size_t g = 0; builtin_geometries[g].name; g++) {
+        size_t len = strlen(builtin_geometries[g].name);
+        if (len > name_width) {
+            name_width = len;
+        }
+    }
+    name_width = 4*((name_width+1+4-1)/4)-1;
+
+    // yes we do need to define a suite, this does a bit of bookeeping
+    // such as setting up the define cache
+    bench_define_suite(&(const struct bench_suite){0});
+
+    printf("%-*s  %7s %7s %7s %7s %11s\n",
+            name_width, "geometry", "read", "prog", "erase", "count", "size");
+    for (size_t g = 0; builtin_geometries[g].name; g++) {
+        bench_define_geometry(&builtin_geometries[g]);
+        bench_define_flush();
+        printf("%-*s  %7ju %7ju %7ju %7ju %11ju\n",
+                name_width,
+                builtin_geometries[g].name,
+                READ_SIZE,
+                PROG_SIZE,
+                ERASE_SIZE,
+                ERASE_COUNT,
+                ERASE_SIZE*ERASE_COUNT);
+    }
+}
+
+
+
+// global bench step count
+size_t bench_step = 0;
+
+void perm_run(
+        void *data,
+        const struct bench_suite *suite,
+        const struct bench_case *case_) {
+    (void)data;
+
+    // skip this step?
+    if (!(bench_step >= bench_step_start
+            && bench_step < bench_step_stop
+            && (bench_step-bench_step_start) % bench_step_step == 0)) {
+        bench_step += 1;
+        return;
+    }
+    bench_step += 1;
+
+    // filter?
+    if (case_->filter && !case_->filter()) {
+        printf("skipped ");
+        perm_printid(suite, case_);
+        printf("\n");
+        return;
+    }
+
+    // create block device and configuration
+    lfs_emubd_t bd;
+
+    struct lfs_config cfg = {
+        .context            = &bd,
+        .read               = lfs_emubd_read,
+        .prog               = lfs_emubd_prog,
+        .erase              = lfs_emubd_erase,
+        .sync               = lfs_emubd_sync,
+        .read_size          = READ_SIZE,
+        .prog_size          = PROG_SIZE,
+        .block_size         = BLOCK_SIZE,
+        .block_count        = BLOCK_COUNT,
+        .block_cycles       = BLOCK_CYCLES,
+        .cache_size         = CACHE_SIZE,
+        .lookahead_size     = LOOKAHEAD_SIZE,
+    };
+
+    struct lfs_emubd_config bdcfg = {
+        .read_size          = READ_SIZE,
+        .prog_size          = PROG_SIZE,
+        .erase_size         = ERASE_SIZE,
+        .erase_count        = ERASE_COUNT,
+        .erase_value        = ERASE_VALUE,
+        .erase_cycles       = ERASE_CYCLES,
+        .badblock_behavior  = BADBLOCK_BEHAVIOR,
+        .disk_path          = bench_disk_path,
+        .read_sleep         = bench_read_sleep,
+        .prog_sleep         = bench_prog_sleep,
+        .erase_sleep        = bench_erase_sleep,
+    };
+
+    int err = lfs_emubd_create(&cfg, &bdcfg);
+    if (err) {
+        fprintf(stderr, "error: could not create block device: %d\n", err);
+        exit(-1);
+    }
+
+    // run the bench
+    bench_cfg = &cfg;
+    bench_reset();
+    printf("running ");
+    perm_printid(suite, case_);
+    printf("\n");
+
+    case_->run(&cfg);
+
+    printf("finished ");
+    perm_printid(suite, case_);
+    printf(" %"PRIu64" %"PRIu64" %"PRIu64,
+        bench_readed,
+        bench_proged,
+        bench_erased);
+    printf("\n");
+
+    // cleanup
+    err = lfs_emubd_destroy(&cfg);
+    if (err) {
+        fprintf(stderr, "error: could not destroy block device: %d\n", err);
+        exit(-1);
+    }
+}
+
+static void run(void) {
+    // ignore disconnected pipes
+    signal(SIGPIPE, SIG_IGN);
+
+    for (size_t t = 0; t < bench_id_count; t++) {
+        for (size_t i = 0; i < BENCH_SUITE_COUNT; i++) {
+            bench_define_suite(&bench_suites[i]);
+
+            for (size_t j = 0; j < bench_suites[i].case_count; j++) {
+                // does neither suite nor case name match?
+                if (bench_ids[t].name && !(
+                        strcmp(bench_ids[t].name,
+                            bench_suites[i].name) == 0
+                        || strcmp(bench_ids[t].name,
+                            bench_suites[i].cases[j].name) == 0)) {
+                    continue;
+                }
+
+                case_forperm(
+                        &bench_suites[i],
+                        &bench_suites[i].cases[j],
+                        bench_ids[t].defines,
+                        bench_ids[t].define_count,
+                        perm_run,
+                        NULL);
+            }
+        }
+    }
+}
+
+
+
+// option handling
+enum opt_flags {
+    OPT_HELP                     = 'h',
+    OPT_SUMMARY                  = 'Y',
+    OPT_LIST_SUITES              = 'l',
+    OPT_LIST_CASES               = 'L',
+    OPT_LIST_SUITE_PATHS         = 1,
+    OPT_LIST_CASE_PATHS          = 2,
+    OPT_LIST_DEFINES             = 3,
+    OPT_LIST_PERMUTATION_DEFINES = 4,
+    OPT_LIST_IMPLICIT_DEFINES    = 5,
+    OPT_LIST_GEOMETRIES          = 6,
+    OPT_DEFINE                   = 'D',
+    OPT_GEOMETRY                 = 'G',
+    OPT_STEP                     = 's',
+    OPT_DISK                     = 'd',
+    OPT_TRACE                    = 't',
+    OPT_TRACE_BACKTRACE          = 7,
+    OPT_TRACE_PERIOD             = 8,
+    OPT_TRACE_FREQ               = 9,
+    OPT_READ_SLEEP               = 10,
+    OPT_PROG_SLEEP               = 11,
+    OPT_ERASE_SLEEP              = 12,
+};
+
+const char *short_opts = "hYlLD:G:s:d:t:";
+
+const struct option long_opts[] = {
+    {"help",             no_argument,       NULL, OPT_HELP},
+    {"summary",          no_argument,       NULL, OPT_SUMMARY},
+    {"list-suites",      no_argument,       NULL, OPT_LIST_SUITES},
+    {"list-cases",       no_argument,       NULL, OPT_LIST_CASES},
+    {"list-suite-paths", no_argument,       NULL, OPT_LIST_SUITE_PATHS},
+    {"list-case-paths",  no_argument,       NULL, OPT_LIST_CASE_PATHS},
+    {"list-defines",     no_argument,       NULL, OPT_LIST_DEFINES},
+    {"list-permutation-defines",
+                         no_argument,       NULL, OPT_LIST_PERMUTATION_DEFINES},
+    {"list-implicit-defines",
+                         no_argument,       NULL, OPT_LIST_IMPLICIT_DEFINES},
+    {"list-geometries",  no_argument,       NULL, OPT_LIST_GEOMETRIES},
+    {"define",           required_argument, NULL, OPT_DEFINE},
+    {"geometry",         required_argument, NULL, OPT_GEOMETRY},
+    {"step",             required_argument, NULL, OPT_STEP},
+    {"disk",             required_argument, NULL, OPT_DISK},
+    {"trace",            required_argument, NULL, OPT_TRACE},
+    {"trace-backtrace",  no_argument,       NULL, OPT_TRACE_BACKTRACE},
+    {"trace-period",     required_argument, NULL, OPT_TRACE_PERIOD},
+    {"trace-freq",       required_argument, NULL, OPT_TRACE_FREQ},
+    {"read-sleep",       required_argument, NULL, OPT_READ_SLEEP},
+    {"prog-sleep",       required_argument, NULL, OPT_PROG_SLEEP},
+    {"erase-sleep",      required_argument, NULL, OPT_ERASE_SLEEP},
+    {NULL, 0, NULL, 0},
+};
+
+const char *const help_text[] = {
+    "Show this help message.",
+    "Show quick summary.",
+    "List bench suites.",
+    "List bench cases.",
+    "List the path for each bench suite.",
+    "List the path and line number for each bench case.",
+    "List all defines in this bench-runner.",
+    "List explicit defines in this bench-runner.",
+    "List implicit defines in this bench-runner.",
+    "List the available disk geometries.",
+    "Override a bench define.",
+    "Comma-separated list of disk geometries to bench.",
+    "Comma-separated range of bench permutations to run (start,stop,step).",
+    "Direct block device operations to this file.",
+    "Direct trace output to this file.",
+    "Include a backtrace with every trace statement.",
+    "Sample trace output at this period in cycles.",
+    "Sample trace output at this frequency in hz.",
+    "Artificial read delay in seconds.",
+    "Artificial prog delay in seconds.",
+    "Artificial erase delay in seconds.",
+};
+
+int main(int argc, char **argv) {
+    void (*op)(void) = run;
+
+    size_t bench_override_capacity = 0;
+    size_t bench_geometry_capacity = 0;
+    size_t bench_id_capacity = 0;
+
+    // parse options
+    while (true) {
+        int c = getopt_long(argc, argv, short_opts, long_opts, NULL);
+        switch (c) {
+            // generate help message
+            case OPT_HELP: {
+                printf("usage: %s [options] [bench_id]\n", argv[0]);
+                printf("\n");
+
+                printf("options:\n");
+                size_t i = 0;
+                while (long_opts[i].name) {
+                    size_t indent;
+                    if (long_opts[i].has_arg == no_argument) {
+                        if (long_opts[i].val >= '0' && long_opts[i].val < 'z') {
+                            indent = printf("  -%c, --%s ",
+                                    long_opts[i].val,
+                                    long_opts[i].name);
+                        } else {
+                            indent = printf("  --%s ",
+                                    long_opts[i].name);
+                        }
+                    } else {
+                        if (long_opts[i].val >= '0' && long_opts[i].val < 'z') {
+                            indent = printf("  -%c %s, --%s %s ",
+                                    long_opts[i].val,
+                                    long_opts[i].name,
+                                    long_opts[i].name,
+                                    long_opts[i].name);
+                        } else {
+                            indent = printf("  --%s %s ",
+                                    long_opts[i].name,
+                                    long_opts[i].name);
+                        }
+                    }
+
+                    // a quick, hacky, byte-level method for text wrapping
+                    size_t len = strlen(help_text[i]);
+                    size_t j = 0;
+                    if (indent < 24) {
+                        printf("%*s %.80s\n",
+                                (int)(24-1-indent),
+                                "",
+                                &help_text[i][j]);
+                        j += 80;
+                    } else {
+                        printf("\n");
+                    }
+
+                    while (j < len) {
+                        printf("%24s%.80s\n", "", &help_text[i][j]);
+                        j += 80;
+                    }
+
+                    i += 1;
+                }
+
+                printf("\n");
+                exit(0);
+            }
+            // summary/list flags
+            case OPT_SUMMARY:
+                op = summary;
+                break;
+            case OPT_LIST_SUITES:
+                op = list_suites;
+                break;
+            case OPT_LIST_CASES:
+                op = list_cases;
+                break;
+            case OPT_LIST_SUITE_PATHS:
+                op = list_suite_paths;
+                break;
+            case OPT_LIST_CASE_PATHS:
+                op = list_case_paths;
+                break;
+            case OPT_LIST_DEFINES:
+                op = list_defines;
+                break;
+            case OPT_LIST_PERMUTATION_DEFINES:
+                op = list_permutation_defines;
+                break;
+            case OPT_LIST_IMPLICIT_DEFINES:
+                op = list_implicit_defines;
+                break;
+            case OPT_LIST_GEOMETRIES:
+                op = list_geometries;
+                break;
+            // configuration
+            case OPT_DEFINE: {
+                // allocate space
+                bench_override_t *override = mappend(
+                        (void**)&bench_overrides,
+                        sizeof(bench_override_t),
+                        &bench_override_count,
+                        &bench_override_capacity);
+
+                // parse into string key/intmax_t value, cannibalizing the
+                // arg in the process
+                char *sep = strchr(optarg, '=');
+                char *parsed = NULL;
+                if (!sep) {
+                    goto invalid_define;
+                }
+                *sep = '\0';
+                override->name = optarg;
+                optarg = sep+1;
+
+                // parse comma-separated permutations
+                {
+                    override->defines = NULL;
+                    override->permutations = 0;
+                    size_t override_capacity = 0;
+                    while (true) {
+                        optarg += strspn(optarg, " ");
+
+                        if (strncmp(optarg, "range", strlen("range")) == 0) {
+                            // range of values
+                            optarg += strlen("range");
+                            optarg += strspn(optarg, " ");
+                            if (*optarg != '(') {
+                                goto invalid_define;
+                            }
+                            optarg += 1;
+
+                            intmax_t start = strtoumax(optarg, &parsed, 0);
+                            intmax_t stop = -1;
+                            intmax_t step = 1;
+                            // allow empty string for start=0
+                            if (parsed == optarg) {
+                                start = 0;
+                            }
+                            optarg = parsed + strspn(parsed, " ");
+
+                            if (*optarg != ',' && *optarg != ')') {
+                                goto invalid_define;
+                            }
+
+                            if (*optarg == ',') {
+                                optarg += 1;
+                                stop = strtoumax(optarg, &parsed, 0);
+                                // allow empty string for stop=end
+                                if (parsed == optarg) {
+                                    stop = -1;
+                                }
+                                optarg = parsed + strspn(parsed, " ");
+
+                                if (*optarg != ',' && *optarg != ')') {
+                                    goto invalid_define;
+                                }
+
+                                if (*optarg == ',') {
+                                    optarg += 1;
+                                    step = strtoumax(optarg, &parsed, 0);
+                                    // allow empty string for stop=1
+                                    if (parsed == optarg) {
+                                        step = 1;
+                                    }
+                                    optarg = parsed + strspn(parsed, " ");
+
+                                    if (*optarg != ')') {
+                                        goto invalid_define;
+                                    }
+                                }
+                            } else {
+                                // single value = stop only
+                                stop = start;
+                                start = 0;
+                            }
+
+                            if (*optarg != ')') {
+                                goto invalid_define;
+                            }
+                            optarg += 1;
+
+                            // calculate the range of values
+                            assert(step != 0);
+                            for (intmax_t i = start;
+                                    (step < 0)
+                                        ? i > stop
+                                        : (uintmax_t)i < (uintmax_t)stop;
+                                    i += step) {
+                                *(intmax_t*)mappend(
+                                        (void**)&override->defines,
+                                        sizeof(intmax_t),
+                                        &override->permutations,
+                                        &override_capacity) = i;
+                            }
+                        } else if (*optarg != '\0') {
+                            // single value
+                            intmax_t define = strtoimax(optarg, &parsed, 0);
+                            if (parsed == optarg) {
+                                goto invalid_define;
+                            }
+                            optarg = parsed + strspn(parsed, " ");
+                            *(intmax_t*)mappend(
+                                    (void**)&override->defines,
+                                    sizeof(intmax_t),
+                                    &override->permutations,
+                                    &override_capacity) = define;
+                        } else {
+                            break;
+                        }
+
+                        if (*optarg == ',') {
+                            optarg += 1;
+                        }
+                    }
+                }
+                assert(override->permutations > 0);
+                break;
+
+invalid_define:
+                fprintf(stderr, "error: invalid define: %s\n", optarg);
+                exit(-1);
+            }
+            case OPT_GEOMETRY: {
+                // reset our geometry scenarios
+                if (bench_geometry_capacity > 0) {
+                    free((bench_geometry_t*)bench_geometries);
+                }
+                bench_geometries = NULL;
+                bench_geometry_count = 0;
+                bench_geometry_capacity = 0;
+
+                // parse the comma separated list of disk geometries
+                while (*optarg) {
+                    // allocate space
+                    bench_geometry_t *geometry = mappend(
+                            (void**)&bench_geometries,
+                            sizeof(bench_geometry_t),
+                            &bench_geometry_count,
+                            &bench_geometry_capacity);
+
+                    // parse the disk geometry
+                    optarg += strspn(optarg, " ");
+
+                    // named disk geometry
+                    size_t len = strcspn(optarg, " ,");
+                    for (size_t i = 0; builtin_geometries[i].name; i++) {
+                        if (len == strlen(builtin_geometries[i].name)
+                                && memcmp(optarg,
+                                    builtin_geometries[i].name,
+                                    len) == 0) {
+                            *geometry = builtin_geometries[i];
+                            optarg += len;
+                            goto geometry_next;
+                        }
+                    }
+
+                    // comma-separated read/prog/erase/count
+                    if (*optarg == '{') {
+                        lfs_size_t sizes[4];
+                        size_t count = 0;
+
+                        char *s = optarg + 1;
+                        while (count < 4) {
+                            char *parsed = NULL;
+                            sizes[count] = strtoumax(s, &parsed, 0);
+                            count += 1;
+
+                            s = parsed + strspn(parsed, " ");
+                            if (*s == ',') {
+                                s += 1;
+                                continue;
+                            } else if (*s == '}') {
+                                s += 1;
+                                break;
+                            } else {
+                                goto geometry_unknown;
+                            }
+                        }
+
+                        // allow implicit r=p and p=e for common geometries
+                        memset(geometry, 0, sizeof(bench_geometry_t));
+                        if (count >= 3) {
+                            geometry->defines[READ_SIZE_i]
+                                    = BENCH_LIT(sizes[0]);
+                            geometry->defines[PROG_SIZE_i]
+                                    = BENCH_LIT(sizes[1]);
+                            geometry->defines[ERASE_SIZE_i]
+                                    = BENCH_LIT(sizes[2]);
+                        } else if (count >= 2) {
+                            geometry->defines[PROG_SIZE_i]
+                                    = BENCH_LIT(sizes[0]);
+                            geometry->defines[ERASE_SIZE_i]
+                                    = BENCH_LIT(sizes[1]);
+                        } else {
+                            geometry->defines[ERASE_SIZE_i]
+                                    = BENCH_LIT(sizes[0]);
+                        }
+                        if (count >= 4) {
+                            geometry->defines[ERASE_COUNT_i]
+                                    = BENCH_LIT(sizes[3]);
+                        }
+                        optarg = s;
+                        goto geometry_next;
+                    }
+
+                    // leb16-encoded read/prog/erase/count
+                    if (*optarg == ':') {
+                        lfs_size_t sizes[4];
+                        size_t count = 0;
+
+                        char *s = optarg + 1;
+                        while (true) {
+                            char *parsed = NULL;
+                            uintmax_t x = leb16_parse(s, &parsed);
+                            if (parsed == s || count >= 4) {
+                                break;
+                            }
+
+                            sizes[count] = x;
+                            count += 1;
+                            s = parsed;
+                        }
+
+                        // allow implicit r=p and p=e for common geometries
+                        memset(geometry, 0, sizeof(bench_geometry_t));
+                        if (count >= 3) {
+                            geometry->defines[READ_SIZE_i]
+                                    = BENCH_LIT(sizes[0]);
+                            geometry->defines[PROG_SIZE_i]
+                                    = BENCH_LIT(sizes[1]);
+                            geometry->defines[ERASE_SIZE_i]
+                                    = BENCH_LIT(sizes[2]);
+                        } else if (count >= 2) {
+                            geometry->defines[PROG_SIZE_i]
+                                    = BENCH_LIT(sizes[0]);
+                            geometry->defines[ERASE_SIZE_i]
+                                    = BENCH_LIT(sizes[1]);
+                        } else {
+                            geometry->defines[ERASE_SIZE_i]
+                                    = BENCH_LIT(sizes[0]);
+                        }
+                        if (count >= 4) {
+                            geometry->defines[ERASE_COUNT_i]
+                                    = BENCH_LIT(sizes[3]);
+                        }
+                        optarg = s;
+                        goto geometry_next;
+                    }
+
+geometry_unknown:
+                    // unknown scenario?
+                    fprintf(stderr, "error: unknown disk geometry: %s\n",
+                            optarg);
+                    exit(-1);
+
+geometry_next:
+                    optarg += strspn(optarg, " ");
+                    if (*optarg == ',') {
+                        optarg += 1;
+                    } else if (*optarg == '\0') {
+                        break;
+                    } else {
+                        goto geometry_unknown;
+                    }
+                }
+                break;
+            }
+            case OPT_STEP: {
+                char *parsed = NULL;
+                bench_step_start = strtoumax(optarg, &parsed, 0);
+                bench_step_stop = -1;
+                bench_step_step = 1;
+                // allow empty string for start=0
+                if (parsed == optarg) {
+                    bench_step_start = 0;
+                }
+                optarg = parsed + strspn(parsed, " ");
+
+                if (*optarg != ',' && *optarg != '\0') {
+                    goto step_unknown;
+                }
+
+                if (*optarg == ',') {
+                    optarg += 1;
+                    bench_step_stop = strtoumax(optarg, &parsed, 0);
+                    // allow empty string for stop=end
+                    if (parsed == optarg) {
+                        bench_step_stop = -1;
+                    }
+                    optarg = parsed + strspn(parsed, " ");
+
+                    if (*optarg != ',' && *optarg != '\0') {
+                        goto step_unknown;
+                    }
+
+                    if (*optarg == ',') {
+                        optarg += 1;
+                        bench_step_step = strtoumax(optarg, &parsed, 0);
+                        // allow empty string for stop=1
+                        if (parsed == optarg) {
+                            bench_step_step = 1;
+                        }
+                        optarg = parsed + strspn(parsed, " ");
+
+                        if (*optarg != '\0') {
+                            goto step_unknown;
+                        }
+                    }
+                } else {
+                    // single value = stop only
+                    bench_step_stop = bench_step_start;
+                    bench_step_start = 0;
+                }
+
+                break;
+step_unknown:
+                fprintf(stderr, "error: invalid step: %s\n", optarg);
+                exit(-1);
+            }
+            case OPT_DISK:
+                bench_disk_path = optarg;
+                break;
+            case OPT_TRACE:
+                bench_trace_path = optarg;
+                break;
+            case OPT_TRACE_BACKTRACE:
+                bench_trace_backtrace = true;
+                break;
+            case OPT_TRACE_PERIOD: {
+                char *parsed = NULL;
+                bench_trace_period = strtoumax(optarg, &parsed, 0);
+                if (parsed == optarg) {
+                    fprintf(stderr, "error: invalid trace-period: %s\n", optarg);
+                    exit(-1);
+                }
+                break;
+            }
+            case OPT_TRACE_FREQ: {
+                char *parsed = NULL;
+                bench_trace_freq = strtoumax(optarg, &parsed, 0);
+                if (parsed == optarg) {
+                    fprintf(stderr, "error: invalid trace-freq: %s\n", optarg);
+                    exit(-1);
+                }
+                break;
+            }
+            case OPT_READ_SLEEP: {
+                char *parsed = NULL;
+                double read_sleep = strtod(optarg, &parsed);
+                if (parsed == optarg) {
+                    fprintf(stderr, "error: invalid read-sleep: %s\n", optarg);
+                    exit(-1);
+                }
+                bench_read_sleep = read_sleep*1.0e9;
+                break;
+            }
+            case OPT_PROG_SLEEP: {
+                char *parsed = NULL;
+                double prog_sleep = strtod(optarg, &parsed);
+                if (parsed == optarg) {
+                    fprintf(stderr, "error: invalid prog-sleep: %s\n", optarg);
+                    exit(-1);
+                }
+                bench_prog_sleep = prog_sleep*1.0e9;
+                break;
+            }
+            case OPT_ERASE_SLEEP: {
+                char *parsed = NULL;
+                double erase_sleep = strtod(optarg, &parsed);
+                if (parsed == optarg) {
+                    fprintf(stderr, "error: invalid erase-sleep: %s\n", optarg);
+                    exit(-1);
+                }
+                bench_erase_sleep = erase_sleep*1.0e9;
+                break;
+            }
+            // done parsing
+            case -1:
+                goto getopt_done;
+            // unknown arg, getopt prints a message for us
+            default:
+                exit(-1);
+        }
+    }
+getopt_done: ;
+
+    if (argc > optind) {
+        // reset our bench identifier list
+        bench_ids = NULL;
+        bench_id_count = 0;
+        bench_id_capacity = 0;
+    }
+
+    // parse bench identifier, if any, cannibalizing the arg in the process
+    for (; argc > optind; optind++) {
+        bench_define_t *defines = NULL;
+        size_t define_count = 0;
+
+        // parse name, can be suite or case
+        char *name = argv[optind];
+        char *defines_ = strchr(name, ':');
+        if (defines_) {
+            *defines_ = '\0';
+            defines_ += 1;
+        }
+
+        // remove optional path and .toml suffix
+        char *slash = strrchr(name, '/');
+        if (slash) {
+            name = slash+1;
+        }
+
+        size_t name_len = strlen(name);
+        if (name_len > 5 && strcmp(&name[name_len-5], ".toml") == 0) {
+            name[name_len-5] = '\0';
+        }
+
+        if (defines_) {
+            // parse defines
+            while (true) {
+                char *parsed;
+                size_t d = leb16_parse(defines_, &parsed);
+                intmax_t v = leb16_parse(parsed, &parsed);
+                if (parsed == defines_) {
+                    break;
+                }
+                defines_ = parsed;
+
+                if (d >= define_count) {
+                    // align to power of two to avoid any superlinear growth
+                    size_t ncount = 1 << lfs_npw2(d+1);
+                    defines = realloc(defines,
+                            ncount*sizeof(bench_define_t));
+                    memset(defines+define_count, 0,
+                            (ncount-define_count)*sizeof(bench_define_t));
+                    define_count = ncount;
+                }
+                defines[d] = BENCH_LIT(v);
+            }
+        }
+
+        // append to identifier list
+        *(bench_id_t*)mappend(
+                (void**)&bench_ids,
+                sizeof(bench_id_t),
+                &bench_id_count,
+                &bench_id_capacity) = (bench_id_t){
+            .name = name,
+            .defines = defines,
+            .define_count = define_count,
+        };
+    }
+
+    // do the thing
+    op();
+
+    // cleanup (need to be done for valgrind benching)
+    bench_define_cleanup();
+    if (bench_overrides) {
+        for (size_t i = 0; i < bench_override_count; i++) {
+            free((void*)bench_overrides[i].defines);
+        }
+        free((void*)bench_overrides);
+    }
+    if (bench_geometry_capacity) {
+        free((void*)bench_geometries);
+    }
+    if (bench_id_capacity) {
+        for (size_t i = 0; i < bench_id_count; i++) {
+            free((void*)bench_ids[i].defines);
+        }
+        free((void*)bench_ids);
+    }
+}
diff --git a/runners/bench_runner.h b/runners/bench_runner.h
new file mode 100644
index 0000000..b072970
--- /dev/null
+++ b/runners/bench_runner.h
@@ -0,0 +1,137 @@
+/*
+ * Runner for littlefs benchmarks
+ *
+ * Copyright (c) 2022, The littlefs authors.
+ * SPDX-License-Identifier: BSD-3-Clause
+ */
+#ifndef BENCH_RUNNER_H
+#define BENCH_RUNNER_H
+
+
+// override LFS_TRACE
+void bench_trace(const char *fmt, ...);
+
+#define LFS_TRACE_(fmt, ...) \
+    bench_trace("%s:%d:trace: " fmt "%s\n", \
+        __FILE__, \
+        __LINE__, \
+        __VA_ARGS__)
+#define LFS_TRACE(...) LFS_TRACE_(__VA_ARGS__, "")
+#define LFS_EMUBD_TRACE(...) LFS_TRACE_(__VA_ARGS__, "")
+
+// provide BENCH_START/BENCH_STOP macros
+void bench_start(void);
+void bench_stop(void);
+
+#define BENCH_START() bench_start()
+#define BENCH_STOP() bench_stop()
+
+
+// note these are indirectly included in any generated files
+#include "bd/lfs_emubd.h"
+#include <stdio.h>
+
+// give source a chance to define feature macros
+#undef _FEATURES_H
+#undef _STDIO_H
+
+
+// generated bench configurations
+struct lfs_config;
+
+enum bench_flags {
+    BENCH_REENTRANT = 0x1,
+};
+typedef uint8_t bench_flags_t;
+
+typedef struct bench_define {
+    intmax_t (*cb)(void *data);
+    void *data;
+} bench_define_t;
+
+struct bench_case {
+    const char *name;
+    const char *path;
+    bench_flags_t flags;
+    size_t permutations;
+
+    const bench_define_t *defines;
+
+    bool (*filter)(void);
+    void (*run)(struct lfs_config *cfg);
+};
+
+struct bench_suite {
+    const char *name;
+    const char *path;
+    bench_flags_t flags;
+
+    const char *const *define_names;
+    size_t define_count;
+
+    const struct bench_case *cases;
+    size_t case_count;
+};
+
+
+// deterministic prng for pseudo-randomness in benches
+uint32_t bench_prng(uint32_t *state);
+
+#define BENCH_PRNG(state) bench_prng(state)
+
+
+// access generated bench defines
+intmax_t bench_define(size_t define);
+
+#define BENCH_DEFINE(i) bench_define(i)
+
+// a few preconfigured defines that control how benches run
+ 
+#define READ_SIZE_i          0
+#define PROG_SIZE_i          1
+#define ERASE_SIZE_i         2
+#define ERASE_COUNT_i        3
+#define BLOCK_SIZE_i         4
+#define BLOCK_COUNT_i        5
+#define CACHE_SIZE_i         6
+#define LOOKAHEAD_SIZE_i     7
+#define BLOCK_CYCLES_i       8
+#define ERASE_VALUE_i        9
+#define ERASE_CYCLES_i       10
+#define BADBLOCK_BEHAVIOR_i  11
+#define POWERLOSS_BEHAVIOR_i 12
+
+#define READ_SIZE           bench_define(READ_SIZE_i)
+#define PROG_SIZE           bench_define(PROG_SIZE_i)
+#define ERASE_SIZE          bench_define(ERASE_SIZE_i)
+#define ERASE_COUNT         bench_define(ERASE_COUNT_i)
+#define BLOCK_SIZE          bench_define(BLOCK_SIZE_i)
+#define BLOCK_COUNT         bench_define(BLOCK_COUNT_i)
+#define CACHE_SIZE          bench_define(CACHE_SIZE_i)
+#define LOOKAHEAD_SIZE      bench_define(LOOKAHEAD_SIZE_i)
+#define BLOCK_CYCLES        bench_define(BLOCK_CYCLES_i)
+#define ERASE_VALUE         bench_define(ERASE_VALUE_i)
+#define ERASE_CYCLES        bench_define(ERASE_CYCLES_i)
+#define BADBLOCK_BEHAVIOR   bench_define(BADBLOCK_BEHAVIOR_i)
+#define POWERLOSS_BEHAVIOR  bench_define(POWERLOSS_BEHAVIOR_i)
+
+#define BENCH_IMPLICIT_DEFINES \
+    BENCH_DEF(READ_SIZE,          PROG_SIZE) \
+    BENCH_DEF(PROG_SIZE,          ERASE_SIZE) \
+    BENCH_DEF(ERASE_SIZE,         0) \
+    BENCH_DEF(ERASE_COUNT,        (1024*1024)/BLOCK_SIZE) \
+    BENCH_DEF(BLOCK_SIZE,         ERASE_SIZE) \
+    BENCH_DEF(BLOCK_COUNT,        ERASE_COUNT/lfs_max(BLOCK_SIZE/ERASE_SIZE,1))\
+    BENCH_DEF(CACHE_SIZE,         lfs_max(64,lfs_max(READ_SIZE,PROG_SIZE))) \
+    BENCH_DEF(LOOKAHEAD_SIZE,     16) \
+    BENCH_DEF(BLOCK_CYCLES,       -1) \
+    BENCH_DEF(ERASE_VALUE,        0xff) \
+    BENCH_DEF(ERASE_CYCLES,       0) \
+    BENCH_DEF(BADBLOCK_BEHAVIOR,  LFS_EMUBD_BADBLOCK_PROGERROR) \
+    BENCH_DEF(POWERLOSS_BEHAVIOR, LFS_EMUBD_POWERLOSS_NOOP)
+
+#define BENCH_GEOMETRY_DEFINE_COUNT 4
+#define BENCH_IMPLICIT_DEFINE_COUNT 13
+
+
+#endif
diff --git a/runners/test_runner.c b/runners/test_runner.c
new file mode 100644
index 0000000..13befdc
--- /dev/null
+++ b/runners/test_runner.c
@@ -0,0 +1,2798 @@
+/*
+ * Runner for littlefs tests
+ *
+ * Copyright (c) 2022, The littlefs authors.
+ * SPDX-License-Identifier: BSD-3-Clause
+ */
+#ifndef _POSIX_C_SOURCE
+#define _POSIX_C_SOURCE 199309L
+#endif
+
+#include "runners/test_runner.h"
+#include "bd/lfs_emubd.h"
+
+#include <getopt.h>
+#include <sys/types.h>
+#include <errno.h>
+#include <setjmp.h>
+#include <fcntl.h>
+#include <stdarg.h>
+#include <stdio.h>
+#include <unistd.h>
+#include <time.h>
+#include <execinfo.h>
+
+
+// some helpers
+
+// append to an array with amortized doubling
+void *mappend(void **p,
+        size_t size,
+        size_t *count,
+        size_t *capacity) {
+    uint8_t *p_ = *p;
+    size_t count_ = *count;
+    size_t capacity_ = *capacity;
+
+    count_ += 1;
+    if (count_ > capacity_) {
+        capacity_ = (2*capacity_ < 4) ? 4 : 2*capacity_;
+
+        p_ = realloc(p_, capacity_*size);
+        if (!p_) {
+            return NULL;
+        }
+    }
+
+    *p = p_;
+    *count = count_;
+    *capacity = capacity_;
+    return &p_[(count_-1)*size];
+}
+
+// a quick self-terminating text-safe varint scheme
+static void leb16_print(uintmax_t x) {
+    // allow 'w' to indicate negative numbers
+    if ((intmax_t)x < 0) {
+        printf("w");
+        x = -x;
+    }
+
+    while (true) {
+        char nibble = (x & 0xf) | (x > 0xf ? 0x10 : 0);
+        printf("%c", (nibble < 10) ? '0'+nibble : 'a'+nibble-10);
+        if (x <= 0xf) {
+            break;
+        }
+        x >>= 4;
+    }
+}
+
+static uintmax_t leb16_parse(const char *s, char **tail) {
+    bool neg = false;
+    uintmax_t x = 0;
+    if (tail) {
+        *tail = (char*)s;
+    }
+
+    if (s[0] == 'w') {
+        neg = true;
+        s = s+1;
+    }
+
+    size_t i = 0;
+    while (true) {
+        uintmax_t nibble = s[i];
+        if (nibble >= '0' && nibble <= '9') {
+            nibble = nibble - '0';
+        } else if (nibble >= 'a' && nibble <= 'v') {
+            nibble = nibble - 'a' + 10;
+        } else {
+            // invalid?
+            return 0;
+        }
+
+        x |= (nibble & 0xf) << (4*i);
+        i += 1;
+        if (!(nibble & 0x10)) {
+            s = s + i;
+            break;
+        }
+    }
+
+    if (tail) {
+        *tail = (char*)s;
+    }
+    return neg ? -x : x;
+}
+
+
+
+// test_runner types
+
+typedef struct test_geometry {
+    const char *name;
+    test_define_t defines[TEST_GEOMETRY_DEFINE_COUNT];
+} test_geometry_t;
+
+typedef struct test_powerloss {
+    const char *name;
+    void (*run)(
+            const lfs_emubd_powercycles_t *cycles,
+            size_t cycle_count,
+            const struct test_suite *suite,
+            const struct test_case *case_);
+    const lfs_emubd_powercycles_t *cycles;
+    size_t cycle_count;
+} test_powerloss_t;
+
+typedef struct test_id {
+    const char *name;
+    const test_define_t *defines;
+    size_t define_count;
+    const lfs_emubd_powercycles_t *cycles;
+    size_t cycle_count;
+} test_id_t;
+
+
+// test suites are linked into a custom ld section
+extern struct test_suite __start__test_suites;
+extern struct test_suite __stop__test_suites;
+
+const struct test_suite *test_suites = &__start__test_suites;
+#define TEST_SUITE_COUNT \
+    ((size_t)(&__stop__test_suites - &__start__test_suites))
+
+
+// test define management
+typedef struct test_define_map {
+    const test_define_t *defines;
+    size_t count;
+} test_define_map_t;
+
+typedef struct test_define_names {
+    const char *const *names;
+    size_t count;
+} test_define_names_t;
+
+intmax_t test_define_lit(void *data) {
+    return (intptr_t)data;
+}
+
+#define TEST_CONST(x) {test_define_lit, (void*)(uintptr_t)(x)}
+#define TEST_LIT(x) ((test_define_t)TEST_CONST(x))
+
+
+#define TEST_DEF(k, v) \
+    intmax_t test_define_##k(void *data) { \
+        (void)data; \
+        return v; \
+    }
+
+    TEST_IMPLICIT_DEFINES
+#undef TEST_DEF
+
+#define TEST_DEFINE_MAP_OVERRIDE    0
+#define TEST_DEFINE_MAP_EXPLICIT    1
+#define TEST_DEFINE_MAP_PERMUTATION 2
+#define TEST_DEFINE_MAP_GEOMETRY    3
+#define TEST_DEFINE_MAP_IMPLICIT    4
+#define TEST_DEFINE_MAP_COUNT       5
+
+test_define_map_t test_define_maps[TEST_DEFINE_MAP_COUNT] = {
+    [TEST_DEFINE_MAP_IMPLICIT] = {
+        (const test_define_t[TEST_IMPLICIT_DEFINE_COUNT]) {
+            #define TEST_DEF(k, v) \
+                [k##_i] = {test_define_##k, NULL},
+
+                TEST_IMPLICIT_DEFINES
+            #undef TEST_DEF
+        },
+        TEST_IMPLICIT_DEFINE_COUNT,
+    },
+};
+
+#define TEST_DEFINE_NAMES_SUITE    0
+#define TEST_DEFINE_NAMES_IMPLICIT 1
+#define TEST_DEFINE_NAMES_COUNT    2
+
+test_define_names_t test_define_names[TEST_DEFINE_NAMES_COUNT] = {
+    [TEST_DEFINE_NAMES_IMPLICIT] = {
+        (const char *const[TEST_IMPLICIT_DEFINE_COUNT]){
+            #define TEST_DEF(k, v) \
+                [k##_i] = #k,
+
+                TEST_IMPLICIT_DEFINES
+            #undef TEST_DEF
+        },
+        TEST_IMPLICIT_DEFINE_COUNT,
+    },
+};
+
+intmax_t *test_define_cache;
+size_t test_define_cache_count;
+unsigned *test_define_cache_mask;
+
+const char *test_define_name(size_t define) {
+    // lookup in our test names
+    for (size_t i = 0; i < TEST_DEFINE_NAMES_COUNT; i++) {
+        if (define < test_define_names[i].count
+                && test_define_names[i].names
+                && test_define_names[i].names[define]) {
+            return test_define_names[i].names[define];
+        }
+    }
+
+    return NULL;
+}
+
+bool test_define_ispermutation(size_t define) {
+    // is this define specific to the permutation?
+    for (size_t i = 0; i < TEST_DEFINE_MAP_IMPLICIT; i++) {
+        if (define < test_define_maps[i].count
+                && test_define_maps[i].defines[define].cb) {
+            return true;
+        }
+    }
+
+    return false;
+}
+
+intmax_t test_define(size_t define) {
+    // is the define in our cache?
+    if (define < test_define_cache_count
+            && (test_define_cache_mask[define/(8*sizeof(unsigned))]
+                & (1 << (define%(8*sizeof(unsigned)))))) {
+        return test_define_cache[define];
+    }
+
+    // lookup in our test defines
+    for (size_t i = 0; i < TEST_DEFINE_MAP_COUNT; i++) {
+        if (define < test_define_maps[i].count
+                && test_define_maps[i].defines[define].cb) {
+            intmax_t v = test_define_maps[i].defines[define].cb(
+                    test_define_maps[i].defines[define].data);
+
+            // insert into cache!
+            test_define_cache[define] = v;
+            test_define_cache_mask[define / (8*sizeof(unsigned))]
+                    |= 1 << (define%(8*sizeof(unsigned)));
+
+            return v;
+        }
+    }
+
+    return 0;
+
+    // not found?
+    const char *name = test_define_name(define);
+    fprintf(stderr, "error: undefined define %s (%zd)\n",
+            name ? name : "(unknown)",
+            define);
+    assert(false);
+    exit(-1);
+}
+
+void test_define_flush(void) {
+    // clear cache between permutations
+    memset(test_define_cache_mask, 0,
+            sizeof(unsigned)*(
+                (test_define_cache_count+(8*sizeof(unsigned))-1)
+                / (8*sizeof(unsigned))));
+}
+
+// geometry updates
+const test_geometry_t *test_geometry = NULL;
+
+void test_define_geometry(const test_geometry_t *geometry) {
+    test_define_maps[TEST_DEFINE_MAP_GEOMETRY] = (test_define_map_t){
+            geometry->defines, TEST_GEOMETRY_DEFINE_COUNT};
+}
+
+// override updates
+typedef struct test_override {
+    const char *name;
+    const intmax_t *defines;
+    size_t permutations;
+} test_override_t;
+
+const test_override_t *test_overrides = NULL;
+size_t test_override_count = 0;
+
+test_define_t *test_override_defines = NULL;
+size_t test_override_define_count = 0;
+size_t test_override_define_permutations = 1;
+size_t test_override_define_capacity = 0;
+
+// suite/perm updates
+void test_define_suite(const struct test_suite *suite) {
+    test_define_names[TEST_DEFINE_NAMES_SUITE] = (test_define_names_t){
+            suite->define_names, suite->define_count};
+
+    // make sure our cache is large enough
+    if (lfs_max(suite->define_count, TEST_IMPLICIT_DEFINE_COUNT)
+            > test_define_cache_count) {
+        // align to power of two to avoid any superlinear growth
+        size_t ncount = 1 << lfs_npw2(
+                lfs_max(suite->define_count, TEST_IMPLICIT_DEFINE_COUNT));
+        test_define_cache = realloc(test_define_cache, ncount*sizeof(intmax_t));
+        test_define_cache_mask = realloc(test_define_cache_mask,
+                sizeof(unsigned)*(
+                    (ncount+(8*sizeof(unsigned))-1)
+                    / (8*sizeof(unsigned))));
+        test_define_cache_count = ncount;
+    }
+
+    // map any overrides
+    if (test_override_count > 0) {
+        // first figure out the total size of override permutations
+        size_t count = 0;
+        size_t permutations = 1;
+        for (size_t i = 0; i < test_override_count; i++) {
+            for (size_t d = 0;
+                    d < lfs_max(
+                        suite->define_count,
+                        TEST_IMPLICIT_DEFINE_COUNT);
+                    d++) {
+                // define name match?
+                const char *name = test_define_name(d);
+                if (name && strcmp(name, test_overrides[i].name) == 0) {
+                    count = lfs_max(count, d+1);
+                    permutations *= test_overrides[i].permutations;
+                    break;
+                }
+            }
+        }
+        test_override_define_count = count;
+        test_override_define_permutations = permutations;
+
+        // make sure our override arrays are big enough
+        if (count * permutations > test_override_define_capacity) {
+            // align to power of two to avoid any superlinear growth
+            size_t ncapacity = 1 << lfs_npw2(count * permutations);
+            test_override_defines = realloc(
+                    test_override_defines,
+                    sizeof(test_define_t)*ncapacity);
+            test_override_define_capacity = ncapacity;
+        }
+
+        // zero unoverridden defines
+        memset(test_override_defines, 0,
+                sizeof(test_define_t) * count * permutations);
+
+        // compute permutations
+        size_t p = 1;
+        for (size_t i = 0; i < test_override_count; i++) {
+            for (size_t d = 0;
+                    d < lfs_max(
+                        suite->define_count,
+                        TEST_IMPLICIT_DEFINE_COUNT);
+                    d++) {
+                // define name match?
+                const char *name = test_define_name(d);
+                if (name && strcmp(name, test_overrides[i].name) == 0) {
+                    // scatter the define permutations based on already
+                    // seen permutations
+                    for (size_t j = 0; j < permutations; j++) {
+                        test_override_defines[j*count + d] = TEST_LIT(
+                                test_overrides[i].defines[(j/p)
+                                    % test_overrides[i].permutations]);
+                    }
+
+                    // keep track of how many permutations we've seen so far
+                    p *= test_overrides[i].permutations;
+                    break;
+                }
+            }
+        }
+    }
+}
+
+void test_define_perm(
+        const struct test_suite *suite,
+        const struct test_case *case_,
+        size_t perm) {
+    if (case_->defines) {
+        test_define_maps[TEST_DEFINE_MAP_PERMUTATION] = (test_define_map_t){
+                case_->defines + perm*suite->define_count,
+                suite->define_count};
+    } else {
+        test_define_maps[TEST_DEFINE_MAP_PERMUTATION] = (test_define_map_t){
+                NULL, 0};
+    }
+}
+
+void test_define_override(size_t perm) {
+    test_define_maps[TEST_DEFINE_MAP_OVERRIDE] = (test_define_map_t){
+            test_override_defines + perm*test_override_define_count,
+            test_override_define_count};
+}
+
+void test_define_explicit(
+        const test_define_t *defines,
+        size_t define_count) {
+    test_define_maps[TEST_DEFINE_MAP_EXPLICIT] = (test_define_map_t){
+            defines, define_count};
+}
+
+void test_define_cleanup(void) {
+    // test define management can allocate a few things
+    free(test_define_cache);
+    free(test_define_cache_mask);
+    free(test_override_defines);
+}
+
+
+
+// test state
+extern const test_geometry_t *test_geometries;
+extern size_t test_geometry_count;
+
+extern const test_powerloss_t *test_powerlosses;
+extern size_t test_powerloss_count;
+
+const test_id_t *test_ids = (const test_id_t[]) {
+    {NULL, NULL, 0, NULL, 0},
+};
+size_t test_id_count = 1;
+
+size_t test_step_start = 0;
+size_t test_step_stop = -1;
+size_t test_step_step = 1;
+
+const char *test_disk_path = NULL;
+const char *test_trace_path = NULL;
+bool test_trace_backtrace = false;
+uint32_t test_trace_period = 0;
+uint32_t test_trace_freq = 0;
+FILE *test_trace_file = NULL;
+uint32_t test_trace_cycles = 0;
+uint64_t test_trace_time = 0;
+uint64_t test_trace_open_time = 0;
+lfs_emubd_sleep_t test_read_sleep = 0.0;
+lfs_emubd_sleep_t test_prog_sleep = 0.0;
+lfs_emubd_sleep_t test_erase_sleep = 0.0;
+
+// this determines both the backtrace buffer and the trace printf buffer, if
+// trace ends up interleaved or truncated this may need to be increased
+#ifndef TEST_TRACE_BACKTRACE_BUFFER_SIZE
+#define TEST_TRACE_BACKTRACE_BUFFER_SIZE 8192
+#endif
+void *test_trace_backtrace_buffer[
+    TEST_TRACE_BACKTRACE_BUFFER_SIZE / sizeof(void*)];
+
+// trace printing
+void test_trace(const char *fmt, ...) {
+    if (test_trace_path) {
+        // sample at a specific period?
+        if (test_trace_period) {
+            if (test_trace_cycles % test_trace_period != 0) {
+                test_trace_cycles += 1;
+                return;
+            }
+            test_trace_cycles += 1;
+        }
+
+        // sample at a specific frequency?
+        if (test_trace_freq) {
+            struct timespec t;
+            clock_gettime(CLOCK_MONOTONIC, &t);
+            uint64_t now = (uint64_t)t.tv_sec*1000*1000*1000
+                    + (uint64_t)t.tv_nsec;
+            if (now - test_trace_time < (1000*1000*1000) / test_trace_freq) {
+                return;
+            }
+            test_trace_time = now;
+        }
+
+        if (!test_trace_file) {
+            // Tracing output is heavy and trying to open every trace
+            // call is slow, so we only try to open the trace file every
+            // so often. Note this doesn't affect successfully opened files
+            struct timespec t;
+            clock_gettime(CLOCK_MONOTONIC, &t);
+            uint64_t now = (uint64_t)t.tv_sec*1000*1000*1000
+                    + (uint64_t)t.tv_nsec;
+            if (now - test_trace_open_time < 100*1000*1000) {
+                return;
+            }
+            test_trace_open_time = now;
+
+            // try to open the trace file
+            int fd;
+            if (strcmp(test_trace_path, "-") == 0) {
+                fd = dup(1);
+                if (fd < 0) {
+                    return;
+                }
+            } else {
+                fd = open(
+                        test_trace_path,
+                        O_WRONLY | O_CREAT | O_APPEND | O_NONBLOCK,
+                        0666);
+                if (fd < 0) {
+                    return;
+                }
+                int err = fcntl(fd, F_SETFL, O_WRONLY | O_CREAT | O_APPEND);
+                assert(!err);
+            }
+
+            FILE *f = fdopen(fd, "a");
+            assert(f);
+            int err = setvbuf(f, NULL, _IOFBF,
+                    TEST_TRACE_BACKTRACE_BUFFER_SIZE);
+            assert(!err);
+            test_trace_file = f;
+        }
+
+        // print trace
+        va_list va;
+        va_start(va, fmt);
+        int res = vfprintf(test_trace_file, fmt, va);
+        va_end(va);
+        if (res < 0) {
+            fclose(test_trace_file);
+            test_trace_file = NULL;
+            return;
+        }
+
+        if (test_trace_backtrace) {
+            // print backtrace
+            size_t count = backtrace(
+                    test_trace_backtrace_buffer,
+                    TEST_TRACE_BACKTRACE_BUFFER_SIZE);
+            // note we skip our own stack frame
+            for (size_t i = 1; i < count; i++) {
+                res = fprintf(test_trace_file, "\tat %p\n",
+                        test_trace_backtrace_buffer[i]);
+                if (res < 0) {
+                    fclose(test_trace_file);
+                    test_trace_file = NULL;
+                    return;
+                }
+            }
+        }
+
+        // flush immediately
+        fflush(test_trace_file);
+    }
+}
+
+
+// test prng
+uint32_t test_prng(uint32_t *state) {
+    // A simple xorshift32 generator, easily reproducible. Keep in mind
+    // determinism is much more important than actual randomness here.
+    uint32_t x = *state;
+    x ^= x << 13;
+    x ^= x >> 17;
+    x ^= x << 5;
+    *state = x;
+    return x;
+}
+
+
+// encode our permutation into a reusable id
+static void perm_printid(
+        const struct test_suite *suite,
+        const struct test_case *case_,
+        const lfs_emubd_powercycles_t *cycles,
+        size_t cycle_count) {
+    (void)suite;
+    // case[:permutation[:powercycles]]
+    printf("%s:", case_->name);
+    for (size_t d = 0;
+            d < lfs_max(
+                suite->define_count,
+                TEST_IMPLICIT_DEFINE_COUNT);
+            d++) {
+        if (test_define_ispermutation(d)) {
+            leb16_print(d);
+            leb16_print(TEST_DEFINE(d));
+        }
+    }
+
+    // only print power-cycles if any occured
+    if (cycles) {
+        printf(":");
+        for (size_t i = 0; i < cycle_count; i++) {
+            leb16_print(cycles[i]);
+        }
+    }
+}
+
+
+// a quick trie for keeping track of permutations we've seen
+typedef struct test_seen {
+    struct test_seen_branch *branches;
+    size_t branch_count;
+    size_t branch_capacity;
+} test_seen_t;
+
+struct test_seen_branch {
+    intmax_t define;
+    struct test_seen branch;
+};
+
+bool test_seen_insert(
+        test_seen_t *seen,
+        const struct test_suite *suite,
+        const struct test_case *case_) {
+    (void)case_;
+    bool was_seen = true;
+
+    // use the currently set defines
+    for (size_t d = 0;
+            d < lfs_max(
+                suite->define_count,
+                TEST_IMPLICIT_DEFINE_COUNT);
+            d++) {
+        // treat unpermuted defines the same as 0
+        intmax_t define = test_define_ispermutation(d) ? TEST_DEFINE(d) : 0;
+
+        // already seen?
+        struct test_seen_branch *branch = NULL;
+        for (size_t i = 0; i < seen->branch_count; i++) {
+            if (seen->branches[i].define == define) {
+                branch = &seen->branches[i];
+                break;
+            }
+        }
+
+        // need to create a new node
+        if (!branch) {
+            was_seen = false;
+            branch = mappend(
+                    (void**)&seen->branches,
+                    sizeof(struct test_seen_branch),
+                    &seen->branch_count,
+                    &seen->branch_capacity);
+            branch->define = define;
+            branch->branch = (test_seen_t){NULL, 0, 0};
+        }
+
+        seen = &branch->branch;
+    }
+
+    return was_seen;
+}
+
+void test_seen_cleanup(test_seen_t *seen) {
+    for (size_t i = 0; i < seen->branch_count; i++) {
+        test_seen_cleanup(&seen->branches[i].branch);
+    }
+    free(seen->branches);
+}
+
+static void run_powerloss_none(
+        const lfs_emubd_powercycles_t *cycles,
+        size_t cycle_count,
+        const struct test_suite *suite,
+        const struct test_case *case_);
+static void run_powerloss_cycles(
+        const lfs_emubd_powercycles_t *cycles,
+        size_t cycle_count,
+        const struct test_suite *suite,
+        const struct test_case *case_);
+
+// iterate through permutations in a test case
+static void case_forperm(
+        const struct test_suite *suite,
+        const struct test_case *case_,
+        const test_define_t *defines,
+        size_t define_count,
+        const lfs_emubd_powercycles_t *cycles,
+        size_t cycle_count,
+        void (*cb)(
+            void *data,
+            const struct test_suite *suite,
+            const struct test_case *case_,
+            const test_powerloss_t *powerloss),
+        void *data) {
+    // explicit permutation?
+    if (defines) {
+        test_define_explicit(defines, define_count);
+
+        for (size_t v = 0; v < test_override_define_permutations; v++) {
+            // define override permutation
+            test_define_override(v);
+            test_define_flush();
+
+            // explicit powerloss cycles?
+            if (cycles) {
+                cb(data, suite, case_, &(test_powerloss_t){
+                        .run=run_powerloss_cycles,
+                        .cycles=cycles,
+                        .cycle_count=cycle_count});
+            } else {
+                for (size_t p = 0; p < test_powerloss_count; p++) {
+                    // skip non-reentrant tests when powerloss testing
+                    if (test_powerlosses[p].run != run_powerloss_none
+                            && !(case_->flags & TEST_REENTRANT)) {
+                        continue;
+                    }
+
+                    cb(data, suite, case_, &test_powerlosses[p]);
+                }
+            }
+        }
+
+        return;
+    }
+
+    test_seen_t seen = {NULL, 0, 0};
+
+    for (size_t k = 0; k < case_->permutations; k++) {
+        // define permutation
+        test_define_perm(suite, case_, k);
+
+        for (size_t v = 0; v < test_override_define_permutations; v++) {
+            // define override permutation
+            test_define_override(v);
+
+            for (size_t g = 0; g < test_geometry_count; g++) {
+                // define geometry
+                test_define_geometry(&test_geometries[g]);
+                test_define_flush();
+
+                // have we seen this permutation before?
+                bool was_seen = test_seen_insert(&seen, suite, case_);
+                if (!(k == 0 && v == 0 && g == 0) && was_seen) {
+                    continue;
+                }
+
+                if (cycles) {
+                    cb(data, suite, case_, &(test_powerloss_t){
+                            .run=run_powerloss_cycles,
+                            .cycles=cycles,
+                            .cycle_count=cycle_count});
+                } else {
+                    for (size_t p = 0; p < test_powerloss_count; p++) {
+                        // skip non-reentrant tests when powerloss testing
+                        if (test_powerlosses[p].run != run_powerloss_none
+                                && !(case_->flags & TEST_REENTRANT)) {
+                            continue;
+                        }
+
+                        cb(data, suite, case_, &test_powerlosses[p]);
+                    }
+                }
+            }
+        }
+    }
+
+    test_seen_cleanup(&seen);
+}
+
+
+// how many permutations are there actually in a test case
+struct perm_count_state {
+    size_t total;
+    size_t filtered;
+};
+
+void perm_count(
+        void *data,
+        const struct test_suite *suite,
+        const struct test_case *case_,
+        const test_powerloss_t *powerloss) {
+    struct perm_count_state *state = data;
+    (void)suite;
+    (void)case_;
+    (void)powerloss;
+
+    state->total += 1;
+
+    if (case_->filter && !case_->filter()) {
+        return;
+    }
+
+    state->filtered += 1;
+}
+
+
+// operations we can do
+static void summary(void) {
+    printf("%-23s  %7s %7s %7s %11s\n",
+            "", "flags", "suites", "cases", "perms");
+    size_t suites = 0;
+    size_t cases = 0;
+    test_flags_t flags = 0;
+    struct perm_count_state perms = {0, 0};
+
+    for (size_t t = 0; t < test_id_count; t++) {
+        for (size_t i = 0; i < TEST_SUITE_COUNT; i++) {
+            test_define_suite(&test_suites[i]);
+
+            for (size_t j = 0; j < test_suites[i].case_count; j++) {
+                // does neither suite nor case name match?
+                if (test_ids[t].name && !(
+                        strcmp(test_ids[t].name,
+                            test_suites[i].name) == 0
+                        || strcmp(test_ids[t].name,
+                            test_suites[i].cases[j].name) == 0)) {
+                    continue;
+                }
+
+                cases += 1;
+                case_forperm(
+                        &test_suites[i],
+                        &test_suites[i].cases[j],
+                        test_ids[t].defines,
+                        test_ids[t].define_count,
+                        test_ids[t].cycles,
+                        test_ids[t].cycle_count,
+                        perm_count,
+                        &perms);
+            }
+
+            suites += 1;
+            flags |= test_suites[i].flags;
+        }
+    }
+
+    char perm_buf[64];
+    sprintf(perm_buf, "%zu/%zu", perms.filtered, perms.total);
+    char flag_buf[64];
+    sprintf(flag_buf, "%s%s",
+            (flags & TEST_REENTRANT) ? "r" : "",
+            (!flags) ? "-" : "");
+    printf("%-23s  %7s %7zu %7zu %11s\n",
+            "TOTAL",
+            flag_buf,
+            suites,
+            cases,
+            perm_buf);
+}
+
+static void list_suites(void) {
+    // at least size so that names fit
+    unsigned name_width = 23;
+    for (size_t i = 0; i < TEST_SUITE_COUNT; i++) {
+        size_t len = strlen(test_suites[i].name);
+        if (len > name_width) {
+            name_width = len;
+        }
+    }
+    name_width = 4*((name_width+1+4-1)/4)-1;
+
+    printf("%-*s  %7s %7s %11s\n",
+            name_width, "suite", "flags", "cases", "perms");
+    for (size_t t = 0; t < test_id_count; t++) {
+        for (size_t i = 0; i < TEST_SUITE_COUNT; i++) {
+            test_define_suite(&test_suites[i]);
+
+            size_t cases = 0;
+            struct perm_count_state perms = {0, 0};
+
+            for (size_t j = 0; j < test_suites[i].case_count; j++) {
+                // does neither suite nor case name match?
+                if (test_ids[t].name && !(
+                        strcmp(test_ids[t].name,
+                            test_suites[i].name) == 0
+                        || strcmp(test_ids[t].name,
+                            test_suites[i].cases[j].name) == 0)) {
+                    continue;
+                }
+
+                cases += 1;
+                case_forperm(
+                        &test_suites[i],
+                        &test_suites[i].cases[j],
+                        test_ids[t].defines,
+                        test_ids[t].define_count,
+                        test_ids[t].cycles,
+                        test_ids[t].cycle_count,
+                        perm_count,
+                        &perms);
+            }
+
+            // no tests found?
+            if (!cases) {
+                continue;
+            }
+
+            char perm_buf[64];
+            sprintf(perm_buf, "%zu/%zu", perms.filtered, perms.total);
+            char flag_buf[64];
+            sprintf(flag_buf, "%s%s",
+                    (test_suites[i].flags & TEST_REENTRANT) ? "r" : "",
+                    (!test_suites[i].flags) ? "-" : "");
+            printf("%-*s  %7s %7zu %11s\n",
+                    name_width,
+                    test_suites[i].name,
+                    flag_buf,
+                    cases,
+                    perm_buf);
+        }
+    }
+}
+
+static void list_cases(void) {
+    // at least size so that names fit
+    unsigned name_width = 23;
+    for (size_t i = 0; i < TEST_SUITE_COUNT; i++) {
+        for (size_t j = 0; j < test_suites[i].case_count; j++) {
+            size_t len = strlen(test_suites[i].cases[j].name);
+            if (len > name_width) {
+                name_width = len;
+            }
+        }
+    }
+    name_width = 4*((name_width+1+4-1)/4)-1;
+
+    printf("%-*s  %7s %11s\n", name_width, "case", "flags", "perms");
+    for (size_t t = 0; t < test_id_count; t++) {
+        for (size_t i = 0; i < TEST_SUITE_COUNT; i++) {
+            test_define_suite(&test_suites[i]);
+
+            for (size_t j = 0; j < test_suites[i].case_count; j++) {
+                // does neither suite nor case name match?
+                if (test_ids[t].name && !(
+                        strcmp(test_ids[t].name,
+                            test_suites[i].name) == 0
+                        || strcmp(test_ids[t].name,
+                            test_suites[i].cases[j].name) == 0)) {
+                    continue;
+                }
+
+                struct perm_count_state perms = {0, 0};
+                case_forperm(
+                        &test_suites[i],
+                        &test_suites[i].cases[j],
+                        test_ids[t].defines,
+                        test_ids[t].define_count,
+                        test_ids[t].cycles,
+                        test_ids[t].cycle_count,
+                        perm_count,
+                        &perms);
+
+                char perm_buf[64];
+                sprintf(perm_buf, "%zu/%zu", perms.filtered, perms.total);
+                char flag_buf[64];
+                sprintf(flag_buf, "%s%s",
+                        (test_suites[i].cases[j].flags & TEST_REENTRANT)
+                            ? "r" : "",
+                        (!test_suites[i].cases[j].flags)
+                            ? "-" : "");
+                printf("%-*s  %7s %11s\n",
+                        name_width,
+                        test_suites[i].cases[j].name,
+                        flag_buf,
+                        perm_buf);
+            }
+        }
+    }
+}
+
+static void list_suite_paths(void) {
+    // at least size so that names fit
+    unsigned name_width = 23;
+    for (size_t i = 0; i < TEST_SUITE_COUNT; i++) {
+        size_t len = strlen(test_suites[i].name);
+        if (len > name_width) {
+            name_width = len;
+        }
+    }
+    name_width = 4*((name_width+1+4-1)/4)-1;
+
+    printf("%-*s  %s\n", name_width, "suite", "path");
+    for (size_t t = 0; t < test_id_count; t++) {
+        for (size_t i = 0; i < TEST_SUITE_COUNT; i++) {
+            size_t cases = 0;
+
+            for (size_t j = 0; j < test_suites[i].case_count; j++) {
+                // does neither suite nor case name match?
+                if (test_ids[t].name && !(
+                        strcmp(test_ids[t].name,
+                            test_suites[i].name) == 0
+                        || strcmp(test_ids[t].name,
+                            test_suites[i].cases[j].name) == 0)) {
+                    continue;
+                }
+
+                cases += 1;
+            }
+
+            // no tests found?
+            if (!cases) {
+                continue;
+            }
+
+            printf("%-*s  %s\n",
+                    name_width,
+                    test_suites[i].name,
+                    test_suites[i].path);
+        }
+    }
+}
+
+static void list_case_paths(void) {
+    // at least size so that names fit
+    unsigned name_width = 23;
+    for (size_t i = 0; i < TEST_SUITE_COUNT; i++) {
+        for (size_t j = 0; j < test_suites[i].case_count; j++) {
+            size_t len = strlen(test_suites[i].cases[j].name);
+            if (len > name_width) {
+                name_width = len;
+            }
+        }
+    }
+    name_width = 4*((name_width+1+4-1)/4)-1;
+
+    printf("%-*s  %s\n", name_width, "case", "path");
+    for (size_t t = 0; t < test_id_count; t++) {
+        for (size_t i = 0; i < TEST_SUITE_COUNT; i++) {
+            for (size_t j = 0; j < test_suites[i].case_count; j++) {
+                // does neither suite nor case name match?
+                if (test_ids[t].name && !(
+                        strcmp(test_ids[t].name,
+                            test_suites[i].name) == 0
+                        || strcmp(test_ids[t].name,
+                            test_suites[i].cases[j].name) == 0)) {
+                    continue;
+                }
+
+                printf("%-*s  %s\n",
+                        name_width,
+                        test_suites[i].cases[j].name,
+                        test_suites[i].cases[j].path);
+            }
+        }
+    }
+}
+
+struct list_defines_define {
+    const char *name;
+    intmax_t *values;
+    size_t value_count;
+    size_t value_capacity;
+};
+
+struct list_defines_defines {
+    struct list_defines_define *defines;
+    size_t define_count;
+    size_t define_capacity;
+};
+
+static void list_defines_add(
+        struct list_defines_defines *defines,
+        size_t d) {
+    const char *name = test_define_name(d);
+    intmax_t value = TEST_DEFINE(d);
+
+    // define already in defines?
+    for (size_t i = 0; i < defines->define_count; i++) {
+        if (strcmp(defines->defines[i].name, name) == 0) {
+            // value already in values?
+            for (size_t j = 0; j < defines->defines[i].value_count; j++) {
+                if (defines->defines[i].values[j] == value) {
+                    return;
+                }
+            }
+
+            *(intmax_t*)mappend(
+                (void**)&defines->defines[i].values,
+                sizeof(intmax_t),
+                &defines->defines[i].value_count,
+                &defines->defines[i].value_capacity) = value;
+
+            return;
+        }
+    }
+
+    // new define?
+    struct list_defines_define *define = mappend(
+            (void**)&defines->defines,
+            sizeof(struct list_defines_define),
+            &defines->define_count,
+            &defines->define_capacity);
+    define->name = name;
+    define->values = malloc(sizeof(intmax_t));
+    define->values[0] = value;
+    define->value_count = 1;
+    define->value_capacity = 1;
+}
+
+void perm_list_defines(
+        void *data,
+        const struct test_suite *suite,
+        const struct test_case *case_,
+        const test_powerloss_t *powerloss) {
+    struct list_defines_defines *defines = data;
+    (void)suite;
+    (void)case_;
+    (void)powerloss;
+
+    // collect defines
+    for (size_t d = 0;
+            d < lfs_max(suite->define_count,
+                TEST_IMPLICIT_DEFINE_COUNT);
+            d++) {
+        if (d < TEST_IMPLICIT_DEFINE_COUNT
+                || test_define_ispermutation(d)) {
+            list_defines_add(defines, d);
+        }
+    }
+}
+
+void perm_list_permutation_defines(
+        void *data,
+        const struct test_suite *suite,
+        const struct test_case *case_,
+        const test_powerloss_t *powerloss) {
+    struct list_defines_defines *defines = data;
+    (void)suite;
+    (void)case_;
+    (void)powerloss;
+
+    // collect permutation_defines
+    for (size_t d = 0;
+            d < lfs_max(suite->define_count,
+                TEST_IMPLICIT_DEFINE_COUNT);
+            d++) {
+        if (test_define_ispermutation(d)) {
+            list_defines_add(defines, d);
+        }
+    }
+}
+
+extern const test_geometry_t builtin_geometries[];
+
+static void list_defines(void) {
+    struct list_defines_defines defines = {NULL, 0, 0};
+
+    // add defines
+    for (size_t t = 0; t < test_id_count; t++) {
+        for (size_t i = 0; i < TEST_SUITE_COUNT; i++) {
+            test_define_suite(&test_suites[i]);
+
+            for (size_t j = 0; j < test_suites[i].case_count; j++) {
+                // does neither suite nor case name match?
+                if (test_ids[t].name && !(
+                        strcmp(test_ids[t].name,
+                            test_suites[i].name) == 0
+                        || strcmp(test_ids[t].name,
+                            test_suites[i].cases[j].name) == 0)) {
+                    continue;
+                }
+
+                case_forperm(
+                        &test_suites[i],
+                        &test_suites[i].cases[j],
+                        test_ids[t].defines,
+                        test_ids[t].define_count,
+                        test_ids[t].cycles,
+                        test_ids[t].cycle_count,
+                        perm_list_defines,
+                        &defines);
+            }
+        }
+    }
+
+    for (size_t i = 0; i < defines.define_count; i++) {
+        printf("%s=", defines.defines[i].name);
+        for (size_t j = 0; j < defines.defines[i].value_count; j++) {
+            printf("%jd", defines.defines[i].values[j]);
+            if (j != defines.defines[i].value_count-1) {
+                printf(",");
+            }
+        }
+        printf("\n");
+    }
+
+    for (size_t i = 0; i < defines.define_count; i++) {
+        free(defines.defines[i].values);
+    }
+    free(defines.defines);
+}
+
+static void list_permutation_defines(void) {
+    struct list_defines_defines defines = {NULL, 0, 0};
+
+    // add permutation defines
+    for (size_t t = 0; t < test_id_count; t++) {
+        for (size_t i = 0; i < TEST_SUITE_COUNT; i++) {
+            test_define_suite(&test_suites[i]);
+
+            for (size_t j = 0; j < test_suites[i].case_count; j++) {
+                // does neither suite nor case name match?
+                if (test_ids[t].name && !(
+                        strcmp(test_ids[t].name,
+                            test_suites[i].name) == 0
+                        || strcmp(test_ids[t].name,
+                            test_suites[i].cases[j].name) == 0)) {
+                    continue;
+                }
+
+                case_forperm(
+                        &test_suites[i],
+                        &test_suites[i].cases[j],
+                        test_ids[t].defines,
+                        test_ids[t].define_count,
+                        test_ids[t].cycles,
+                        test_ids[t].cycle_count,
+                        perm_list_permutation_defines,
+                        &defines);
+            }
+        }
+    }
+
+    for (size_t i = 0; i < defines.define_count; i++) {
+        printf("%s=", defines.defines[i].name);
+        for (size_t j = 0; j < defines.defines[i].value_count; j++) {
+            printf("%jd", defines.defines[i].values[j]);
+            if (j != defines.defines[i].value_count-1) {
+                printf(",");
+            }
+        }
+        printf("\n");
+    }
+
+    for (size_t i = 0; i < defines.define_count; i++) {
+        free(defines.defines[i].values);
+    }
+    free(defines.defines);
+}
+
+static void list_implicit_defines(void) {
+    struct list_defines_defines defines = {NULL, 0, 0};
+
+    // yes we do need to define a suite, this does a bit of bookeeping
+    // such as setting up the define cache
+    test_define_suite(&(const struct test_suite){0});
+
+    // make sure to include builtin geometries here
+    extern const test_geometry_t builtin_geometries[];
+    for (size_t g = 0; builtin_geometries[g].name; g++) {
+        test_define_geometry(&builtin_geometries[g]);
+        test_define_flush();
+
+        // add implicit defines
+        for (size_t d = 0; d < TEST_IMPLICIT_DEFINE_COUNT; d++) {
+            list_defines_add(&defines, d);
+        }
+    }
+
+    for (size_t i = 0; i < defines.define_count; i++) {
+        printf("%s=", defines.defines[i].name);
+        for (size_t j = 0; j < defines.defines[i].value_count; j++) {
+            printf("%jd", defines.defines[i].values[j]);
+            if (j != defines.defines[i].value_count-1) {
+                printf(",");
+            }
+        }
+        printf("\n");
+    }
+
+    for (size_t i = 0; i < defines.define_count; i++) {
+        free(defines.defines[i].values);
+    }
+    free(defines.defines);
+}
+
+
+
+// geometries to test
+
+const test_geometry_t builtin_geometries[] = {
+    {"default", {{0}, TEST_CONST(16),   TEST_CONST(512),   {0}}},
+    {"eeprom",  {{0}, TEST_CONST(1),    TEST_CONST(512),   {0}}},
+    {"emmc",    {{0}, {0},              TEST_CONST(512),   {0}}},
+    {"nor",     {{0}, TEST_CONST(1),    TEST_CONST(4096),  {0}}},
+    {"nand",    {{0}, TEST_CONST(4096), TEST_CONST(32768), {0}}},
+    {NULL, {{0}, {0}, {0}, {0}}},
+};
+
+const test_geometry_t *test_geometries = builtin_geometries;
+size_t test_geometry_count = 5;
+
+static void list_geometries(void) {
+    // at least size so that names fit
+    unsigned name_width = 23;
+    for (size_t g = 0; builtin_geometries[g].name; g++) {
+        size_t len = strlen(builtin_geometries[g].name);
+        if (len > name_width) {
+            name_width = len;
+        }
+    }
+    name_width = 4*((name_width+1+4-1)/4)-1;
+
+    // yes we do need to define a suite, this does a bit of bookeeping
+    // such as setting up the define cache
+    test_define_suite(&(const struct test_suite){0});
+
+    printf("%-*s  %7s %7s %7s %7s %11s\n",
+            name_width, "geometry", "read", "prog", "erase", "count", "size");
+    for (size_t g = 0; builtin_geometries[g].name; g++) {
+        test_define_geometry(&builtin_geometries[g]);
+        test_define_flush();
+        printf("%-*s  %7ju %7ju %7ju %7ju %11ju\n",
+                name_width,
+                builtin_geometries[g].name,
+                READ_SIZE,
+                PROG_SIZE,
+                ERASE_SIZE,
+                ERASE_COUNT,
+                ERASE_SIZE*ERASE_COUNT);
+    }
+}
+
+
+// scenarios to run tests under power-loss
+
+static void run_powerloss_none(
+        const lfs_emubd_powercycles_t *cycles,
+        size_t cycle_count,
+        const struct test_suite *suite,
+        const struct test_case *case_) {
+    (void)cycles;
+    (void)cycle_count;
+    (void)suite;
+
+    // create block device and configuration
+    lfs_emubd_t bd;
+
+    struct lfs_config cfg = {
+        .context            = &bd,
+        .read               = lfs_emubd_read,
+        .prog               = lfs_emubd_prog,
+        .erase              = lfs_emubd_erase,
+        .sync               = lfs_emubd_sync,
+        .read_size          = READ_SIZE,
+        .prog_size          = PROG_SIZE,
+        .block_size         = BLOCK_SIZE,
+        .block_count        = BLOCK_COUNT,
+        .block_cycles       = BLOCK_CYCLES,
+        .cache_size         = CACHE_SIZE,
+        .lookahead_size     = LOOKAHEAD_SIZE,
+    #ifdef LFS_MULTIVERSION
+        .disk_version       = DISK_VERSION,
+    #endif
+    };
+
+    struct lfs_emubd_config bdcfg = {
+        .read_size          = READ_SIZE,
+        .prog_size          = PROG_SIZE,
+        .erase_size         = ERASE_SIZE,
+        .erase_count        = ERASE_COUNT,
+        .erase_value        = ERASE_VALUE,
+        .erase_cycles       = ERASE_CYCLES,
+        .badblock_behavior  = BADBLOCK_BEHAVIOR,
+        .disk_path          = test_disk_path,
+        .read_sleep         = test_read_sleep,
+        .prog_sleep         = test_prog_sleep,
+        .erase_sleep        = test_erase_sleep,
+    };
+
+    int err = lfs_emubd_create(&cfg, &bdcfg);
+    if (err) {
+        fprintf(stderr, "error: could not create block device: %d\n", err);
+        exit(-1);
+    }
+
+    // run the test
+    printf("running ");
+    perm_printid(suite, case_, NULL, 0);
+    printf("\n");
+
+    case_->run(&cfg);
+
+    printf("finished ");
+    perm_printid(suite, case_, NULL, 0);
+    printf("\n");
+
+    // cleanup
+    err = lfs_emubd_destroy(&cfg);
+    if (err) {
+        fprintf(stderr, "error: could not destroy block device: %d\n", err);
+        exit(-1);
+    }
+}
+
+static void powerloss_longjmp(void *c) {
+    jmp_buf *powerloss_jmp = c;
+    longjmp(*powerloss_jmp, 1);
+}
+
+static void run_powerloss_linear(
+        const lfs_emubd_powercycles_t *cycles,
+        size_t cycle_count,
+        const struct test_suite *suite,
+        const struct test_case *case_) {
+    (void)cycles;
+    (void)cycle_count;
+    (void)suite;
+
+    // create block device and configuration
+    lfs_emubd_t bd;
+    jmp_buf powerloss_jmp;
+    volatile lfs_emubd_powercycles_t i = 1;
+
+    struct lfs_config cfg = {
+        .context            = &bd,
+        .read               = lfs_emubd_read,
+        .prog               = lfs_emubd_prog,
+        .erase              = lfs_emubd_erase,
+        .sync               = lfs_emubd_sync,
+        .read_size          = READ_SIZE,
+        .prog_size          = PROG_SIZE,
+        .block_size         = BLOCK_SIZE,
+        .block_count        = BLOCK_COUNT,
+        .block_cycles       = BLOCK_CYCLES,
+        .cache_size         = CACHE_SIZE,
+        .lookahead_size     = LOOKAHEAD_SIZE,
+    #ifdef LFS_MULTIVERSION
+        .disk_version       = DISK_VERSION,
+    #endif
+    };
+
+    struct lfs_emubd_config bdcfg = {
+        .read_size          = READ_SIZE,
+        .prog_size          = PROG_SIZE,
+        .erase_size         = ERASE_SIZE,
+        .erase_count        = ERASE_COUNT,
+        .erase_value        = ERASE_VALUE,
+        .erase_cycles       = ERASE_CYCLES,
+        .badblock_behavior  = BADBLOCK_BEHAVIOR,
+        .disk_path          = test_disk_path,
+        .read_sleep         = test_read_sleep,
+        .prog_sleep         = test_prog_sleep,
+        .erase_sleep        = test_erase_sleep,
+        .power_cycles       = i,
+        .powerloss_behavior = POWERLOSS_BEHAVIOR,
+        .powerloss_cb       = powerloss_longjmp,
+        .powerloss_data     = &powerloss_jmp,
+    };
+
+    int err = lfs_emubd_create(&cfg, &bdcfg);
+    if (err) {
+        fprintf(stderr, "error: could not create block device: %d\n", err);
+        exit(-1);
+    }
+
+    // run the test, increasing power-cycles as power-loss events occur
+    printf("running ");
+    perm_printid(suite, case_, NULL, 0);
+    printf("\n");
+
+    while (true) {
+        if (!setjmp(powerloss_jmp)) {
+            // run the test
+            case_->run(&cfg);
+            break;
+        }
+
+        // power-loss!
+        printf("powerloss ");
+        perm_printid(suite, case_, NULL, 0);
+        printf(":");
+        for (lfs_emubd_powercycles_t j = 1; j <= i; j++) {
+            leb16_print(j);
+        }
+        printf("\n");
+
+        i += 1;
+        lfs_emubd_setpowercycles(&cfg, i);
+    }
+
+    printf("finished ");
+    perm_printid(suite, case_, NULL, 0);
+    printf("\n");
+
+    // cleanup
+    err = lfs_emubd_destroy(&cfg);
+    if (err) {
+        fprintf(stderr, "error: could not destroy block device: %d\n", err);
+        exit(-1);
+    }
+}
+
+static void run_powerloss_log(
+        const lfs_emubd_powercycles_t *cycles,
+        size_t cycle_count,
+        const struct test_suite *suite,
+        const struct test_case *case_) {
+    (void)cycles;
+    (void)cycle_count;
+    (void)suite;
+
+    // create block device and configuration
+    lfs_emubd_t bd;
+    jmp_buf powerloss_jmp;
+    volatile lfs_emubd_powercycles_t i = 1;
+
+    struct lfs_config cfg = {
+        .context            = &bd,
+        .read               = lfs_emubd_read,
+        .prog               = lfs_emubd_prog,
+        .erase              = lfs_emubd_erase,
+        .sync               = lfs_emubd_sync,
+        .read_size          = READ_SIZE,
+        .prog_size          = PROG_SIZE,
+        .block_size         = BLOCK_SIZE,
+        .block_count        = BLOCK_COUNT,
+        .block_cycles       = BLOCK_CYCLES,
+        .cache_size         = CACHE_SIZE,
+        .lookahead_size     = LOOKAHEAD_SIZE,
+    #ifdef LFS_MULTIVERSION
+        .disk_version       = DISK_VERSION,
+    #endif
+    };
+
+    struct lfs_emubd_config bdcfg = {
+        .read_size          = READ_SIZE,
+        .prog_size          = PROG_SIZE,
+        .erase_size         = ERASE_SIZE,
+        .erase_count        = ERASE_COUNT,
+        .erase_value        = ERASE_VALUE,
+        .erase_cycles       = ERASE_CYCLES,
+        .badblock_behavior  = BADBLOCK_BEHAVIOR,
+        .disk_path          = test_disk_path,
+        .read_sleep         = test_read_sleep,
+        .prog_sleep         = test_prog_sleep,
+        .erase_sleep        = test_erase_sleep,
+        .power_cycles       = i,
+        .powerloss_behavior = POWERLOSS_BEHAVIOR,
+        .powerloss_cb       = powerloss_longjmp,
+        .powerloss_data     = &powerloss_jmp,
+    };
+
+    int err = lfs_emubd_create(&cfg, &bdcfg);
+    if (err) {
+        fprintf(stderr, "error: could not create block device: %d\n", err);
+        exit(-1);
+    }
+
+    // run the test, increasing power-cycles as power-loss events occur
+    printf("running ");
+    perm_printid(suite, case_, NULL, 0);
+    printf("\n");
+
+    while (true) {
+        if (!setjmp(powerloss_jmp)) {
+            // run the test
+            case_->run(&cfg);
+            break;
+        }
+
+        // power-loss!
+        printf("powerloss ");
+        perm_printid(suite, case_, NULL, 0);
+        printf(":");
+        for (lfs_emubd_powercycles_t j = 1; j <= i; j *= 2) {
+            leb16_print(j);
+        }
+        printf("\n");
+
+        i *= 2;
+        lfs_emubd_setpowercycles(&cfg, i);
+    }
+
+    printf("finished ");
+    perm_printid(suite, case_, NULL, 0);
+    printf("\n");
+
+    // cleanup
+    err = lfs_emubd_destroy(&cfg);
+    if (err) {
+        fprintf(stderr, "error: could not destroy block device: %d\n", err);
+        exit(-1);
+    }
+}
+
+static void run_powerloss_cycles(
+        const lfs_emubd_powercycles_t *cycles,
+        size_t cycle_count,
+        const struct test_suite *suite,
+        const struct test_case *case_) {
+    (void)suite;
+
+    // create block device and configuration
+    lfs_emubd_t bd;
+    jmp_buf powerloss_jmp;
+    volatile size_t i = 0;
+
+    struct lfs_config cfg = {
+        .context            = &bd,
+        .read               = lfs_emubd_read,
+        .prog               = lfs_emubd_prog,
+        .erase              = lfs_emubd_erase,
+        .sync               = lfs_emubd_sync,
+        .read_size          = READ_SIZE,
+        .prog_size          = PROG_SIZE,
+        .block_size         = BLOCK_SIZE,
+        .block_count        = BLOCK_COUNT,
+        .block_cycles       = BLOCK_CYCLES,
+        .cache_size         = CACHE_SIZE,
+        .lookahead_size     = LOOKAHEAD_SIZE,
+    #ifdef LFS_MULTIVERSION
+        .disk_version       = DISK_VERSION,
+    #endif
+    };
+
+    struct lfs_emubd_config bdcfg = {
+        .read_size          = READ_SIZE,
+        .prog_size          = PROG_SIZE,
+        .erase_size         = ERASE_SIZE,
+        .erase_count        = ERASE_COUNT,
+        .erase_value        = ERASE_VALUE,
+        .erase_cycles       = ERASE_CYCLES,
+        .badblock_behavior  = BADBLOCK_BEHAVIOR,
+        .disk_path          = test_disk_path,
+        .read_sleep         = test_read_sleep,
+        .prog_sleep         = test_prog_sleep,
+        .erase_sleep        = test_erase_sleep,
+        .power_cycles       = (i < cycle_count) ? cycles[i] : 0,
+        .powerloss_behavior = POWERLOSS_BEHAVIOR,
+        .powerloss_cb       = powerloss_longjmp,
+        .powerloss_data     = &powerloss_jmp,
+    };
+
+    int err = lfs_emubd_create(&cfg, &bdcfg);
+    if (err) {
+        fprintf(stderr, "error: could not create block device: %d\n", err);
+        exit(-1);
+    }
+
+    // run the test, increasing power-cycles as power-loss events occur
+    printf("running ");
+    perm_printid(suite, case_, NULL, 0);
+    printf("\n");
+
+    while (true) {
+        if (!setjmp(powerloss_jmp)) {
+            // run the test
+            case_->run(&cfg);
+            break;
+        }
+
+        // power-loss!
+        assert(i <= cycle_count);
+        printf("powerloss ");
+        perm_printid(suite, case_, cycles, i+1);
+        printf("\n");
+
+        i += 1;
+        lfs_emubd_setpowercycles(&cfg,
+                (i < cycle_count) ? cycles[i] : 0);
+    }
+
+    printf("finished ");
+    perm_printid(suite, case_, NULL, 0);
+    printf("\n");
+
+    // cleanup
+    err = lfs_emubd_destroy(&cfg);
+    if (err) {
+        fprintf(stderr, "error: could not destroy block device: %d\n", err);
+        exit(-1);
+    }
+}
+
+struct powerloss_exhaustive_state {
+    struct lfs_config *cfg;
+
+    lfs_emubd_t *branches;
+    size_t branch_count;
+    size_t branch_capacity;
+};
+
+struct powerloss_exhaustive_cycles {
+    lfs_emubd_powercycles_t *cycles;
+    size_t cycle_count;
+    size_t cycle_capacity;
+};
+
+static void powerloss_exhaustive_branch(void *c) {
+    struct powerloss_exhaustive_state *state = c;
+    // append to branches
+    lfs_emubd_t *branch = mappend(
+            (void**)&state->branches,
+            sizeof(lfs_emubd_t),
+            &state->branch_count,
+            &state->branch_capacity);
+    if (!branch) {
+        fprintf(stderr, "error: exhaustive: out of memory\n");
+        exit(-1);
+    }
+
+    // create copy-on-write copy
+    int err = lfs_emubd_copy(state->cfg, branch);
+    if (err) {
+        fprintf(stderr, "error: exhaustive: could not create bd copy\n");
+        exit(-1);
+    }
+
+    // also trigger on next power cycle
+    lfs_emubd_setpowercycles(state->cfg, 1);
+}
+
+static void run_powerloss_exhaustive_layer(
+        struct powerloss_exhaustive_cycles *cycles,
+        const struct test_suite *suite,
+        const struct test_case *case_,
+        struct lfs_config *cfg,
+        struct lfs_emubd_config *bdcfg,
+        size_t depth) {
+    (void)suite;
+
+    struct powerloss_exhaustive_state state = {
+        .cfg = cfg,
+        .branches = NULL,
+        .branch_count = 0,
+        .branch_capacity = 0,
+    };
+
+    // run through the test without additional powerlosses, collecting possible
+    // branches as we do so
+    lfs_emubd_setpowercycles(state.cfg, depth > 0 ? 1 : 0);
+    bdcfg->powerloss_data = &state;
+
+    // run the tests
+    case_->run(cfg);
+
+    // aggressively clean up memory here to try to keep our memory usage low
+    int err = lfs_emubd_destroy(cfg);
+    if (err) {
+        fprintf(stderr, "error: could not destroy block device: %d\n", err);
+        exit(-1);
+    }
+
+    // recurse into each branch
+    for (size_t i = 0; i < state.branch_count; i++) {
+        // first push and print the branch
+        lfs_emubd_powercycles_t *cycle = mappend(
+                (void**)&cycles->cycles,
+                sizeof(lfs_emubd_powercycles_t),
+                &cycles->cycle_count,
+                &cycles->cycle_capacity);
+        if (!cycle) {
+            fprintf(stderr, "error: exhaustive: out of memory\n");
+            exit(-1);
+        }
+        *cycle = i+1;
+
+        printf("powerloss ");
+        perm_printid(suite, case_, cycles->cycles, cycles->cycle_count);
+        printf("\n");
+
+        // now recurse
+        cfg->context = &state.branches[i];
+        run_powerloss_exhaustive_layer(cycles,
+                suite, case_,
+                cfg, bdcfg, depth-1);
+
+        // pop the cycle
+        cycles->cycle_count -= 1;
+    }
+
+    // clean up memory
+    free(state.branches);
+}
+
+static void run_powerloss_exhaustive(
+        const lfs_emubd_powercycles_t *cycles,
+        size_t cycle_count,
+        const struct test_suite *suite,
+        const struct test_case *case_) {
+    (void)cycles;
+    (void)suite;
+
+    // create block device and configuration
+    lfs_emubd_t bd;
+
+    struct lfs_config cfg = {
+        .context            = &bd,
+        .read               = lfs_emubd_read,
+        .prog               = lfs_emubd_prog,
+        .erase              = lfs_emubd_erase,
+        .sync               = lfs_emubd_sync,
+        .read_size          = READ_SIZE,
+        .prog_size          = PROG_SIZE,
+        .block_size         = BLOCK_SIZE,
+        .block_count        = BLOCK_COUNT,
+        .block_cycles       = BLOCK_CYCLES,
+        .cache_size         = CACHE_SIZE,
+        .lookahead_size     = LOOKAHEAD_SIZE,
+    #ifdef LFS_MULTIVERSION
+        .disk_version       = DISK_VERSION,
+    #endif
+    };
+
+    struct lfs_emubd_config bdcfg = {
+        .read_size          = READ_SIZE,
+        .prog_size          = PROG_SIZE,
+        .erase_size         = ERASE_SIZE,
+        .erase_count        = ERASE_COUNT,
+        .erase_value        = ERASE_VALUE,
+        .erase_cycles       = ERASE_CYCLES,
+        .badblock_behavior  = BADBLOCK_BEHAVIOR,
+        .disk_path          = test_disk_path,
+        .read_sleep         = test_read_sleep,
+        .prog_sleep         = test_prog_sleep,
+        .erase_sleep        = test_erase_sleep,
+        .powerloss_behavior = POWERLOSS_BEHAVIOR,
+        .powerloss_cb       = powerloss_exhaustive_branch,
+        .powerloss_data     = NULL,
+    };
+
+    int err = lfs_emubd_create(&cfg, &bdcfg);
+    if (err) {
+        fprintf(stderr, "error: could not create block device: %d\n", err);
+        exit(-1);
+    }
+
+    // run the test, increasing power-cycles as power-loss events occur
+    printf("running ");
+    perm_printid(suite, case_, NULL, 0);
+    printf("\n");
+
+    // recursively exhaust each layer of powerlosses
+    run_powerloss_exhaustive_layer(
+            &(struct powerloss_exhaustive_cycles){NULL, 0, 0},
+            suite, case_,
+            &cfg, &bdcfg, cycle_count);
+
+    printf("finished ");
+    perm_printid(suite, case_, NULL, 0);
+    printf("\n");
+}
+
+
+const test_powerloss_t builtin_powerlosses[] = {
+    {"none",       run_powerloss_none,       NULL, 0},
+    {"log",        run_powerloss_log,        NULL, 0},
+    {"linear",     run_powerloss_linear,     NULL, 0},
+    {"exhaustive", run_powerloss_exhaustive, NULL, SIZE_MAX},
+    {NULL, NULL, NULL, 0},
+};
+
+const char *const builtin_powerlosses_help[] = {
+    "Run with no power-losses.",
+    "Run with exponentially-decreasing power-losses.",
+    "Run with linearly-decreasing power-losses.",
+    "Run a all permutations of power-losses, this may take a while.",
+    "Run a all permutations of n power-losses.",
+    "Run a custom comma-separated set of power-losses.",
+    "Run a custom leb16-encoded set of power-losses.",
+};
+
+// default to -Pnone,linear, which provides a good heuristic while still
+// running quickly
+const test_powerloss_t *test_powerlosses = (const test_powerloss_t[]){
+    {"none",   run_powerloss_none,   NULL, 0},
+    {"linear", run_powerloss_linear, NULL, 0},
+};
+size_t test_powerloss_count = 2;
+
+static void list_powerlosses(void) {
+    // at least size so that names fit
+    unsigned name_width = 23;
+    for (size_t i = 0; builtin_powerlosses[i].name; i++) {
+        size_t len = strlen(builtin_powerlosses[i].name);
+        if (len > name_width) {
+            name_width = len;
+        }
+    }
+    name_width = 4*((name_width+1+4-1)/4)-1;
+
+    printf("%-*s %s\n", name_width, "scenario", "description");
+    size_t i = 0;
+    for (; builtin_powerlosses[i].name; i++) {
+        printf("%-*s %s\n",
+                name_width,
+                builtin_powerlosses[i].name,
+                builtin_powerlosses_help[i]);
+    }
+
+    // a couple more options with special parsing
+    printf("%-*s %s\n", name_width, "1,2,3",   builtin_powerlosses_help[i+0]);
+    printf("%-*s %s\n", name_width, "{1,2,3}", builtin_powerlosses_help[i+1]);
+    printf("%-*s %s\n", name_width, ":1248g1", builtin_powerlosses_help[i+2]);
+}
+
+
+// global test step count
+size_t test_step = 0;
+
+void perm_run(
+        void *data,
+        const struct test_suite *suite,
+        const struct test_case *case_,
+        const test_powerloss_t *powerloss) {
+    (void)data;
+
+    // skip this step?
+    if (!(test_step >= test_step_start
+            && test_step < test_step_stop
+            && (test_step-test_step_start) % test_step_step == 0)) {
+        test_step += 1;
+        return;
+    }
+    test_step += 1;
+
+    // filter?
+    if (case_->filter && !case_->filter()) {
+        printf("skipped ");
+        perm_printid(suite, case_, NULL, 0);
+        printf("\n");
+        return;
+    }
+
+    powerloss->run(
+            powerloss->cycles, powerloss->cycle_count,
+            suite, case_);
+}
+
+static void run(void) {
+    // ignore disconnected pipes
+    signal(SIGPIPE, SIG_IGN);
+
+    for (size_t t = 0; t < test_id_count; t++) {
+        for (size_t i = 0; i < TEST_SUITE_COUNT; i++) {
+            test_define_suite(&test_suites[i]);
+
+            for (size_t j = 0; j < test_suites[i].case_count; j++) {
+                // does neither suite nor case name match?
+                if (test_ids[t].name && !(
+                        strcmp(test_ids[t].name,
+                            test_suites[i].name) == 0
+                        || strcmp(test_ids[t].name,
+                            test_suites[i].cases[j].name) == 0)) {
+                    continue;
+                }
+
+                case_forperm(
+                        &test_suites[i],
+                        &test_suites[i].cases[j],
+                        test_ids[t].defines,
+                        test_ids[t].define_count,
+                        test_ids[t].cycles,
+                        test_ids[t].cycle_count,
+                        perm_run,
+                        NULL);
+            }
+        }
+    }
+}
+
+
+
+// option handling
+enum opt_flags {
+    OPT_HELP                     = 'h',
+    OPT_SUMMARY                  = 'Y',
+    OPT_LIST_SUITES              = 'l',
+    OPT_LIST_CASES               = 'L',
+    OPT_LIST_SUITE_PATHS         = 1,
+    OPT_LIST_CASE_PATHS          = 2,
+    OPT_LIST_DEFINES             = 3,
+    OPT_LIST_PERMUTATION_DEFINES = 4,
+    OPT_LIST_IMPLICIT_DEFINES    = 5,
+    OPT_LIST_GEOMETRIES          = 6,
+    OPT_LIST_POWERLOSSES         = 7,
+    OPT_DEFINE                   = 'D',
+    OPT_GEOMETRY                 = 'G',
+    OPT_POWERLOSS                = 'P',
+    OPT_STEP                     = 's',
+    OPT_DISK                     = 'd',
+    OPT_TRACE                    = 't',
+    OPT_TRACE_BACKTRACE          = 8,
+    OPT_TRACE_PERIOD             = 9,
+    OPT_TRACE_FREQ               = 10,
+    OPT_READ_SLEEP               = 11,
+    OPT_PROG_SLEEP               = 12,
+    OPT_ERASE_SLEEP              = 13,
+};
+
+const char *short_opts = "hYlLD:G:P:s:d:t:";
+
+const struct option long_opts[] = {
+    {"help",             no_argument,       NULL, OPT_HELP},
+    {"summary",          no_argument,       NULL, OPT_SUMMARY},
+    {"list-suites",      no_argument,       NULL, OPT_LIST_SUITES},
+    {"list-cases",       no_argument,       NULL, OPT_LIST_CASES},
+    {"list-suite-paths", no_argument,       NULL, OPT_LIST_SUITE_PATHS},
+    {"list-case-paths",  no_argument,       NULL, OPT_LIST_CASE_PATHS},
+    {"list-defines",     no_argument,       NULL, OPT_LIST_DEFINES},
+    {"list-permutation-defines",
+                         no_argument,       NULL, OPT_LIST_PERMUTATION_DEFINES},
+    {"list-implicit-defines",
+                         no_argument,       NULL, OPT_LIST_IMPLICIT_DEFINES},
+    {"list-geometries",  no_argument,       NULL, OPT_LIST_GEOMETRIES},
+    {"list-powerlosses", no_argument,       NULL, OPT_LIST_POWERLOSSES},
+    {"define",           required_argument, NULL, OPT_DEFINE},
+    {"geometry",         required_argument, NULL, OPT_GEOMETRY},
+    {"powerloss",        required_argument, NULL, OPT_POWERLOSS},
+    {"step",             required_argument, NULL, OPT_STEP},
+    {"disk",             required_argument, NULL, OPT_DISK},
+    {"trace",            required_argument, NULL, OPT_TRACE},
+    {"trace-backtrace",  no_argument,       NULL, OPT_TRACE_BACKTRACE},
+    {"trace-period",     required_argument, NULL, OPT_TRACE_PERIOD},
+    {"trace-freq",       required_argument, NULL, OPT_TRACE_FREQ},
+    {"read-sleep",       required_argument, NULL, OPT_READ_SLEEP},
+    {"prog-sleep",       required_argument, NULL, OPT_PROG_SLEEP},
+    {"erase-sleep",      required_argument, NULL, OPT_ERASE_SLEEP},
+    {NULL, 0, NULL, 0},
+};
+
+const char *const help_text[] = {
+    "Show this help message.",
+    "Show quick summary.",
+    "List test suites.",
+    "List test cases.",
+    "List the path for each test suite.",
+    "List the path and line number for each test case.",
+    "List all defines in this test-runner.",
+    "List explicit defines in this test-runner.",
+    "List implicit defines in this test-runner.",
+    "List the available disk geometries.",
+    "List the available power-loss scenarios.",
+    "Override a test define.",
+    "Comma-separated list of disk geometries to test.",
+    "Comma-separated list of power-loss scenarios to test.",
+    "Comma-separated range of test permutations to run (start,stop,step).",
+    "Direct block device operations to this file.",
+    "Direct trace output to this file.",
+    "Include a backtrace with every trace statement.",
+    "Sample trace output at this period in cycles.",
+    "Sample trace output at this frequency in hz.",
+    "Artificial read delay in seconds.",
+    "Artificial prog delay in seconds.",
+    "Artificial erase delay in seconds.",
+};
+
+int main(int argc, char **argv) {
+    void (*op)(void) = run;
+
+    size_t test_override_capacity = 0;
+    size_t test_geometry_capacity = 0;
+    size_t test_powerloss_capacity = 0;
+    size_t test_id_capacity = 0;
+
+    // parse options
+    while (true) {
+        int c = getopt_long(argc, argv, short_opts, long_opts, NULL);
+        switch (c) {
+            // generate help message
+            case OPT_HELP: {
+                printf("usage: %s [options] [test_id]\n", argv[0]);
+                printf("\n");
+
+                printf("options:\n");
+                size_t i = 0;
+                while (long_opts[i].name) {
+                    size_t indent;
+                    if (long_opts[i].has_arg == no_argument) {
+                        if (long_opts[i].val >= '0' && long_opts[i].val < 'z') {
+                            indent = printf("  -%c, --%s ",
+                                    long_opts[i].val,
+                                    long_opts[i].name);
+                        } else {
+                            indent = printf("  --%s ",
+                                    long_opts[i].name);
+                        }
+                    } else {
+                        if (long_opts[i].val >= '0' && long_opts[i].val < 'z') {
+                            indent = printf("  -%c %s, --%s %s ",
+                                    long_opts[i].val,
+                                    long_opts[i].name,
+                                    long_opts[i].name,
+                                    long_opts[i].name);
+                        } else {
+                            indent = printf("  --%s %s ",
+                                    long_opts[i].name,
+                                    long_opts[i].name);
+                        }
+                    }
+
+                    // a quick, hacky, byte-level method for text wrapping
+                    size_t len = strlen(help_text[i]);
+                    size_t j = 0;
+                    if (indent < 24) {
+                        printf("%*s %.80s\n",
+                                (int)(24-1-indent),
+                                "",
+                                &help_text[i][j]);
+                        j += 80;
+                    } else {
+                        printf("\n");
+                    }
+
+                    while (j < len) {
+                        printf("%24s%.80s\n", "", &help_text[i][j]);
+                        j += 80;
+                    }
+
+                    i += 1;
+                }
+
+                printf("\n");
+                exit(0);
+            }
+            // summary/list flags
+            case OPT_SUMMARY:
+                op = summary;
+                break;
+            case OPT_LIST_SUITES:
+                op = list_suites;
+                break;
+            case OPT_LIST_CASES:
+                op = list_cases;
+                break;
+            case OPT_LIST_SUITE_PATHS:
+                op = list_suite_paths;
+                break;
+            case OPT_LIST_CASE_PATHS:
+                op = list_case_paths;
+                break;
+            case OPT_LIST_DEFINES:
+                op = list_defines;
+                break;
+            case OPT_LIST_PERMUTATION_DEFINES:
+                op = list_permutation_defines;
+                break;
+            case OPT_LIST_IMPLICIT_DEFINES:
+                op = list_implicit_defines;
+                break;
+            case OPT_LIST_GEOMETRIES:
+                op = list_geometries;
+                break;
+            case OPT_LIST_POWERLOSSES:
+                op = list_powerlosses;
+                break;
+            // configuration
+            case OPT_DEFINE: {
+                // allocate space
+                test_override_t *override = mappend(
+                        (void**)&test_overrides,
+                        sizeof(test_override_t),
+                        &test_override_count,
+                        &test_override_capacity);
+
+                // parse into string key/intmax_t value, cannibalizing the
+                // arg in the process
+                char *sep = strchr(optarg, '=');
+                char *parsed = NULL;
+                if (!sep) {
+                    goto invalid_define;
+                }
+                *sep = '\0';
+                override->name = optarg;
+                optarg = sep+1;
+
+                // parse comma-separated permutations
+                {
+                    override->defines = NULL;
+                    override->permutations = 0;
+                    size_t override_capacity = 0;
+                    while (true) {
+                        optarg += strspn(optarg, " ");
+
+                        if (strncmp(optarg, "range", strlen("range")) == 0) {
+                            // range of values
+                            optarg += strlen("range");
+                            optarg += strspn(optarg, " ");
+                            if (*optarg != '(') {
+                                goto invalid_define;
+                            }
+                            optarg += 1;
+
+                            intmax_t start = strtoumax(optarg, &parsed, 0);
+                            intmax_t stop = -1;
+                            intmax_t step = 1;
+                            // allow empty string for start=0
+                            if (parsed == optarg) {
+                                start = 0;
+                            }
+                            optarg = parsed + strspn(parsed, " ");
+
+                            if (*optarg != ',' && *optarg != ')') {
+                                goto invalid_define;
+                            }
+
+                            if (*optarg == ',') {
+                                optarg += 1;
+                                stop = strtoumax(optarg, &parsed, 0);
+                                // allow empty string for stop=end
+                                if (parsed == optarg) {
+                                    stop = -1;
+                                }
+                                optarg = parsed + strspn(parsed, " ");
+
+                                if (*optarg != ',' && *optarg != ')') {
+                                    goto invalid_define;
+                                }
+
+                                if (*optarg == ',') {
+                                    optarg += 1;
+                                    step = strtoumax(optarg, &parsed, 0);
+                                    // allow empty string for stop=1
+                                    if (parsed == optarg) {
+                                        step = 1;
+                                    }
+                                    optarg = parsed + strspn(parsed, " ");
+
+                                    if (*optarg != ')') {
+                                        goto invalid_define;
+                                    }
+                                }
+                            } else {
+                                // single value = stop only
+                                stop = start;
+                                start = 0;
+                            }
+                            
+                            if (*optarg != ')') {
+                                goto invalid_define;
+                            }
+                            optarg += 1;
+
+                            // calculate the range of values
+                            assert(step != 0);
+                            for (intmax_t i = start;
+                                    (step < 0)
+                                        ? i > stop
+                                        : (uintmax_t)i < (uintmax_t)stop;
+                                    i += step) {
+                                *(intmax_t*)mappend(
+                                        (void**)&override->defines,
+                                        sizeof(intmax_t),
+                                        &override->permutations,
+                                        &override_capacity) = i;
+                            }
+                        } else if (*optarg != '\0') {
+                            // single value
+                            intmax_t define = strtoimax(optarg, &parsed, 0);
+                            if (parsed == optarg) {
+                                goto invalid_define;
+                            }
+                            optarg = parsed + strspn(parsed, " ");
+                            *(intmax_t*)mappend(
+                                    (void**)&override->defines,
+                                    sizeof(intmax_t),
+                                    &override->permutations,
+                                    &override_capacity) = define;
+                        } else {
+                            break;
+                        }
+
+                        if (*optarg == ',') {
+                            optarg += 1;
+                        }
+                    }
+                }
+                assert(override->permutations > 0);
+                break;
+
+invalid_define:
+                fprintf(stderr, "error: invalid define: %s\n", optarg);
+                exit(-1);
+            }
+            case OPT_GEOMETRY: {
+                // reset our geometry scenarios
+                if (test_geometry_capacity > 0) {
+                    free((test_geometry_t*)test_geometries);
+                }
+                test_geometries = NULL;
+                test_geometry_count = 0;
+                test_geometry_capacity = 0;
+
+                // parse the comma separated list of disk geometries
+                while (*optarg) {
+                    // allocate space
+                    test_geometry_t *geometry = mappend(
+                            (void**)&test_geometries,
+                            sizeof(test_geometry_t),
+                            &test_geometry_count,
+                            &test_geometry_capacity);
+
+                    // parse the disk geometry
+                    optarg += strspn(optarg, " ");
+
+                    // named disk geometry
+                    size_t len = strcspn(optarg, " ,");
+                    for (size_t i = 0; builtin_geometries[i].name; i++) {
+                        if (len == strlen(builtin_geometries[i].name)
+                                && memcmp(optarg,
+                                    builtin_geometries[i].name,
+                                    len) == 0)  {
+                            *geometry = builtin_geometries[i];
+                            optarg += len;
+                            goto geometry_next;
+                        }
+                    }
+
+                    // comma-separated read/prog/erase/count
+                    if (*optarg == '{') {
+                        lfs_size_t sizes[4];
+                        size_t count = 0;
+
+                        char *s = optarg + 1;
+                        while (count < 4) {
+                            char *parsed = NULL;
+                            sizes[count] = strtoumax(s, &parsed, 0);
+                            count += 1;
+
+                            s = parsed + strspn(parsed, " ");
+                            if (*s == ',') {
+                                s += 1;
+                                continue;
+                            } else if (*s == '}') {
+                                s += 1;
+                                break;
+                            } else {
+                                goto geometry_unknown;
+                            }
+                        }
+
+                        // allow implicit r=p and p=e for common geometries
+                        memset(geometry, 0, sizeof(test_geometry_t));
+                        if (count >= 3) {
+                            geometry->defines[READ_SIZE_i]
+                                    = TEST_LIT(sizes[0]);
+                            geometry->defines[PROG_SIZE_i]
+                                    = TEST_LIT(sizes[1]);
+                            geometry->defines[ERASE_SIZE_i]
+                                    = TEST_LIT(sizes[2]);
+                        } else if (count >= 2) {
+                            geometry->defines[PROG_SIZE_i]
+                                    = TEST_LIT(sizes[0]);
+                            geometry->defines[ERASE_SIZE_i]
+                                    = TEST_LIT(sizes[1]);
+                        } else {
+                            geometry->defines[ERASE_SIZE_i]
+                                    = TEST_LIT(sizes[0]);
+                        }
+                        if (count >= 4) {
+                            geometry->defines[ERASE_COUNT_i]
+                                    = TEST_LIT(sizes[3]);
+                        }
+                        optarg = s;
+                        goto geometry_next;
+                    }
+
+                    // leb16-encoded read/prog/erase/count
+                    if (*optarg == ':') {
+                        lfs_size_t sizes[4];
+                        size_t count = 0;
+
+                        char *s = optarg + 1;
+                        while (true) {
+                            char *parsed = NULL;
+                            uintmax_t x = leb16_parse(s, &parsed);
+                            if (parsed == s || count >= 4) {
+                                break;
+                            }
+
+                            sizes[count] = x;
+                            count += 1;
+                            s = parsed;
+                        }
+
+                        // allow implicit r=p and p=e for common geometries
+                        memset(geometry, 0, sizeof(test_geometry_t));
+                        if (count >= 3) {
+                            geometry->defines[READ_SIZE_i]
+                                    = TEST_LIT(sizes[0]);
+                            geometry->defines[PROG_SIZE_i]
+                                    = TEST_LIT(sizes[1]);
+                            geometry->defines[ERASE_SIZE_i]
+                                    = TEST_LIT(sizes[2]);
+                        } else if (count >= 2) {
+                            geometry->defines[PROG_SIZE_i]
+                                    = TEST_LIT(sizes[0]);
+                            geometry->defines[ERASE_SIZE_i]
+                                    = TEST_LIT(sizes[1]);
+                        } else {
+                            geometry->defines[ERASE_SIZE_i]
+                                    = TEST_LIT(sizes[0]);
+                        }
+                        if (count >= 4) {
+                            geometry->defines[ERASE_COUNT_i]
+                                    = TEST_LIT(sizes[3]);
+                        }
+                        optarg = s;
+                        goto geometry_next;
+                    }
+
+geometry_unknown:
+                    // unknown scenario?
+                    fprintf(stderr, "error: unknown disk geometry: %s\n",
+                            optarg);
+                    exit(-1);
+
+geometry_next:
+                    optarg += strspn(optarg, " ");
+                    if (*optarg == ',') {
+                        optarg += 1;
+                    } else if (*optarg == '\0') {
+                        break;
+                    } else {
+                        goto geometry_unknown;
+                    }
+                }
+                break;
+            }
+            case OPT_POWERLOSS: {
+                // reset our powerloss scenarios
+                if (test_powerloss_capacity > 0) {
+                    free((test_powerloss_t*)test_powerlosses);
+                }
+                test_powerlosses = NULL;
+                test_powerloss_count = 0;
+                test_powerloss_capacity = 0;
+
+                // parse the comma separated list of power-loss scenarios
+                while (*optarg) {
+                    // allocate space
+                    test_powerloss_t *powerloss = mappend(
+                            (void**)&test_powerlosses,
+                            sizeof(test_powerloss_t),
+                            &test_powerloss_count,
+                            &test_powerloss_capacity);
+
+                    // parse the power-loss scenario
+                    optarg += strspn(optarg, " ");
+
+                    // named power-loss scenario
+                    size_t len = strcspn(optarg, " ,");
+                    for (size_t i = 0; builtin_powerlosses[i].name; i++) {
+                        if (len == strlen(builtin_powerlosses[i].name)
+                                && memcmp(optarg,
+                                    builtin_powerlosses[i].name,
+                                    len) == 0) {
+                            *powerloss = builtin_powerlosses[i];
+                            optarg += len;
+                            goto powerloss_next;
+                        }
+                    }
+
+                    // comma-separated permutation
+                    if (*optarg == '{') {
+                        lfs_emubd_powercycles_t *cycles = NULL;
+                        size_t cycle_count = 0;
+                        size_t cycle_capacity = 0;
+
+                        char *s = optarg + 1;
+                        while (true) {
+                            char *parsed = NULL;
+                            *(lfs_emubd_powercycles_t*)mappend(
+                                    (void**)&cycles,
+                                    sizeof(lfs_emubd_powercycles_t),
+                                    &cycle_count,
+                                    &cycle_capacity)
+                                    = strtoumax(s, &parsed, 0);
+
+                            s = parsed + strspn(parsed, " ");
+                            if (*s == ',') {
+                                s += 1;
+                                continue;
+                            } else if (*s == '}') {
+                                s += 1;
+                                break;
+                            } else {
+                                goto powerloss_unknown;
+                            }
+                        }
+
+                        *powerloss = (test_powerloss_t){
+                            .run = run_powerloss_cycles,
+                            .cycles = cycles,
+                            .cycle_count = cycle_count,
+                        };
+                        optarg = s;
+                        goto powerloss_next;
+                    }
+
+                    // leb16-encoded permutation
+                    if (*optarg == ':') {
+                        lfs_emubd_powercycles_t *cycles = NULL;
+                        size_t cycle_count = 0;
+                        size_t cycle_capacity = 0;
+
+                        char *s = optarg + 1;
+                        while (true) {
+                            char *parsed = NULL;
+                            uintmax_t x = leb16_parse(s, &parsed);
+                            if (parsed == s) {
+                                break;
+                            }
+
+                            *(lfs_emubd_powercycles_t*)mappend(
+                                    (void**)&cycles,
+                                    sizeof(lfs_emubd_powercycles_t),
+                                    &cycle_count,
+                                    &cycle_capacity) = x;
+                            s = parsed;
+                        }
+
+                        *powerloss = (test_powerloss_t){
+                            .run = run_powerloss_cycles,
+                            .cycles = cycles,
+                            .cycle_count = cycle_count,
+                        };
+                        optarg = s;
+                        goto powerloss_next;
+                    }
+
+                    // exhaustive permutations
+                    {
+                        char *parsed = NULL;
+                        size_t count = strtoumax(optarg, &parsed, 0);
+                        if (parsed == optarg) {
+                            goto powerloss_unknown;
+                        }
+                        *powerloss = (test_powerloss_t){
+                            .run = run_powerloss_exhaustive,
+                            .cycles = NULL,
+                            .cycle_count = count,
+                        };
+                        optarg = (char*)parsed;
+                        goto powerloss_next;
+                    }
+
+powerloss_unknown:
+                    // unknown scenario?
+                    fprintf(stderr, "error: unknown power-loss scenario: %s\n",
+                            optarg);
+                    exit(-1);
+
+powerloss_next:
+                    optarg += strspn(optarg, " ");
+                    if (*optarg == ',') {
+                        optarg += 1;
+                    } else if (*optarg == '\0') {
+                        break;
+                    } else {
+                        goto powerloss_unknown;
+                    }
+                }
+                break;
+            }
+            case OPT_STEP: {
+                char *parsed = NULL;
+                test_step_start = strtoumax(optarg, &parsed, 0);
+                test_step_stop = -1;
+                test_step_step = 1;
+                // allow empty string for start=0
+                if (parsed == optarg) {
+                    test_step_start = 0;
+                }
+                optarg = parsed + strspn(parsed, " ");
+
+                if (*optarg != ',' && *optarg != '\0') {
+                    goto step_unknown;
+                }
+
+                if (*optarg == ',') {
+                    optarg += 1;
+                    test_step_stop = strtoumax(optarg, &parsed, 0);
+                    // allow empty string for stop=end
+                    if (parsed == optarg) {
+                        test_step_stop = -1;
+                    }
+                    optarg = parsed + strspn(parsed, " ");
+
+                    if (*optarg != ',' && *optarg != '\0') {
+                        goto step_unknown;
+                    }
+
+                    if (*optarg == ',') {
+                        optarg += 1;
+                        test_step_step = strtoumax(optarg, &parsed, 0);
+                        // allow empty string for stop=1
+                        if (parsed == optarg) {
+                            test_step_step = 1;
+                        }
+                        optarg = parsed + strspn(parsed, " ");
+
+                        if (*optarg != '\0') {
+                            goto step_unknown;
+                        }
+                    }
+                } else {
+                    // single value = stop only
+                    test_step_stop = test_step_start;
+                    test_step_start = 0;
+                }
+
+                break;
+step_unknown:
+                fprintf(stderr, "error: invalid step: %s\n", optarg);
+                exit(-1);
+            }
+            case OPT_DISK:
+                test_disk_path = optarg;
+                break;
+            case OPT_TRACE:
+                test_trace_path = optarg;
+                break;
+            case OPT_TRACE_BACKTRACE:
+                test_trace_backtrace = true;
+                break;
+            case OPT_TRACE_PERIOD: {
+                char *parsed = NULL;
+                test_trace_period = strtoumax(optarg, &parsed, 0);
+                if (parsed == optarg) {
+                    fprintf(stderr, "error: invalid trace-period: %s\n", optarg);
+                    exit(-1);
+                }
+                break;
+            }
+            case OPT_TRACE_FREQ: {
+                char *parsed = NULL;
+                test_trace_freq = strtoumax(optarg, &parsed, 0);
+                if (parsed == optarg) {
+                    fprintf(stderr, "error: invalid trace-freq: %s\n", optarg);
+                    exit(-1);
+                }
+                break;
+            }
+            case OPT_READ_SLEEP: {
+                char *parsed = NULL;
+                double read_sleep = strtod(optarg, &parsed);
+                if (parsed == optarg) {
+                    fprintf(stderr, "error: invalid read-sleep: %s\n", optarg);
+                    exit(-1);
+                }
+                test_read_sleep = read_sleep*1.0e9;
+                break;
+            }
+            case OPT_PROG_SLEEP: {
+                char *parsed = NULL;
+                double prog_sleep = strtod(optarg, &parsed);
+                if (parsed == optarg) {
+                    fprintf(stderr, "error: invalid prog-sleep: %s\n", optarg);
+                    exit(-1);
+                }
+                test_prog_sleep = prog_sleep*1.0e9;
+                break;
+            }
+            case OPT_ERASE_SLEEP: {
+                char *parsed = NULL;
+                double erase_sleep = strtod(optarg, &parsed);
+                if (parsed == optarg) {
+                    fprintf(stderr, "error: invalid erase-sleep: %s\n", optarg);
+                    exit(-1);
+                }
+                test_erase_sleep = erase_sleep*1.0e9;
+                break;
+            }
+            // done parsing
+            case -1:
+                goto getopt_done;
+            // unknown arg, getopt prints a message for us
+            default:
+                exit(-1);
+        }
+    }
+getopt_done: ;
+
+    if (argc > optind) {
+        // reset our test identifier list
+        test_ids = NULL;
+        test_id_count = 0;
+        test_id_capacity = 0;
+    }
+
+    // parse test identifier, if any, cannibalizing the arg in the process
+    for (; argc > optind; optind++) {
+        test_define_t *defines = NULL;
+        size_t define_count = 0;
+        lfs_emubd_powercycles_t *cycles = NULL;
+        size_t cycle_count = 0;
+
+        // parse name, can be suite or case
+        char *name = argv[optind];
+        char *defines_ = strchr(name, ':');
+        if (defines_) {
+            *defines_ = '\0';
+            defines_ += 1;
+        }
+
+        // remove optional path and .toml suffix
+        char *slash = strrchr(name, '/');
+        if (slash) {
+            name = slash+1;
+        }
+
+        size_t name_len = strlen(name);
+        if (name_len > 5 && strcmp(&name[name_len-5], ".toml") == 0) {
+            name[name_len-5] = '\0';
+        }
+
+        if (defines_) {
+            // parse defines
+            char *cycles_ = strchr(defines_, ':');
+            if (cycles_) {
+                *cycles_ = '\0';
+                cycles_ += 1;
+            }
+
+            while (true) {
+                char *parsed;
+                size_t d = leb16_parse(defines_, &parsed);
+                intmax_t v = leb16_parse(parsed, &parsed);
+                if (parsed == defines_) {
+                    break;
+                }
+                defines_ = parsed;
+
+                if (d >= define_count) {
+                    // align to power of two to avoid any superlinear growth
+                    size_t ncount = 1 << lfs_npw2(d+1);
+                    defines = realloc(defines,
+                            ncount*sizeof(test_define_t));
+                    memset(defines+define_count, 0,
+                            (ncount-define_count)*sizeof(test_define_t));
+                    define_count = ncount;
+                }
+                defines[d] = TEST_LIT(v);
+            }
+
+            if (cycles_) {
+                // parse power cycles
+                size_t cycle_capacity = 0;
+                while (*cycles_ != '\0') {
+                    char *parsed = NULL;
+                    *(lfs_emubd_powercycles_t*)mappend(
+                            (void**)&cycles,
+                            sizeof(lfs_emubd_powercycles_t),
+                            &cycle_count,
+                            &cycle_capacity)
+                            = leb16_parse(cycles_, &parsed);
+                    if (parsed == cycles_) {
+                        fprintf(stderr, "error: "
+                                "could not parse test cycles: %s\n",
+                                cycles_);
+                        exit(-1);
+                    }
+                    cycles_ = parsed;
+                }
+            }
+        }
+
+        // append to identifier list
+        *(test_id_t*)mappend(
+                (void**)&test_ids,
+                sizeof(test_id_t),
+                &test_id_count,
+                &test_id_capacity) = (test_id_t){
+            .name = name,
+            .defines = defines,
+            .define_count = define_count,
+            .cycles = cycles,
+            .cycle_count = cycle_count,
+        };
+    }
+
+    // do the thing
+    op();
+
+    // cleanup (need to be done for valgrind testing)
+    test_define_cleanup();
+    if (test_overrides) {
+        for (size_t i = 0; i < test_override_count; i++) {
+            free((void*)test_overrides[i].defines);
+        }
+        free((void*)test_overrides);
+    }
+    if (test_geometry_capacity) {
+        free((void*)test_geometries);
+    }
+    if (test_powerloss_capacity) {
+        for (size_t i = 0; i < test_powerloss_count; i++) {
+            free((void*)test_powerlosses[i].cycles);
+        }
+        free((void*)test_powerlosses);
+    }
+    if (test_id_capacity) {
+        for (size_t i = 0; i < test_id_count; i++) {
+            free((void*)test_ids[i].defines);
+            free((void*)test_ids[i].cycles);
+        }
+        free((void*)test_ids);
+    }
+}
diff --git a/runners/test_runner.h b/runners/test_runner.h
new file mode 100644
index 0000000..4be72e4
--- /dev/null
+++ b/runners/test_runner.h
@@ -0,0 +1,133 @@
+/*
+ * Runner for littlefs tests
+ *
+ * Copyright (c) 2022, The littlefs authors.
+ * SPDX-License-Identifier: BSD-3-Clause
+ */
+#ifndef TEST_RUNNER_H
+#define TEST_RUNNER_H
+
+
+// override LFS_TRACE
+void test_trace(const char *fmt, ...);
+
+#define LFS_TRACE_(fmt, ...) \
+    test_trace("%s:%d:trace: " fmt "%s\n", \
+        __FILE__, \
+        __LINE__, \
+        __VA_ARGS__)
+#define LFS_TRACE(...) LFS_TRACE_(__VA_ARGS__, "")
+#define LFS_EMUBD_TRACE(...) LFS_TRACE_(__VA_ARGS__, "")
+
+
+// note these are indirectly included in any generated files
+#include "bd/lfs_emubd.h"
+#include <stdio.h>
+
+// give source a chance to define feature macros
+#undef _FEATURES_H
+#undef _STDIO_H
+
+
+// generated test configurations
+struct lfs_config;
+
+enum test_flags {
+    TEST_REENTRANT = 0x1,
+};
+typedef uint8_t test_flags_t;
+
+typedef struct test_define {
+    intmax_t (*cb)(void *data);
+    void *data;
+} test_define_t;
+
+struct test_case {
+    const char *name;
+    const char *path;
+    test_flags_t flags;
+    size_t permutations;
+
+    const test_define_t *defines;
+
+    bool (*filter)(void);
+    void (*run)(struct lfs_config *cfg);
+};
+
+struct test_suite {
+    const char *name;
+    const char *path;
+    test_flags_t flags;
+
+    const char *const *define_names;
+    size_t define_count;
+
+    const struct test_case *cases;
+    size_t case_count;
+};
+
+
+// deterministic prng for pseudo-randomness in testes
+uint32_t test_prng(uint32_t *state);
+
+#define TEST_PRNG(state) test_prng(state)
+
+
+// access generated test defines
+intmax_t test_define(size_t define);
+
+#define TEST_DEFINE(i) test_define(i)
+
+// a few preconfigured defines that control how tests run
+ 
+#define READ_SIZE_i          0
+#define PROG_SIZE_i          1
+#define ERASE_SIZE_i         2
+#define ERASE_COUNT_i        3
+#define BLOCK_SIZE_i         4
+#define BLOCK_COUNT_i        5
+#define CACHE_SIZE_i         6
+#define LOOKAHEAD_SIZE_i     7
+#define BLOCK_CYCLES_i       8
+#define ERASE_VALUE_i        9
+#define ERASE_CYCLES_i       10
+#define BADBLOCK_BEHAVIOR_i  11
+#define POWERLOSS_BEHAVIOR_i 12
+#define DISK_VERSION_i       13
+
+#define READ_SIZE           TEST_DEFINE(READ_SIZE_i)
+#define PROG_SIZE           TEST_DEFINE(PROG_SIZE_i)
+#define ERASE_SIZE          TEST_DEFINE(ERASE_SIZE_i)
+#define ERASE_COUNT         TEST_DEFINE(ERASE_COUNT_i)
+#define BLOCK_SIZE          TEST_DEFINE(BLOCK_SIZE_i)
+#define BLOCK_COUNT         TEST_DEFINE(BLOCK_COUNT_i)
+#define CACHE_SIZE          TEST_DEFINE(CACHE_SIZE_i)
+#define LOOKAHEAD_SIZE      TEST_DEFINE(LOOKAHEAD_SIZE_i)
+#define BLOCK_CYCLES        TEST_DEFINE(BLOCK_CYCLES_i)
+#define ERASE_VALUE         TEST_DEFINE(ERASE_VALUE_i)
+#define ERASE_CYCLES        TEST_DEFINE(ERASE_CYCLES_i)
+#define BADBLOCK_BEHAVIOR   TEST_DEFINE(BADBLOCK_BEHAVIOR_i)
+#define POWERLOSS_BEHAVIOR  TEST_DEFINE(POWERLOSS_BEHAVIOR_i)
+#define DISK_VERSION        TEST_DEFINE(DISK_VERSION_i)
+
+#define TEST_IMPLICIT_DEFINES \
+    TEST_DEF(READ_SIZE,          PROG_SIZE) \
+    TEST_DEF(PROG_SIZE,          ERASE_SIZE) \
+    TEST_DEF(ERASE_SIZE,         0) \
+    TEST_DEF(ERASE_COUNT,        (1024*1024)/ERASE_SIZE) \
+    TEST_DEF(BLOCK_SIZE,         ERASE_SIZE) \
+    TEST_DEF(BLOCK_COUNT,        ERASE_COUNT/lfs_max(BLOCK_SIZE/ERASE_SIZE,1)) \
+    TEST_DEF(CACHE_SIZE,         lfs_max(64,lfs_max(READ_SIZE,PROG_SIZE))) \
+    TEST_DEF(LOOKAHEAD_SIZE,     16) \
+    TEST_DEF(BLOCK_CYCLES,       -1) \
+    TEST_DEF(ERASE_VALUE,        0xff) \
+    TEST_DEF(ERASE_CYCLES,       0) \
+    TEST_DEF(BADBLOCK_BEHAVIOR,  LFS_EMUBD_BADBLOCK_PROGERROR) \
+    TEST_DEF(POWERLOSS_BEHAVIOR, LFS_EMUBD_POWERLOSS_NOOP) \
+    TEST_DEF(DISK_VERSION,       0)
+
+#define TEST_GEOMETRY_DEFINE_COUNT 4
+#define TEST_IMPLICIT_DEFINE_COUNT 14
+
+
+#endif
diff --git a/scripts/bench.py b/scripts/bench.py
new file mode 100644
index 0000000..f22841e
--- /dev/null
+++ b/scripts/bench.py
@@ -0,0 +1,1430 @@
+#!/usr/bin/env python3
+#
+# Script to compile and runs benches.
+#
+# Example:
+# ./scripts/bench.py runners/bench_runner -b
+#
+# Copyright (c) 2022, The littlefs authors.
+# SPDX-License-Identifier: BSD-3-Clause
+#
+
+import collections as co
+import csv
+import errno
+import glob
+import itertools as it
+import math as m
+import os
+import pty
+import re
+import shlex
+import shutil
+import signal
+import subprocess as sp
+import threading as th
+import time
+import toml
+
+
+RUNNER_PATH = './runners/bench_runner'
+HEADER_PATH = 'runners/bench_runner.h'
+
+GDB_PATH = ['gdb']
+VALGRIND_PATH = ['valgrind']
+PERF_SCRIPT = ['./scripts/perf.py']
+
+
+def openio(path, mode='r', buffering=-1):
+    # allow '-' for stdin/stdout
+    if path == '-':
+        if mode == 'r':
+            return os.fdopen(os.dup(sys.stdin.fileno()), mode, buffering)
+        else:
+            return os.fdopen(os.dup(sys.stdout.fileno()), mode, buffering)
+    else:
+        return open(path, mode, buffering)
+
+class BenchCase:
+    # create a BenchCase object from a config
+    def __init__(self, config, args={}):
+        self.name = config.pop('name')
+        self.path = config.pop('path')
+        self.suite = config.pop('suite')
+        self.lineno = config.pop('lineno', None)
+        self.if_ = config.pop('if', None)
+        if isinstance(self.if_, bool):
+            self.if_ = 'true' if self.if_ else 'false'
+        self.code = config.pop('code')
+        self.code_lineno = config.pop('code_lineno', None)
+        self.in_ = config.pop('in',
+            config.pop('suite_in', None))
+
+        # figure out defines and build possible permutations
+        self.defines = set()
+        self.permutations = []
+
+        # defines can be a dict or a list or dicts
+        suite_defines = config.pop('suite_defines', {})
+        if not isinstance(suite_defines, list):
+            suite_defines = [suite_defines]
+        defines = config.pop('defines', {})
+        if not isinstance(defines, list):
+            defines = [defines]
+
+        def csplit(v):
+            # split commas but only outside of parens
+            parens = 0
+            i_ = 0
+            for i in range(len(v)):
+                if v[i] == ',' and parens == 0:
+                    yield v[i_:i]
+                    i_ = i+1
+                elif v[i] in '([{':
+                    parens += 1
+                elif v[i] in '}])':
+                    parens -= 1
+            if v[i_:].strip():
+                yield v[i_:]
+
+        def parse_define(v):
+            # a define entry can be a list
+            if isinstance(v, list):
+                for v_ in v:
+                    yield from parse_define(v_)
+            # or a string
+            elif isinstance(v, str):
+                # which can be comma-separated values, with optional
+                # range statements. This matches the runtime define parser in
+                # the runner itself.
+                for v_ in csplit(v):
+                    m = re.search(r'\brange\b\s*\('
+                        '(?P<start>[^,\s]*)'
+                        '\s*(?:,\s*(?P<stop>[^,\s]*)'
+                        '\s*(?:,\s*(?P<step>[^,\s]*)\s*)?)?\)',
+                        v_)
+                    if m:
+                        start = (int(m.group('start'), 0)
+                            if m.group('start') else 0)
+                        stop = (int(m.group('stop'), 0)
+                            if m.group('stop') else None)
+                        step = (int(m.group('step'), 0)
+                            if m.group('step') else 1)
+                        if m.lastindex <= 1:
+                            start, stop = 0, start
+                        for x in range(start, stop, step):
+                            yield from parse_define('%s(%d)%s' % (
+                                v_[:m.start()], x, v_[m.end():]))
+                    else:
+                        yield v_
+            # or a literal value
+            elif isinstance(v, bool):
+                yield 'true' if v else 'false'
+            else:
+                yield v
+
+        # build possible permutations
+        for suite_defines_ in suite_defines:
+            self.defines |= suite_defines_.keys()
+            for defines_ in defines:
+                self.defines |= defines_.keys()
+                self.permutations.extend(dict(perm) for perm in it.product(*(
+                    [(k, v) for v in parse_define(vs)]
+                    for k, vs in sorted((suite_defines_ | defines_).items()))))
+
+        for k in config.keys():
+            print('%swarning:%s in %s, found unused key %r' % (
+                '\x1b[01;33m' if args['color'] else '',
+                '\x1b[m' if args['color'] else '',
+                self.name,
+                k),
+                file=sys.stderr)
+
+
+class BenchSuite:
+    # create a BenchSuite object from a toml file
+    def __init__(self, path, args={}):
+        self.path = path
+        self.name = os.path.basename(path)
+        if self.name.endswith('.toml'):
+            self.name = self.name[:-len('.toml')]
+
+        # load toml file and parse bench cases
+        with open(self.path) as f:
+            # load benches
+            config = toml.load(f)
+
+            # find line numbers
+            f.seek(0)
+            case_linenos = []
+            code_linenos = []
+            for i, line in enumerate(f):
+                match = re.match(
+                    '(?P<case>\[\s*cases\s*\.\s*(?P<name>\w+)\s*\])'
+                        '|' '(?P<code>code\s*=)',
+                    line)
+                if match and match.group('case'):
+                    case_linenos.append((i+1, match.group('name')))
+                elif match and match.group('code'):
+                    code_linenos.append(i+2)
+
+            # sort in case toml parsing did not retain order
+            case_linenos.sort()
+
+            cases = config.pop('cases')
+            for (lineno, name), (nlineno, _) in it.zip_longest(
+                    case_linenos, case_linenos[1:],
+                    fillvalue=(float('inf'), None)):
+                code_lineno = min(
+                    (l for l in code_linenos if l >= lineno and l < nlineno),
+                    default=None)
+                cases[name]['lineno'] = lineno
+                cases[name]['code_lineno'] = code_lineno
+
+            self.if_ = config.pop('if', None)
+            if isinstance(self.if_, bool):
+                self.if_ = 'true' if self.if_ else 'false'
+
+            self.code = config.pop('code', None)
+            self.code_lineno = min(
+                (l for l in code_linenos
+                    if not case_linenos or l < case_linenos[0][0]),
+                default=None)
+
+            # a couple of these we just forward to all cases
+            defines = config.pop('defines', {})
+            in_ = config.pop('in', None)
+
+            self.cases = []
+            for name, case in sorted(cases.items(),
+                    key=lambda c: c[1].get('lineno')):
+                self.cases.append(BenchCase(config={
+                    'name': name,
+                    'path': path + (':%d' % case['lineno']
+                        if 'lineno' in case else ''),
+                    'suite': self.name,
+                    'suite_defines': defines,
+                    'suite_in': in_,
+                    **case},
+                    args=args))
+
+            # combine per-case defines
+            self.defines = set.union(*(
+                set(case.defines) for case in self.cases))
+
+        for k in config.keys():
+            print('%swarning:%s in %s, found unused key %r' % (
+                '\x1b[01;33m' if args['color'] else '',
+                '\x1b[m' if args['color'] else '',
+                self.name,
+                k),
+                file=sys.stderr)
+
+
+
+def compile(bench_paths, **args):
+    # find .toml files
+    paths = []
+    for path in bench_paths:
+        if os.path.isdir(path):
+            path = path + '/*.toml'
+
+        for path in glob.glob(path):
+            paths.append(path)
+
+    if not paths:
+        print('no bench suites found in %r?' % bench_paths)
+        sys.exit(-1)
+
+    # load the suites
+    suites = [BenchSuite(path, args) for path in paths]
+    suites.sort(key=lambda s: s.name)
+
+    # check for name conflicts, these will cause ambiguity problems later
+    # when running benches
+    seen = {}
+    for suite in suites:
+        if suite.name in seen:
+            print('%swarning:%s conflicting suite %r, %s and %s' % (
+                '\x1b[01;33m' if args['color'] else '',
+                '\x1b[m' if args['color'] else '',
+                suite.name,
+                suite.path,
+                seen[suite.name].path),
+                file=sys.stderr)
+        seen[suite.name] = suite
+
+        for case in suite.cases:
+            # only allow conflicts if a case and its suite share a name
+            if case.name in seen and not (
+                    isinstance(seen[case.name], BenchSuite)
+                    and seen[case.name].cases == [case]):
+                print('%swarning:%s conflicting case %r, %s and %s' % (
+                    '\x1b[01;33m' if args['color'] else '',
+                    '\x1b[m' if args['color'] else '',
+                    case.name,
+                    case.path,
+                    seen[case.name].path),
+                    file=sys.stderr)
+            seen[case.name] = case
+
+    # we can only compile one bench suite at a time
+    if not args.get('source'):
+        if len(suites) > 1:
+            print('more than one bench suite for compilation? (%r)' % bench_paths)
+            sys.exit(-1)
+
+        suite = suites[0]
+
+    # write generated bench source
+    if 'output' in args:
+        with openio(args['output'], 'w') as f:
+            _write = f.write
+            def write(s):
+                f.lineno += s.count('\n')
+                _write(s)
+            def writeln(s=''):
+                f.lineno += s.count('\n') + 1
+                _write(s)
+                _write('\n')
+            f.lineno = 1
+            f.write = write
+            f.writeln = writeln
+
+            f.writeln("// Generated by %s:" % sys.argv[0])
+            f.writeln("//")
+            f.writeln("// %s" % ' '.join(sys.argv))
+            f.writeln("//")
+            f.writeln()
+
+            # include bench_runner.h in every generated file
+            f.writeln("#include \"%s\"" % args['include'])
+            f.writeln()
+
+            # write out generated functions, this can end up in different
+            # files depending on the "in" attribute
+            #
+            # note it's up to the specific generated file to declare
+            # the bench defines
+            def write_case_functions(f, suite, case):
+                    # create case define functions
+                    if case.defines:
+                        # deduplicate defines by value to try to reduce the
+                        # number of functions we generate
+                        define_cbs = {}
+                        for i, defines in enumerate(case.permutations):
+                            for k, v in sorted(defines.items()):
+                                if v not in define_cbs:
+                                    name = ('__bench__%s__%s__%d'
+                                        % (case.name, k, i))
+                                    define_cbs[v] = name
+                                    f.writeln('intmax_t %s('
+                                        '__attribute__((unused)) '
+                                        'void *data) {' % name)
+                                    f.writeln(4*' '+'return %s;' % v)
+                                    f.writeln('}')
+                                    f.writeln()
+                        f.writeln('const bench_define_t '
+                            '__bench__%s__defines[]['
+                            'BENCH_IMPLICIT_DEFINE_COUNT+%d] = {'
+                            % (case.name, len(suite.defines)))
+                        for defines in case.permutations:
+                            f.writeln(4*' '+'{')
+                            for k, v in sorted(defines.items()):
+                                f.writeln(8*' '+'[%-24s] = {%s, NULL},' % (
+                                    k+'_i', define_cbs[v]))
+                            f.writeln(4*' '+'},')
+                        f.writeln('};')
+                        f.writeln()
+
+                    # create case filter function
+                    if suite.if_ is not None or case.if_ is not None:
+                        f.writeln('bool __bench__%s__filter(void) {'
+                            % (case.name))
+                        f.writeln(4*' '+'return %s;'
+                            % ' && '.join('(%s)' % if_
+                                for if_ in [suite.if_, case.if_]
+                                if if_ is not None))
+                        f.writeln('}')
+                        f.writeln()
+
+                    # create case run function
+                    f.writeln('void __bench__%s__run('
+                        '__attribute__((unused)) struct lfs_config *cfg) {'
+                        % (case.name))
+                    f.writeln(4*' '+'// bench case %s' % case.name)
+                    if case.code_lineno is not None:
+                        f.writeln(4*' '+'#line %d "%s"'
+                            % (case.code_lineno, suite.path))
+                    f.write(case.code)
+                    if case.code_lineno is not None:
+                        f.writeln(4*' '+'#line %d "%s"'
+                            % (f.lineno+1, args['output']))
+                    f.writeln('}')
+                    f.writeln()
+
+            if not args.get('source'):
+                if suite.code is not None:
+                    if suite.code_lineno is not None:
+                        f.writeln('#line %d "%s"'
+                            % (suite.code_lineno, suite.path))
+                    f.write(suite.code)
+                    if suite.code_lineno is not None:
+                        f.writeln('#line %d "%s"'
+                            % (f.lineno+1, args['output']))
+                    f.writeln()
+
+                if suite.defines:
+                    for i, define in enumerate(sorted(suite.defines)):
+                        f.writeln('#ifndef %s' % define)
+                        f.writeln('#define %-24s '
+                            'BENCH_IMPLICIT_DEFINE_COUNT+%d' % (define+'_i', i))
+                        f.writeln('#define %-24s '
+                            'BENCH_DEFINE(%s)' % (define, define+'_i'))
+                        f.writeln('#endif')
+                    f.writeln()
+
+                # create case functions
+                for case in suite.cases:
+                    if case.in_ is None:
+                        write_case_functions(f, suite, case)
+                    else:
+                        if case.defines:
+                            f.writeln('extern const bench_define_t '
+                                '__bench__%s__defines[]['
+                                'BENCH_IMPLICIT_DEFINE_COUNT+%d];'
+                                % (case.name, len(suite.defines)))
+                        if suite.if_ is not None or case.if_ is not None:
+                            f.writeln('extern bool __bench__%s__filter('
+                                'void);'
+                                % (case.name))
+                        f.writeln('extern void __bench__%s__run('
+                            'struct lfs_config *cfg);'
+                            % (case.name))
+                        f.writeln()
+
+                # create suite struct
+                #
+                # note we place this in the custom bench_suites section with
+                # minimum alignment, otherwise GCC ups the alignment to
+                # 32-bytes for some reason
+                f.writeln('__attribute__((section("_bench_suites"), '
+                    'aligned(1)))')
+                f.writeln('const struct bench_suite __bench__%s__suite = {'
+                    % suite.name)
+                f.writeln(4*' '+'.name = "%s",' % suite.name)
+                f.writeln(4*' '+'.path = "%s",' % suite.path)
+                f.writeln(4*' '+'.flags = 0,')
+                if suite.defines:
+                    # create suite define names
+                    f.writeln(4*' '+'.define_names = (const char *const['
+                        'BENCH_IMPLICIT_DEFINE_COUNT+%d]){' % (
+                        len(suite.defines)))
+                    for k in sorted(suite.defines):
+                        f.writeln(8*' '+'[%-24s] = "%s",' % (k+'_i', k))
+                    f.writeln(4*' '+'},')
+                    f.writeln(4*' '+'.define_count = '
+                        'BENCH_IMPLICIT_DEFINE_COUNT+%d,' % len(suite.defines))
+                f.writeln(4*' '+'.cases = (const struct bench_case[]){')
+                for case in suite.cases:
+                    # create case structs
+                    f.writeln(8*' '+'{')
+                    f.writeln(12*' '+'.name = "%s",' % case.name)
+                    f.writeln(12*' '+'.path = "%s",' % case.path)
+                    f.writeln(12*' '+'.flags = 0,')
+                    f.writeln(12*' '+'.permutations = %d,'
+                        % len(case.permutations))
+                    if case.defines:
+                        f.writeln(12*' '+'.defines '
+                            '= (const bench_define_t*)__bench__%s__defines,'
+                            % (case.name))
+                    if suite.if_ is not None or case.if_ is not None:
+                        f.writeln(12*' '+'.filter = __bench__%s__filter,'
+                            % (case.name))
+                    f.writeln(12*' '+'.run = __bench__%s__run,'
+                        % (case.name))
+                    f.writeln(8*' '+'},')
+                f.writeln(4*' '+'},')
+                f.writeln(4*' '+'.case_count = %d,' % len(suite.cases))
+                f.writeln('};')
+                f.writeln()
+
+            else:
+                # copy source
+                f.writeln('#line 1 "%s"' % args['source'])
+                with open(args['source']) as sf:
+                    shutil.copyfileobj(sf, f)
+                f.writeln()
+
+                # write any internal benches
+                for suite in suites:
+                    for case in suite.cases:
+                        if (case.in_ is not None
+                                and os.path.normpath(case.in_)
+                                    == os.path.normpath(args['source'])):
+                            # write defines, but note we need to undef any
+                            # new defines since we're in someone else's file
+                            if suite.defines:
+                                for i, define in enumerate(
+                                        sorted(suite.defines)):
+                                    f.writeln('#ifndef %s' % define)
+                                    f.writeln('#define %-24s '
+                                        'BENCH_IMPLICIT_DEFINE_COUNT+%d' % (
+                                        define+'_i', i))
+                                    f.writeln('#define %-24s '
+                                        'BENCH_DEFINE(%s)' % (
+                                        define, define+'_i'))
+                                    f.writeln('#define '
+                                        '__BENCH__%s__NEEDS_UNDEF' % (
+                                        define))
+                                    f.writeln('#endif')
+                                f.writeln()
+
+                            write_case_functions(f, suite, case)
+
+                            if suite.defines:
+                                for define in sorted(suite.defines):
+                                    f.writeln('#ifdef __BENCH__%s__NEEDS_UNDEF'
+                                        % define)
+                                    f.writeln('#undef __BENCH__%s__NEEDS_UNDEF'
+                                        % define)
+                                    f.writeln('#undef %s' % define)
+                                    f.writeln('#undef %s' % (define+'_i'))
+                                    f.writeln('#endif')
+                                f.writeln()
+
+def find_runner(runner, **args):
+    cmd = runner.copy()
+
+    # run under some external command?
+    if args.get('exec'):
+        cmd[:0] = args['exec']
+
+    # run under valgrind?
+    if args.get('valgrind'):
+        cmd[:0] = args['valgrind_path'] + [
+            '--leak-check=full',
+            '--track-origins=yes',
+            '--error-exitcode=4',
+            '-q']
+
+    # run under perf?
+    if args.get('perf'):
+        cmd[:0] = args['perf_script'] + list(filter(None, [
+            '-R',
+            '--perf-freq=%s' % args['perf_freq']
+                if args.get('perf_freq') else None,
+            '--perf-period=%s' % args['perf_period']
+                if args.get('perf_period') else None,
+            '--perf-events=%s' % args['perf_events']
+                if args.get('perf_events') else None,
+            '--perf-path=%s' % args['perf_path']
+                if args.get('perf_path') else None,
+            '-o%s' % args['perf']]))
+
+    # other context
+    if args.get('geometry'):
+        cmd.append('-G%s' % args['geometry'])
+    if args.get('disk'):
+        cmd.append('-d%s' % args['disk'])
+    if args.get('trace'):
+        cmd.append('-t%s' % args['trace'])
+    if args.get('trace_backtrace'):
+        cmd.append('--trace-backtrace')
+    if args.get('trace_period'):
+        cmd.append('--trace-period=%s' % args['trace_period'])
+    if args.get('trace_freq'):
+        cmd.append('--trace-freq=%s' % args['trace_freq'])
+    if args.get('read_sleep'):
+        cmd.append('--read-sleep=%s' % args['read_sleep'])
+    if args.get('prog_sleep'):
+        cmd.append('--prog-sleep=%s' % args['prog_sleep'])
+    if args.get('erase_sleep'):
+        cmd.append('--erase-sleep=%s' % args['erase_sleep'])
+
+    # defines?
+    if args.get('define'):
+        for define in args.get('define'):
+            cmd.append('-D%s' % define)
+
+    return cmd
+
+def list_(runner, bench_ids=[], **args):
+    cmd = find_runner(runner, **args) + bench_ids
+    if args.get('summary'):          cmd.append('--summary')
+    if args.get('list_suites'):      cmd.append('--list-suites')
+    if args.get('list_cases'):       cmd.append('--list-cases')
+    if args.get('list_suite_paths'): cmd.append('--list-suite-paths')
+    if args.get('list_case_paths'):  cmd.append('--list-case-paths')
+    if args.get('list_defines'):     cmd.append('--list-defines')
+    if args.get('list_permutation_defines'):
+                                     cmd.append('--list-permutation-defines')
+    if args.get('list_implicit_defines'):
+                                     cmd.append('--list-implicit-defines')
+    if args.get('list_geometries'):  cmd.append('--list-geometries')
+
+    if args.get('verbose'):
+        print(' '.join(shlex.quote(c) for c in cmd))
+    return sp.call(cmd)
+
+
+def find_perms(runner_, ids=[], **args):
+    case_suites = {}
+    expected_case_perms = co.defaultdict(lambda: 0)
+    expected_perms = 0
+    total_perms = 0
+
+    # query cases from the runner
+    cmd = runner_ + ['--list-cases'] + ids
+    if args.get('verbose'):
+        print(' '.join(shlex.quote(c) for c in cmd))
+    proc = sp.Popen(cmd,
+        stdout=sp.PIPE,
+        stderr=sp.PIPE if not args.get('verbose') else None,
+        universal_newlines=True,
+        errors='replace',
+        close_fds=False)
+    pattern = re.compile(
+        '^(?P<case>[^\s]+)'
+            '\s+(?P<flags>[^\s]+)'
+            '\s+(?P<filtered>\d+)/(?P<perms>\d+)')
+    # skip the first line
+    for line in it.islice(proc.stdout, 1, None):
+        m = pattern.match(line)
+        if m:
+            filtered = int(m.group('filtered'))
+            perms = int(m.group('perms'))
+            expected_case_perms[m.group('case')] += filtered
+            expected_perms += filtered
+            total_perms += perms
+    proc.wait()
+    if proc.returncode != 0:
+        if not args.get('verbose'):
+            for line in proc.stderr:
+                sys.stdout.write(line)
+        sys.exit(-1)
+
+    # get which suite each case belongs to via paths
+    cmd = runner_ + ['--list-case-paths'] + ids
+    if args.get('verbose'):
+        print(' '.join(shlex.quote(c) for c in cmd))
+    proc = sp.Popen(cmd,
+        stdout=sp.PIPE,
+        stderr=sp.PIPE if not args.get('verbose') else None,
+        universal_newlines=True,
+        errors='replace',
+        close_fds=False)
+    pattern = re.compile(
+        '^(?P<case>[^\s]+)'
+            '\s+(?P<path>[^:]+):(?P<lineno>\d+)')
+    # skip the first line
+    for line in it.islice(proc.stdout, 1, None):
+        m = pattern.match(line)
+        if m:
+            path = m.group('path')
+            # strip path/suffix here
+            suite = os.path.basename(path)
+            if suite.endswith('.toml'):
+                suite = suite[:-len('.toml')]
+            case_suites[m.group('case')] = suite
+    proc.wait()
+    if proc.returncode != 0:
+        if not args.get('verbose'):
+            for line in proc.stderr:
+                sys.stdout.write(line)
+        sys.exit(-1)
+
+    # figure out expected suite perms
+    expected_suite_perms = co.defaultdict(lambda: 0)
+    for case, suite in case_suites.items():
+        expected_suite_perms[suite] += expected_case_perms[case]
+
+    return (
+        case_suites,
+        expected_suite_perms,
+        expected_case_perms,
+        expected_perms,
+        total_perms)
+
+def find_path(runner_, id, **args):
+    path = None
+    # query from runner
+    cmd = runner_ + ['--list-case-paths', id]
+    if args.get('verbose'):
+        print(' '.join(shlex.quote(c) for c in cmd))
+    proc = sp.Popen(cmd,
+        stdout=sp.PIPE,
+        stderr=sp.PIPE if not args.get('verbose') else None,
+        universal_newlines=True,
+        errors='replace',
+        close_fds=False)
+    pattern = re.compile(
+        '^(?P<case>[^\s]+)'
+            '\s+(?P<path>[^:]+):(?P<lineno>\d+)')
+    # skip the first line
+    for line in it.islice(proc.stdout, 1, None):
+        m = pattern.match(line)
+        if m and path is None:
+            path_ = m.group('path')
+            lineno = int(m.group('lineno'))
+            path = (path_, lineno)
+    proc.wait()
+    if proc.returncode != 0:
+        if not args.get('verbose'):
+            for line in proc.stderr:
+                sys.stdout.write(line)
+        sys.exit(-1)
+
+    return path
+
+def find_defines(runner_, id, **args):
+    # query permutation defines from runner
+    cmd = runner_ + ['--list-permutation-defines', id]
+    if args.get('verbose'):
+        print(' '.join(shlex.quote(c) for c in cmd))
+    proc = sp.Popen(cmd,
+        stdout=sp.PIPE,
+        stderr=sp.PIPE if not args.get('verbose') else None,
+        universal_newlines=True,
+        errors='replace',
+        close_fds=False)
+    defines = co.OrderedDict()
+    pattern = re.compile('^(?P<define>\w+)=(?P<value>.+)')
+    for line in proc.stdout:
+        m = pattern.match(line)
+        if m:
+            define = m.group('define')
+            value = m.group('value')
+            defines[define] = value
+    proc.wait()
+    if proc.returncode != 0:
+        if not args.get('verbose'):
+            for line in proc.stderr:
+                sys.stdout.write(line)
+        sys.exit(-1)
+
+    return defines
+
+
+# Thread-safe CSV writer
+class BenchOutput:
+    def __init__(self, path, head=None, tail=None):
+        self.f = openio(path, 'w+', 1)
+        self.lock = th.Lock()
+        self.head = head or []
+        self.tail = tail or []
+        self.writer = csv.DictWriter(self.f, self.head + self.tail)
+        self.rows = []
+
+    def close(self):
+        self.f.close()
+
+    def __enter__(self):
+        return self
+
+    def __exit__(self, *_):
+        self.f.close()
+
+    def writerow(self, row):
+        with self.lock:
+            self.rows.append(row)
+            if all(k in self.head or k in self.tail for k in row.keys()):
+                # can simply append
+                self.writer.writerow(row)
+            else:
+                # need to rewrite the file
+                self.head.extend(row.keys() - (self.head + self.tail))
+                self.f.seek(0)
+                self.f.truncate()
+                self.writer = csv.DictWriter(self.f, self.head + self.tail)
+                self.writer.writeheader()
+                for row in self.rows:
+                    self.writer.writerow(row)
+
+# A bench failure
+class BenchFailure(Exception):
+    def __init__(self, id, returncode, stdout, assert_=None):
+        self.id = id
+        self.returncode = returncode
+        self.stdout = stdout
+        self.assert_ = assert_
+
+def run_stage(name, runner_, ids, stdout_, trace_, output_, **args):
+    # get expected suite/case/perm counts
+    (case_suites,
+        expected_suite_perms,
+        expected_case_perms,
+        expected_perms,
+        total_perms) = find_perms(runner_, ids, **args)
+
+    passed_suite_perms = co.defaultdict(lambda: 0)
+    passed_case_perms = co.defaultdict(lambda: 0)
+    passed_perms = 0
+    readed = 0
+    proged = 0
+    erased = 0
+    failures = []
+    killed = False
+
+    pattern = re.compile('^(?:'
+            '(?P<op>running|finished|skipped|powerloss)'
+                ' (?P<id>(?P<case>[^:]+)[^\s]*)'
+                '(?: (?P<readed>\d+))?'
+                '(?: (?P<proged>\d+))?'
+                '(?: (?P<erased>\d+))?'
+            '|' '(?P<path>[^:]+):(?P<lineno>\d+):(?P<op_>assert):'
+                ' *(?P<message>.*)'
+        ')$')
+    locals = th.local()
+    children = set()
+
+    def run_runner(runner_, ids=[]):
+        nonlocal passed_suite_perms
+        nonlocal passed_case_perms
+        nonlocal passed_perms
+        nonlocal readed
+        nonlocal proged
+        nonlocal erased
+        nonlocal locals
+
+        # run the benches!
+        cmd = runner_ + ids
+        if args.get('verbose'):
+            print(' '.join(shlex.quote(c) for c in cmd))
+
+        mpty, spty = pty.openpty()
+        proc = sp.Popen(cmd, stdout=spty, stderr=spty, close_fds=False)
+        os.close(spty)
+        children.add(proc)
+        mpty = os.fdopen(mpty, 'r', 1)
+
+        last_id = None
+        last_stdout = co.deque(maxlen=args.get('context', 5) + 1)
+        last_assert = None
+        try:
+            while True:
+                # parse a line for state changes
+                try:
+                    line = mpty.readline()
+                except OSError as e:
+                    if e.errno != errno.EIO:
+                        raise
+                    break
+                if not line:
+                    break
+                last_stdout.append(line)
+                if stdout_:
+                    try:
+                        stdout_.write(line)
+                        stdout_.flush()
+                    except BrokenPipeError:
+                        pass
+
+                m = pattern.match(line)
+                if m:
+                    op = m.group('op') or m.group('op_')
+                    if op == 'running':
+                        locals.seen_perms += 1
+                        last_id = m.group('id')
+                        last_stdout.clear()
+                        last_assert = None
+                    elif op == 'finished':
+                        case = m.group('case')
+                        suite = case_suites[case]
+                        readed_ = int(m.group('readed'))
+                        proged_ = int(m.group('proged'))
+                        erased_ = int(m.group('erased'))
+                        passed_suite_perms[suite] += 1
+                        passed_case_perms[case] += 1
+                        passed_perms += 1
+                        readed += readed_
+                        proged += proged_
+                        erased += erased_
+                        if output_:
+                            # get defines and write to csv
+                            defines = find_defines(
+                                runner_, m.group('id'), **args)
+                            output_.writerow({
+                                'suite': suite,
+                                'case': case,
+                                'bench_readed': readed_,
+                                'bench_proged': proged_,
+                                'bench_erased': erased_,
+                                **defines})
+                    elif op == 'skipped':
+                        locals.seen_perms += 1
+                    elif op == 'assert':
+                        last_assert = (
+                            m.group('path'),
+                            int(m.group('lineno')),
+                            m.group('message'))
+                        # go ahead and kill the process, aborting takes a while
+                        if args.get('keep_going'):
+                            proc.kill()
+        except KeyboardInterrupt:
+            raise BenchFailure(last_id, 1, list(last_stdout))
+        finally:
+            children.remove(proc)
+            mpty.close()
+
+        proc.wait()
+        if proc.returncode != 0:
+            raise BenchFailure(
+                last_id,
+                proc.returncode,
+                list(last_stdout),
+                last_assert)
+
+    def run_job(runner_, ids=[], start=None, step=None):
+        nonlocal failures
+        nonlocal killed
+        nonlocal locals
+
+        start = start or 0
+        step = step or 1
+        while start < total_perms:
+            job_runner = runner_.copy()
+            if args.get('isolate') or args.get('valgrind'):
+                job_runner.append('-s%s,%s,%s' % (start, start+step, step))
+            else:
+                job_runner.append('-s%s,,%s' % (start, step))
+
+            try:
+                # run the benches
+                locals.seen_perms = 0
+                run_runner(job_runner, ids)
+                assert locals.seen_perms > 0
+                start += locals.seen_perms*step
+
+            except BenchFailure as failure:
+                # keep track of failures
+                if output_:
+                    case, _ = failure.id.split(':', 1)
+                    suite = case_suites[case]
+                    # get defines and write to csv
+                    defines = find_defines(runner_, failure.id, **args)
+                    output_.writerow({
+                        'suite': suite,
+                        'case': case,
+                        **defines})
+
+                # race condition for multiple failures?
+                if failures and not args.get('keep_going'):
+                    break
+
+                failures.append(failure)
+
+                if args.get('keep_going') and not killed:
+                    # resume after failed bench
+                    assert locals.seen_perms > 0
+                    start += locals.seen_perms*step
+                    continue
+                else:
+                    # stop other benches
+                    killed = True
+                    for child in children.copy():
+                        child.kill()
+                    break
+
+
+    # parallel jobs?
+    runners = []
+    if 'jobs' in args:
+        for job in range(args['jobs']):
+            runners.append(th.Thread(
+                target=run_job, args=(runner_, ids, job, args['jobs']),
+                daemon=True))
+    else:
+        runners.append(th.Thread(
+            target=run_job, args=(runner_, ids, None, None),
+            daemon=True))
+
+    def print_update(done):
+        if not args.get('verbose') and (args['color'] or done):
+            sys.stdout.write('%s%srunning %s%s:%s %s%s' % (
+                '\r\x1b[K' if args['color'] else '',
+                '\x1b[?7l' if not done else '',
+                ('\x1b[34m' if not failures else '\x1b[31m')
+                    if args['color'] else '',
+                name,
+                '\x1b[m' if args['color'] else '',
+                ', '.join(filter(None, [
+                    '%d/%d suites' % (
+                        sum(passed_suite_perms[k] == v
+                            for k, v in expected_suite_perms.items()),
+                        len(expected_suite_perms))
+                        if (not args.get('by_suites')
+                            and not args.get('by_cases')) else None,
+                    '%d/%d cases' % (
+                        sum(passed_case_perms[k] == v
+                            for k, v in expected_case_perms.items()),
+                        len(expected_case_perms))
+                        if not args.get('by_cases') else None,
+                    '%d/%d perms' % (passed_perms, expected_perms),
+                    '%s%d/%d failures%s' % (
+                            '\x1b[31m' if args['color'] else '',
+                            len(failures),
+                            expected_perms,
+                            '\x1b[m' if args['color'] else '')
+                        if failures else None])),
+                '\x1b[?7h' if not done else '\n'))
+            sys.stdout.flush()
+
+    for r in runners:
+        r.start()
+
+    try:
+        while any(r.is_alive() for r in runners):
+            time.sleep(0.01)
+            print_update(False)
+    except KeyboardInterrupt:
+        # this is handled by the runner threads, we just
+        # need to not abort here
+        killed = True
+    finally:
+        print_update(True)
+
+    for r in runners:
+        r.join()
+
+    return (
+        expected_perms,
+        passed_perms,
+        readed,
+        proged,
+        erased,
+        failures,
+        killed)
+
+
+def run(runner, bench_ids=[], **args):
+    # query runner for benches
+    runner_ = find_runner(runner, **args)
+    print('using runner: %s' % ' '.join(shlex.quote(c) for c in runner_))
+    (_,
+        expected_suite_perms,
+        expected_case_perms,
+        expected_perms,
+        total_perms) = find_perms(runner_, bench_ids, **args)
+    print('found %d suites, %d cases, %d/%d permutations' % (
+        len(expected_suite_perms),
+        len(expected_case_perms),
+        expected_perms,
+        total_perms))
+    print()
+
+    # automatic job detection?
+    if args.get('jobs') == 0:
+        args['jobs'] = len(os.sched_getaffinity(0))
+
+    # truncate and open logs here so they aren't disconnected between benches
+    stdout = None
+    if args.get('stdout'):
+        stdout = openio(args['stdout'], 'w', 1)
+    trace = None
+    if args.get('trace'):
+        trace = openio(args['trace'], 'w', 1)
+    output = None
+    if args.get('output'):
+        output = BenchOutput(args['output'],
+            ['suite', 'case'],
+            ['bench_readed', 'bench_proged', 'bench_erased'])
+
+    # measure runtime
+    start = time.time()
+
+    # spawn runners
+    expected = 0
+    passed = 0
+    readed = 0
+    proged = 0
+    erased = 0
+    failures = []
+    for by in (bench_ids if bench_ids
+            else expected_case_perms.keys() if args.get('by_cases')
+            else expected_suite_perms.keys() if args.get('by_suites')
+            else [None]):
+        # spawn jobs for stage
+        (expected_,
+            passed_,
+            readed_,
+            proged_,
+            erased_,
+            failures_,
+            killed) = run_stage(
+                by or 'benches',
+                runner_,
+                [by] if by is not None else [],
+                stdout,
+                trace,
+                output,
+                **args)
+        # collect passes/failures
+        expected += expected_
+        passed += passed_
+        readed += readed_
+        proged += proged_
+        erased += erased_
+        failures.extend(failures_)
+        if (failures and not args.get('keep_going')) or killed:
+            break
+
+    stop = time.time()
+
+    if stdout:
+        try:
+            stdout.close()
+        except BrokenPipeError:
+            pass
+    if trace:
+        try:
+            trace.close()
+        except BrokenPipeError:
+            pass
+    if output:
+        output.close()
+
+    # show summary
+    print()
+    print('%sdone:%s %s' % (
+        ('\x1b[34m' if not failures else '\x1b[31m')
+            if args['color'] else '',
+        '\x1b[m' if args['color'] else '',
+        ', '.join(filter(None, [
+            '%d readed' % readed,
+            '%d proged' % proged,
+            '%d erased' % erased,
+            'in %.2fs' % (stop-start)]))))
+    print()
+
+    # print each failure
+    for failure in failures:
+        assert failure.id is not None, '%s broken? %r' % (
+            ' '.join(shlex.quote(c) for c in runner_),
+            failure)
+
+        # get some extra info from runner
+        path, lineno = find_path(runner_, failure.id, **args)
+        defines = find_defines(runner_, failure.id, **args)
+
+        # show summary of failure
+        print('%s%s:%d:%sfailure:%s %s%s failed' % (
+            '\x1b[01m' if args['color'] else '',
+            path, lineno,
+            '\x1b[01;31m' if args['color'] else '',
+            '\x1b[m' if args['color'] else '',
+            failure.id,
+            ' (%s)' % ', '.join('%s=%s' % (k,v) for k,v in defines.items())
+                if defines else ''))
+
+        if failure.stdout:
+            stdout = failure.stdout
+            if failure.assert_ is not None:
+                stdout = stdout[:-1]
+            for line in stdout[-args.get('context', 5):]:
+                sys.stdout.write(line)
+
+        if failure.assert_ is not None:
+            path, lineno, message = failure.assert_
+            print('%s%s:%d:%sassert:%s %s' % (
+                '\x1b[01m' if args['color'] else '',
+                path, lineno,
+                '\x1b[01;31m' if args['color'] else '',
+                '\x1b[m' if args['color'] else '',
+                message))
+            with open(path) as f:
+                line = next(it.islice(f, lineno-1, None)).strip('\n')
+                print(line)
+        print()
+
+    # drop into gdb?
+    if failures and (args.get('gdb')
+            or args.get('gdb_case')
+            or args.get('gdb_main')):
+        failure = failures[0]
+        cmd = runner_ + [failure.id]
+
+        if args.get('gdb_main'):
+            # we don't really need the case breakpoint here, but it
+            # can be helpful
+            path, lineno = find_path(runner_, failure.id, **args)
+            cmd[:0] = args['gdb_path'] + [
+                '-ex', 'break main',
+                '-ex', 'break %s:%d' % (path, lineno),
+                '-ex', 'run',
+                '--args']
+        elif args.get('gdb_case'):
+            path, lineno = find_path(runner_, failure.id, **args)
+            cmd[:0] = args['gdb_path'] + [
+                '-ex', 'break %s:%d' % (path, lineno),
+                '-ex', 'run',
+                '--args']
+        elif failure.assert_ is not None:
+            cmd[:0] = args['gdb_path'] + [
+                '-ex', 'run',
+                '-ex', 'frame function raise',
+                '-ex', 'up 2',
+                '--args']
+        else:
+            cmd[:0] = args['gdb_path'] + [
+                '-ex', 'run',
+                '--args']
+
+        # exec gdb interactively
+        if args.get('verbose'):
+            print(' '.join(shlex.quote(c) for c in cmd))
+        os.execvp(cmd[0], cmd)
+
+    return 1 if failures else 0
+
+
+def main(**args):
+    # figure out what color should be
+    if args.get('color') == 'auto':
+        args['color'] = sys.stdout.isatty()
+    elif args.get('color') == 'always':
+        args['color'] = True
+    else:
+        args['color'] = False
+
+    if args.get('compile'):
+        return compile(**args)
+    elif (args.get('summary')
+            or args.get('list_suites')
+            or args.get('list_cases')
+            or args.get('list_suite_paths')
+            or args.get('list_case_paths')
+            or args.get('list_defines')
+            or args.get('list_permutation_defines')
+            or args.get('list_implicit_defines')
+            or args.get('list_geometries')):
+        return list_(**args)
+    else:
+        return run(**args)
+
+
+if __name__ == "__main__":
+    import argparse
+    import sys
+    argparse.ArgumentParser._handle_conflict_ignore = lambda *_: None
+    argparse._ArgumentGroup._handle_conflict_ignore = lambda *_: None
+    parser = argparse.ArgumentParser(
+        description="Build and run benches.",
+        allow_abbrev=False,
+        conflict_handler='ignore')
+    parser.add_argument(
+        '-v', '--verbose',
+        action='store_true',
+        help="Output commands that run behind the scenes.")
+    parser.add_argument(
+        '--color',
+        choices=['never', 'always', 'auto'],
+        default='auto',
+        help="When to use terminal colors. Defaults to 'auto'.")
+
+    # bench flags
+    bench_parser = parser.add_argument_group('bench options')
+    bench_parser.add_argument(
+        'runner',
+        nargs='?',
+        type=lambda x: x.split(),
+        help="Bench runner to use for benching. Defaults to %r." % RUNNER_PATH)
+    bench_parser.add_argument(
+        'bench_ids',
+        nargs='*',
+        help="Description of benches to run.")
+    bench_parser.add_argument(
+        '-Y', '--summary',
+        action='store_true',
+        help="Show quick summary.")
+    bench_parser.add_argument(
+        '-l', '--list-suites',
+        action='store_true',
+        help="List bench suites.")
+    bench_parser.add_argument(
+        '-L', '--list-cases',
+        action='store_true',
+        help="List bench cases.")
+    bench_parser.add_argument(
+        '--list-suite-paths',
+        action='store_true',
+        help="List the path for each bench suite.")
+    bench_parser.add_argument(
+        '--list-case-paths',
+        action='store_true',
+        help="List the path and line number for each bench case.")
+    bench_parser.add_argument(
+        '--list-defines',
+        action='store_true',
+        help="List all defines in this bench-runner.")
+    bench_parser.add_argument(
+        '--list-permutation-defines',
+        action='store_true',
+        help="List explicit defines in this bench-runner.")
+    bench_parser.add_argument(
+        '--list-implicit-defines',
+        action='store_true',
+        help="List implicit defines in this bench-runner.")
+    bench_parser.add_argument(
+        '--list-geometries',
+        action='store_true',
+        help="List the available disk geometries.")
+    bench_parser.add_argument(
+        '-D', '--define',
+        action='append',
+        help="Override a bench define.")
+    bench_parser.add_argument(
+        '-G', '--geometry',
+        help="Comma-separated list of disk geometries to bench.")
+    bench_parser.add_argument(
+        '-d', '--disk',
+        help="Direct block device operations to this file.")
+    bench_parser.add_argument(
+        '-t', '--trace',
+        help="Direct trace output to this file.")
+    bench_parser.add_argument(
+        '--trace-backtrace',
+        action='store_true',
+        help="Include a backtrace with every trace statement.")
+    bench_parser.add_argument(
+        '--trace-period',
+        help="Sample trace output at this period in cycles.")
+    bench_parser.add_argument(
+        '--trace-freq',
+        help="Sample trace output at this frequency in hz.")
+    bench_parser.add_argument(
+        '-O', '--stdout',
+        help="Direct stdout to this file. Note stderr is already merged here.")
+    bench_parser.add_argument(
+        '-o', '--output',
+        help="CSV file to store results.")
+    bench_parser.add_argument(
+        '--read-sleep',
+        help="Artificial read delay in seconds.")
+    bench_parser.add_argument(
+        '--prog-sleep',
+        help="Artificial prog delay in seconds.")
+    bench_parser.add_argument(
+        '--erase-sleep',
+        help="Artificial erase delay in seconds.")
+    bench_parser.add_argument(
+        '-j', '--jobs',
+        nargs='?',
+        type=lambda x: int(x, 0),
+        const=0,
+        help="Number of parallel runners to run. 0 runs one runner per core.")
+    bench_parser.add_argument(
+        '-k', '--keep-going',
+        action='store_true',
+        help="Don't stop on first error.")
+    bench_parser.add_argument(
+        '-i', '--isolate',
+        action='store_true',
+        help="Run each bench permutation in a separate process.")
+    bench_parser.add_argument(
+        '-b', '--by-suites',
+        action='store_true',
+        help="Step through benches by suite.")
+    bench_parser.add_argument(
+        '-B', '--by-cases',
+        action='store_true',
+        help="Step through benches by case.")
+    bench_parser.add_argument(
+        '--context',
+        type=lambda x: int(x, 0),
+        default=5,
+        help="Show this many lines of stdout on bench failure. "
+            "Defaults to 5.")
+    bench_parser.add_argument(
+        '--gdb',
+        action='store_true',
+        help="Drop into gdb on bench failure.")
+    bench_parser.add_argument(
+        '--gdb-case',
+        action='store_true',
+        help="Drop into gdb on bench failure but stop at the beginning "
+            "of the failing bench case.")
+    bench_parser.add_argument(
+        '--gdb-main',
+        action='store_true',
+        help="Drop into gdb on bench failure but stop at the beginning "
+            "of main.")
+    bench_parser.add_argument(
+        '--gdb-path',
+        type=lambda x: x.split(),
+        default=GDB_PATH,
+        help="Path to the gdb executable, may include flags. "
+            "Defaults to %r." % GDB_PATH)
+    bench_parser.add_argument(
+        '--exec',
+        type=lambda e: e.split(),
+        help="Run under another executable.")
+    bench_parser.add_argument(
+        '--valgrind',
+        action='store_true',
+        help="Run under Valgrind to find memory errors. Implicitly sets "
+            "--isolate.")
+    bench_parser.add_argument(
+        '--valgrind-path',
+        type=lambda x: x.split(),
+        default=VALGRIND_PATH,
+        help="Path to the Valgrind executable, may include flags. "
+            "Defaults to %r." % VALGRIND_PATH)
+    bench_parser.add_argument(
+        '-p', '--perf',
+        help="Run under Linux's perf to sample performance counters, writing "
+            "samples to this file.")
+    bench_parser.add_argument(
+        '--perf-freq',
+        help="perf sampling frequency. This is passed directly to the perf "
+            "script.")
+    bench_parser.add_argument(
+        '--perf-period',
+        help="perf sampling period. This is passed directly to the perf "
+            "script.")
+    bench_parser.add_argument(
+        '--perf-events',
+        help="perf events to record. This is passed directly to the perf "
+            "script.")
+    bench_parser.add_argument(
+        '--perf-script',
+        type=lambda x: x.split(),
+        default=PERF_SCRIPT,
+        help="Path to the perf script to use. Defaults to %r." % PERF_SCRIPT)
+    bench_parser.add_argument(
+        '--perf-path',
+        type=lambda x: x.split(),
+        help="Path to the perf executable, may include flags. This is passed "
+            "directly to the perf script")
+
+    # compilation flags
+    comp_parser = parser.add_argument_group('compilation options')
+    comp_parser.add_argument(
+        'bench_paths',
+        nargs='*',
+        help="Description of *.toml files to compile. May be a directory "
+            "or a list of paths.")
+    comp_parser.add_argument(
+        '-c', '--compile',
+        action='store_true',
+        help="Compile a bench suite or source file.")
+    comp_parser.add_argument(
+        '-s', '--source',
+        help="Source file to compile, possibly injecting internal benches.")
+    comp_parser.add_argument(
+        '--include',
+        default=HEADER_PATH,
+        help="Inject this header file into every compiled bench file. "
+            "Defaults to %r." % HEADER_PATH)
+    comp_parser.add_argument(
+        '-o', '--output',
+        help="Output file.")
+
+    # runner/bench_paths overlap, so need to do some munging here
+    args = parser.parse_intermixed_args()
+    args.bench_paths = [' '.join(args.runner or [])] + args.bench_ids
+    args.runner = args.runner or [RUNNER_PATH]
+
+    sys.exit(main(**{k: v
+        for k, v in vars(args).items()
+        if v is not None}))
diff --git a/scripts/changeprefix.py b/scripts/changeprefix.py
new file mode 100644
index 0000000..381a456
--- /dev/null
+++ b/scripts/changeprefix.py
@@ -0,0 +1,181 @@
+#!/usr/bin/env python3
+#
+# Change prefixes in files/filenames. Useful for creating different versions
+# of a codebase that don't conflict at compile time.
+#
+# Example:
+# $ ./scripts/changeprefix.py lfs lfs3
+#
+# Copyright (c) 2022, The littlefs authors.
+# Copyright (c) 2019, Arm Limited. All rights reserved.
+# SPDX-License-Identifier: BSD-3-Clause
+#
+
+import glob
+import itertools
+import os
+import os.path
+import re
+import shlex
+import shutil
+import subprocess
+import tempfile
+
+GIT_PATH = ['git']
+
+
+def openio(path, mode='r', buffering=-1):
+    # allow '-' for stdin/stdout
+    if path == '-':
+        if mode == 'r':
+            return os.fdopen(os.dup(sys.stdin.fileno()), mode, buffering)
+        else:
+            return os.fdopen(os.dup(sys.stdout.fileno()), mode, buffering)
+    else:
+        return open(path, mode, buffering)
+
+def changeprefix(from_prefix, to_prefix, line):
+    line, count1 = re.subn(
+        '\\b'+from_prefix,
+        to_prefix,
+        line)
+    line, count2 = re.subn(
+        '\\b'+from_prefix.upper(),
+        to_prefix.upper(),
+        line)
+    line, count3 = re.subn(
+        '\\B-D'+from_prefix.upper(),
+        '-D'+to_prefix.upper(),
+        line)
+    return line, count1+count2+count3
+
+def changefile(from_prefix, to_prefix, from_path, to_path, *,
+        no_replacements=False):
+    # rename any prefixes in file
+    count = 0
+
+    # create a temporary file to avoid overwriting ourself
+    if from_path == to_path and to_path != '-':
+        to_path_temp = tempfile.NamedTemporaryFile('w', delete=False)
+        to_path = to_path_temp.name
+    else:
+        to_path_temp = None
+
+    with openio(from_path) as from_f:
+        with openio(to_path, 'w') as to_f:
+            for line in from_f:
+                if not no_replacements:
+                    line, n = changeprefix(from_prefix, to_prefix, line)
+                    count += n
+                to_f.write(line)
+
+    if from_path != '-' and to_path != '-':
+        shutil.copystat(from_path, to_path)
+
+    if to_path_temp:
+        os.rename(to_path, from_path)
+    elif from_path != '-':
+        os.remove(from_path)
+
+    # Summary
+    print('%s: %d replacements' % (
+        '%s -> %s' % (from_path, to_path) if not to_path_temp else from_path,
+        count))
+
+def main(from_prefix, to_prefix, paths=[], *,
+        verbose=False,
+        output=None,
+        no_replacements=False,
+        no_renames=False,
+        git=False,
+        no_stage=False,
+        git_path=GIT_PATH):
+    if not paths:
+        if git:
+            cmd = git_path + ['ls-tree', '-r', '--name-only', 'HEAD']
+            if verbose:
+                print(' '.join(shlex.quote(c) for c in cmd))
+            paths = subprocess.check_output(cmd, encoding='utf8').split()
+        else:
+            print('no paths?', file=sys.stderr)
+            sys.exit(1)
+        
+    for from_path in paths:
+        # rename filename?
+        if output:
+            to_path = output
+        elif no_renames:
+            to_path = from_path
+        else:
+            to_path = os.path.join(
+                os.path.dirname(from_path),
+                changeprefix(from_prefix, to_prefix,
+                    os.path.basename(from_path))[0])
+
+        # rename contents
+        changefile(from_prefix, to_prefix, from_path, to_path,
+            no_replacements=no_replacements)
+
+        # stage?
+        if git and not no_stage:
+            if from_path != to_path:
+                cmd = git_path + ['rm', '-q', from_path]
+                if verbose:
+                    print(' '.join(shlex.quote(c) for c in cmd))
+                subprocess.check_call(cmd)
+            cmd = git_path + ['add', to_path]
+            if verbose:
+                print(' '.join(shlex.quote(c) for c in cmd))
+            subprocess.check_call(cmd)
+
+
+if __name__ == "__main__":
+    import argparse
+    import sys
+    parser = argparse.ArgumentParser(
+        description="Change prefixes in files/filenames. Useful for creating "
+            "different versions of a codebase that don't conflict at compile "
+            "time.",
+        allow_abbrev=False)
+    parser.add_argument(
+        'from_prefix',
+        help="Prefix to replace.")
+    parser.add_argument(
+        'to_prefix',
+        help="Prefix to replace with.")
+    parser.add_argument(
+        'paths',
+        nargs='*',
+        help="Files to operate on.")
+    parser.add_argument(
+        '-v', '--verbose',
+        action='store_true',
+        help="Output commands that run behind the scenes.")
+    parser.add_argument(
+        '-o', '--output',
+        help="Output file.")
+    parser.add_argument(
+        '-N', '--no-replacements',
+        action='store_true',
+        help="Don't change prefixes in files")
+    parser.add_argument(
+        '-R', '--no-renames',
+        action='store_true',
+        help="Don't rename files")
+    parser.add_argument(
+        '--git',
+        action='store_true',
+        help="Use git to find/update files.")
+    parser.add_argument(
+        '--no-stage',
+        action='store_true',
+        help="Don't stage changes with git.")
+    parser.add_argument(
+        '--git-path',
+        type=lambda x: x.split(),
+        default=GIT_PATH,
+        help="Path to git executable, may include flags. "
+            "Defaults to %r." % GIT_PATH)
+    sys.exit(main(**{k: v
+        for k, v in vars(parser.parse_intermixed_args()).items()
+        if v is not None}))
diff --git a/scripts/code.py b/scripts/code.py
index b394e9c..ba8bd1e 100755
--- a/scripts/code.py
+++ b/scripts/code.py
@@ -1,42 +1,188 @@
 #!/usr/bin/env python3
 #
-# Script to find code size at the function level. Basically just a bit wrapper
+# Script to find code size at the function level. Basically just a big wrapper
 # around nm with some extra conveniences for comparing builds. Heavily inspired
 # by Linux's Bloat-O-Meter.
 #
+# Example:
+# ./scripts/code.py lfs.o lfs_util.o -Ssize
+#
+# Copyright (c) 2022, The littlefs authors.
+# Copyright (c) 2020, Arm Limited. All rights reserved.
+# SPDX-License-Identifier: BSD-3-Clause
+#
 
-import os
-import glob
-import itertools as it
-import subprocess as sp
-import shlex
-import re
-import csv
 import collections as co
+import csv
+import difflib
+import itertools as it
+import math as m
+import os
+import re
+import shlex
+import subprocess as sp
 
 
-OBJ_PATHS = ['*.o']
+NM_PATH = ['nm']
+NM_TYPES = 'tTrRdD'
+OBJDUMP_PATH = ['objdump']
 
-def collect(paths, **args):
-    results = co.defaultdict(lambda: 0)
-    pattern = re.compile(
+
+# integer fields
+class Int(co.namedtuple('Int', 'x')):
+    __slots__ = ()
+    def __new__(cls, x=0):
+        if isinstance(x, Int):
+            return x
+        if isinstance(x, str):
+            try:
+                x = int(x, 0)
+            except ValueError:
+                # also accept +-∞ and +-inf
+                if re.match('^\s*\+?\s*(?:∞|inf)\s*$', x):
+                    x = m.inf
+                elif re.match('^\s*-\s*(?:∞|inf)\s*$', x):
+                    x = -m.inf
+                else:
+                    raise
+        assert isinstance(x, int) or m.isinf(x), x
+        return super().__new__(cls, x)
+
+    def __str__(self):
+        if self.x == m.inf:
+            return '∞'
+        elif self.x == -m.inf:
+            return '-∞'
+        else:
+            return str(self.x)
+
+    def __int__(self):
+        assert not m.isinf(self.x)
+        return self.x
+
+    def __float__(self):
+        return float(self.x)
+
+    none = '%7s' % '-'
+    def table(self):
+        return '%7s' % (self,)
+
+    diff_none = '%7s' % '-'
+    diff_table = table
+
+    def diff_diff(self, other):
+        new = self.x if self else 0
+        old = other.x if other else 0
+        diff = new - old
+        if diff == +m.inf:
+            return '%7s' % '+∞'
+        elif diff == -m.inf:
+            return '%7s' % '-∞'
+        else:
+            return '%+7d' % diff
+
+    def ratio(self, other):
+        new = self.x if self else 0
+        old = other.x if other else 0
+        if m.isinf(new) and m.isinf(old):
+            return 0.0
+        elif m.isinf(new):
+            return +m.inf
+        elif m.isinf(old):
+            return -m.inf
+        elif not old and not new:
+            return 0.0
+        elif not old:
+            return 1.0
+        else:
+            return (new-old) / old
+
+    def __add__(self, other):
+        return self.__class__(self.x + other.x)
+
+    def __sub__(self, other):
+        return self.__class__(self.x - other.x)
+
+    def __mul__(self, other):
+        return self.__class__(self.x * other.x)
+
+# code size results
+class CodeResult(co.namedtuple('CodeResult', [
+        'file', 'function',
+        'size'])):
+    _by = ['file', 'function']
+    _fields = ['size']
+    _sort = ['size']
+    _types = {'size': Int}
+
+    __slots__ = ()
+    def __new__(cls, file='', function='', size=0):
+        return super().__new__(cls, file, function,
+            Int(size))
+
+    def __add__(self, other):
+        return CodeResult(self.file, self.function,
+            self.size + other.size)
+
+
+def openio(path, mode='r', buffering=-1):
+    # allow '-' for stdin/stdout
+    if path == '-':
+        if mode == 'r':
+            return os.fdopen(os.dup(sys.stdin.fileno()), mode, buffering)
+        else:
+            return os.fdopen(os.dup(sys.stdout.fileno()), mode, buffering)
+    else:
+        return open(path, mode, buffering)
+
+def collect(obj_paths, *,
+        nm_path=NM_PATH,
+        nm_types=NM_TYPES,
+        objdump_path=OBJDUMP_PATH,
+        sources=None,
+        everything=False,
+        **args):
+    size_pattern = re.compile(
         '^(?P<size>[0-9a-fA-F]+)' +
-        ' (?P<type>[%s])' % re.escape(args['type']) +
+        ' (?P<type>[%s])' % re.escape(nm_types) +
         ' (?P<func>.+?)$')
-    for path in paths:
-        # note nm-tool may contain extra args
-        cmd = args['nm_tool'] + ['--size-sort', path]
+    line_pattern = re.compile(
+        '^\s+(?P<no>[0-9]+)'
+            '(?:\s+(?P<dir>[0-9]+))?'
+            '\s+.*'
+            '\s+(?P<path>[^\s]+)$')
+    info_pattern = re.compile(
+        '^(?:.*(?P<tag>DW_TAG_[a-z_]+).*'
+            '|.*DW_AT_name.*:\s*(?P<name>[^:\s]+)\s*'
+            '|.*DW_AT_decl_file.*:\s*(?P<file>[0-9]+)\s*)$')
+
+    results = []
+    for path in obj_paths:
+        # guess the source, if we have debug-info we'll replace this later
+        file = re.sub('(\.o)?$', '.c', path, 1)
+
+        # find symbol sizes
+        results_ = []
+        # note nm-path may contain extra args
+        cmd = nm_path + ['--size-sort', path]
         if args.get('verbose'):
             print(' '.join(shlex.quote(c) for c in cmd))
         proc = sp.Popen(cmd,
             stdout=sp.PIPE,
             stderr=sp.PIPE if not args.get('verbose') else None,
             universal_newlines=True,
-            errors='replace')
+            errors='replace',
+            close_fds=False)
         for line in proc.stdout:
-            m = pattern.match(line)
+            m = size_pattern.match(line)
             if m:
-                results[(path, m.group('func'))] += int(m.group('size'), 16)
+                func = m.group('func')
+                # discard internal functions
+                if not everything and func.startswith('__'):
+                    continue
+                results_.append(CodeResult(
+                    file, func,
+                    int(m.group('size'), 16)))
         proc.wait()
         if proc.returncode != 0:
             if not args.get('verbose'):
@@ -44,241 +190,518 @@ def collect(paths, **args):
                     sys.stdout.write(line)
             sys.exit(-1)
 
-    flat_results = []
-    for (file, func), size in results.items():
-        # map to source files
-        if args.get('build_dir'):
-            file = re.sub('%s/*' % re.escape(args['build_dir']), '', file)
-        # replace .o with .c, different scripts report .o/.c, we need to
-        # choose one if we want to deduplicate csv files
-        file = re.sub('\.o$', '.c', file)
-        # discard internal functions
-        if not args.get('everything'):
-            if func.startswith('__'):
-                continue
-        # discard .8449 suffixes created by optimizer
-        func = re.sub('\.[0-9]+', '', func)
 
-        flat_results.append((file, func, size))
+        # try to figure out the source file if we have debug-info
+        dirs = {}
+        files = {}
+        # note objdump-path may contain extra args
+        cmd = objdump_path + ['--dwarf=rawline', path]
+        if args.get('verbose'):
+            print(' '.join(shlex.quote(c) for c in cmd))
+        proc = sp.Popen(cmd,
+            stdout=sp.PIPE,
+            stderr=sp.PIPE if not args.get('verbose') else None,
+            universal_newlines=True,
+            errors='replace',
+            close_fds=False)
+        for line in proc.stdout:
+            # note that files contain references to dirs, which we
+            # dereference as soon as we see them as each file table follows a
+            # dir table
+            m = line_pattern.match(line)
+            if m:
+                if not m.group('dir'):
+                    # found a directory entry
+                    dirs[int(m.group('no'))] = m.group('path')
+                else:
+                    # found a file entry
+                    dir = int(m.group('dir'))
+                    if dir in dirs:
+                        files[int(m.group('no'))] = os.path.join(
+                            dirs[dir],
+                            m.group('path'))
+                    else:
+                        files[int(m.group('no'))] = m.group('path')
+        proc.wait()
+        if proc.returncode != 0:
+            if not args.get('verbose'):
+                for line in proc.stderr:
+                    sys.stdout.write(line)
+            # do nothing on error, we don't need objdump to work, source files
+            # may just be inaccurate
+            pass
 
-    return flat_results
+        defs = {}
+        is_func = False
+        f_name = None
+        f_file = None
+        # note objdump-path may contain extra args
+        cmd = objdump_path + ['--dwarf=info', path]
+        if args.get('verbose'):
+            print(' '.join(shlex.quote(c) for c in cmd))
+        proc = sp.Popen(cmd,
+            stdout=sp.PIPE,
+            stderr=sp.PIPE if not args.get('verbose') else None,
+            universal_newlines=True,
+            errors='replace',
+            close_fds=False)
+        for line in proc.stdout:
+            # state machine here to find definitions
+            m = info_pattern.match(line)
+            if m:
+                if m.group('tag'):
+                    if is_func:
+                        defs[f_name] = files.get(f_file, '?')
+                    is_func = (m.group('tag') == 'DW_TAG_subprogram')
+                elif m.group('name'):
+                    f_name = m.group('name')
+                elif m.group('file'):
+                    f_file = int(m.group('file'))
+        if is_func:
+            defs[f_name] = files.get(f_file, '?')
+        proc.wait()
+        if proc.returncode != 0:
+            if not args.get('verbose'):
+                for line in proc.stderr:
+                    sys.stdout.write(line)
+            # do nothing on error, we don't need objdump to work, source files
+            # may just be inaccurate
+            pass
 
-def main(**args):
-    def openio(path, mode='r'):
-        if path == '-':
-            if 'r' in mode:
-                return os.fdopen(os.dup(sys.stdin.fileno()), 'r')
+        for r in results_:
+            # find best matching debug symbol, this may be slightly different
+            # due to optimizations
+            if defs:
+                # exact match? avoid difflib if we can for speed
+                if r.function in defs:
+                    file = defs[r.function]
+                else:
+                    _, file = max(
+                        defs.items(),
+                        key=lambda d: difflib.SequenceMatcher(None,
+                            d[0],
+                            r.function, False).ratio())
             else:
-                return os.fdopen(os.dup(sys.stdout.fileno()), 'w')
-        else:
-            return open(path, mode)
+                file = r.file
 
-    # find sizes
-    if not args.get('use', None):
-        # find .o files
-        paths = []
-        for path in args['obj_paths']:
-            if os.path.isdir(path):
-                path = path + '/*.o'
+            # ignore filtered sources
+            if sources is not None:
+                if not any(
+                        os.path.abspath(file) == os.path.abspath(s)
+                        for s in sources):
+                    continue
+            else:
+                # default to only cwd
+                if not everything and not os.path.commonpath([
+                        os.getcwd(),
+                        os.path.abspath(file)]) == os.getcwd():
+                    continue
 
-            for path in glob.glob(path):
-                paths.append(path)
+            # simplify path
+            if os.path.commonpath([
+                    os.getcwd(),
+                    os.path.abspath(file)]) == os.getcwd():
+                file = os.path.relpath(file)
+            else:
+                file = os.path.abspath(file)
 
-        if not paths:
-            print('no .obj files found in %r?' % args['obj_paths'])
+            results.append(r._replace(file=file))
+
+    return results
+
+
+def fold(Result, results, *,
+        by=None,
+        defines=None,
+        **_):
+    if by is None:
+        by = Result._by
+
+    for k in it.chain(by or [], (k for k, _ in defines or [])):
+        if k not in Result._by and k not in Result._fields:
+            print("error: could not find field %r?" % k)
             sys.exit(-1)
 
-        results = collect(paths, **args)
+    # filter by matching defines
+    if defines is not None:
+        results_ = []
+        for r in results:
+            if all(getattr(r, k) in vs for k, vs in defines):
+                results_.append(r)
+        results = results_
+
+    # organize results into conflicts
+    folding = co.OrderedDict()
+    for r in results:
+        name = tuple(getattr(r, k) for k in by)
+        if name not in folding:
+            folding[name] = []
+        folding[name].append(r)
+
+    # merge conflicts
+    folded = []
+    for name, rs in folding.items():
+        folded.append(sum(rs[1:], start=rs[0]))
+
+    return folded
+
+def table(Result, results, diff_results=None, *,
+        by=None,
+        fields=None,
+        sort=None,
+        summary=False,
+        all=False,
+        percent=False,
+        **_):
+    all_, all = all, __builtins__.all
+
+    if by is None:
+        by = Result._by
+    if fields is None:
+        fields = Result._fields
+    types = Result._types
+
+    # fold again
+    results = fold(Result, results, by=by)
+    if diff_results is not None:
+        diff_results = fold(Result, diff_results, by=by)
+
+    # organize by name
+    table = {
+        ','.join(str(getattr(r, k) or '') for k in by): r
+        for r in results}
+    diff_table = {
+        ','.join(str(getattr(r, k) or '') for k in by): r
+        for r in diff_results or []}
+    names = list(table.keys() | diff_table.keys())
+
+    # sort again, now with diff info, note that python's sort is stable
+    names.sort()
+    if diff_results is not None:
+        names.sort(key=lambda n: tuple(
+            types[k].ratio(
+                getattr(table.get(n), k, None),
+                getattr(diff_table.get(n), k, None))
+            for k in fields),
+            reverse=True)
+    if sort:
+        for k, reverse in reversed(sort):
+            names.sort(
+                key=lambda n: tuple(
+                    (getattr(table[n], k),)
+                    if getattr(table.get(n), k, None) is not None else ()
+                    for k in ([k] if k else [
+                        k for k in Result._sort if k in fields])),
+                reverse=reverse ^ (not k or k in Result._fields))
+
+
+    # build up our lines
+    lines = []
+
+    # header
+    header = []
+    header.append('%s%s' % (
+        ','.join(by),
+        ' (%d added, %d removed)' % (
+            sum(1 for n in table if n not in diff_table),
+            sum(1 for n in diff_table if n not in table))
+            if diff_results is not None and not percent else '')
+        if not summary else '')
+    if diff_results is None:
+        for k in fields:
+            header.append(k)
+    elif percent:
+        for k in fields:
+            header.append(k)
     else:
+        for k in fields:
+            header.append('o'+k)
+        for k in fields:
+            header.append('n'+k)
+        for k in fields:
+            header.append('d'+k)
+    header.append('')
+    lines.append(header)
+
+    def table_entry(name, r, diff_r=None, ratios=[]):
+        entry = []
+        entry.append(name)
+        if diff_results is None:
+            for k in fields:
+                entry.append(getattr(r, k).table()
+                    if getattr(r, k, None) is not None
+                    else types[k].none)
+        elif percent:
+            for k in fields:
+                entry.append(getattr(r, k).diff_table()
+                    if getattr(r, k, None) is not None
+                    else types[k].diff_none)
+        else:
+            for k in fields:
+                entry.append(getattr(diff_r, k).diff_table()
+                    if getattr(diff_r, k, None) is not None
+                    else types[k].diff_none)
+            for k in fields:
+                entry.append(getattr(r, k).diff_table()
+                    if getattr(r, k, None) is not None
+                    else types[k].diff_none)
+            for k in fields:
+                entry.append(types[k].diff_diff(
+                        getattr(r, k, None),
+                        getattr(diff_r, k, None)))
+        if diff_results is None:
+            entry.append('')
+        elif percent:
+            entry.append(' (%s)' % ', '.join(
+                '+∞%' if t == +m.inf
+                else '-∞%' if t == -m.inf
+                else '%+.1f%%' % (100*t)
+                for t in ratios))
+        else:
+            entry.append(' (%s)' % ', '.join(
+                    '+∞%' if t == +m.inf
+                    else '-∞%' if t == -m.inf
+                    else '%+.1f%%' % (100*t)
+                    for t in ratios
+                    if t)
+                if any(ratios) else '')
+        return entry
+
+    # entries
+    if not summary:
+        for name in names:
+            r = table.get(name)
+            if diff_results is None:
+                diff_r = None
+                ratios = None
+            else:
+                diff_r = diff_table.get(name)
+                ratios = [
+                    types[k].ratio(
+                        getattr(r, k, None),
+                        getattr(diff_r, k, None))
+                    for k in fields]
+                if not all_ and not any(ratios):
+                    continue
+            lines.append(table_entry(name, r, diff_r, ratios))
+
+    # total
+    r = next(iter(fold(Result, results, by=[])), None)
+    if diff_results is None:
+        diff_r = None
+        ratios = None
+    else:
+        diff_r = next(iter(fold(Result, diff_results, by=[])), None)
+        ratios = [
+            types[k].ratio(
+                getattr(r, k, None),
+                getattr(diff_r, k, None))
+            for k in fields]
+    lines.append(table_entry('TOTAL', r, diff_r, ratios))
+
+    # find the best widths, note that column 0 contains the names and column -1
+    # the ratios, so those are handled a bit differently
+    widths = [
+        ((max(it.chain([w], (len(l[i]) for l in lines)))+1+4-1)//4)*4-1
+        for w, i in zip(
+            it.chain([23], it.repeat(7)),
+            range(len(lines[0])-1))]
+
+    # print our table
+    for line in lines:
+        print('%-*s  %s%s' % (
+            widths[0], line[0],
+            ' '.join('%*s' % (w, x)
+                for w, x in zip(widths[1:], line[1:-1])),
+            line[-1]))
+
+
+def main(obj_paths, *,
+        by=None,
+        fields=None,
+        defines=None,
+        sort=None,
+        **args):
+    # find sizes
+    if not args.get('use', None):
+        results = collect(obj_paths, **args)
+    else:
+        results = []
         with openio(args['use']) as f:
-            r = csv.DictReader(f)
-            results = [
-                (   result['file'],
-                    result['name'],
-                    int(result['code_size']))
-                for result in r
-                if result.get('code_size') not in {None, ''}]
+            reader = csv.DictReader(f, restval='')
+            for r in reader:
+                if not any('code_'+k in r and r['code_'+k].strip()
+                        for k in CodeResult._fields):
+                    continue
+                try:
+                    results.append(CodeResult(
+                        **{k: r[k] for k in CodeResult._by
+                            if k in r and r[k].strip()},
+                        **{k: r['code_'+k] for k in CodeResult._fields
+                            if 'code_'+k in r and r['code_'+k].strip()}))
+                except TypeError:
+                    pass
 
-    total = 0
-    for _, _, size in results:
-        total += size
+    # fold
+    results = fold(CodeResult, results, by=by, defines=defines)
 
-    # find previous results?
-    if args.get('diff'):
-        try:
-            with openio(args['diff']) as f:
-                r = csv.DictReader(f)
-                prev_results = [
-                    (   result['file'],
-                        result['name'],
-                        int(result['code_size']))
-                    for result in r
-                    if result.get('code_size') not in {None, ''}]
-        except FileNotFoundError:
-            prev_results = []
-
-        prev_total = 0
-        for _, _, size in prev_results:
-            prev_total += size
+    # sort, note that python's sort is stable
+    results.sort()
+    if sort:
+        for k, reverse in reversed(sort):
+            results.sort(
+                key=lambda r: tuple(
+                    (getattr(r, k),) if getattr(r, k) is not None else ()
+                    for k in ([k] if k else CodeResult._sort)),
+                reverse=reverse ^ (not k or k in CodeResult._fields))
 
     # write results to CSV
     if args.get('output'):
-        merged_results = co.defaultdict(lambda: {})
-        other_fields = []
-
-        # merge?
-        if args.get('merge'):
-            try:
-                with openio(args['merge']) as f:
-                    r = csv.DictReader(f)
-                    for result in r:
-                        file = result.pop('file', '')
-                        func = result.pop('name', '')
-                        result.pop('code_size', None)
-                        merged_results[(file, func)] = result
-                        other_fields = result.keys()
-            except FileNotFoundError:
-                pass
-
-        for file, func, size in results:
-            merged_results[(file, func)]['code_size'] = size
-
         with openio(args['output'], 'w') as f:
-            w = csv.DictWriter(f, ['file', 'name', *other_fields, 'code_size'])
-            w.writeheader()
-            for (file, func), result in sorted(merged_results.items()):
-                w.writerow({'file': file, 'name': func, **result})
+            writer = csv.DictWriter(f,
+                (by if by is not None else CodeResult._by)
+                + ['code_'+k for k in (
+                    fields if fields is not None else CodeResult._fields)])
+            writer.writeheader()
+            for r in results:
+                writer.writerow(
+                    {k: getattr(r, k) for k in (
+                        by if by is not None else CodeResult._by)}
+                    | {'code_'+k: getattr(r, k) for k in (
+                        fields if fields is not None else CodeResult._fields)})
 
-    # print results
-    def dedup_entries(results, by='name'):
-        entries = co.defaultdict(lambda: 0)
-        for file, func, size in results:
-            entry = (file if by == 'file' else func)
-            entries[entry] += size
-        return entries
+    # find previous results?
+    if args.get('diff'):
+        diff_results = []
+        try:
+            with openio(args['diff']) as f:
+                reader = csv.DictReader(f, restval='')
+                for r in reader:
+                    if not any('code_'+k in r and r['code_'+k].strip()
+                            for k in CodeResult._fields):
+                        continue
+                    try:
+                        diff_results.append(CodeResult(
+                            **{k: r[k] for k in CodeResult._by
+                                if k in r and r[k].strip()},
+                            **{k: r['code_'+k] for k in CodeResult._fields
+                                if 'code_'+k in r and r['code_'+k].strip()}))
+                    except TypeError:
+                        pass
+        except FileNotFoundError:
+            pass
 
-    def diff_entries(olds, news):
-        diff = co.defaultdict(lambda: (0, 0, 0, 0))
-        for name, new in news.items():
-            diff[name] = (0, new, new, 1.0)
-        for name, old in olds.items():
-            _, new, _, _ = diff[name]
-            diff[name] = (old, new, new-old, (new-old)/old if old else 1.0)
-        return diff
+        # fold
+        diff_results = fold(CodeResult, diff_results, by=by, defines=defines)
 
-    def sorted_entries(entries):
-        if args.get('size_sort'):
-            return sorted(entries, key=lambda x: (-x[1], x))
-        elif args.get('reverse_size_sort'):
-            return sorted(entries, key=lambda x: (+x[1], x))
-        else:
-            return sorted(entries)
+    # print table
+    if not args.get('quiet'):
+        table(CodeResult, results,
+            diff_results if args.get('diff') else None,
+            by=by if by is not None else ['function'],
+            fields=fields,
+            sort=sort,
+            **args)
 
-    def sorted_diff_entries(entries):
-        if args.get('size_sort'):
-            return sorted(entries, key=lambda x: (-x[1][1], x))
-        elif args.get('reverse_size_sort'):
-            return sorted(entries, key=lambda x: (+x[1][1], x))
-        else:
-            return sorted(entries, key=lambda x: (-x[1][3], x))
-
-    def print_header(by=''):
-        if not args.get('diff'):
-            print('%-36s %7s' % (by, 'size'))
-        else:
-            print('%-36s %7s %7s %7s' % (by, 'old', 'new', 'diff'))
-
-    def print_entry(name, size):
-        print("%-36s %7d" % (name, size))
-
-    def print_diff_entry(name, old, new, diff, ratio):
-        print("%-36s %7s %7s %+7d%s" % (name,
-            old or "-",
-            new or "-",
-            diff,
-            ' (%+.1f%%)' % (100*ratio) if ratio else ''))
-
-    def print_entries(by='name'):
-        entries = dedup_entries(results, by=by)
-
-        if not args.get('diff'):
-            print_header(by=by)
-            for name, size in sorted_entries(entries.items()):
-                print_entry(name, size)
-        else:
-            prev_entries = dedup_entries(prev_results, by=by)
-            diff = diff_entries(prev_entries, entries)
-            print_header(by='%s (%d added, %d removed)' % (by,
-                sum(1 for old, _, _, _ in diff.values() if not old),
-                sum(1 for _, new, _, _ in diff.values() if not new)))
-            for name, (old, new, diff, ratio) in sorted_diff_entries(
-                    diff.items()):
-                if ratio or args.get('all'):
-                    print_diff_entry(name, old, new, diff, ratio)
-
-    def print_totals():
-        if not args.get('diff'):
-            print_entry('TOTAL', total)
-        else:
-            ratio = (0.0 if not prev_total and not total
-                else 1.0 if not prev_total
-                else (total-prev_total)/prev_total)
-            print_diff_entry('TOTAL',
-                prev_total, total,
-                total-prev_total,
-                ratio)
-
-    if args.get('quiet'):
-        pass
-    elif args.get('summary'):
-        print_header()
-        print_totals()
-    elif args.get('files'):
-        print_entries(by='file')
-        print_totals()
-    else:
-        print_entries(by='name')
-        print_totals()
 
 if __name__ == "__main__":
     import argparse
     import sys
     parser = argparse.ArgumentParser(
-        description="Find code size at the function level.")
-    parser.add_argument('obj_paths', nargs='*', default=OBJ_PATHS,
-        help="Description of where to find *.o files. May be a directory \
-            or a list of paths. Defaults to %r." % OBJ_PATHS)
-    parser.add_argument('-v', '--verbose', action='store_true',
+        description="Find code size at the function level.",
+        allow_abbrev=False)
+    parser.add_argument(
+        'obj_paths',
+        nargs='*',
+        help="Input *.o files.")
+    parser.add_argument(
+        '-v', '--verbose',
+        action='store_true',
         help="Output commands that run behind the scenes.")
-    parser.add_argument('-q', '--quiet', action='store_true',
+    parser.add_argument(
+        '-q', '--quiet',
+        action='store_true',
         help="Don't show anything, useful with -o.")
-    parser.add_argument('-o', '--output',
+    parser.add_argument(
+        '-o', '--output',
         help="Specify CSV file to store results.")
-    parser.add_argument('-u', '--use',
-        help="Don't compile and find code sizes, instead use this CSV file.")
-    parser.add_argument('-d', '--diff',
-        help="Specify CSV file to diff code size against.")
-    parser.add_argument('-m', '--merge',
-        help="Merge with an existing CSV file when writing to output.")
-    parser.add_argument('-a', '--all', action='store_true',
-        help="Show all functions, not just the ones that changed.")
-    parser.add_argument('-A', '--everything', action='store_true',
+    parser.add_argument(
+        '-u', '--use',
+        help="Don't parse anything, use this CSV file.")
+    parser.add_argument(
+        '-d', '--diff',
+        help="Specify CSV file to diff against.")
+    parser.add_argument(
+        '-a', '--all',
+        action='store_true',
+        help="Show all, not just the ones that changed.")
+    parser.add_argument(
+        '-p', '--percent',
+        action='store_true',
+        help="Only show percentage change, not a full diff.")
+    parser.add_argument(
+        '-b', '--by',
+        action='append',
+        choices=CodeResult._by,
+        help="Group by this field.")
+    parser.add_argument(
+        '-f', '--field',
+        dest='fields',
+        action='append',
+        choices=CodeResult._fields,
+        help="Show this field.")
+    parser.add_argument(
+        '-D', '--define',
+        dest='defines',
+        action='append',
+        type=lambda x: (lambda k,v: (k, set(v.split(','))))(*x.split('=', 1)),
+        help="Only include results where this field is this value.")
+    class AppendSort(argparse.Action):
+        def __call__(self, parser, namespace, value, option):
+            if namespace.sort is None:
+                namespace.sort = []
+            namespace.sort.append((value, True if option == '-S' else False))
+    parser.add_argument(
+        '-s', '--sort',
+        nargs='?',
+        action=AppendSort,
+        help="Sort by this field.")
+    parser.add_argument(
+        '-S', '--reverse-sort',
+        nargs='?',
+        action=AppendSort,
+        help="Sort by this field, but backwards.")
+    parser.add_argument(
+        '-Y', '--summary',
+        action='store_true',
+        help="Only show the total.")
+    parser.add_argument(
+        '-F', '--source',
+        dest='sources',
+        action='append',
+        help="Only consider definitions in this file. Defaults to anything "
+            "in the current directory.")
+    parser.add_argument(
+        '--everything',
+        action='store_true',
         help="Include builtin and libc specific symbols.")
-    parser.add_argument('-s', '--size-sort', action='store_true',
-        help="Sort by size.")
-    parser.add_argument('-S', '--reverse-size-sort', action='store_true',
-        help="Sort by size, but backwards.")
-    parser.add_argument('-F', '--files', action='store_true',
-        help="Show file-level code sizes. Note this does not include padding! "
-            "So sizes may differ from other tools.")
-    parser.add_argument('-Y', '--summary', action='store_true',
-        help="Only show the total code size.")
-    parser.add_argument('--type', default='tTrRdD',
+    parser.add_argument(
+        '--nm-types',
+        default=NM_TYPES,
         help="Type of symbols to report, this uses the same single-character "
-            "type-names emitted by nm. Defaults to %(default)r.")
-    parser.add_argument('--nm-tool', default=['nm'], type=lambda x: x.split(),
-        help="Path to the nm tool to use.")
-    parser.add_argument('--build-dir',
-        help="Specify the relative build directory. Used to map object files \
-            to the correct source files.")
-    sys.exit(main(**vars(parser.parse_args())))
+            "type-names emitted by nm. Defaults to %r." % NM_TYPES)
+    parser.add_argument(
+        '--nm-path',
+        type=lambda x: x.split(),
+        default=NM_PATH,
+        help="Path to the nm executable, may include flags. "
+            "Defaults to %r." % NM_PATH)
+    parser.add_argument(
+        '--objdump-path',
+        type=lambda x: x.split(),
+        default=OBJDUMP_PATH,
+        help="Path to the objdump executable, may include flags. "
+            "Defaults to %r." % OBJDUMP_PATH)
+    sys.exit(main(**{k: v
+        for k, v in vars(parser.parse_intermixed_args()).items()
+        if v is not None}))
diff --git a/scripts/cov.py b/scripts/cov.py
new file mode 100644
index 0000000..b61b2e5
--- /dev/null
+++ b/scripts/cov.py
@@ -0,0 +1,828 @@
+#!/usr/bin/env python3
+#
+# Script to find coverage info after running tests.
+#
+# Example:
+# ./scripts/cov.py \
+#     lfs.t.a.gcda lfs_util.t.a.gcda \
+#     -Flfs.c -Flfs_util.c -slines
+#
+# Copyright (c) 2022, The littlefs authors.
+# Copyright (c) 2020, Arm Limited. All rights reserved.
+# SPDX-License-Identifier: BSD-3-Clause
+#
+
+import collections as co
+import csv
+import itertools as it
+import json
+import math as m
+import os
+import re
+import shlex
+import subprocess as sp
+
+# TODO use explode_asserts to avoid counting assert branches?
+# TODO use dwarf=info to find functions for inline functions?
+
+GCOV_PATH = ['gcov']
+
+
+# integer fields
+class Int(co.namedtuple('Int', 'x')):
+    __slots__ = ()
+    def __new__(cls, x=0):
+        if isinstance(x, Int):
+            return x
+        if isinstance(x, str):
+            try:
+                x = int(x, 0)
+            except ValueError:
+                # also accept +-∞ and +-inf
+                if re.match('^\s*\+?\s*(?:∞|inf)\s*$', x):
+                    x = m.inf
+                elif re.match('^\s*-\s*(?:∞|inf)\s*$', x):
+                    x = -m.inf
+                else:
+                    raise
+        assert isinstance(x, int) or m.isinf(x), x
+        return super().__new__(cls, x)
+
+    def __str__(self):
+        if self.x == m.inf:
+            return '∞'
+        elif self.x == -m.inf:
+            return '-∞'
+        else:
+            return str(self.x)
+
+    def __int__(self):
+        assert not m.isinf(self.x)
+        return self.x
+
+    def __float__(self):
+        return float(self.x)
+
+    none = '%7s' % '-'
+    def table(self):
+        return '%7s' % (self,)
+
+    diff_none = '%7s' % '-'
+    diff_table = table
+
+    def diff_diff(self, other):
+        new = self.x if self else 0
+        old = other.x if other else 0
+        diff = new - old
+        if diff == +m.inf:
+            return '%7s' % '+∞'
+        elif diff == -m.inf:
+            return '%7s' % '-∞'
+        else:
+            return '%+7d' % diff
+
+    def ratio(self, other):
+        new = self.x if self else 0
+        old = other.x if other else 0
+        if m.isinf(new) and m.isinf(old):
+            return 0.0
+        elif m.isinf(new):
+            return +m.inf
+        elif m.isinf(old):
+            return -m.inf
+        elif not old and not new:
+            return 0.0
+        elif not old:
+            return 1.0
+        else:
+            return (new-old) / old
+
+    def __add__(self, other):
+        return self.__class__(self.x + other.x)
+
+    def __sub__(self, other):
+        return self.__class__(self.x - other.x)
+
+    def __mul__(self, other):
+        return self.__class__(self.x * other.x)
+
+# fractional fields, a/b
+class Frac(co.namedtuple('Frac', 'a,b')):
+    __slots__ = ()
+    def __new__(cls, a=0, b=None):
+        if isinstance(a, Frac) and b is None:
+            return a
+        if isinstance(a, str) and b is None:
+            a, b = a.split('/', 1)
+        if b is None:
+            b = a
+        return super().__new__(cls, Int(a), Int(b))
+
+    def __str__(self):
+        return '%s/%s' % (self.a, self.b)
+
+    def __float__(self):
+        return float(self.a)
+
+    none = '%11s %7s' % ('-', '-')
+    def table(self):
+        t = self.a.x/self.b.x if self.b.x else 1.0
+        return '%11s %7s' % (
+            self,
+            '∞%' if t == +m.inf
+            else '-∞%' if t == -m.inf
+            else '%.1f%%' % (100*t))
+
+    diff_none = '%11s' % '-'
+    def diff_table(self):
+        return '%11s' % (self,)
+
+    def diff_diff(self, other):
+        new_a, new_b = self if self else (Int(0), Int(0))
+        old_a, old_b = other if other else (Int(0), Int(0))
+        return '%11s' % ('%s/%s' % (
+            new_a.diff_diff(old_a).strip(),
+            new_b.diff_diff(old_b).strip()))
+
+    def ratio(self, other):
+        new_a, new_b = self if self else (Int(0), Int(0))
+        old_a, old_b = other if other else (Int(0), Int(0))
+        new = new_a.x/new_b.x if new_b.x else 1.0
+        old = old_a.x/old_b.x if old_b.x else 1.0
+        return new - old
+
+    def __add__(self, other):
+        return self.__class__(self.a + other.a, self.b + other.b)
+
+    def __sub__(self, other):
+        return self.__class__(self.a - other.a, self.b - other.b)
+
+    def __mul__(self, other):
+        return self.__class__(self.a * other.a, self.b + other.b)
+
+    def __lt__(self, other):
+        self_t = self.a.x/self.b.x if self.b.x else 1.0
+        other_t = other.a.x/other.b.x if other.b.x else 1.0
+        return (self_t, self.a.x) < (other_t, other.a.x)
+
+    def __gt__(self, other):
+        return self.__class__.__lt__(other, self)
+
+    def __le__(self, other):
+        return not self.__gt__(other)
+
+    def __ge__(self, other):
+        return not self.__lt__(other)
+
+# coverage results
+class CovResult(co.namedtuple('CovResult', [
+        'file', 'function', 'line',
+        'calls', 'hits', 'funcs', 'lines', 'branches'])):
+    _by = ['file', 'function', 'line']
+    _fields = ['calls', 'hits', 'funcs', 'lines', 'branches']
+    _sort = ['funcs', 'lines', 'branches', 'hits', 'calls']
+    _types = {
+        'calls': Int, 'hits': Int,
+        'funcs': Frac, 'lines': Frac, 'branches': Frac}
+
+    __slots__ = ()
+    def __new__(cls, file='', function='', line=0,
+            calls=0, hits=0, funcs=0, lines=0, branches=0):
+        return super().__new__(cls, file, function, int(Int(line)),
+            Int(calls), Int(hits), Frac(funcs), Frac(lines), Frac(branches))
+
+    def __add__(self, other):
+        return CovResult(self.file, self.function, self.line,
+            max(self.calls, other.calls),
+            max(self.hits, other.hits),
+            self.funcs + other.funcs,
+            self.lines + other.lines,
+            self.branches + other.branches)
+
+
+def openio(path, mode='r', buffering=-1):
+    # allow '-' for stdin/stdout
+    if path == '-':
+        if mode == 'r':
+            return os.fdopen(os.dup(sys.stdin.fileno()), mode, buffering)
+        else:
+            return os.fdopen(os.dup(sys.stdout.fileno()), mode, buffering)
+    else:
+        return open(path, mode, buffering)
+
+def collect(gcda_paths, *,
+        gcov_path=GCOV_PATH,
+        sources=None,
+        everything=False,
+        **args):
+    results = []
+    for path in gcda_paths:
+        # get coverage info through gcov's json output
+        # note, gcov-path may contain extra args
+        cmd = GCOV_PATH + ['-b', '-t', '--json-format', path]
+        if args.get('verbose'):
+            print(' '.join(shlex.quote(c) for c in cmd))
+        proc = sp.Popen(cmd,
+            stdout=sp.PIPE,
+            stderr=sp.PIPE if not args.get('verbose') else None,
+            universal_newlines=True,
+            errors='replace',
+            close_fds=False)
+        data = json.load(proc.stdout)
+        proc.wait()
+        if proc.returncode != 0:
+            if not args.get('verbose'):
+                for line in proc.stderr:
+                    sys.stdout.write(line)
+            sys.exit(-1)
+
+        # collect line/branch coverage
+        for file in data['files']:
+            # ignore filtered sources
+            if sources is not None:
+                if not any(
+                        os.path.abspath(file['file']) == os.path.abspath(s)
+                        for s in sources):
+                    continue
+            else:
+                # default to only cwd
+                if not everything and not os.path.commonpath([
+                        os.getcwd(),
+                        os.path.abspath(file['file'])]) == os.getcwd():
+                    continue
+
+            # simplify path
+            if os.path.commonpath([
+                    os.getcwd(),
+                    os.path.abspath(file['file'])]) == os.getcwd():
+                file_name = os.path.relpath(file['file'])
+            else:
+                file_name = os.path.abspath(file['file'])
+
+            for func in file['functions']:
+                func_name = func.get('name', '(inlined)')
+                # discard internal functions (this includes injected test cases)
+                if not everything:
+                    if func_name.startswith('__'):
+                        continue
+
+                # go ahead and add functions, later folding will merge this if
+                # there are other hits on this line
+                results.append(CovResult(
+                    file_name, func_name, func['start_line'],
+                    func['execution_count'], 0,
+                    Frac(1 if func['execution_count'] > 0 else 0, 1),
+                    0,
+                    0))
+
+            for line in file['lines']:
+                func_name = line.get('function_name', '(inlined)')
+                # discard internal function (this includes injected test cases)
+                if not everything:
+                    if func_name.startswith('__'):
+                        continue
+
+                # go ahead and add lines, later folding will merge this if
+                # there are other hits on this line
+                results.append(CovResult(
+                    file_name, func_name, line['line_number'],
+                    0, line['count'],
+                    0,
+                    Frac(1 if line['count'] > 0 else 0, 1),
+                    Frac(
+                        sum(1 if branch['count'] > 0 else 0
+                            for branch in line['branches']),
+                        len(line['branches']))))
+
+    return results
+
+
+def fold(Result, results, *,
+        by=None,
+        defines=None,
+        **_):
+    if by is None:
+        by = Result._by
+
+    for k in it.chain(by or [], (k for k, _ in defines or [])):
+        if k not in Result._by and k not in Result._fields:
+            print("error: could not find field %r?" % k)
+            sys.exit(-1)
+
+    # filter by matching defines
+    if defines is not None:
+        results_ = []
+        for r in results:
+            if all(getattr(r, k) in vs for k, vs in defines):
+                results_.append(r)
+        results = results_
+
+    # organize results into conflicts
+    folding = co.OrderedDict()
+    for r in results:
+        name = tuple(getattr(r, k) for k in by)
+        if name not in folding:
+            folding[name] = []
+        folding[name].append(r)
+
+    # merge conflicts
+    folded = []
+    for name, rs in folding.items():
+        folded.append(sum(rs[1:], start=rs[0]))
+
+    return folded
+
+def table(Result, results, diff_results=None, *,
+        by=None,
+        fields=None,
+        sort=None,
+        summary=False,
+        all=False,
+        percent=False,
+        **_):
+    all_, all = all, __builtins__.all
+
+    if by is None:
+        by = Result._by
+    if fields is None:
+        fields = Result._fields
+    types = Result._types
+
+    # fold again
+    results = fold(Result, results, by=by)
+    if diff_results is not None:
+        diff_results = fold(Result, diff_results, by=by)
+
+    # organize by name
+    table = {
+        ','.join(str(getattr(r, k) or '') for k in by): r
+        for r in results}
+    diff_table = {
+        ','.join(str(getattr(r, k) or '') for k in by): r
+        for r in diff_results or []}
+    names = list(table.keys() | diff_table.keys())
+
+    # sort again, now with diff info, note that python's sort is stable
+    names.sort()
+    if diff_results is not None:
+        names.sort(key=lambda n: tuple(
+            types[k].ratio(
+                getattr(table.get(n), k, None),
+                getattr(diff_table.get(n), k, None))
+            for k in fields),
+            reverse=True)
+    if sort:
+        for k, reverse in reversed(sort):
+            names.sort(
+                key=lambda n: tuple(
+                    (getattr(table[n], k),)
+                    if getattr(table.get(n), k, None) is not None else ()
+                    for k in ([k] if k else [
+                        k for k in Result._sort if k in fields])),
+                reverse=reverse ^ (not k or k in Result._fields))
+
+
+    # build up our lines
+    lines = []
+
+    # header
+    header = []
+    header.append('%s%s' % (
+        ','.join(by),
+        ' (%d added, %d removed)' % (
+            sum(1 for n in table if n not in diff_table),
+            sum(1 for n in diff_table if n not in table))
+            if diff_results is not None and not percent else '')
+        if not summary else '')
+    if diff_results is None:
+        for k in fields:
+            header.append(k)
+    elif percent:
+        for k in fields:
+            header.append(k)
+    else:
+        for k in fields:
+            header.append('o'+k)
+        for k in fields:
+            header.append('n'+k)
+        for k in fields:
+            header.append('d'+k)
+    header.append('')
+    lines.append(header)
+
+    def table_entry(name, r, diff_r=None, ratios=[]):
+        entry = []
+        entry.append(name)
+        if diff_results is None:
+            for k in fields:
+                entry.append(getattr(r, k).table()
+                    if getattr(r, k, None) is not None
+                    else types[k].none)
+        elif percent:
+            for k in fields:
+                entry.append(getattr(r, k).diff_table()
+                    if getattr(r, k, None) is not None
+                    else types[k].diff_none)
+        else:
+            for k in fields:
+                entry.append(getattr(diff_r, k).diff_table()
+                    if getattr(diff_r, k, None) is not None
+                    else types[k].diff_none)
+            for k in fields:
+                entry.append(getattr(r, k).diff_table()
+                    if getattr(r, k, None) is not None
+                    else types[k].diff_none)
+            for k in fields:
+                entry.append(types[k].diff_diff(
+                        getattr(r, k, None),
+                        getattr(diff_r, k, None)))
+        if diff_results is None:
+            entry.append('')
+        elif percent:
+            entry.append(' (%s)' % ', '.join(
+                '+∞%' if t == +m.inf
+                else '-∞%' if t == -m.inf
+                else '%+.1f%%' % (100*t)
+                for t in ratios))
+        else:
+            entry.append(' (%s)' % ', '.join(
+                    '+∞%' if t == +m.inf
+                    else '-∞%' if t == -m.inf
+                    else '%+.1f%%' % (100*t)
+                    for t in ratios
+                    if t)
+                if any(ratios) else '')
+        return entry
+
+    # entries
+    if not summary:
+        for name in names:
+            r = table.get(name)
+            if diff_results is None:
+                diff_r = None
+                ratios = None
+            else:
+                diff_r = diff_table.get(name)
+                ratios = [
+                    types[k].ratio(
+                        getattr(r, k, None),
+                        getattr(diff_r, k, None))
+                    for k in fields]
+                if not all_ and not any(ratios):
+                    continue
+            lines.append(table_entry(name, r, diff_r, ratios))
+
+    # total
+    r = next(iter(fold(Result, results, by=[])), None)
+    if diff_results is None:
+        diff_r = None
+        ratios = None
+    else:
+        diff_r = next(iter(fold(Result, diff_results, by=[])), None)
+        ratios = [
+            types[k].ratio(
+                getattr(r, k, None),
+                getattr(diff_r, k, None))
+            for k in fields]
+    lines.append(table_entry('TOTAL', r, diff_r, ratios))
+
+    # find the best widths, note that column 0 contains the names and column -1
+    # the ratios, so those are handled a bit differently
+    widths = [
+        ((max(it.chain([w], (len(l[i]) for l in lines)))+1+4-1)//4)*4-1
+        for w, i in zip(
+            it.chain([23], it.repeat(7)),
+            range(len(lines[0])-1))]
+
+    # print our table
+    for line in lines:
+        print('%-*s  %s%s' % (
+            widths[0], line[0],
+            ' '.join('%*s' % (w, x)
+                for w, x in zip(widths[1:], line[1:-1])),
+            line[-1]))
+
+
+def annotate(Result, results, *,
+        annotate=False,
+        lines=False,
+        branches=False,
+        **args):
+    # if neither branches/lines specified, color both
+    if annotate and not lines and not branches:
+        lines, branches = True, True
+
+    for path in co.OrderedDict.fromkeys(r.file for r in results).keys():
+        # flatten to line info
+        results = fold(Result, results, by=['file', 'line'])
+        table = {r.line: r for r in results if r.file == path}
+
+        # calculate spans to show
+        if not annotate:
+            spans = []
+            last = None
+            func = None
+            for line, r in sorted(table.items()):
+                if ((lines and int(r.hits) == 0)
+                        or (branches and r.branches.a < r.branches.b)):
+                    if last is not None and line - last.stop <= args['context']:
+                        last = range(
+                            last.start,
+                            line+1+args['context'])
+                    else:
+                        if last is not None:
+                            spans.append((last, func))
+                        last = range(
+                            line-args['context'],
+                            line+1+args['context'])
+                        func = r.function
+            if last is not None:
+                spans.append((last, func))
+
+        with open(path) as f:
+            skipped = False
+            for i, line in enumerate(f):
+                # skip lines not in spans?
+                if not annotate and not any(i+1 in s for s, _ in spans):
+                    skipped = True
+                    continue
+
+                if skipped:
+                    skipped = False
+                    print('%s@@ %s:%d: %s @@%s' % (
+                        '\x1b[36m' if args['color'] else '',
+                        path,
+                        i+1,
+                        next(iter(f for _, f in spans)),
+                        '\x1b[m' if args['color'] else ''))
+
+                # build line
+                if line.endswith('\n'):
+                    line = line[:-1]
+
+                if i+1 in table:
+                    r = table[i+1]
+                    line = '%-*s // %s hits%s' % (
+                        args['width'],
+                        line,
+                        r.hits,
+                        ', %s branches' % (r.branches,)
+                            if int(r.branches.b) else '')
+
+                    if args['color']:
+                        if lines and int(r.hits) == 0:
+                            line = '\x1b[1;31m%s\x1b[m' % line
+                        elif branches and r.branches.a < r.branches.b:
+                            line = '\x1b[35m%s\x1b[m' % line
+
+                print(line)
+
+
+def main(gcda_paths, *,
+        by=None,
+        fields=None,
+        defines=None,
+        sort=None,
+        hits=False,
+        **args):
+    # figure out what color should be
+    if args.get('color') == 'auto':
+        args['color'] = sys.stdout.isatty()
+    elif args.get('color') == 'always':
+        args['color'] = True
+    else:
+        args['color'] = False
+
+    # find sizes
+    if not args.get('use', None):
+        results = collect(gcda_paths, **args)
+    else:
+        results = []
+        with openio(args['use']) as f:
+            reader = csv.DictReader(f, restval='')
+            for r in reader:
+                if not any('cov_'+k in r and r['cov_'+k].strip()
+                        for k in CovResult._fields):
+                    continue
+                try:
+                    results.append(CovResult(
+                        **{k: r[k] for k in CovResult._by
+                            if k in r and r[k].strip()},
+                        **{k: r['cov_'+k]
+                            for k in CovResult._fields
+                            if 'cov_'+k in r
+                                and r['cov_'+k].strip()}))
+                except TypeError:
+                    pass
+
+    # fold
+    results = fold(CovResult, results, by=by, defines=defines)
+
+    # sort, note that python's sort is stable
+    results.sort()
+    if sort:
+        for k, reverse in reversed(sort):
+            results.sort(
+                key=lambda r: tuple(
+                    (getattr(r, k),) if getattr(r, k) is not None else ()
+                    for k in ([k] if k else CovResult._sort)),
+                reverse=reverse ^ (not k or k in CovResult._fields))
+
+    # write results to CSV
+    if args.get('output'):
+        with openio(args['output'], 'w') as f:
+            writer = csv.DictWriter(f,
+                (by if by is not None else CovResult._by)
+                + ['cov_'+k for k in (
+                    fields if fields is not None else CovResult._fields)])
+            writer.writeheader()
+            for r in results:
+                writer.writerow(
+                    {k: getattr(r, k) for k in (
+                        by if by is not None else CovResult._by)}
+                    | {'cov_'+k: getattr(r, k) for k in (
+                        fields if fields is not None else CovResult._fields)})
+
+    # find previous results?
+    if args.get('diff'):
+        diff_results = []
+        try:
+            with openio(args['diff']) as f:
+                reader = csv.DictReader(f, restval='')
+                for r in reader:
+                    if not any('cov_'+k in r and r['cov_'+k].strip()
+                            for k in CovResult._fields):
+                        continue
+                    try:
+                        diff_results.append(CovResult(
+                            **{k: r[k] for k in CovResult._by
+                                if k in r and r[k].strip()},
+                            **{k: r['cov_'+k]
+                                for k in CovResult._fields
+                                if 'cov_'+k in r
+                                    and r['cov_'+k].strip()}))
+                    except TypeError:
+                        pass
+        except FileNotFoundError:
+            pass
+
+        # fold
+        diff_results = fold(CovResult, diff_results,
+            by=by, defines=defines)
+
+    # print table
+    if not args.get('quiet'):
+        if (args.get('annotate')
+                or args.get('lines')
+                or args.get('branches')):
+            # annotate sources
+            annotate(CovResult, results, **args)
+        else:
+            # print table
+            table(CovResult, results,
+                diff_results if args.get('diff') else None,
+                by=by if by is not None else ['function'],
+                fields=fields if fields is not None
+                    else ['lines', 'branches'] if not hits
+                    else ['calls', 'hits'],
+                sort=sort,
+                **args)
+
+    # catch lack of coverage
+    if args.get('error_on_lines') and any(
+            r.lines.a < r.lines.b for r in results):
+        sys.exit(2)
+    elif args.get('error_on_branches') and any(
+            r.branches.a < r.branches.b for r in results):
+        sys.exit(3)
+
+
+if __name__ == "__main__":
+    import argparse
+    import sys
+    parser = argparse.ArgumentParser(
+        description="Find coverage info after running tests.",
+        allow_abbrev=False)
+    parser.add_argument(
+        'gcda_paths',
+        nargs='*',
+        help="Input *.gcda files.")
+    parser.add_argument(
+        '-v', '--verbose',
+        action='store_true',
+        help="Output commands that run behind the scenes.")
+    parser.add_argument(
+        '-q', '--quiet',
+        action='store_true',
+        help="Don't show anything, useful with -o.")
+    parser.add_argument(
+        '-o', '--output',
+        help="Specify CSV file to store results.")
+    parser.add_argument(
+        '-u', '--use',
+        help="Don't parse anything, use this CSV file.")
+    parser.add_argument(
+        '-d', '--diff',
+        help="Specify CSV file to diff against.")
+    parser.add_argument(
+        '-a', '--all',
+        action='store_true',
+        help="Show all, not just the ones that changed.")
+    parser.add_argument(
+        '-p', '--percent',
+        action='store_true',
+        help="Only show percentage change, not a full diff.")
+    parser.add_argument(
+        '-b', '--by',
+        action='append',
+        choices=CovResult._by,
+        help="Group by this field.")
+    parser.add_argument(
+        '-f', '--field',
+        dest='fields',
+        action='append',
+        choices=CovResult._fields,
+        help="Show this field.")
+    parser.add_argument(
+        '-D', '--define',
+        dest='defines',
+        action='append',
+        type=lambda x: (lambda k,v: (k, set(v.split(','))))(*x.split('=', 1)),
+        help="Only include results where this field is this value.")
+    class AppendSort(argparse.Action):
+        def __call__(self, parser, namespace, value, option):
+            if namespace.sort is None:
+                namespace.sort = []
+            namespace.sort.append((value, True if option == '-S' else False))
+    parser.add_argument(
+        '-s', '--sort',
+        nargs='?',
+        action=AppendSort,
+        help="Sort by this field.")
+    parser.add_argument(
+        '-S', '--reverse-sort',
+        nargs='?',
+        action=AppendSort,
+        help="Sort by this field, but backwards.")
+    parser.add_argument(
+        '-Y', '--summary',
+        action='store_true',
+        help="Only show the total.")
+    parser.add_argument(
+        '-F', '--source',
+        dest='sources',
+        action='append',
+        help="Only consider definitions in this file. Defaults to anything "
+            "in the current directory.")
+    parser.add_argument(
+        '--everything',
+        action='store_true',
+        help="Include builtin and libc specific symbols.")
+    parser.add_argument(
+        '--hits',
+        action='store_true',
+        help="Show total hits instead of coverage.")
+    parser.add_argument(
+        '-A', '--annotate',
+        action='store_true',
+        help="Show source files annotated with coverage info.")
+    parser.add_argument(
+        '-L', '--lines',
+        action='store_true',
+        help="Show uncovered lines.")
+    parser.add_argument(
+        '-B', '--branches',
+        action='store_true',
+        help="Show uncovered branches.")
+    parser.add_argument(
+        '-c', '--context',
+        type=lambda x: int(x, 0),
+        default=3,
+        help="Show n additional lines of context. Defaults to 3.")
+    parser.add_argument(
+        '-W', '--width',
+        type=lambda x: int(x, 0),
+        default=80,
+        help="Assume source is styled with this many columns. Defaults to 80.")
+    parser.add_argument(
+        '--color',
+        choices=['never', 'always', 'auto'],
+        default='auto',
+        help="When to use terminal colors. Defaults to 'auto'.")
+    parser.add_argument(
+        '-e', '--error-on-lines',
+        action='store_true',
+        help="Error if any lines are not covered.")
+    parser.add_argument(
+        '-E', '--error-on-branches',
+        action='store_true',
+        help="Error if any branches are not covered.")
+    parser.add_argument(
+        '--gcov-path',
+        default=GCOV_PATH,
+        type=lambda x: x.split(),
+        help="Path to the gcov executable, may include paths. "
+            "Defaults to %r." % GCOV_PATH)
+    sys.exit(main(**{k: v
+        for k, v in vars(parser.parse_intermixed_args()).items()
+        if v is not None}))
diff --git a/scripts/data.py b/scripts/data.py
index 4b8e00d..e9770aa 100644
--- a/scripts/data.py
+++ b/scripts/data.py
@@ -1,42 +1,188 @@
 #!/usr/bin/env python3
 #
-# Script to find data size at the function level. Basically just a bit wrapper
+# Script to find data size at the function level. Basically just a big wrapper
 # around nm with some extra conveniences for comparing builds. Heavily inspired
 # by Linux's Bloat-O-Meter.
 #
+# Example:
+# ./scripts/data.py lfs.o lfs_util.o -Ssize
+#
+# Copyright (c) 2022, The littlefs authors.
+# Copyright (c) 2020, Arm Limited. All rights reserved.
+# SPDX-License-Identifier: BSD-3-Clause
+#
 
-import os
-import glob
-import itertools as it
-import subprocess as sp
-import shlex
-import re
-import csv
 import collections as co
+import csv
+import difflib
+import itertools as it
+import math as m
+import os
+import re
+import shlex
+import subprocess as sp
 
 
-OBJ_PATHS = ['*.o']
+NM_PATH = ['nm']
+NM_TYPES = 'dDbB'
+OBJDUMP_PATH = ['objdump']
 
-def collect(paths, **args):
-    results = co.defaultdict(lambda: 0)
-    pattern = re.compile(
+
+# integer fields
+class Int(co.namedtuple('Int', 'x')):
+    __slots__ = ()
+    def __new__(cls, x=0):
+        if isinstance(x, Int):
+            return x
+        if isinstance(x, str):
+            try:
+                x = int(x, 0)
+            except ValueError:
+                # also accept +-∞ and +-inf
+                if re.match('^\s*\+?\s*(?:∞|inf)\s*$', x):
+                    x = m.inf
+                elif re.match('^\s*-\s*(?:∞|inf)\s*$', x):
+                    x = -m.inf
+                else:
+                    raise
+        assert isinstance(x, int) or m.isinf(x), x
+        return super().__new__(cls, x)
+
+    def __str__(self):
+        if self.x == m.inf:
+            return '∞'
+        elif self.x == -m.inf:
+            return '-∞'
+        else:
+            return str(self.x)
+
+    def __int__(self):
+        assert not m.isinf(self.x)
+        return self.x
+
+    def __float__(self):
+        return float(self.x)
+
+    none = '%7s' % '-'
+    def table(self):
+        return '%7s' % (self,)
+
+    diff_none = '%7s' % '-'
+    diff_table = table
+
+    def diff_diff(self, other):
+        new = self.x if self else 0
+        old = other.x if other else 0
+        diff = new - old
+        if diff == +m.inf:
+            return '%7s' % '+∞'
+        elif diff == -m.inf:
+            return '%7s' % '-∞'
+        else:
+            return '%+7d' % diff
+
+    def ratio(self, other):
+        new = self.x if self else 0
+        old = other.x if other else 0
+        if m.isinf(new) and m.isinf(old):
+            return 0.0
+        elif m.isinf(new):
+            return +m.inf
+        elif m.isinf(old):
+            return -m.inf
+        elif not old and not new:
+            return 0.0
+        elif not old:
+            return 1.0
+        else:
+            return (new-old) / old
+
+    def __add__(self, other):
+        return self.__class__(self.x + other.x)
+
+    def __sub__(self, other):
+        return self.__class__(self.x - other.x)
+
+    def __mul__(self, other):
+        return self.__class__(self.x * other.x)
+
+# data size results
+class DataResult(co.namedtuple('DataResult', [
+        'file', 'function',
+        'size'])):
+    _by = ['file', 'function']
+    _fields = ['size']
+    _sort = ['size']
+    _types = {'size': Int}
+
+    __slots__ = ()
+    def __new__(cls, file='', function='', size=0):
+        return super().__new__(cls, file, function,
+            Int(size))
+
+    def __add__(self, other):
+        return DataResult(self.file, self.function,
+            self.size + other.size)
+
+
+def openio(path, mode='r', buffering=-1):
+    # allow '-' for stdin/stdout
+    if path == '-':
+        if mode == 'r':
+            return os.fdopen(os.dup(sys.stdin.fileno()), mode, buffering)
+        else:
+            return os.fdopen(os.dup(sys.stdout.fileno()), mode, buffering)
+    else:
+        return open(path, mode, buffering)
+
+def collect(obj_paths, *,
+        nm_path=NM_PATH,
+        nm_types=NM_TYPES,
+        objdump_path=OBJDUMP_PATH,
+        sources=None,
+        everything=False,
+        **args):
+    size_pattern = re.compile(
         '^(?P<size>[0-9a-fA-F]+)' +
-        ' (?P<type>[%s])' % re.escape(args['type']) +
+        ' (?P<type>[%s])' % re.escape(nm_types) +
         ' (?P<func>.+?)$')
-    for path in paths:
-        # note nm-tool may contain extra args
-        cmd = args['nm_tool'] + ['--size-sort', path]
+    line_pattern = re.compile(
+        '^\s+(?P<no>[0-9]+)'
+            '(?:\s+(?P<dir>[0-9]+))?'
+            '\s+.*'
+            '\s+(?P<path>[^\s]+)$')
+    info_pattern = re.compile(
+        '^(?:.*(?P<tag>DW_TAG_[a-z_]+).*'
+            '|.*DW_AT_name.*:\s*(?P<name>[^:\s]+)\s*'
+            '|.*DW_AT_decl_file.*:\s*(?P<file>[0-9]+)\s*)$')
+
+    results = []
+    for path in obj_paths:
+        # guess the source, if we have debug-info we'll replace this later
+        file = re.sub('(\.o)?$', '.c', path, 1)
+
+        # find symbol sizes
+        results_ = []
+        # note nm-path may contain extra args
+        cmd = nm_path + ['--size-sort', path]
         if args.get('verbose'):
             print(' '.join(shlex.quote(c) for c in cmd))
         proc = sp.Popen(cmd,
             stdout=sp.PIPE,
             stderr=sp.PIPE if not args.get('verbose') else None,
             universal_newlines=True,
-            errors='replace')
+            errors='replace',
+            close_fds=False)
         for line in proc.stdout:
-            m = pattern.match(line)
+            m = size_pattern.match(line)
             if m:
-                results[(path, m.group('func'))] += int(m.group('size'), 16)
+                func = m.group('func')
+                # discard internal functions
+                if not everything and func.startswith('__'):
+                    continue
+                results_.append(DataResult(
+                    file, func,
+                    int(m.group('size'), 16)))
         proc.wait()
         if proc.returncode != 0:
             if not args.get('verbose'):
@@ -44,240 +190,515 @@ def collect(paths, **args):
                     sys.stdout.write(line)
             sys.exit(-1)
 
-    flat_results = []
-    for (file, func), size in results.items():
-        # map to source files
-        if args.get('build_dir'):
-            file = re.sub('%s/*' % re.escape(args['build_dir']), '', file)
-        # replace .o with .c, different scripts report .o/.c, we need to
-        # choose one if we want to deduplicate csv files
-        file = re.sub('\.o$', '.c', file)
-        # discard internal functions
-        if not args.get('everything'):
-            if func.startswith('__'):
-                continue
-        # discard .8449 suffixes created by optimizer
-        func = re.sub('\.[0-9]+', '', func)
-        flat_results.append((file, func, size))
 
-    return flat_results
+        # try to figure out the source file if we have debug-info
+        dirs = {}
+        files = {}
+        # note objdump-path may contain extra args
+        cmd = objdump_path + ['--dwarf=rawline', path]
+        if args.get('verbose'):
+            print(' '.join(shlex.quote(c) for c in cmd))
+        proc = sp.Popen(cmd,
+            stdout=sp.PIPE,
+            stderr=sp.PIPE if not args.get('verbose') else None,
+            universal_newlines=True,
+            errors='replace',
+            close_fds=False)
+        for line in proc.stdout:
+            # note that files contain references to dirs, which we
+            # dereference as soon as we see them as each file table follows a
+            # dir table
+            m = line_pattern.match(line)
+            if m:
+                if not m.group('dir'):
+                    # found a directory entry
+                    dirs[int(m.group('no'))] = m.group('path')
+                else:
+                    # found a file entry
+                    dir = int(m.group('dir'))
+                    if dir in dirs:
+                        files[int(m.group('no'))] = os.path.join(
+                            dirs[dir],
+                            m.group('path'))
+                    else:
+                        files[int(m.group('no'))] = m.group('path')
+        proc.wait()
+        if proc.returncode != 0:
+            if not args.get('verbose'):
+                for line in proc.stderr:
+                    sys.stdout.write(line)
+            # do nothing on error, we don't need objdump to work, source files
+            # may just be inaccurate
+            pass
 
-def main(**args):
-    def openio(path, mode='r'):
-        if path == '-':
-            if 'r' in mode:
-                return os.fdopen(os.dup(sys.stdin.fileno()), 'r')
+        defs = {}
+        is_func = False
+        f_name = None
+        f_file = None
+        # note objdump-path may contain extra args
+        cmd = objdump_path + ['--dwarf=info', path]
+        if args.get('verbose'):
+            print(' '.join(shlex.quote(c) for c in cmd))
+        proc = sp.Popen(cmd,
+            stdout=sp.PIPE,
+            stderr=sp.PIPE if not args.get('verbose') else None,
+            universal_newlines=True,
+            errors='replace',
+            close_fds=False)
+        for line in proc.stdout:
+            # state machine here to find definitions
+            m = info_pattern.match(line)
+            if m:
+                if m.group('tag'):
+                    if is_func:
+                        defs[f_name] = files.get(f_file, '?')
+                    is_func = (m.group('tag') == 'DW_TAG_subprogram')
+                elif m.group('name'):
+                    f_name = m.group('name')
+                elif m.group('file'):
+                    f_file = int(m.group('file'))
+        if is_func:
+            defs[f_name] = files.get(f_file, '?')
+        proc.wait()
+        if proc.returncode != 0:
+            if not args.get('verbose'):
+                for line in proc.stderr:
+                    sys.stdout.write(line)
+            # do nothing on error, we don't need objdump to work, source files
+            # may just be inaccurate
+            pass
+
+        for r in results_:
+            # find best matching debug symbol, this may be slightly different
+            # due to optimizations
+            if defs:
+                # exact match? avoid difflib if we can for speed
+                if r.function in defs:
+                    file = defs[r.function]
+                else:
+                    _, file = max(
+                        defs.items(),
+                        key=lambda d: difflib.SequenceMatcher(None,
+                            d[0],
+                            r.function, False).ratio())
             else:
-                return os.fdopen(os.dup(sys.stdout.fileno()), 'w')
-        else:
-            return open(path, mode)
+                file = r.file
 
-    # find sizes
-    if not args.get('use', None):
-        # find .o files
-        paths = []
-        for path in args['obj_paths']:
-            if os.path.isdir(path):
-                path = path + '/*.o'
+            # ignore filtered sources
+            if sources is not None:
+                if not any(
+                        os.path.abspath(file) == os.path.abspath(s)
+                        for s in sources):
+                    continue
+            else:
+                # default to only cwd
+                if not everything and not os.path.commonpath([
+                        os.getcwd(),
+                        os.path.abspath(file)]) == os.getcwd():
+                    continue
 
-            for path in glob.glob(path):
-                paths.append(path)
+            # simplify path
+            if os.path.commonpath([
+                    os.getcwd(),
+                    os.path.abspath(file)]) == os.getcwd():
+                file = os.path.relpath(file)
+            else:
+                file = os.path.abspath(file)
 
-        if not paths:
-            print('no .obj files found in %r?' % args['obj_paths'])
+            results.append(r._replace(file=file))
+
+    return results
+
+
+def fold(Result, results, *,
+        by=None,
+        defines=None,
+        **_):
+    if by is None:
+        by = Result._by
+
+    for k in it.chain(by or [], (k for k, _ in defines or [])):
+        if k not in Result._by and k not in Result._fields:
+            print("error: could not find field %r?" % k)
             sys.exit(-1)
 
-        results = collect(paths, **args)
+    # filter by matching defines
+    if defines is not None:
+        results_ = []
+        for r in results:
+            if all(getattr(r, k) in vs for k, vs in defines):
+                results_.append(r)
+        results = results_
+
+    # organize results into conflicts
+    folding = co.OrderedDict()
+    for r in results:
+        name = tuple(getattr(r, k) for k in by)
+        if name not in folding:
+            folding[name] = []
+        folding[name].append(r)
+
+    # merge conflicts
+    folded = []
+    for name, rs in folding.items():
+        folded.append(sum(rs[1:], start=rs[0]))
+
+    return folded
+
+def table(Result, results, diff_results=None, *,
+        by=None,
+        fields=None,
+        sort=None,
+        summary=False,
+        all=False,
+        percent=False,
+        **_):
+    all_, all = all, __builtins__.all
+
+    if by is None:
+        by = Result._by
+    if fields is None:
+        fields = Result._fields
+    types = Result._types
+
+    # fold again
+    results = fold(Result, results, by=by)
+    if diff_results is not None:
+        diff_results = fold(Result, diff_results, by=by)
+
+    # organize by name
+    table = {
+        ','.join(str(getattr(r, k) or '') for k in by): r
+        for r in results}
+    diff_table = {
+        ','.join(str(getattr(r, k) or '') for k in by): r
+        for r in diff_results or []}
+    names = list(table.keys() | diff_table.keys())
+
+    # sort again, now with diff info, note that python's sort is stable
+    names.sort()
+    if diff_results is not None:
+        names.sort(key=lambda n: tuple(
+            types[k].ratio(
+                getattr(table.get(n), k, None),
+                getattr(diff_table.get(n), k, None))
+            for k in fields),
+            reverse=True)
+    if sort:
+        for k, reverse in reversed(sort):
+            names.sort(
+                key=lambda n: tuple(
+                    (getattr(table[n], k),)
+                    if getattr(table.get(n), k, None) is not None else ()
+                    for k in ([k] if k else [
+                        k for k in Result._sort if k in fields])),
+                reverse=reverse ^ (not k or k in Result._fields))
+
+
+    # build up our lines
+    lines = []
+
+    # header
+    header = []
+    header.append('%s%s' % (
+        ','.join(by),
+        ' (%d added, %d removed)' % (
+            sum(1 for n in table if n not in diff_table),
+            sum(1 for n in diff_table if n not in table))
+            if diff_results is not None and not percent else '')
+        if not summary else '')
+    if diff_results is None:
+        for k in fields:
+            header.append(k)
+    elif percent:
+        for k in fields:
+            header.append(k)
     else:
+        for k in fields:
+            header.append('o'+k)
+        for k in fields:
+            header.append('n'+k)
+        for k in fields:
+            header.append('d'+k)
+    header.append('')
+    lines.append(header)
+
+    def table_entry(name, r, diff_r=None, ratios=[]):
+        entry = []
+        entry.append(name)
+        if diff_results is None:
+            for k in fields:
+                entry.append(getattr(r, k).table()
+                    if getattr(r, k, None) is not None
+                    else types[k].none)
+        elif percent:
+            for k in fields:
+                entry.append(getattr(r, k).diff_table()
+                    if getattr(r, k, None) is not None
+                    else types[k].diff_none)
+        else:
+            for k in fields:
+                entry.append(getattr(diff_r, k).diff_table()
+                    if getattr(diff_r, k, None) is not None
+                    else types[k].diff_none)
+            for k in fields:
+                entry.append(getattr(r, k).diff_table()
+                    if getattr(r, k, None) is not None
+                    else types[k].diff_none)
+            for k in fields:
+                entry.append(types[k].diff_diff(
+                        getattr(r, k, None),
+                        getattr(diff_r, k, None)))
+        if diff_results is None:
+            entry.append('')
+        elif percent:
+            entry.append(' (%s)' % ', '.join(
+                '+∞%' if t == +m.inf
+                else '-∞%' if t == -m.inf
+                else '%+.1f%%' % (100*t)
+                for t in ratios))
+        else:
+            entry.append(' (%s)' % ', '.join(
+                    '+∞%' if t == +m.inf
+                    else '-∞%' if t == -m.inf
+                    else '%+.1f%%' % (100*t)
+                    for t in ratios
+                    if t)
+                if any(ratios) else '')
+        return entry
+
+    # entries
+    if not summary:
+        for name in names:
+            r = table.get(name)
+            if diff_results is None:
+                diff_r = None
+                ratios = None
+            else:
+                diff_r = diff_table.get(name)
+                ratios = [
+                    types[k].ratio(
+                        getattr(r, k, None),
+                        getattr(diff_r, k, None))
+                    for k in fields]
+                if not all_ and not any(ratios):
+                    continue
+            lines.append(table_entry(name, r, diff_r, ratios))
+
+    # total
+    r = next(iter(fold(Result, results, by=[])), None)
+    if diff_results is None:
+        diff_r = None
+        ratios = None
+    else:
+        diff_r = next(iter(fold(Result, diff_results, by=[])), None)
+        ratios = [
+            types[k].ratio(
+                getattr(r, k, None),
+                getattr(diff_r, k, None))
+            for k in fields]
+    lines.append(table_entry('TOTAL', r, diff_r, ratios))
+
+    # find the best widths, note that column 0 contains the names and column -1
+    # the ratios, so those are handled a bit differently
+    widths = [
+        ((max(it.chain([w], (len(l[i]) for l in lines)))+1+4-1)//4)*4-1
+        for w, i in zip(
+            it.chain([23], it.repeat(7)),
+            range(len(lines[0])-1))]
+
+    # print our table
+    for line in lines:
+        print('%-*s  %s%s' % (
+            widths[0], line[0],
+            ' '.join('%*s' % (w, x)
+                for w, x in zip(widths[1:], line[1:-1])),
+            line[-1]))
+
+
+def main(obj_paths, *,
+        by=None,
+        fields=None,
+        defines=None,
+        sort=None,
+        **args):
+    # find sizes
+    if not args.get('use', None):
+        results = collect(obj_paths, **args)
+    else:
+        results = []
         with openio(args['use']) as f:
-            r = csv.DictReader(f)
-            results = [
-                (   result['file'],
-                    result['name'],
-                    int(result['data_size']))
-                for result in r
-                if result.get('data_size') not in {None, ''}]
+            reader = csv.DictReader(f, restval='')
+            for r in reader:
+                try:
+                    results.append(DataResult(
+                        **{k: r[k] for k in DataResult._by
+                            if k in r and r[k].strip()},
+                        **{k: r['data_'+k] for k in DataResult._fields
+                            if 'data_'+k in r and r['data_'+k].strip()}))
+                except TypeError:
+                    pass
 
-    total = 0
-    for _, _, size in results:
-        total += size
+    # fold
+    results = fold(DataResult, results, by=by, defines=defines)
 
-    # find previous results?
-    if args.get('diff'):
-        try:
-            with openio(args['diff']) as f:
-                r = csv.DictReader(f)
-                prev_results = [
-                    (   result['file'],
-                        result['name'],
-                        int(result['data_size']))
-                    for result in r
-                    if result.get('data_size') not in {None, ''}]
-        except FileNotFoundError:
-            prev_results = []
-
-        prev_total = 0
-        for _, _, size in prev_results:
-            prev_total += size
+    # sort, note that python's sort is stable
+    results.sort()
+    if sort:
+        for k, reverse in reversed(sort):
+            results.sort(
+                key=lambda r: tuple(
+                    (getattr(r, k),) if getattr(r, k) is not None else ()
+                    for k in ([k] if k else DataResult._sort)),
+                reverse=reverse ^ (not k or k in DataResult._fields))
 
     # write results to CSV
     if args.get('output'):
-        merged_results = co.defaultdict(lambda: {})
-        other_fields = []
-
-        # merge?
-        if args.get('merge'):
-            try:
-                with openio(args['merge']) as f:
-                    r = csv.DictReader(f)
-                    for result in r:
-                        file = result.pop('file', '')
-                        func = result.pop('name', '')
-                        result.pop('data_size', None)
-                        merged_results[(file, func)] = result
-                        other_fields = result.keys()
-            except FileNotFoundError:
-                pass
-
-        for file, func, size in results:
-            merged_results[(file, func)]['data_size'] = size
-
         with openio(args['output'], 'w') as f:
-            w = csv.DictWriter(f, ['file', 'name', *other_fields, 'data_size'])
-            w.writeheader()
-            for (file, func), result in sorted(merged_results.items()):
-                w.writerow({'file': file, 'name': func, **result})
+            writer = csv.DictWriter(f,
+                (by if by is not None else DataResult._by)
+                + ['data_'+k for k in (
+                    fields if fields is not None else DataResult._fields)])
+            writer.writeheader()
+            for r in results:
+                writer.writerow(
+                    {k: getattr(r, k) for k in (
+                        by if by is not None else DataResult._by)}
+                    | {'data_'+k: getattr(r, k) for k in (
+                        fields if fields is not None else DataResult._fields)})
 
-    # print results
-    def dedup_entries(results, by='name'):
-        entries = co.defaultdict(lambda: 0)
-        for file, func, size in results:
-            entry = (file if by == 'file' else func)
-            entries[entry] += size
-        return entries
+    # find previous results?
+    if args.get('diff'):
+        diff_results = []
+        try:
+            with openio(args['diff']) as f:
+                reader = csv.DictReader(f, restval='')
+                for r in reader:
+                    if not any('data_'+k in r and r['data_'+k].strip()
+                            for k in DataResult._fields):
+                        continue
+                    try:
+                        diff_results.append(DataResult(
+                            **{k: r[k] for k in DataResult._by
+                                if k in r and r[k].strip()},
+                            **{k: r['data_'+k] for k in DataResult._fields
+                                if 'data_'+k in r and r['data_'+k].strip()}))
+                    except TypeError:
+                        pass
+        except FileNotFoundError:
+            pass
 
-    def diff_entries(olds, news):
-        diff = co.defaultdict(lambda: (0, 0, 0, 0))
-        for name, new in news.items():
-            diff[name] = (0, new, new, 1.0)
-        for name, old in olds.items():
-            _, new, _, _ = diff[name]
-            diff[name] = (old, new, new-old, (new-old)/old if old else 1.0)
-        return diff
+        # fold
+        diff_results = fold(DataResult, diff_results, by=by, defines=defines)
 
-    def sorted_entries(entries):
-        if args.get('size_sort'):
-            return sorted(entries, key=lambda x: (-x[1], x))
-        elif args.get('reverse_size_sort'):
-            return sorted(entries, key=lambda x: (+x[1], x))
-        else:
-            return sorted(entries)
+    # print table
+    if not args.get('quiet'):
+        table(DataResult, results,
+            diff_results if args.get('diff') else None,
+            by=by if by is not None else ['function'],
+            fields=fields,
+            sort=sort,
+            **args)
 
-    def sorted_diff_entries(entries):
-        if args.get('size_sort'):
-            return sorted(entries, key=lambda x: (-x[1][1], x))
-        elif args.get('reverse_size_sort'):
-            return sorted(entries, key=lambda x: (+x[1][1], x))
-        else:
-            return sorted(entries, key=lambda x: (-x[1][3], x))
-
-    def print_header(by=''):
-        if not args.get('diff'):
-            print('%-36s %7s' % (by, 'size'))
-        else:
-            print('%-36s %7s %7s %7s' % (by, 'old', 'new', 'diff'))
-
-    def print_entry(name, size):
-        print("%-36s %7d" % (name, size))
-
-    def print_diff_entry(name, old, new, diff, ratio):
-        print("%-36s %7s %7s %+7d%s" % (name,
-            old or "-",
-            new or "-",
-            diff,
-            ' (%+.1f%%)' % (100*ratio) if ratio else ''))
-
-    def print_entries(by='name'):
-        entries = dedup_entries(results, by=by)
-
-        if not args.get('diff'):
-            print_header(by=by)
-            for name, size in sorted_entries(entries.items()):
-                print_entry(name, size)
-        else:
-            prev_entries = dedup_entries(prev_results, by=by)
-            diff = diff_entries(prev_entries, entries)
-            print_header(by='%s (%d added, %d removed)' % (by,
-                sum(1 for old, _, _, _ in diff.values() if not old),
-                sum(1 for _, new, _, _ in diff.values() if not new)))
-            for name, (old, new, diff, ratio) in sorted_diff_entries(
-                    diff.items()):
-                if ratio or args.get('all'):
-                    print_diff_entry(name, old, new, diff, ratio)
-
-    def print_totals():
-        if not args.get('diff'):
-            print_entry('TOTAL', total)
-        else:
-            ratio = (0.0 if not prev_total and not total
-                else 1.0 if not prev_total
-                else (total-prev_total)/prev_total)
-            print_diff_entry('TOTAL',
-                prev_total, total,
-                total-prev_total,
-                ratio)
-
-    if args.get('quiet'):
-        pass
-    elif args.get('summary'):
-        print_header()
-        print_totals()
-    elif args.get('files'):
-        print_entries(by='file')
-        print_totals()
-    else:
-        print_entries(by='name')
-        print_totals()
 
 if __name__ == "__main__":
     import argparse
     import sys
     parser = argparse.ArgumentParser(
-        description="Find data size at the function level.")
-    parser.add_argument('obj_paths', nargs='*', default=OBJ_PATHS,
-        help="Description of where to find *.o files. May be a directory \
-            or a list of paths. Defaults to %r." % OBJ_PATHS)
-    parser.add_argument('-v', '--verbose', action='store_true',
+        description="Find data size at the function level.",
+        allow_abbrev=False)
+    parser.add_argument(
+        'obj_paths',
+        nargs='*',
+        help="Input *.o files.")
+    parser.add_argument(
+        '-v', '--verbose',
+        action='store_true',
         help="Output commands that run behind the scenes.")
-    parser.add_argument('-q', '--quiet', action='store_true',
+    parser.add_argument(
+        '-q', '--quiet',
+        action='store_true',
         help="Don't show anything, useful with -o.")
-    parser.add_argument('-o', '--output',
+    parser.add_argument(
+        '-o', '--output',
         help="Specify CSV file to store results.")
-    parser.add_argument('-u', '--use',
-        help="Don't compile and find data sizes, instead use this CSV file.")
-    parser.add_argument('-d', '--diff',
-        help="Specify CSV file to diff data size against.")
-    parser.add_argument('-m', '--merge',
-        help="Merge with an existing CSV file when writing to output.")
-    parser.add_argument('-a', '--all', action='store_true',
-        help="Show all functions, not just the ones that changed.")
-    parser.add_argument('-A', '--everything', action='store_true',
+    parser.add_argument(
+        '-u', '--use',
+        help="Don't parse anything, use this CSV file.")
+    parser.add_argument(
+        '-d', '--diff',
+        help="Specify CSV file to diff against.")
+    parser.add_argument(
+        '-a', '--all',
+        action='store_true',
+        help="Show all, not just the ones that changed.")
+    parser.add_argument(
+        '-p', '--percent',
+        action='store_true',
+        help="Only show percentage change, not a full diff.")
+    parser.add_argument(
+        '-b', '--by',
+        action='append',
+        choices=DataResult._by,
+        help="Group by this field.")
+    parser.add_argument(
+        '-f', '--field',
+        dest='fields',
+        action='append',
+        choices=DataResult._fields,
+        help="Show this field.")
+    parser.add_argument(
+        '-D', '--define',
+        dest='defines',
+        action='append',
+        type=lambda x: (lambda k,v: (k, set(v.split(','))))(*x.split('=', 1)),
+        help="Only include results where this field is this value.")
+    class AppendSort(argparse.Action):
+        def __call__(self, parser, namespace, value, option):
+            if namespace.sort is None:
+                namespace.sort = []
+            namespace.sort.append((value, True if option == '-S' else False))
+    parser.add_argument(
+        '-s', '--sort',
+        nargs='?',
+        action=AppendSort,
+        help="Sort by this field.")
+    parser.add_argument(
+        '-S', '--reverse-sort',
+        nargs='?',
+        action=AppendSort,
+        help="Sort by this field, but backwards.")
+    parser.add_argument(
+        '-Y', '--summary',
+        action='store_true',
+        help="Only show the total.")
+    parser.add_argument(
+        '-F', '--source',
+        dest='sources',
+        action='append',
+        help="Only consider definitions in this file. Defaults to anything "
+            "in the current directory.")
+    parser.add_argument(
+        '--everything',
+        action='store_true',
         help="Include builtin and libc specific symbols.")
-    parser.add_argument('-s', '--size-sort', action='store_true',
-        help="Sort by size.")
-    parser.add_argument('-S', '--reverse-size-sort', action='store_true',
-        help="Sort by size, but backwards.")
-    parser.add_argument('-F', '--files', action='store_true',
-        help="Show file-level data sizes. Note this does not include padding! "
-            "So sizes may differ from other tools.")
-    parser.add_argument('-Y', '--summary', action='store_true',
-        help="Only show the total data size.")
-    parser.add_argument('--type', default='dDbB',
+    parser.add_argument(
+        '--nm-types',
+        default=NM_TYPES,
         help="Type of symbols to report, this uses the same single-character "
-            "type-names emitted by nm. Defaults to %(default)r.")
-    parser.add_argument('--nm-tool', default=['nm'], type=lambda x: x.split(),
-        help="Path to the nm tool to use.")
-    parser.add_argument('--build-dir',
-        help="Specify the relative build directory. Used to map object files \
-            to the correct source files.")
-    sys.exit(main(**vars(parser.parse_args())))
+            "type-names emitted by nm. Defaults to %r." % NM_TYPES)
+    parser.add_argument(
+        '--nm-path',
+        type=lambda x: x.split(),
+        default=NM_PATH,
+        help="Path to the nm executable, may include flags. "
+            "Defaults to %r." % NM_PATH)
+    parser.add_argument(
+        '--objdump-path',
+        type=lambda x: x.split(),
+        default=OBJDUMP_PATH,
+        help="Path to the objdump executable, may include flags. "
+            "Defaults to %r." % OBJDUMP_PATH)
+    sys.exit(main(**{k: v
+        for k, v in vars(parser.parse_intermixed_args()).items()
+        if v is not None}))
diff --git a/scripts/perf.py b/scripts/perf.py
new file mode 100644
index 0000000..2ee006c
--- /dev/null
+++ b/scripts/perf.py
@@ -0,0 +1,1344 @@
+#!/usr/bin/env python3
+#
+# Script to aggregate and report Linux perf results.
+#
+# Example:
+# ./scripts/perf.py -R -obench.perf ./runners/bench_runner
+# ./scripts/perf.py bench.perf -j -Flfs.c -Flfs_util.c -Scycles
+#
+# Copyright (c) 2022, The littlefs authors.
+# SPDX-License-Identifier: BSD-3-Clause
+#
+
+import bisect
+import collections as co
+import csv
+import errno
+import fcntl
+import functools as ft
+import itertools as it
+import math as m
+import multiprocessing as mp
+import os
+import re
+import shlex
+import shutil
+import subprocess as sp
+import tempfile
+import zipfile
+
+# TODO support non-zip perf results?
+
+
+PERF_PATH = ['perf']
+PERF_EVENTS = 'cycles,branch-misses,branches,cache-misses,cache-references'
+PERF_FREQ = 100
+OBJDUMP_PATH = ['objdump']
+THRESHOLD = (0.5, 0.85)
+
+
+# integer fields
+class Int(co.namedtuple('Int', 'x')):
+    __slots__ = ()
+    def __new__(cls, x=0):
+        if isinstance(x, Int):
+            return x
+        if isinstance(x, str):
+            try:
+                x = int(x, 0)
+            except ValueError:
+                # also accept +-∞ and +-inf
+                if re.match('^\s*\+?\s*(?:∞|inf)\s*$', x):
+                    x = m.inf
+                elif re.match('^\s*-\s*(?:∞|inf)\s*$', x):
+                    x = -m.inf
+                else:
+                    raise
+        assert isinstance(x, int) or m.isinf(x), x
+        return super().__new__(cls, x)
+
+    def __str__(self):
+        if self.x == m.inf:
+            return '∞'
+        elif self.x == -m.inf:
+            return '-∞'
+        else:
+            return str(self.x)
+
+    def __int__(self):
+        assert not m.isinf(self.x)
+        return self.x
+
+    def __float__(self):
+        return float(self.x)
+
+    none = '%7s' % '-'
+    def table(self):
+        return '%7s' % (self,)
+
+    diff_none = '%7s' % '-'
+    diff_table = table
+
+    def diff_diff(self, other):
+        new = self.x if self else 0
+        old = other.x if other else 0
+        diff = new - old
+        if diff == +m.inf:
+            return '%7s' % '+∞'
+        elif diff == -m.inf:
+            return '%7s' % '-∞'
+        else:
+            return '%+7d' % diff
+
+    def ratio(self, other):
+        new = self.x if self else 0
+        old = other.x if other else 0
+        if m.isinf(new) and m.isinf(old):
+            return 0.0
+        elif m.isinf(new):
+            return +m.inf
+        elif m.isinf(old):
+            return -m.inf
+        elif not old and not new:
+            return 0.0
+        elif not old:
+            return 1.0
+        else:
+            return (new-old) / old
+
+    def __add__(self, other):
+        return self.__class__(self.x + other.x)
+
+    def __sub__(self, other):
+        return self.__class__(self.x - other.x)
+
+    def __mul__(self, other):
+        return self.__class__(self.x * other.x)
+
+# perf results
+class PerfResult(co.namedtuple('PerfResult', [
+        'file', 'function', 'line',
+        'cycles', 'bmisses', 'branches', 'cmisses', 'caches',
+        'children'])):
+    _by = ['file', 'function', 'line']
+    _fields = ['cycles', 'bmisses', 'branches', 'cmisses', 'caches']
+    _sort = ['cycles', 'bmisses', 'cmisses', 'branches', 'caches']
+    _types = {
+        'cycles': Int,
+        'bmisses': Int, 'branches': Int,
+        'cmisses': Int, 'caches': Int}
+
+    __slots__ = ()
+    def __new__(cls, file='', function='', line=0,
+            cycles=0, bmisses=0, branches=0, cmisses=0, caches=0,
+            children=[]):
+        return super().__new__(cls, file, function, int(Int(line)),
+            Int(cycles), Int(bmisses), Int(branches), Int(cmisses), Int(caches),
+            children)
+
+    def __add__(self, other):
+        return PerfResult(self.file, self.function, self.line,
+            self.cycles + other.cycles,
+            self.bmisses + other.bmisses,
+            self.branches + other.branches,
+            self.cmisses + other.cmisses,
+            self.caches + other.caches,
+            self.children + other.children)
+
+
+def openio(path, mode='r', buffering=-1):
+    # allow '-' for stdin/stdout
+    if path == '-':
+        if mode == 'r':
+            return os.fdopen(os.dup(sys.stdin.fileno()), mode, buffering)
+        else:
+            return os.fdopen(os.dup(sys.stdout.fileno()), mode, buffering)
+    else:
+        return open(path, mode, buffering)
+
+# run perf as a subprocess, storing measurements into a zip file
+def record(command, *,
+        output=None,
+        perf_freq=PERF_FREQ,
+        perf_period=None,
+        perf_events=PERF_EVENTS,
+        perf_path=PERF_PATH,
+        **args):
+    # create a temporary file for perf to write to, as far as I can tell
+    # this is strictly needed because perf's pipe-mode only works with stdout
+    with tempfile.NamedTemporaryFile('rb') as f:
+        # figure out our perf invocation
+        perf = perf_path + list(filter(None, [
+            'record',
+            '-F%s' % perf_freq
+                if perf_freq is not None
+                and perf_period is None else None,
+            '-c%s' % perf_period
+                if perf_period is not None else None,
+            '-B',
+            '-g',
+            '--all-user',
+            '-e%s' % perf_events,
+            '-o%s' % f.name]))
+
+        # run our command
+        try:
+            if args.get('verbose'):
+                print(' '.join(shlex.quote(c) for c in perf + command))
+            err = sp.call(perf + command, close_fds=False)
+
+        except KeyboardInterrupt:
+            err = errno.EOWNERDEAD
+
+        # synchronize access
+        z = os.open(output, os.O_RDWR | os.O_CREAT)
+        fcntl.flock(z, fcntl.LOCK_EX)
+
+        # copy measurements into our zip file
+        with os.fdopen(z, 'r+b') as z:
+            with zipfile.ZipFile(z, 'a',
+                    compression=zipfile.ZIP_DEFLATED,
+                    compresslevel=1) as z:
+                with z.open('perf.%d' % os.getpid(), 'w') as g:
+                    shutil.copyfileobj(f, g)
+
+    # forward the return code
+    return err
+
+
+# try to only process each dso onceS
+#
+# note this only caches with the non-keyword arguments
+def multiprocessing_cache(f):
+    local_cache = {}
+    manager = mp.Manager()
+    global_cache = manager.dict()
+    lock = mp.Lock()
+
+    def multiprocessing_cache(*args, **kwargs):
+        # check local cache?
+        if args in local_cache:
+            return local_cache[args]
+        # check global cache?
+        with lock:
+            if args in global_cache:
+                v = global_cache[args]
+                local_cache[args] = v
+                return v
+            # fall back to calling the function
+            v = f(*args, **kwargs)
+            global_cache[args] = v
+            local_cache[args] = v
+            return v
+
+    return multiprocessing_cache
+
+@multiprocessing_cache
+def collect_syms_and_lines(obj_path, *,
+        objdump_path=None,
+        **args):
+    symbol_pattern = re.compile(
+        '^(?P<addr>[0-9a-fA-F]+)'
+            '\s+.*'
+            '\s+(?P<size>[0-9a-fA-F]+)'
+            '\s+(?P<name>[^\s]+)\s*$')
+    line_pattern = re.compile(
+        '^\s+(?:'
+            # matches dir/file table
+            '(?P<no>[0-9]+)'
+                '(?:\s+(?P<dir>[0-9]+))?'
+                '\s+.*'
+                '\s+(?P<path>[^\s]+)'
+            # matches line opcodes
+            '|' '\[[^\]]*\]\s+'
+                '(?:'
+                    '(?P<op_special>Special)'
+                    '|' '(?P<op_copy>Copy)'
+                    '|' '(?P<op_end>End of Sequence)'
+                    '|' 'File .*?to (?:entry )?(?P<op_file>\d+)'
+                    '|' 'Line .*?to (?P<op_line>[0-9]+)'
+                    '|' '(?:Address|PC) .*?to (?P<op_addr>[0x0-9a-fA-F]+)'
+                    '|' '.' ')*'
+            ')$', re.IGNORECASE)
+
+    # figure out symbol addresses and file+line ranges
+    syms = {}
+    sym_at = []
+    cmd = objdump_path + ['-t', obj_path]
+    if args.get('verbose'):
+        print(' '.join(shlex.quote(c) for c in cmd))
+    proc = sp.Popen(cmd,
+        stdout=sp.PIPE,
+        stderr=sp.PIPE if not args.get('verbose') else None,
+        universal_newlines=True,
+        errors='replace',
+        close_fds=False)
+    for line in proc.stdout:
+        m = symbol_pattern.match(line)
+        if m:
+            name = m.group('name')
+            addr = int(m.group('addr'), 16)
+            size = int(m.group('size'), 16)
+            # ignore zero-sized symbols
+            if not size:
+                continue
+            # note multiple symbols can share a name
+            if name not in syms:
+                syms[name] = set()
+            syms[name].add((addr, size))
+            sym_at.append((addr, name, size))
+    proc.wait()
+    if proc.returncode != 0:
+        if not args.get('verbose'):
+            for line in proc.stderr:
+                sys.stdout.write(line)
+        # assume no debug-info on failure
+        pass
+
+    # sort and keep largest/first when duplicates
+    sym_at.sort(key=lambda x: (x[0], -x[2], x[1]))
+    sym_at_ = []
+    for addr, name, size in sym_at:
+        if len(sym_at_) == 0 or sym_at_[-1][0] != addr:
+            sym_at_.append((addr, name, size))
+    sym_at = sym_at_
+
+    # state machine for dwarf line numbers, note that objdump's
+    # decodedline seems to have issues with multiple dir/file
+    # tables, which is why we need this
+    lines = []
+    line_at = []
+    dirs = {}
+    files = {}
+    op_file = 1
+    op_line = 1
+    op_addr = 0
+    cmd = objdump_path + ['--dwarf=rawline', obj_path]
+    if args.get('verbose'):
+        print(' '.join(shlex.quote(c) for c in cmd))
+    proc = sp.Popen(cmd,
+        stdout=sp.PIPE,
+        stderr=sp.PIPE if not args.get('verbose') else None,
+        universal_newlines=True,
+        errors='replace',
+        close_fds=False)
+    for line in proc.stdout:
+        m = line_pattern.match(line)
+        if m:
+            if m.group('no') and not m.group('dir'):
+                # found a directory entry
+                dirs[int(m.group('no'))] = m.group('path')
+            elif m.group('no'):
+                # found a file entry
+                dir = int(m.group('dir'))
+                if dir in dirs:
+                    files[int(m.group('no'))] = os.path.join(
+                        dirs[dir],
+                        m.group('path'))
+                else:
+                    files[int(m.group('no'))] = m.group('path')
+            else:
+                # found a state machine update
+                if m.group('op_file'):
+                    op_file = int(m.group('op_file'), 0)
+                if m.group('op_line'):
+                    op_line = int(m.group('op_line'), 0)
+                if m.group('op_addr'):
+                    op_addr = int(m.group('op_addr'), 0)
+
+                if (m.group('op_special')
+                        or m.group('op_copy')
+                        or m.group('op_end')):
+                    file = os.path.abspath(files.get(op_file, '?'))
+                    lines.append((file, op_line, op_addr))
+                    line_at.append((op_addr, file, op_line))
+
+                if m.group('op_end'):
+                    op_file = 1
+                    op_line = 1
+                    op_addr = 0
+    proc.wait()
+    if proc.returncode != 0:
+        if not args.get('verbose'):
+            for line in proc.stderr:
+                sys.stdout.write(line)
+        # assume no debug-info on failure
+        pass
+
+    # sort and keep first when duplicates
+    lines.sort()
+    lines_ = []
+    for file, line, addr in lines:
+        if len(lines_) == 0 or lines_[-1][0] != file or lines[-1][1] != line:
+            lines_.append((file, line, addr))
+    lines = lines_
+
+    # sort and keep first when duplicates
+    line_at.sort()
+    line_at_ = []
+    for addr, file, line in line_at:
+        if len(line_at_) == 0 or line_at_[-1][0] != addr:
+            line_at_.append((addr, file, line))
+    line_at = line_at_
+
+    return syms, sym_at, lines, line_at
+
+
+def collect_decompressed(path, *,
+        perf_path=PERF_PATH,
+        sources=None,
+        everything=False,
+        propagate=0,
+        depth=1,
+        **args):
+    sample_pattern = re.compile(
+        '(?P<comm>\w+)'
+        '\s+(?P<pid>\w+)'
+        '\s+(?P<time>[\w.]+):'
+        '\s*(?P<period>\w+)'
+        '\s+(?P<event>[^:]+):')
+    frame_pattern = re.compile(
+        '\s+(?P<addr>\w+)'
+        '\s+(?P<sym>[^\s\+]+)(?:\+(?P<off>\w+))?'
+        '\s+\((?P<dso>[^\)]+)\)')
+    events = {
+        'cycles':           'cycles',
+        'branch-misses':    'bmisses',
+        'branches':         'branches',
+        'cache-misses':     'cmisses',
+        'cache-references': 'caches'}
+
+    # note perf_path may contain extra args
+    cmd = perf_path + [
+        'script',
+        '-i%s' % path]
+    if args.get('verbose'):
+        print(' '.join(shlex.quote(c) for c in cmd))
+    proc = sp.Popen(cmd,
+        stdout=sp.PIPE,
+        stderr=sp.PIPE if not args.get('verbose') else None,
+        universal_newlines=True,
+        errors='replace',
+        close_fds=False)
+
+    last_filtered = False
+    last_event = ''
+    last_period = 0
+    last_stack = []
+    deltas = co.defaultdict(lambda: {})
+    syms_ = co.defaultdict(lambda: {})
+    at_cache = {}
+    results = {}
+
+    def commit():
+        # tail-recursively propagate measurements
+        for i in range(len(last_stack)):
+            results_ = results
+            for j in reversed(range(i+1)):
+                if i+1-j > depth:
+                    break
+
+                # propagate
+                name = last_stack[j]
+                if name not in results_:
+                    results_[name] = (co.defaultdict(lambda: 0), {})
+                results_[name][0][last_event] += last_period
+
+                # recurse
+                results_ = results_[name][1]
+
+    for line in proc.stdout:
+        # we need to process a lot of data, so wait to use regex as late
+        # as possible
+        if not line.startswith('\t'):
+            if last_filtered:
+                commit()
+            last_filtered = False
+
+            if line:
+                m = sample_pattern.match(line)
+                if m and m.group('event') in events:
+                    last_filtered = True
+                    last_event = m.group('event')
+                    last_period = int(m.group('period'), 0)
+                    last_stack = []
+
+        elif last_filtered:
+            m = frame_pattern.match(line)
+            if m:
+                # filter out internal/kernel functions
+                if not everything and (
+                        m.group('sym').startswith('__')
+                        or m.group('sym').startswith('0')
+                        or m.group('sym').startswith('-')
+                        or m.group('sym').startswith('[')
+                        or m.group('dso').startswith('/usr/lib')):
+                    continue
+
+                dso = m.group('dso')
+                sym = m.group('sym')
+                off = int(m.group('off'), 0) if m.group('off') else 0
+                addr_ = int(m.group('addr'), 16)
+
+                # get the syms/lines for the dso, this is cached
+                syms, sym_at, lines, line_at = collect_syms_and_lines(
+                    dso,
+                    **args)
+
+                # ASLR is tricky, we have symbols+offsets, but static symbols
+                # means we may have multiple options for each symbol.
+                #
+                # To try to solve this, we use previous seen symbols to build
+                # confidence for the correct ASLR delta. This means we may
+                # guess incorrectly for early symbols, but this will only affect
+                # a few samples.
+                if sym in syms:
+                    sym_addr_ = addr_ - off
+
+                    # track possible deltas?
+                    for sym_addr, size in syms[sym]:
+                        delta = sym_addr - sym_addr_
+                        if delta not in deltas[dso]:
+                            deltas[dso][delta] = sum(
+                                abs(a_+delta - a)
+                                for s, (a_, _) in syms_[dso].items()
+                                for a, _ in syms[s])
+                    for delta in deltas[dso].keys():
+                        deltas[dso][delta] += abs(sym_addr_+delta - sym_addr)
+                    syms_[dso][sym] = sym_addr_, size
+
+                    # guess the best delta
+                    delta, _ = min(deltas[dso].items(),
+                        key=lambda x: (x[1], x[0]))
+                    addr = addr_ + delta
+
+                    # cached?
+                    if (dso,addr) in at_cache:
+                        cached = at_cache[(dso,addr)]
+                        if cached is None:
+                            # cache says to skip
+                            continue
+                        file, line = cached
+                    else:
+                        # find file+line
+                        i = bisect.bisect(line_at, addr, key=lambda x: x[0])
+                        if i > 0:
+                            _, file, line = line_at[i-1]
+                        else:
+                            file, line = re.sub('(\.o)?$', '.c', dso, 1), 0
+
+                        # ignore filtered sources
+                        if sources is not None:
+                            if not any(
+                                    os.path.abspath(file) == os.path.abspath(s)
+                                    for s in sources):
+                                at_cache[(dso,addr)] = None
+                                continue
+                        else:
+                            # default to only cwd
+                            if not everything and not os.path.commonpath([
+                                    os.getcwd(),
+                                    os.path.abspath(file)]) == os.getcwd():
+                                at_cache[(dso,addr)] = None
+                                continue
+
+                        # simplify path
+                        if os.path.commonpath([
+                                os.getcwd(),
+                                os.path.abspath(file)]) == os.getcwd():
+                            file = os.path.relpath(file)
+                        else:
+                            file = os.path.abspath(file)
+
+                        at_cache[(dso,addr)] = file, line
+                else:
+                    file, line = re.sub('(\.o)?$', '.c', dso, 1), 0
+
+                last_stack.append((file, sym, line))
+
+                # stop propogating?
+                if propagate and len(last_stack) >= propagate:
+                    commit()
+                    last_filtered = False
+    if last_filtered:
+        commit()
+
+    proc.wait()
+    if proc.returncode != 0:
+        if not args.get('verbose'):
+            for line in proc.stderr:
+                sys.stdout.write(line)
+        sys.exit(-1)
+
+    # rearrange results into result type
+    def to_results(results):
+        results_ = []
+        for name, (r, children) in results.items():
+            results_.append(PerfResult(*name,
+                **{events[k]: v for k, v in r.items()},
+                children=to_results(children)))
+        return results_
+
+    return to_results(results)
+
+def collect_job(path, i, **args):
+    # decompress into a temporary file, this is to work around
+    # some limitations of perf
+    with zipfile.ZipFile(path) as z:
+        with z.open(i) as f:
+            with tempfile.NamedTemporaryFile('wb') as g:
+                shutil.copyfileobj(f, g)
+                g.flush()
+
+                return collect_decompressed(g.name, **args)
+
+def starapply(args):
+    f, args, kwargs = args
+    return f(*args, **kwargs)
+
+def collect(perf_paths, *,
+        jobs=None,
+        **args):
+    # automatic job detection?
+    if jobs == 0:
+        jobs = len(os.sched_getaffinity(0))
+
+    records = []
+    for path in perf_paths:
+        # each .perf file is actually a zip file containing perf files from
+        # multiple runs
+        with zipfile.ZipFile(path) as z:
+            records.extend((path, i) for i in z.infolist())
+
+    # we're dealing with a lot of data but also surprisingly
+    # parallelizable
+    if jobs is not None:
+        results = []
+        with mp.Pool(jobs) as p:
+            for results_ in p.imap_unordered(
+                    starapply,
+                    ((collect_job, (path, i), args) for path, i in records)):
+                results.extend(results_)
+    else:
+        results = []
+        for path, i in records:
+            results.extend(collect_job(path, i, **args))
+
+    return results
+
+
+def fold(Result, results, *,
+        by=None,
+        defines=None,
+        **_):
+    if by is None:
+        by = Result._by
+
+    for k in it.chain(by or [], (k for k, _ in defines or [])):
+        if k not in Result._by and k not in Result._fields:
+            print("error: could not find field %r?" % k)
+            sys.exit(-1)
+
+    # filter by matching defines
+    if defines is not None:
+        results_ = []
+        for r in results:
+            if all(getattr(r, k) in vs for k, vs in defines):
+                results_.append(r)
+        results = results_
+
+    # organize results into conflicts
+    folding = co.OrderedDict()
+    for r in results:
+        name = tuple(getattr(r, k) for k in by)
+        if name not in folding:
+            folding[name] = []
+        folding[name].append(r)
+
+    # merge conflicts
+    folded = []
+    for name, rs in folding.items():
+        folded.append(sum(rs[1:], start=rs[0]))
+
+    # fold recursively
+    folded_ = []
+    for r in folded:
+        folded_.append(r._replace(children=fold(
+            Result, r.children,
+            by=by,
+            defines=defines)))
+    folded = folded_
+
+    return folded
+
+def table(Result, results, diff_results=None, *,
+        by=None,
+        fields=None,
+        sort=None,
+        summary=False,
+        all=False,
+        percent=False,
+        depth=1,
+        **_):
+    all_, all = all, __builtins__.all
+
+    if by is None:
+        by = Result._by
+    if fields is None:
+        fields = Result._fields
+    types = Result._types
+
+    # fold again
+    results = fold(Result, results, by=by)
+    if diff_results is not None:
+        diff_results = fold(Result, diff_results, by=by)
+
+    # organize by name
+    table = {
+        ','.join(str(getattr(r, k) or '') for k in by): r
+        for r in results}
+    diff_table = {
+        ','.join(str(getattr(r, k) or '') for k in by): r
+        for r in diff_results or []}
+    names = list(table.keys() | diff_table.keys())
+
+    # sort again, now with diff info, note that python's sort is stable
+    names.sort()
+    if diff_results is not None:
+        names.sort(key=lambda n: tuple(
+            types[k].ratio(
+                getattr(table.get(n), k, None),
+                getattr(diff_table.get(n), k, None))
+            for k in fields),
+            reverse=True)
+    if sort:
+        for k, reverse in reversed(sort):
+            names.sort(
+                key=lambda n: tuple(
+                    (getattr(table[n], k),)
+                    if getattr(table.get(n), k, None) is not None else ()
+                    for k in ([k] if k else [
+                        k for k in Result._sort if k in fields])),
+                reverse=reverse ^ (not k or k in Result._fields))
+
+
+    # build up our lines
+    lines = []
+
+    # header
+    header = []
+    header.append('%s%s' % (
+        ','.join(by),
+        ' (%d added, %d removed)' % (
+            sum(1 for n in table if n not in diff_table),
+            sum(1 for n in diff_table if n not in table))
+            if diff_results is not None and not percent else '')
+        if not summary else '')
+    if diff_results is None:
+        for k in fields:
+            header.append(k)
+    elif percent:
+        for k in fields:
+            header.append(k)
+    else:
+        for k in fields:
+            header.append('o'+k)
+        for k in fields:
+            header.append('n'+k)
+        for k in fields:
+            header.append('d'+k)
+    header.append('')
+    lines.append(header)
+
+    def table_entry(name, r, diff_r=None, ratios=[]):
+        entry = []
+        entry.append(name)
+        if diff_results is None:
+            for k in fields:
+                entry.append(getattr(r, k).table()
+                    if getattr(r, k, None) is not None
+                    else types[k].none)
+        elif percent:
+            for k in fields:
+                entry.append(getattr(r, k).diff_table()
+                    if getattr(r, k, None) is not None
+                    else types[k].diff_none)
+        else:
+            for k in fields:
+                entry.append(getattr(diff_r, k).diff_table()
+                    if getattr(diff_r, k, None) is not None
+                    else types[k].diff_none)
+            for k in fields:
+                entry.append(getattr(r, k).diff_table()
+                    if getattr(r, k, None) is not None
+                    else types[k].diff_none)
+            for k in fields:
+                entry.append(types[k].diff_diff(
+                        getattr(r, k, None),
+                        getattr(diff_r, k, None)))
+        if diff_results is None:
+            entry.append('')
+        elif percent:
+            entry.append(' (%s)' % ', '.join(
+                '+∞%' if t == +m.inf
+                else '-∞%' if t == -m.inf
+                else '%+.1f%%' % (100*t)
+                for t in ratios))
+        else:
+            entry.append(' (%s)' % ', '.join(
+                    '+∞%' if t == +m.inf
+                    else '-∞%' if t == -m.inf
+                    else '%+.1f%%' % (100*t)
+                    for t in ratios
+                    if t)
+                if any(ratios) else '')
+        return entry
+
+    # entries
+    if not summary:
+        for name in names:
+            r = table.get(name)
+            if diff_results is None:
+                diff_r = None
+                ratios = None
+            else:
+                diff_r = diff_table.get(name)
+                ratios = [
+                    types[k].ratio(
+                        getattr(r, k, None),
+                        getattr(diff_r, k, None))
+                    for k in fields]
+                if not all_ and not any(ratios):
+                    continue
+            lines.append(table_entry(name, r, diff_r, ratios))
+
+    # total
+    r = next(iter(fold(Result, results, by=[])), None)
+    if diff_results is None:
+        diff_r = None
+        ratios = None
+    else:
+        diff_r = next(iter(fold(Result, diff_results, by=[])), None)
+        ratios = [
+            types[k].ratio(
+                getattr(r, k, None),
+                getattr(diff_r, k, None))
+            for k in fields]
+    lines.append(table_entry('TOTAL', r, diff_r, ratios))
+
+    # find the best widths, note that column 0 contains the names and column -1
+    # the ratios, so those are handled a bit differently
+    widths = [
+        ((max(it.chain([w], (len(l[i]) for l in lines)))+1+4-1)//4)*4-1
+        for w, i in zip(
+            it.chain([23], it.repeat(7)),
+            range(len(lines[0])-1))]
+
+    # adjust the name width based on the expected call depth, though
+    # note this doesn't really work with unbounded recursion
+    if not summary and not m.isinf(depth):
+        widths[0] += 4*(depth-1)
+
+    # print the tree recursively
+    print('%-*s  %s%s' % (
+        widths[0], lines[0][0],
+        ' '.join('%*s' % (w, x)
+            for w, x in zip(widths[1:], lines[0][1:-1])),
+        lines[0][-1]))
+
+    if not summary:
+        def recurse(results_, depth_, prefixes=('', '', '', '')):
+            # rebuild our tables at each layer
+            table_ = {
+                ','.join(str(getattr(r, k) or '') for k in by): r
+                for r in results_}
+            names_ = list(table_.keys())
+
+            # sort again at each layer, keep in mind the numbers are
+            # changing as we descend
+            names_.sort()
+            if sort:
+                for k, reverse in reversed(sort):
+                    names_.sort(
+                        key=lambda n: tuple(
+                            (getattr(table_[n], k),)
+                            if getattr(table_.get(n), k, None) is not None
+                            else ()
+                            for k in ([k] if k else [
+                                k for k in Result._sort if k in fields])),
+                        reverse=reverse ^ (not k or k in Result._fields))
+
+            for i, name in enumerate(names_):
+                r = table_[name]
+                is_last = (i == len(names_)-1)
+
+                print('%s%-*s  %s' % (
+                    prefixes[0+is_last],
+                    widths[0] - (
+                        len(prefixes[0+is_last])
+                        if not m.isinf(depth) else 0),
+                    name,
+                    ' '.join('%*s' % (w, x)
+                        for w, x in zip(
+                            widths[1:],
+                            table_entry(name, r)[1:]))))
+
+                # recurse?
+                if depth_ > 1:
+                    recurse(
+                        r.children,
+                        depth_-1,
+                        (prefixes[2+is_last] + "|-> ",
+                         prefixes[2+is_last] + "'-> ",
+                         prefixes[2+is_last] + "|   ",
+                         prefixes[2+is_last] + "    "))
+
+        # we have enough going on with diffing to make the top layer
+        # a special case
+        for name, line in zip(names, lines[1:-1]):
+            print('%-*s  %s%s' % (
+                widths[0], line[0],
+                ' '.join('%*s' % (w, x)
+                    for w, x in zip(widths[1:], line[1:-1])),
+                line[-1]))
+
+            if name in table and depth > 1:
+                recurse(
+                    table[name].children,
+                    depth-1,
+                    ("|-> ",
+                     "'-> ",
+                     "|   ",
+                     "    "))
+
+    print('%-*s  %s%s' % (
+        widths[0], lines[-1][0],
+        ' '.join('%*s' % (w, x)
+            for w, x in zip(widths[1:], lines[-1][1:-1])),
+        lines[-1][-1]))
+
+
+def annotate(Result, results, *,
+        annotate=None,
+        threshold=None,
+        branches=False,
+        caches=False,
+        **args):
+    # figure out the threshold
+    if threshold is None:
+        t0, t1 = THRESHOLD
+    elif len(threshold) == 1:
+        t0, t1 = threshold[0], threshold[0]
+    else:
+        t0, t1 = threshold
+    t0, t1 = min(t0, t1), max(t0, t1)
+
+    if not branches and not caches:
+        tk = 'cycles'
+    elif branches:
+        tk = 'bmisses'
+    else:
+        tk = 'cmisses'
+
+    # find max cycles
+    max_ = max(it.chain((float(getattr(r, tk)) for r in results), [1]))
+
+    for path in co.OrderedDict.fromkeys(r.file for r in results).keys():
+        # flatten to line info
+        results = fold(Result, results, by=['file', 'line'])
+        table = {r.line: r for r in results if r.file == path}
+
+        # calculate spans to show
+        if not annotate:
+            spans = []
+            last = None
+            func = None
+            for line, r in sorted(table.items()):
+                if float(getattr(r, tk)) / max_ >= t0:
+                    if last is not None and line - last.stop <= args['context']:
+                        last = range(
+                            last.start,
+                            line+1+args['context'])
+                    else:
+                        if last is not None:
+                            spans.append((last, func))
+                        last = range(
+                            line-args['context'],
+                            line+1+args['context'])
+                        func = r.function
+            if last is not None:
+                spans.append((last, func))
+
+        with open(path) as f:
+            skipped = False
+            for i, line in enumerate(f):
+                # skip lines not in spans?
+                if not annotate and not any(i+1 in s for s, _ in spans):
+                    skipped = True
+                    continue
+
+                if skipped:
+                    skipped = False
+                    print('%s@@ %s:%d: %s @@%s' % (
+                        '\x1b[36m' if args['color'] else '',
+                        path,
+                        i+1,
+                        next(iter(f for _, f in spans)),
+                        '\x1b[m' if args['color'] else ''))
+
+                # build line
+                if line.endswith('\n'):
+                    line = line[:-1]
+
+                r = table.get(i+1)
+                if r is not None and (
+                        float(r.cycles) > 0
+                        if not branches and not caches
+                        else float(r.bmisses) > 0 or float(r.branches) > 0
+                        if branches
+                        else float(r.cmisses) > 0 or float(r.caches) > 0):
+                    line = '%-*s // %s' % (
+                        args['width'],
+                        line,
+                        '%s cycles' % r.cycles
+                        if not branches and not caches
+                        else '%s bmisses, %s branches' % (r.bmisses, r.branches)
+                        if branches
+                        else '%s cmisses, %s caches' % (r.cmisses, r.caches))
+
+                    if args['color']:
+                        if float(getattr(r, tk)) / max_ >= t1:
+                            line = '\x1b[1;31m%s\x1b[m' % line
+                        elif float(getattr(r, tk)) / max_ >= t0:
+                            line = '\x1b[35m%s\x1b[m' % line
+
+                print(line)
+
+
+def report(perf_paths, *,
+        by=None,
+        fields=None,
+        defines=None,
+        sort=None,
+        branches=False,
+        caches=False,
+        **args):
+    # figure out what color should be
+    if args.get('color') == 'auto':
+        args['color'] = sys.stdout.isatty()
+    elif args.get('color') == 'always':
+        args['color'] = True
+    else:
+        args['color'] = False
+
+    # depth of 0 == m.inf
+    if args.get('depth') == 0:
+        args['depth'] = m.inf
+
+    # find sizes
+    if not args.get('use', None):
+        results = collect(perf_paths, **args)
+    else:
+        results = []
+        with openio(args['use']) as f:
+            reader = csv.DictReader(f, restval='')
+            for r in reader:
+                if not any('perf_'+k in r and r['perf_'+k].strip()
+                        for k in PerfResult._fields):
+                    continue
+                try:
+                    results.append(PerfResult(
+                        **{k: r[k] for k in PerfResult._by
+                            if k in r and r[k].strip()},
+                        **{k: r['perf_'+k] for k in PerfResult._fields
+                            if 'perf_'+k in r and r['perf_'+k].strip()}))
+                except TypeError:
+                    pass
+
+    # fold
+    results = fold(PerfResult, results, by=by, defines=defines)
+
+    # sort, note that python's sort is stable
+    results.sort()
+    if sort:
+        for k, reverse in reversed(sort):
+            results.sort(
+                key=lambda r: tuple(
+                    (getattr(r, k),) if getattr(r, k) is not None else ()
+                    for k in ([k] if k else PerfResult._sort)),
+                reverse=reverse ^ (not k or k in PerfResult._fields))
+
+    # write results to CSV
+    if args.get('output'):
+        with openio(args['output'], 'w') as f:
+            writer = csv.DictWriter(f,
+                (by if by is not None else PerfResult._by)
+                + ['perf_'+k for k in (
+                    fields if fields is not None else PerfResult._fields)])
+            writer.writeheader()
+            for r in results:
+                writer.writerow(
+                    {k: getattr(r, k) for k in (
+                        by if by is not None else PerfResult._by)}
+                    | {'perf_'+k: getattr(r, k) for k in (
+                        fields if fields is not None else PerfResult._fields)})
+
+    # find previous results?
+    if args.get('diff'):
+        diff_results = []
+        try:
+            with openio(args['diff']) as f:
+                reader = csv.DictReader(f, restval='')
+                for r in reader:
+                    if not any('perf_'+k in r and r['perf_'+k].strip()
+                            for k in PerfResult._fields):
+                        continue
+                    try:
+                        diff_results.append(PerfResult(
+                            **{k: r[k] for k in PerfResult._by
+                                if k in r and r[k].strip()},
+                            **{k: r['perf_'+k] for k in PerfResult._fields
+                                if 'perf_'+k in r and r['perf_'+k].strip()}))
+                    except TypeError:
+                        pass
+        except FileNotFoundError:
+            pass
+
+        # fold
+        diff_results = fold(PerfResult, diff_results, by=by, defines=defines)
+
+    # print table
+    if not args.get('quiet'):
+        if args.get('annotate') or args.get('threshold'):
+            # annotate sources
+            annotate(PerfResult, results,
+                branches=branches,
+                caches=caches,
+                **args)
+        else:
+            # print table
+            table(PerfResult, results,
+                diff_results if args.get('diff') else None,
+                by=by if by is not None else ['function'],
+                fields=fields if fields is not None
+                    else ['cycles'] if not branches and not caches
+                    else ['bmisses', 'branches'] if branches
+                    else ['cmisses', 'caches'],
+                sort=sort,
+                **args)
+
+
+def main(**args):
+    if args.get('record'):
+        return record(**args)
+    else:
+        return report(**args)
+
+
+if __name__ == "__main__":
+    import argparse
+    import sys
+
+    # bit of a hack, but parse_intermixed_args and REMAINDER are
+    # incompatible, so we need to figure out what we want before running
+    # argparse
+    if '-R' in sys.argv or '--record' in sys.argv:
+        nargs = argparse.REMAINDER
+    else:
+        nargs = '*'
+
+    argparse.ArgumentParser._handle_conflict_ignore = lambda *_: None
+    argparse._ArgumentGroup._handle_conflict_ignore = lambda *_: None
+    parser = argparse.ArgumentParser(
+        description="Aggregate and report Linux perf results.",
+        allow_abbrev=False,
+        conflict_handler='ignore')
+    parser.add_argument(
+        'perf_paths',
+        nargs=nargs,
+        help="Input *.perf files.")
+    parser.add_argument(
+        '-v', '--verbose',
+        action='store_true',
+        help="Output commands that run behind the scenes.")
+    parser.add_argument(
+        '-q', '--quiet',
+        action='store_true',
+        help="Don't show anything, useful with -o.")
+    parser.add_argument(
+        '-o', '--output',
+        help="Specify CSV file to store results.")
+    parser.add_argument(
+        '-u', '--use',
+        help="Don't parse anything, use this CSV file.")
+    parser.add_argument(
+        '-d', '--diff',
+        help="Specify CSV file to diff against.")
+    parser.add_argument(
+        '-a', '--all',
+        action='store_true',
+        help="Show all, not just the ones that changed.")
+    parser.add_argument(
+        '-p', '--percent',
+        action='store_true',
+        help="Only show percentage change, not a full diff.")
+    parser.add_argument(
+        '-b', '--by',
+        action='append',
+        choices=PerfResult._by,
+        help="Group by this field.")
+    parser.add_argument(
+        '-f', '--field',
+        dest='fields',
+        action='append',
+        choices=PerfResult._fields,
+        help="Show this field.")
+    parser.add_argument(
+        '-D', '--define',
+        dest='defines',
+        action='append',
+        type=lambda x: (lambda k,v: (k, set(v.split(','))))(*x.split('=', 1)),
+        help="Only include results where this field is this value.")
+    class AppendSort(argparse.Action):
+        def __call__(self, parser, namespace, value, option):
+            if namespace.sort is None:
+                namespace.sort = []
+            namespace.sort.append((value, True if option == '-S' else False))
+    parser.add_argument(
+        '-s', '--sort',
+        nargs='?',
+        action=AppendSort,
+        help="Sort by this field.")
+    parser.add_argument(
+        '-S', '--reverse-sort',
+        nargs='?',
+        action=AppendSort,
+        help="Sort by this field, but backwards.")
+    parser.add_argument(
+        '-Y', '--summary',
+        action='store_true',
+        help="Only show the total.")
+    parser.add_argument(
+        '-F', '--source',
+        dest='sources',
+        action='append',
+        help="Only consider definitions in this file. Defaults to anything "
+            "in the current directory.")
+    parser.add_argument(
+        '--everything',
+        action='store_true',
+        help="Include builtin and libc specific symbols.")
+    parser.add_argument(
+        '--branches',
+        action='store_true',
+        help="Show branches and branch misses.")
+    parser.add_argument(
+        '--caches',
+        action='store_true',
+        help="Show cache accesses and cache misses.")
+    parser.add_argument(
+        '-P', '--propagate',
+        type=lambda x: int(x, 0),
+        help="Depth to propagate samples up the call-stack. 0 propagates up "
+            "to the entry point, 1 does no propagation. Defaults to 0.")
+    parser.add_argument(
+        '-Z', '--depth',
+        nargs='?',
+        type=lambda x: int(x, 0),
+        const=0,
+        help="Depth of function calls to show. 0 shows all calls but may not "
+            "terminate!")
+    parser.add_argument(
+        '-A', '--annotate',
+        action='store_true',
+        help="Show source files annotated with coverage info.")
+    parser.add_argument(
+        '-T', '--threshold',
+        nargs='?',
+        type=lambda x: tuple(float(x) for x in x.split(',')),
+        const=THRESHOLD,
+        help="Show lines with samples above this threshold as a percent of "
+            "all lines. Defaults to %s." % ','.join(str(t) for t in THRESHOLD))
+    parser.add_argument(
+        '-c', '--context',
+        type=lambda x: int(x, 0),
+        default=3,
+        help="Show n additional lines of context. Defaults to 3.")
+    parser.add_argument(
+        '-W', '--width',
+        type=lambda x: int(x, 0),
+        default=80,
+        help="Assume source is styled with this many columns. Defaults to 80.")
+    parser.add_argument(
+        '--color',
+        choices=['never', 'always', 'auto'],
+        default='auto',
+        help="When to use terminal colors. Defaults to 'auto'.")
+    parser.add_argument(
+        '-j', '--jobs',
+        nargs='?',
+        type=lambda x: int(x, 0),
+        const=0,
+        help="Number of processes to use. 0 spawns one process per core.")
+    parser.add_argument(
+        '--perf-path',
+        type=lambda x: x.split(),
+        help="Path to the perf executable, may include flags. "
+            "Defaults to %r." % PERF_PATH)
+    parser.add_argument(
+        '--objdump-path',
+        type=lambda x: x.split(),
+        default=OBJDUMP_PATH,
+        help="Path to the objdump executable, may include flags. "
+            "Defaults to %r." % OBJDUMP_PATH)
+
+    # record flags
+    record_parser = parser.add_argument_group('record options')
+    record_parser.add_argument(
+        'command',
+        nargs=nargs,
+        help="Command to run.")
+    record_parser.add_argument(
+        '-R', '--record',
+        action='store_true',
+        help="Run a command and aggregate perf measurements.")
+    record_parser.add_argument(
+        '-o', '--output',
+        help="Output file. Uses flock to synchronize. This is stored as a "
+            "zip-file of multiple perf results.")
+    record_parser.add_argument(
+        '--perf-freq',
+        help="perf sampling frequency. This is passed directly to perf. "
+            "Defaults to %r." % PERF_FREQ)
+    record_parser.add_argument(
+        '--perf-period',
+        help="perf sampling period. This is passed directly to perf.")
+    record_parser.add_argument(
+        '--perf-events',
+        help="perf events to record. This is passed directly to perf. "
+            "Defaults to %r." % PERF_EVENTS)
+    record_parser.add_argument(
+        '--perf-path',
+        type=lambda x: x.split(),
+        help="Path to the perf executable, may include flags. "
+            "Defaults to %r." % PERF_PATH)
+
+    # avoid intermixed/REMAINDER conflict, see above
+    if nargs == argparse.REMAINDER:
+        args = parser.parse_args()
+    else:
+        args = parser.parse_intermixed_args()
+
+    # perf_paths/command overlap, so need to do some munging here
+    args.command = args.perf_paths
+    if args.record:
+        if not args.command:
+            print('error: no command specified?')
+            sys.exit(-1)
+        if not args.output:
+            print('error: no output file specified?')
+            sys.exit(-1)
+
+    sys.exit(main(**{k: v
+        for k, v in vars(args).items()
+        if v is not None}))
diff --git a/scripts/perfbd.py b/scripts/perfbd.py
new file mode 100644
index 0000000..bf57f60
--- /dev/null
+++ b/scripts/perfbd.py
@@ -0,0 +1,1276 @@
+#!/usr/bin/env python3
+#
+# Aggregate and report call-stack propagated block-device operations
+# from trace output.
+#
+# Example:
+# ./scripts/bench.py -ttrace
+# ./scripts/perfbd.py trace -j -Flfs.c -Flfs_util.c -Serased -Sproged -Sreaded
+#
+# Copyright (c) 2022, The littlefs authors.
+# SPDX-License-Identifier: BSD-3-Clause
+#
+
+import bisect
+import collections as co
+import csv
+import functools as ft
+import itertools as it
+import math as m
+import multiprocessing as mp
+import os
+import re
+import shlex
+import subprocess as sp
+
+
+OBJDUMP_PATH = ['objdump']
+THRESHOLD = (0.5, 0.85)
+
+
+# integer fields
+class Int(co.namedtuple('Int', 'x')):
+    __slots__ = ()
+    def __new__(cls, x=0):
+        if isinstance(x, Int):
+            return x
+        if isinstance(x, str):
+            try:
+                x = int(x, 0)
+            except ValueError:
+                # also accept +-∞ and +-inf
+                if re.match('^\s*\+?\s*(?:∞|inf)\s*$', x):
+                    x = m.inf
+                elif re.match('^\s*-\s*(?:∞|inf)\s*$', x):
+                    x = -m.inf
+                else:
+                    raise
+        assert isinstance(x, int) or m.isinf(x), x
+        return super().__new__(cls, x)
+
+    def __str__(self):
+        if self.x == m.inf:
+            return '∞'
+        elif self.x == -m.inf:
+            return '-∞'
+        else:
+            return str(self.x)
+
+    def __int__(self):
+        assert not m.isinf(self.x)
+        return self.x
+
+    def __float__(self):
+        return float(self.x)
+
+    none = '%7s' % '-'
+    def table(self):
+        return '%7s' % (self,)
+
+    diff_none = '%7s' % '-'
+    diff_table = table
+
+    def diff_diff(self, other):
+        new = self.x if self else 0
+        old = other.x if other else 0
+        diff = new - old
+        if diff == +m.inf:
+            return '%7s' % '+∞'
+        elif diff == -m.inf:
+            return '%7s' % '-∞'
+        else:
+            return '%+7d' % diff
+
+    def ratio(self, other):
+        new = self.x if self else 0
+        old = other.x if other else 0
+        if m.isinf(new) and m.isinf(old):
+            return 0.0
+        elif m.isinf(new):
+            return +m.inf
+        elif m.isinf(old):
+            return -m.inf
+        elif not old and not new:
+            return 0.0
+        elif not old:
+            return 1.0
+        else:
+            return (new-old) / old
+
+    def __add__(self, other):
+        return self.__class__(self.x + other.x)
+
+    def __sub__(self, other):
+        return self.__class__(self.x - other.x)
+
+    def __mul__(self, other):
+        return self.__class__(self.x * other.x)
+
+# perf results
+class PerfBdResult(co.namedtuple('PerfBdResult', [
+        'file', 'function', 'line',
+        'readed', 'proged', 'erased',
+        'children'])):
+    _by = ['file', 'function', 'line']
+    _fields = ['readed', 'proged', 'erased']
+    _sort = ['erased', 'proged', 'readed']
+    _types = {'readed': Int, 'proged': Int, 'erased': Int}
+
+    __slots__ = ()
+    def __new__(cls, file='', function='', line=0,
+            readed=0, proged=0, erased=0,
+            children=[]):
+        return super().__new__(cls, file, function, int(Int(line)),
+            Int(readed), Int(proged), Int(erased),
+            children)
+
+    def __add__(self, other):
+        return PerfBdResult(self.file, self.function, self.line,
+            self.readed + other.readed,
+            self.proged + other.proged,
+            self.erased + other.erased,
+            self.children + other.children)
+
+
+def openio(path, mode='r', buffering=-1):
+    # allow '-' for stdin/stdout
+    if path == '-':
+        if mode == 'r':
+            return os.fdopen(os.dup(sys.stdin.fileno()), mode, buffering)
+        else:
+            return os.fdopen(os.dup(sys.stdout.fileno()), mode, buffering)
+    else:
+        return open(path, mode, buffering)
+
+def collect_syms_and_lines(obj_path, *,
+        objdump_path=None,
+        **args):
+    symbol_pattern = re.compile(
+        '^(?P<addr>[0-9a-fA-F]+)'
+            '\s+.*'
+            '\s+(?P<size>[0-9a-fA-F]+)'
+            '\s+(?P<name>[^\s]+)\s*$')
+    line_pattern = re.compile(
+        '^\s+(?:'
+            # matches dir/file table
+            '(?P<no>[0-9]+)'
+                '(?:\s+(?P<dir>[0-9]+))?'
+                '\s+.*'
+                '\s+(?P<path>[^\s]+)'
+            # matches line opcodes
+            '|' '\[[^\]]*\]\s+'
+                '(?:'
+                    '(?P<op_special>Special)'
+                    '|' '(?P<op_copy>Copy)'
+                    '|' '(?P<op_end>End of Sequence)'
+                    '|' 'File .*?to (?:entry )?(?P<op_file>\d+)'
+                    '|' 'Line .*?to (?P<op_line>[0-9]+)'
+                    '|' '(?:Address|PC) .*?to (?P<op_addr>[0x0-9a-fA-F]+)'
+                    '|' '.' ')*'
+            ')$', re.IGNORECASE)
+
+    # figure out symbol addresses
+    syms = {}
+    sym_at = []
+    cmd = objdump_path + ['-t', obj_path]
+    if args.get('verbose'):
+        print(' '.join(shlex.quote(c) for c in cmd))
+    proc = sp.Popen(cmd,
+        stdout=sp.PIPE,
+        stderr=sp.PIPE if not args.get('verbose') else None,
+        universal_newlines=True,
+        errors='replace',
+        close_fds=False)
+    for line in proc.stdout:
+        m = symbol_pattern.match(line)
+        if m:
+            name = m.group('name')
+            addr = int(m.group('addr'), 16)
+            size = int(m.group('size'), 16)
+            # ignore zero-sized symbols
+            if not size:
+                continue
+            # note multiple symbols can share a name
+            if name not in syms:
+                syms[name] = set()
+            syms[name].add((addr, size))
+            sym_at.append((addr, name, size))
+    proc.wait()
+    if proc.returncode != 0:
+        if not args.get('verbose'):
+            for line in proc.stderr:
+                sys.stdout.write(line)
+        # assume no debug-info on failure
+        pass
+
+    # sort and keep largest/first when duplicates
+    sym_at.sort(key=lambda x: (x[0], -x[2], x[1]))
+    sym_at_ = []
+    for addr, name, size in sym_at:
+        if len(sym_at_) == 0 or sym_at_[-1][0] != addr:
+            sym_at_.append((addr, name, size))
+    sym_at = sym_at_
+
+    # state machine for dwarf line numbers, note that objdump's
+    # decodedline seems to have issues with multiple dir/file
+    # tables, which is why we need this
+    lines = []
+    line_at = []
+    dirs = {}
+    files = {}
+    op_file = 1
+    op_line = 1
+    op_addr = 0
+    cmd = objdump_path + ['--dwarf=rawline', obj_path]
+    if args.get('verbose'):
+        print(' '.join(shlex.quote(c) for c in cmd))
+    proc = sp.Popen(cmd,
+        stdout=sp.PIPE,
+        stderr=sp.PIPE if not args.get('verbose') else None,
+        universal_newlines=True,
+        errors='replace',
+        close_fds=False)
+    for line in proc.stdout:
+        m = line_pattern.match(line)
+        if m:
+            if m.group('no') and not m.group('dir'):
+                # found a directory entry
+                dirs[int(m.group('no'))] = m.group('path')
+            elif m.group('no'):
+                # found a file entry
+                dir = int(m.group('dir'))
+                if dir in dirs:
+                    files[int(m.group('no'))] = os.path.join(
+                        dirs[dir],
+                        m.group('path'))
+                else:
+                    files[int(m.group('no'))] = m.group('path')
+            else:
+                # found a state machine update
+                if m.group('op_file'):
+                    op_file = int(m.group('op_file'), 0)
+                if m.group('op_line'):
+                    op_line = int(m.group('op_line'), 0)
+                if m.group('op_addr'):
+                    op_addr = int(m.group('op_addr'), 0)
+
+                if (m.group('op_special')
+                        or m.group('op_copy')
+                        or m.group('op_end')):
+                    file = os.path.abspath(files.get(op_file, '?'))
+                    lines.append((file, op_line, op_addr))
+                    line_at.append((op_addr, file, op_line))
+
+                if m.group('op_end'):
+                    op_file = 1
+                    op_line = 1
+                    op_addr = 0
+    proc.wait()
+    if proc.returncode != 0:
+        if not args.get('verbose'):
+            for line in proc.stderr:
+                sys.stdout.write(line)
+        # assume no debug-info on failure
+        pass
+
+    # sort and keep first when duplicates
+    lines.sort()
+    lines_ = []
+    for file, line, addr in lines:
+        if len(lines_) == 0 or lines_[-1][0] != file or lines[-1][1] != line:
+            lines_.append((file, line, addr))
+    lines = lines_
+
+    # sort and keep first when duplicates
+    line_at.sort()
+    line_at_ = []
+    for addr, file, line in line_at:
+        if len(line_at_) == 0 or line_at_[-1][0] != addr:
+            line_at_.append((addr, file, line))
+    line_at = line_at_
+
+    return syms, sym_at, lines, line_at
+
+
+def collect_job(path, start, stop, syms, sym_at, lines, line_at, *,
+        sources=None,
+        everything=False,
+        propagate=0,
+        depth=1,
+        **args):
+    trace_pattern = re.compile(
+        '^(?P<file>[^:]*):(?P<line>[0-9]+):trace:\s*(?P<prefix>[^\s]*?bd_)(?:'
+            '(?P<read>read)\('
+                '\s*(?P<read_ctx>\w+)' '\s*,'
+                '\s*(?P<read_block>\w+)' '\s*,'
+                '\s*(?P<read_off>\w+)' '\s*,'
+                '\s*(?P<read_buffer>\w+)' '\s*,'
+                '\s*(?P<read_size>\w+)' '\s*\)'
+            '|' '(?P<prog>prog)\('
+                '\s*(?P<prog_ctx>\w+)' '\s*,'
+                '\s*(?P<prog_block>\w+)' '\s*,'
+                '\s*(?P<prog_off>\w+)' '\s*,'
+                '\s*(?P<prog_buffer>\w+)' '\s*,'
+                '\s*(?P<prog_size>\w+)' '\s*\)'
+            '|' '(?P<erase>erase)\('
+                '\s*(?P<erase_ctx>\w+)' '\s*,'
+                '\s*(?P<erase_block>\w+)'
+                '\s*\(\s*(?P<erase_size>\w+)\s*\)' '\s*\)' ')\s*$')
+    frame_pattern = re.compile(
+        '^\s+at (?P<addr>\w+)\s*$')
+
+    # parse all of the trace files for read/prog/erase operations
+    last_filtered = False
+    last_file = None
+    last_line = None
+    last_sym = None
+    last_readed = 0
+    last_proged = 0
+    last_erased = 0
+    last_stack = []
+    last_delta = None
+    at_cache = {}
+    results = {}
+
+    def commit():
+        # fallback to just capturing top-level measurements
+        if not last_stack:
+            file = last_file
+            sym = last_sym
+            line = last_line
+
+            # ignore filtered sources
+            if sources is not None:
+                if not any(
+                        os.path.abspath(file)
+                            == os.path.abspath(s)
+                        for s in sources):
+                    return
+            else:
+                # default to only cwd
+                if not everything and not os.path.commonpath([
+                        os.getcwd(),
+                        os.path.abspath(file)]) == os.getcwd():
+                    return
+
+            # simplify path
+            if os.path.commonpath([
+                    os.getcwd(),
+                    os.path.abspath(file)]) == os.getcwd():
+                file = os.path.relpath(file)
+            else:
+                file = os.path.abspath(file)
+
+            results[(file, sym, line)] = (
+                last_readed,
+                last_proged,
+                last_erased,
+                {})
+        else:
+            # tail-recursively propagate measurements
+            for i in range(len(last_stack)):
+                results_ = results
+                for j in reversed(range(i+1)):
+                    if i+1-j > depth:
+                        break
+
+                    # propagate
+                    name = last_stack[j]
+                    if name in results_:
+                        r, p, e, children = results_[name]
+                    else:
+                        r, p, e, children = 0, 0, 0, {}
+                    results_[name] = (
+                        r+last_readed,
+                        p+last_proged,
+                        e+last_erased,
+                        children)
+
+                    # recurse
+                    results_ = results_[name][-1]
+
+    with openio(path) as f:
+        # try to jump to middle of file? need step out of utf8-safe mode and
+        # then resync up with the next newline to avoid parsing half a line
+        if start is not None and start > 0:
+            fd = f.fileno()
+            os.lseek(fd, start, os.SEEK_SET)
+            while os.read(fd, 1) not in {b'\n', b'\r', b''}:
+                pass
+            f = os.fdopen(fd)
+
+        for line in f:
+            # we have a lot of data, try to take a few shortcuts,
+            # string search is much faster than regex so try to use
+            # regex as late as possible.
+            if not line.startswith('\t'):
+                if last_filtered:
+                    commit()
+                last_filtered = False
+
+                # done processing our slice?
+                if stop is not None:
+                    if os.lseek(f.fileno(), 0, os.SEEK_CUR) > stop:
+                        break
+
+                if 'trace' in line and 'bd' in line:
+                    m = trace_pattern.match(line)
+                    if m:
+                        last_filtered = True
+                        last_file = os.path.abspath(m.group('file'))
+                        last_line = int(m.group('line'), 0)
+                        last_sym = m.group('prefix')
+                        last_readed = 0
+                        last_proged = 0
+                        last_erased = 0
+                        last_stack = []
+                        last_delta = None
+
+                        if m.group('read'):
+                            last_sym += m.group('read')
+                            last_readed += int(m.group('read_size'))
+                        elif m.group('prog'):
+                            last_sym += m.group('prog')
+                            last_proged += int(m.group('prog_size'))
+                        elif m.group('erase'):
+                            last_sym += m.group('erase')
+                            last_erased += int(m.group('erase_size'))
+
+            elif last_filtered:
+                m = frame_pattern.match(line)
+                if m:
+                    addr_ = int(m.group('addr'), 0)
+
+                    # before we can do anything with addr, we need to
+                    # reverse ASLR, fortunately we know the file+line of
+                    # the first stack frame, so we can use that as a point
+                    # of reference
+                    if last_delta is None:
+                        i = bisect.bisect(lines, (last_file, last_line),
+                            key=lambda x: (x[0], x[1]))
+                        if i > 0:
+                            last_delta = lines[i-1][2] - addr_
+                        else:
+                            # can't reverse ASLR, give up on backtrace
+                            commit()
+                            last_filtered = False
+                            continue
+
+                    addr = addr_ + last_delta
+
+                    # cached?
+                    if addr in at_cache:
+                        cached = at_cache[addr]
+                        if cached is None:
+                            # cache says to skip
+                            continue
+                        file, sym, line = cached
+                    else:
+                        # find sym
+                        i = bisect.bisect(sym_at, addr, key=lambda x: x[0])
+                        # check that we're actually in the sym's size
+                        if i > 0 and addr < sym_at[i-1][0] + sym_at[i-1][2]:
+                            _, sym, _ = sym_at[i-1]
+                        else:
+                            sym = hex(addr)
+
+                        # filter out internal/unknown functions
+                        if not everything and (
+                                sym.startswith('__')
+                                or sym.startswith('0')
+                                or sym.startswith('-')
+                                or sym == '_start'):
+                            at_cache[addr] = None
+                            continue
+
+                        # find file+line
+                        i = bisect.bisect(line_at, addr, key=lambda x: x[0])
+                        if i > 0:
+                            _, file, line = line_at[i-1]
+                        elif len(last_stack) == 0:
+                            file, line = last_file, last_line
+                        else:
+                            file, line = re.sub('(\.o)?$', '.c', obj_path, 1), 0
+
+                        # ignore filtered sources
+                        if sources is not None:
+                            if not any(
+                                    os.path.abspath(file)
+                                        == os.path.abspath(s)
+                                    for s in sources):
+                                at_cache[addr] = None
+                                continue
+                        else:
+                            # default to only cwd
+                            if not everything and not os.path.commonpath([
+                                    os.getcwd(),
+                                    os.path.abspath(file)]) == os.getcwd():
+                                at_cache[addr] = None
+                                continue
+
+                        # simplify path
+                        if os.path.commonpath([
+                                os.getcwd(),
+                                os.path.abspath(file)]) == os.getcwd():
+                            file = os.path.relpath(file)
+                        else:
+                            file = os.path.abspath(file)
+
+                        at_cache[addr] = file, sym, line
+
+                    last_stack.append((file, sym, line))
+
+                    # stop propagating?
+                    if propagate and len(last_stack) >= propagate:
+                        commit()
+                        last_filtered = False
+        if last_filtered:
+            commit()
+
+    # rearrange results into result type
+    def to_results(results):
+        results_ = []
+        for name, (r, p, e, children) in results.items():
+            results_.append(PerfBdResult(*name,
+                r, p, e,
+                children=to_results(children)))
+        return results_
+
+    return to_results(results)
+
+def starapply(args):
+    f, args, kwargs = args
+    return f(*args, **kwargs)
+
+def collect(obj_path, trace_paths, *,
+        jobs=None,
+        **args):
+    # automatic job detection?
+    if jobs == 0:
+        jobs = len(os.sched_getaffinity(0))
+
+    # find sym/line info to reverse ASLR
+    syms, sym_at, lines, line_at = collect_syms_and_lines(obj_path, **args)
+
+    if jobs is not None:
+        # try to split up files so that even single files can be processed
+        # in parallel
+        #
+        # this looks naive, since we're splitting up text files by bytes, but
+        # we do proper backtrace delimination in collect_job
+        trace_ranges = []
+        for path in trace_paths:
+            if path == '-':
+                trace_ranges.append([(None, None)])
+                continue
+
+            size = os.path.getsize(path)
+            if size == 0:
+                trace_ranges.append([(None, None)])
+                continue
+
+            perjob = m.ceil(size // jobs)
+            trace_ranges.append([(i, i+perjob) for i in range(0, size, perjob)])
+
+        results = []
+        with mp.Pool(jobs) as p:
+            for results_ in p.imap_unordered(
+                    starapply,
+                    ((collect_job, (path, start, stop,
+                        syms, sym_at, lines, line_at),
+                        args)
+                        for path, ranges in zip(trace_paths, trace_ranges)
+                        for start, stop in ranges)):
+                results.extend(results_)
+
+    else:
+        results = []
+        for path in trace_paths:
+            results.extend(collect_job(path, None, None,
+                syms, sym_at, lines, line_at,
+                **args))
+
+    return results
+
+
+def fold(Result, results, *,
+        by=None,
+        defines=None,
+        **_):
+    if by is None:
+        by = Result._by
+
+    for k in it.chain(by or [], (k for k, _ in defines or [])):
+        if k not in Result._by and k not in Result._fields:
+            print("error: could not find field %r?" % k)
+            sys.exit(-1)
+
+    # filter by matching defines
+    if defines is not None:
+        results_ = []
+        for r in results:
+            if all(getattr(r, k) in vs for k, vs in defines):
+                results_.append(r)
+        results = results_
+
+    # organize results into conflicts
+    folding = co.OrderedDict()
+    for r in results:
+        name = tuple(getattr(r, k) for k in by)
+        if name not in folding:
+            folding[name] = []
+        folding[name].append(r)
+
+    # merge conflicts
+    folded = []
+    for name, rs in folding.items():
+        folded.append(sum(rs[1:], start=rs[0]))
+
+    # fold recursively
+    folded_ = []
+    for r in folded:
+        folded_.append(r._replace(children=fold(
+            Result, r.children,
+            by=by,
+            defines=defines)))
+    folded = folded_
+
+    return folded
+
+def table(Result, results, diff_results=None, *,
+        by=None,
+        fields=None,
+        sort=None,
+        summary=False,
+        all=False,
+        percent=False,
+        depth=1,
+        **_):
+    all_, all = all, __builtins__.all
+
+    if by is None:
+        by = Result._by
+    if fields is None:
+        fields = Result._fields
+    types = Result._types
+
+    # fold again
+    results = fold(Result, results, by=by)
+    if diff_results is not None:
+        diff_results = fold(Result, diff_results, by=by)
+
+    # organize by name
+    table = {
+        ','.join(str(getattr(r, k) or '') for k in by): r
+        for r in results}
+    diff_table = {
+        ','.join(str(getattr(r, k) or '') for k in by): r
+        for r in diff_results or []}
+    names = list(table.keys() | diff_table.keys())
+
+    # sort again, now with diff info, note that python's sort is stable
+    names.sort()
+    if diff_results is not None:
+        names.sort(key=lambda n: tuple(
+            types[k].ratio(
+                getattr(table.get(n), k, None),
+                getattr(diff_table.get(n), k, None))
+            for k in fields),
+            reverse=True)
+    if sort:
+        for k, reverse in reversed(sort):
+            names.sort(
+                key=lambda n: tuple(
+                    (getattr(table[n], k),)
+                    if getattr(table.get(n), k, None) is not None else ()
+                    for k in ([k] if k else [
+                        k for k in Result._sort if k in fields])),
+                reverse=reverse ^ (not k or k in Result._fields))
+
+
+    # build up our lines
+    lines = []
+
+    # header
+    header = []
+    header.append('%s%s' % (
+        ','.join(by),
+        ' (%d added, %d removed)' % (
+            sum(1 for n in table if n not in diff_table),
+            sum(1 for n in diff_table if n not in table))
+            if diff_results is not None and not percent else '')
+        if not summary else '')
+    if diff_results is None:
+        for k in fields:
+            header.append(k)
+    elif percent:
+        for k in fields:
+            header.append(k)
+    else:
+        for k in fields:
+            header.append('o'+k)
+        for k in fields:
+            header.append('n'+k)
+        for k in fields:
+            header.append('d'+k)
+    header.append('')
+    lines.append(header)
+
+    def table_entry(name, r, diff_r=None, ratios=[]):
+        entry = []
+        entry.append(name)
+        if diff_results is None:
+            for k in fields:
+                entry.append(getattr(r, k).table()
+                    if getattr(r, k, None) is not None
+                    else types[k].none)
+        elif percent:
+            for k in fields:
+                entry.append(getattr(r, k).diff_table()
+                    if getattr(r, k, None) is not None
+                    else types[k].diff_none)
+        else:
+            for k in fields:
+                entry.append(getattr(diff_r, k).diff_table()
+                    if getattr(diff_r, k, None) is not None
+                    else types[k].diff_none)
+            for k in fields:
+                entry.append(getattr(r, k).diff_table()
+                    if getattr(r, k, None) is not None
+                    else types[k].diff_none)
+            for k in fields:
+                entry.append(types[k].diff_diff(
+                        getattr(r, k, None),
+                        getattr(diff_r, k, None)))
+        if diff_results is None:
+            entry.append('')
+        elif percent:
+            entry.append(' (%s)' % ', '.join(
+                '+∞%' if t == +m.inf
+                else '-∞%' if t == -m.inf
+                else '%+.1f%%' % (100*t)
+                for t in ratios))
+        else:
+            entry.append(' (%s)' % ', '.join(
+                    '+∞%' if t == +m.inf
+                    else '-∞%' if t == -m.inf
+                    else '%+.1f%%' % (100*t)
+                    for t in ratios
+                    if t)
+                if any(ratios) else '')
+        return entry
+
+    # entries
+    if not summary:
+        for name in names:
+            r = table.get(name)
+            if diff_results is None:
+                diff_r = None
+                ratios = None
+            else:
+                diff_r = diff_table.get(name)
+                ratios = [
+                    types[k].ratio(
+                        getattr(r, k, None),
+                        getattr(diff_r, k, None))
+                    for k in fields]
+                if not all_ and not any(ratios):
+                    continue
+            lines.append(table_entry(name, r, diff_r, ratios))
+
+    # total
+    r = next(iter(fold(Result, results, by=[])), None)
+    if diff_results is None:
+        diff_r = None
+        ratios = None
+    else:
+        diff_r = next(iter(fold(Result, diff_results, by=[])), None)
+        ratios = [
+            types[k].ratio(
+                getattr(r, k, None),
+                getattr(diff_r, k, None))
+            for k in fields]
+    lines.append(table_entry('TOTAL', r, diff_r, ratios))
+
+    # find the best widths, note that column 0 contains the names and column -1
+    # the ratios, so those are handled a bit differently
+    widths = [
+        ((max(it.chain([w], (len(l[i]) for l in lines)))+1+4-1)//4)*4-1
+        for w, i in zip(
+            it.chain([23], it.repeat(7)),
+            range(len(lines[0])-1))]
+
+    # adjust the name width based on the expected call depth, though
+    # note this doesn't really work with unbounded recursion
+    if not summary and not m.isinf(depth):
+        widths[0] += 4*(depth-1)
+
+    # print the tree recursively
+    print('%-*s  %s%s' % (
+        widths[0], lines[0][0],
+        ' '.join('%*s' % (w, x)
+            for w, x in zip(widths[1:], lines[0][1:-1])),
+        lines[0][-1]))
+
+    if not summary:
+        def recurse(results_, depth_, prefixes=('', '', '', '')):
+            # rebuild our tables at each layer
+            table_ = {
+                ','.join(str(getattr(r, k) or '') for k in by): r
+                for r in results_}
+            names_ = list(table_.keys())
+
+            # sort again at each layer, keep in mind the numbers are
+            # changing as we descend
+            names_.sort()
+            if sort:
+                for k, reverse in reversed(sort):
+                    names_.sort(
+                        key=lambda n: tuple(
+                            (getattr(table_[n], k),)
+                            if getattr(table_.get(n), k, None) is not None
+                            else ()
+                            for k in ([k] if k else [
+                                k for k in Result._sort if k in fields])),
+                        reverse=reverse ^ (not k or k in Result._fields))
+
+            for i, name in enumerate(names_):
+                r = table_[name]
+                is_last = (i == len(names_)-1)
+
+                print('%s%-*s  %s' % (
+                    prefixes[0+is_last],
+                    widths[0] - (
+                        len(prefixes[0+is_last])
+                        if not m.isinf(depth) else 0),
+                    name,
+                    ' '.join('%*s' % (w, x)
+                        for w, x in zip(
+                            widths[1:],
+                            table_entry(name, r)[1:]))))
+
+                # recurse?
+                if depth_ > 1:
+                    recurse(
+                        r.children,
+                        depth_-1,
+                        (prefixes[2+is_last] + "|-> ",
+                         prefixes[2+is_last] + "'-> ",
+                         prefixes[2+is_last] + "|   ",
+                         prefixes[2+is_last] + "    "))
+
+        # we have enough going on with diffing to make the top layer
+        # a special case
+        for name, line in zip(names, lines[1:-1]):
+            print('%-*s  %s%s' % (
+                widths[0], line[0],
+                ' '.join('%*s' % (w, x)
+                    for w, x in zip(widths[1:], line[1:-1])),
+                line[-1]))
+
+            if name in table and depth > 1:
+                recurse(
+                    table[name].children,
+                    depth-1,
+                    ("|-> ",
+                     "'-> ",
+                     "|   ",
+                     "    "))
+
+    print('%-*s  %s%s' % (
+        widths[0], lines[-1][0],
+        ' '.join('%*s' % (w, x)
+            for w, x in zip(widths[1:], lines[-1][1:-1])),
+        lines[-1][-1]))
+
+
+def annotate(Result, results, *,
+        annotate=None,
+        threshold=None,
+        read_threshold=None,
+        prog_threshold=None,
+        erase_threshold=None,
+        **args):
+    # figure out the thresholds
+    if threshold is None:
+        threshold = THRESHOLD
+    elif len(threshold) == 1:
+        threshold = threshold[0], threshold[0]
+
+    if read_threshold is None:
+        read_t0, read_t1 = threshold
+    elif len(read_threshold) == 1:
+        read_t0, read_t1 = read_threshold[0], read_threshold[0]
+    else:
+        read_t0, read_t1 = read_threshold
+    read_t0, read_t1 = min(read_t0, read_t1), max(read_t0, read_t1)
+
+    if prog_threshold is None:
+        prog_t0, prog_t1 = threshold
+    elif len(prog_threshold) == 1:
+        prog_t0, prog_t1 = prog_threshold[0], prog_threshold[0]
+    else:
+        prog_t0, prog_t1 = prog_threshold
+    prog_t0, prog_t1 = min(prog_t0, prog_t1), max(prog_t0, prog_t1)
+
+    if erase_threshold is None:
+        erase_t0, erase_t1 = threshold
+    elif len(erase_threshold) == 1:
+        erase_t0, erase_t1 = erase_threshold[0], erase_threshold[0]
+    else:
+        erase_t0, erase_t1 = erase_threshold
+    erase_t0, erase_t1 = min(erase_t0, erase_t1), max(erase_t0, erase_t1)
+
+    # find maxs
+    max_readed = max(it.chain((float(r.readed) for r in results), [1]))
+    max_proged = max(it.chain((float(r.proged) for r in results), [1]))
+    max_erased = max(it.chain((float(r.erased) for r in results), [1]))
+
+    for path in co.OrderedDict.fromkeys(r.file for r in results).keys():
+        # flatten to line info
+        results = fold(Result, results, by=['file', 'line'])
+        table = {r.line: r for r in results if r.file == path}
+
+        # calculate spans to show
+        if not annotate:
+            spans = []
+            last = None
+            func = None
+            for line, r in sorted(table.items()):
+                if (float(r.readed) / max_readed >= read_t0
+                        or float(r.proged) / max_proged >= prog_t0
+                        or float(r.erased) / max_erased >= erase_t0):
+                    if last is not None and line - last.stop <= args['context']:
+                        last = range(
+                            last.start,
+                            line+1+args['context'])
+                    else:
+                        if last is not None:
+                            spans.append((last, func))
+                        last = range(
+                            line-args['context'],
+                            line+1+args['context'])
+                        func = r.function
+            if last is not None:
+                spans.append((last, func))
+
+        with open(path) as f:
+            skipped = False
+            for i, line in enumerate(f):
+                # skip lines not in spans?
+                if not annotate and not any(i+1 in s for s, _ in spans):
+                    skipped = True
+                    continue
+
+                if skipped:
+                    skipped = False
+                    print('%s@@ %s:%d: %s @@%s' % (
+                        '\x1b[36m' if args['color'] else '',
+                        path,
+                        i+1,
+                        next(iter(f for _, f in spans)),
+                        '\x1b[m' if args['color'] else ''))
+
+                # build line
+                if line.endswith('\n'):
+                    line = line[:-1]
+
+                if i+1 in table:
+                    r = table[i+1]
+                    line = '%-*s // %s readed, %s proged, %s erased' % (
+                        args['width'],
+                        line,
+                        r.readed,
+                        r.proged,
+                        r.erased)
+
+                    if args['color']:
+                        if (float(r.readed) / max_readed >= read_t1
+                                or float(r.proged) / max_proged >= prog_t1
+                                or float(r.erased) / max_erased >= erase_t1):
+                            line = '\x1b[1;31m%s\x1b[m' % line
+                        elif (float(r.readed) / max_readed >= read_t0
+                                or float(r.proged) / max_proged >= prog_t0
+                                or float(r.erased) / max_erased >= erase_t0):
+                            line = '\x1b[35m%s\x1b[m' % line
+
+                print(line)
+
+
+def report(obj_path='', trace_paths=[], *,
+        by=None,
+        fields=None,
+        defines=None,
+        sort=None,
+        **args):
+    # figure out what color should be
+    if args.get('color') == 'auto':
+        args['color'] = sys.stdout.isatty()
+    elif args.get('color') == 'always':
+        args['color'] = True
+    else:
+        args['color'] = False
+
+    # depth of 0 == m.inf
+    if args.get('depth') == 0:
+        args['depth'] = m.inf
+
+    # find sizes
+    if not args.get('use', None):
+        results = collect(obj_path, trace_paths, **args)
+    else:
+        results = []
+        with openio(args['use']) as f:
+            reader = csv.DictReader(f, restval='')
+            for r in reader:
+                if not any('perfbd_'+k in r and r['perfbd_'+k].strip()
+                        for k in PerfBdResult._fields):
+                    continue
+                try:
+                    results.append(PerfBdResult(
+                        **{k: r[k] for k in PerfBdResult._by
+                            if k in r and r[k].strip()},
+                        **{k: r['perfbd_'+k] for k in PerfBdResult._fields
+                            if 'perfbd_'+k in r and r['perfbd_'+k].strip()}))
+                except TypeError:
+                    pass
+
+    # fold
+    results = fold(PerfBdResult, results, by=by, defines=defines)
+
+    # sort, note that python's sort is stable
+    results.sort()
+    if sort:
+        for k, reverse in reversed(sort):
+            results.sort(
+                key=lambda r: tuple(
+                    (getattr(r, k),) if getattr(r, k) is not None else ()
+                    for k in ([k] if k else PerfBdResult._sort)),
+                reverse=reverse ^ (not k or k in PerfBdResult._fields))
+
+    # write results to CSV
+    if args.get('output'):
+        with openio(args['output'], 'w') as f:
+            writer = csv.DictWriter(f,
+                (by if by is not None else PerfBdResult._by)
+                + ['perfbd_'+k for k in (
+                    fields if fields is not None else PerfBdResult._fields)])
+            writer.writeheader()
+            for r in results:
+                writer.writerow(
+                    {k: getattr(r, k) for k in (
+                        by if by is not None else PerfBdResult._by)}
+                    | {'perfbd_'+k: getattr(r, k) for k in (
+                        fields if fields is not None else PerfBdResult._fields)})
+
+    # find previous results?
+    if args.get('diff'):
+        diff_results = []
+        try:
+            with openio(args['diff']) as f:
+                reader = csv.DictReader(f, restval='')
+                for r in reader:
+                    if not any('perfbd_'+k in r and r['perfbd_'+k].strip()
+                            for k in PerfBdResult._fields):
+                        continue
+                    try:
+                        diff_results.append(PerfBdResult(
+                            **{k: r[k] for k in PerfBdResult._by
+                                if k in r and r[k].strip()},
+                            **{k: r['perfbd_'+k] for k in PerfBdResult._fields
+                                if 'perfbd_'+k in r
+                                    and r['perfbd_'+k].strip()}))
+                    except TypeError:
+                        pass
+        except FileNotFoundError:
+            pass
+
+        # fold
+        diff_results = fold(PerfBdResult, diff_results, by=by, defines=defines)
+
+    # print table
+    if not args.get('quiet'):
+        if (args.get('annotate')
+                or args.get('threshold')
+                or args.get('read_threshold')
+                or args.get('prog_threshold')
+                or args.get('erase_threshold')):
+            # annotate sources
+            annotate(PerfBdResult, results, **args)
+        else:
+            # print table
+            table(PerfBdResult, results,
+                diff_results if args.get('diff') else None,
+                by=by if by is not None else ['function'],
+                fields=fields,
+                sort=sort,
+                **args)
+
+
+def main(**args):
+    if args.get('record'):
+        return record(**args)
+    else:
+        return report(**args)
+
+
+if __name__ == "__main__":
+    import argparse
+    import sys
+    parser = argparse.ArgumentParser(
+        description="Aggregate and report call-stack propagated "
+            "block-device operations from trace output.",
+        allow_abbrev=False)
+    parser.add_argument(
+        'obj_path',
+        nargs='?',
+        help="Input executable for mapping addresses to symbols.")
+    parser.add_argument(
+        'trace_paths',
+        nargs='*',
+        help="Input *.trace files.")
+    parser.add_argument(
+        '-v', '--verbose',
+        action='store_true',
+        help="Output commands that run behind the scenes.")
+    parser.add_argument(
+        '-q', '--quiet',
+        action='store_true',
+        help="Don't show anything, useful with -o.")
+    parser.add_argument(
+        '-o', '--output',
+        help="Specify CSV file to store results.")
+    parser.add_argument(
+        '-u', '--use',
+        help="Don't parse anything, use this CSV file.")
+    parser.add_argument(
+        '-d', '--diff',
+        help="Specify CSV file to diff against.")
+    parser.add_argument(
+        '-a', '--all',
+        action='store_true',
+        help="Show all, not just the ones that changed.")
+    parser.add_argument(
+        '-p', '--percent',
+        action='store_true',
+        help="Only show percentage change, not a full diff.")
+    parser.add_argument(
+        '-b', '--by',
+        action='append',
+        choices=PerfBdResult._by,
+        help="Group by this field.")
+    parser.add_argument(
+        '-f', '--field',
+        dest='fields',
+        action='append',
+        choices=PerfBdResult._fields,
+        help="Show this field.")
+    parser.add_argument(
+        '-D', '--define',
+        dest='defines',
+        action='append',
+        type=lambda x: (lambda k,v: (k, set(v.split(','))))(*x.split('=', 1)),
+        help="Only include results where this field is this value.")
+    class AppendSort(argparse.Action):
+        def __call__(self, parser, namespace, value, option):
+            if namespace.sort is None:
+                namespace.sort = []
+            namespace.sort.append((value, True if option == '-S' else False))
+    parser.add_argument(
+        '-s', '--sort',
+        nargs='?',
+        action=AppendSort,
+        help="Sort by this field.")
+    parser.add_argument(
+        '-S', '--reverse-sort',
+        nargs='?',
+        action=AppendSort,
+        help="Sort by this field, but backwards.")
+    parser.add_argument(
+        '-Y', '--summary',
+        action='store_true',
+        help="Only show the total.")
+    parser.add_argument(
+        '-F', '--source',
+        dest='sources',
+        action='append',
+        help="Only consider definitions in this file. Defaults to anything "
+            "in the current directory.")
+    parser.add_argument(
+        '--everything',
+        action='store_true',
+        help="Include builtin and libc specific symbols.")
+    parser.add_argument(
+        '-P', '--propagate',
+        type=lambda x: int(x, 0),
+        help="Depth to propagate samples up the call-stack. 0 propagates up "
+            "to the entry point, 1 does no propagation. Defaults to 0.")
+    parser.add_argument(
+        '-Z', '--depth',
+        nargs='?',
+        type=lambda x: int(x, 0),
+        const=0,
+        help="Depth of function calls to show. 0 shows all calls but may not "
+            "terminate!")
+    parser.add_argument(
+        '-A', '--annotate',
+        action='store_true',
+        help="Show source files annotated with coverage info.")
+    parser.add_argument(
+        '-T', '--threshold',
+        nargs='?',
+        type=lambda x: tuple(float(x) for x in x.split(',')),
+        const=THRESHOLD,
+        help="Show lines with any ops above this threshold as a percent of "
+            "all lines. Defaults to %s." % ','.join(str(t) for t in THRESHOLD))
+    parser.add_argument(
+        '--read-threshold',
+        nargs='?',
+        type=lambda x: tuple(float(x) for x in x.split(',')),
+        const=THRESHOLD,
+        help="Show lines with reads above this threshold as a percent of "
+            "all lines. Defaults to %s." % ','.join(str(t) for t in THRESHOLD))
+    parser.add_argument(
+        '--prog-threshold',
+        nargs='?',
+        type=lambda x: tuple(float(x) for x in x.split(',')),
+        const=THRESHOLD,
+        help="Show lines with progs above this threshold as a percent of "
+            "all lines. Defaults to %s." % ','.join(str(t) for t in THRESHOLD))
+    parser.add_argument(
+        '--erase-threshold',
+        nargs='?',
+        type=lambda x: tuple(float(x) for x in x.split(',')),
+        const=THRESHOLD,
+        help="Show lines with erases above this threshold as a percent of "
+            "all lines. Defaults to %s." % ','.join(str(t) for t in THRESHOLD))
+    parser.add_argument(
+        '-c', '--context',
+        type=lambda x: int(x, 0),
+        default=3,
+        help="Show n additional lines of context. Defaults to 3.")
+    parser.add_argument(
+        '-W', '--width',
+        type=lambda x: int(x, 0),
+        default=80,
+        help="Assume source is styled with this many columns. Defaults to 80.")
+    parser.add_argument(
+        '--color',
+        choices=['never', 'always', 'auto'],
+        default='auto',
+        help="When to use terminal colors. Defaults to 'auto'.")
+    parser.add_argument(
+        '-j', '--jobs',
+        nargs='?',
+        type=lambda x: int(x, 0),
+        const=0,
+        help="Number of processes to use. 0 spawns one process per core.")
+    parser.add_argument(
+        '--objdump-path',
+        type=lambda x: x.split(),
+        default=OBJDUMP_PATH,
+        help="Path to the objdump executable, may include flags. "
+            "Defaults to %r." % OBJDUMP_PATH)
+    sys.exit(main(**{k: v
+        for k, v in vars(parser.parse_intermixed_args()).items()
+        if v is not None}))
diff --git a/scripts/plot.py b/scripts/plot.py
new file mode 100644
index 0000000..2f980f4
--- /dev/null
+++ b/scripts/plot.py
@@ -0,0 +1,1592 @@
+#!/usr/bin/env python3
+#
+# Plot CSV files in terminal.
+#
+# Example:
+# ./scripts/plot.py bench.csv -xSIZE -ybench_read -W80 -H17
+#
+# Copyright (c) 2022, The littlefs authors.
+# SPDX-License-Identifier: BSD-3-Clause
+#
+
+import bisect
+import codecs
+import collections as co
+import csv
+import io
+import itertools as it
+import math as m
+import os
+import shlex
+import shutil
+import time
+
+try:
+    import inotify_simple
+except ModuleNotFoundError:
+    inotify_simple = None
+
+
+COLORS = [
+    '1;34', # bold blue
+    '1;31', # bold red
+    '1;32', # bold green
+    '1;35', # bold purple
+    '1;33', # bold yellow
+    '1;36', # bold cyan
+    '34',   # blue
+    '31',   # red
+    '32',   # green
+    '35',   # purple
+    '33',   # yellow
+    '36',   # cyan
+]
+
+CHARS_DOTS = " .':"
+CHARS_BRAILLE = (
+    '⠀⢀⡀⣀⠠⢠⡠⣠⠄⢄⡄⣄⠤⢤⡤⣤' '⠐⢐⡐⣐⠰⢰⡰⣰⠔⢔⡔⣔⠴⢴⡴⣴'
+    '⠂⢂⡂⣂⠢⢢⡢⣢⠆⢆⡆⣆⠦⢦⡦⣦' '⠒⢒⡒⣒⠲⢲⡲⣲⠖⢖⡖⣖⠶⢶⡶⣶'
+    '⠈⢈⡈⣈⠨⢨⡨⣨⠌⢌⡌⣌⠬⢬⡬⣬' '⠘⢘⡘⣘⠸⢸⡸⣸⠜⢜⡜⣜⠼⢼⡼⣼'
+    '⠊⢊⡊⣊⠪⢪⡪⣪⠎⢎⡎⣎⠮⢮⡮⣮' '⠚⢚⡚⣚⠺⢺⡺⣺⠞⢞⡞⣞⠾⢾⡾⣾'
+    '⠁⢁⡁⣁⠡⢡⡡⣡⠅⢅⡅⣅⠥⢥⡥⣥' '⠑⢑⡑⣑⠱⢱⡱⣱⠕⢕⡕⣕⠵⢵⡵⣵'
+    '⠃⢃⡃⣃⠣⢣⡣⣣⠇⢇⡇⣇⠧⢧⡧⣧' '⠓⢓⡓⣓⠳⢳⡳⣳⠗⢗⡗⣗⠷⢷⡷⣷'
+    '⠉⢉⡉⣉⠩⢩⡩⣩⠍⢍⡍⣍⠭⢭⡭⣭' '⠙⢙⡙⣙⠹⢹⡹⣹⠝⢝⡝⣝⠽⢽⡽⣽'
+    '⠋⢋⡋⣋⠫⢫⡫⣫⠏⢏⡏⣏⠯⢯⡯⣯' '⠛⢛⡛⣛⠻⢻⡻⣻⠟⢟⡟⣟⠿⢿⡿⣿')
+CHARS_POINTS_AND_LINES = 'o'
+
+SI_PREFIXES = {
+    18:  'E',
+    15:  'P',
+    12:  'T',
+    9:   'G',
+    6:   'M',
+    3:   'K',
+    0:   '',
+    -3:  'm',
+    -6:  'u',
+    -9:  'n',
+    -12: 'p',
+    -15: 'f',
+    -18: 'a',
+}
+
+SI2_PREFIXES = {
+    60:  'Ei',
+    50:  'Pi',
+    40:  'Ti',
+    30:  'Gi',
+    20:  'Mi',
+    10:  'Ki',
+    0:   '',
+    -10: 'mi',
+    -20: 'ui',
+    -30: 'ni',
+    -40: 'pi',
+    -50: 'fi',
+    -60: 'ai',
+}
+
+
+# format a number to a strict character width using SI prefixes
+def si(x, w=4):
+    if x == 0:
+        return '0'
+    # figure out prefix and scale
+    #
+    # note we adjust this so that 100K = .1M, which has more info
+    # per character
+    p = 3*int(m.log(abs(x)*10, 10**3))
+    p = min(18, max(-18, p))
+    # format with enough digits
+    s = '%.*f' % (w, abs(x) / (10.0**p))
+    s = s.lstrip('0')
+    # truncate but only digits that follow the dot
+    if '.' in s:
+        s = s[:max(s.find('.'), w-(2 if x < 0 else 1))]
+        s = s.rstrip('0')
+        s = s.rstrip('.')
+    return '%s%s%s' % ('-' if x < 0 else '', s, SI_PREFIXES[p])
+
+def si2(x, w=5):
+    if x == 0:
+        return '0'
+    # figure out prefix and scale
+    #
+    # note we adjust this so that 128Ki = .1Mi, which has more info
+    # per character
+    p = 10*int(m.log(abs(x)*10, 2**10))
+    p = min(30, max(-30, p))
+    # format with enough digits
+    s = '%.*f' % (w, abs(x) / (2.0**p))
+    s = s.lstrip('0')
+    # truncate but only digits that follow the dot
+    if '.' in s:
+        s = s[:max(s.find('.'), w-(3 if x < 0 else 2))]
+        s = s.rstrip('0')
+        s = s.rstrip('.')
+    return '%s%s%s' % ('-' if x < 0 else '', s, SI2_PREFIXES[p])
+
+# parse escape strings
+def escape(s):
+    return codecs.escape_decode(s.encode('utf8'))[0].decode('utf8')
+
+def openio(path, mode='r', buffering=-1):
+    # allow '-' for stdin/stdout
+    if path == '-':
+        if mode == 'r':
+            return os.fdopen(os.dup(sys.stdin.fileno()), mode, buffering)
+        else:
+            return os.fdopen(os.dup(sys.stdout.fileno()), mode, buffering)
+    else:
+        return open(path, mode, buffering)
+
+def inotifywait(paths):
+    # wait for interesting events
+    inotify = inotify_simple.INotify()
+    flags = (inotify_simple.flags.ATTRIB
+        | inotify_simple.flags.CREATE
+        | inotify_simple.flags.DELETE
+        | inotify_simple.flags.DELETE_SELF
+        | inotify_simple.flags.MODIFY
+        | inotify_simple.flags.MOVED_FROM
+        | inotify_simple.flags.MOVED_TO
+        | inotify_simple.flags.MOVE_SELF)
+
+    # recurse into directories
+    for path in paths:
+        if os.path.isdir(path):
+            for dir, _, files in os.walk(path):
+                inotify.add_watch(dir, flags)
+                for f in files:
+                    inotify.add_watch(os.path.join(dir, f), flags)
+        else:
+            inotify.add_watch(path, flags)
+
+    # wait for event
+    inotify.read()
+
+class LinesIO:
+    def __init__(self, maxlen=None):
+        self.maxlen = maxlen
+        self.lines = co.deque(maxlen=maxlen)
+        self.tail = io.StringIO()
+
+        # trigger automatic sizing
+        if maxlen == 0:
+            self.resize(0)
+
+    def write(self, s):
+        # note using split here ensures the trailing string has no newline
+        lines = s.split('\n')
+
+        if len(lines) > 1 and self.tail.getvalue():
+            self.tail.write(lines[0])
+            lines[0] = self.tail.getvalue()
+            self.tail = io.StringIO()
+
+        self.lines.extend(lines[:-1])
+
+        if lines[-1]:
+            self.tail.write(lines[-1])
+
+    def resize(self, maxlen):
+        self.maxlen = maxlen
+        if maxlen == 0:
+            maxlen = shutil.get_terminal_size((80, 5))[1]
+        if maxlen != self.lines.maxlen:
+            self.lines = co.deque(self.lines, maxlen=maxlen)
+
+    canvas_lines = 1
+    def draw(self):
+        # did terminal size change?
+        if self.maxlen == 0:
+            self.resize(0)
+
+        # first thing first, give ourself a canvas
+        while LinesIO.canvas_lines < len(self.lines):
+            sys.stdout.write('\n')
+            LinesIO.canvas_lines += 1
+
+        # clear the bottom of the canvas if we shrink
+        shrink = LinesIO.canvas_lines - len(self.lines)
+        if shrink > 0:
+            for i in range(shrink):
+                sys.stdout.write('\r')
+                if shrink-1-i > 0:
+                    sys.stdout.write('\x1b[%dA' % (shrink-1-i))
+                sys.stdout.write('\x1b[K')
+                if shrink-1-i > 0:
+                    sys.stdout.write('\x1b[%dB' % (shrink-1-i))
+            sys.stdout.write('\x1b[%dA' % shrink)
+            LinesIO.canvas_lines = len(self.lines)
+
+        for i, line in enumerate(self.lines):
+            # move cursor, clear line, disable/reenable line wrapping
+            sys.stdout.write('\r')
+            if len(self.lines)-1-i > 0:
+                sys.stdout.write('\x1b[%dA' % (len(self.lines)-1-i))
+            sys.stdout.write('\x1b[K')
+            sys.stdout.write('\x1b[?7l')
+            sys.stdout.write(line)
+            sys.stdout.write('\x1b[?7h')
+            if len(self.lines)-1-i > 0:
+                sys.stdout.write('\x1b[%dB' % (len(self.lines)-1-i))
+        sys.stdout.flush()
+
+
+# parse different data representations
+def dat(x):
+    # allow the first part of an a/b fraction
+    if '/' in x:
+        x, _ = x.split('/', 1)
+
+    # first try as int
+    try:
+        return int(x, 0)
+    except ValueError:
+        pass
+
+    # then try as float
+    try:
+        return float(x)
+        # just don't allow infinity or nan
+        if m.isinf(x) or m.isnan(x):
+            raise ValueError("invalid dat %r" % x)
+    except ValueError:
+        pass
+
+    # else give up
+    raise ValueError("invalid dat %r" % x)
+
+
+# a hack log that preserves sign, with a linear region between -1 and 1
+def symlog(x):
+    if x > 1:
+        return m.log(x)+1
+    elif x < -1:
+        return -m.log(-x)-1
+    else:
+        return x
+
+class Plot:
+    def __init__(self, width, height, *,
+            xlim=None,
+            ylim=None,
+            xlog=False,
+            ylog=False,
+            braille=False,
+            dots=False):
+        # scale if we're printing with dots or braille
+        self.width = 2*width if braille else width
+        self.height = (4*height if braille
+            else 2*height if dots
+            else height)
+
+        self.xlim = xlim or (0, width)
+        self.ylim = ylim or (0, height)
+        self.xlog = xlog
+        self.ylog = ylog
+        self.braille = braille
+        self.dots = dots
+
+        self.grid = [('',False)]*(self.width*self.height)
+
+    def scale(self, x, y):
+        # scale and clamp
+        try:
+            if self.xlog:
+                x = int(self.width * (
+                    (symlog(x)-symlog(self.xlim[0]))
+                    / (symlog(self.xlim[1])-symlog(self.xlim[0]))))
+            else:
+                x = int(self.width * (
+                    (x-self.xlim[0])
+                    / (self.xlim[1]-self.xlim[0])))
+            if self.ylog:
+                y = int(self.height * (
+                    (symlog(y)-symlog(self.ylim[0]))
+                    / (symlog(self.ylim[1])-symlog(self.ylim[0]))))
+            else:
+                y = int(self.height * (
+                    (y-self.ylim[0])
+                    / (self.ylim[1]-self.ylim[0])))
+        except ZeroDivisionError:
+            x = 0
+            y = 0
+        return x, y
+
+    def point(self, x, y, *,
+            color=COLORS[0],
+            char=True):
+        # scale
+        x, y = self.scale(x, y)
+
+        # ignore out of bounds points
+        if x >= 0 and x < self.width and y >= 0 and y < self.height:
+            self.grid[x + y*self.width] = (color, char)
+
+    def line(self, x1, y1, x2, y2, *,
+            color=COLORS[0],
+            char=True):
+        # scale
+        x1, y1 = self.scale(x1, y1)
+        x2, y2 = self.scale(x2, y2)
+
+        # incremental error line algorithm
+        ex = abs(x2 - x1)
+        ey = -abs(y2 - y1)
+        dx = +1 if x1 < x2 else -1
+        dy = +1 if y1 < y2 else -1
+        e = ex + ey
+
+        while True:
+            if x1 >= 0 and x1 < self.width and y1 >= 0 and y1 < self.height:
+                self.grid[x1 + y1*self.width] = (color, char)
+            e2 = 2*e
+
+            if x1 == x2 and y1 == y2:
+                break
+
+            if e2 > ey:
+                e += ey
+                x1 += dx
+
+            if x1 == x2 and y1 == y2:
+                break
+
+            if e2 < ex:
+                e += ex
+                y1 += dy
+
+        if x2 >= 0 and x2 < self.width and y2 >= 0 and y2 < self.height:
+            self.grid[x2 + y2*self.width] = (color, char)
+
+    def plot(self, coords, *,
+            color=COLORS[0],
+            char=True,
+            line_char=True):
+        # draw lines
+        if line_char:
+            for (x1, y1), (x2, y2) in zip(coords, coords[1:]):
+                if y1 is not None and y2 is not None:
+                    self.line(x1, y1, x2, y2,
+                        color=color,
+                        char=line_char)
+
+        # draw points
+        if char and (not line_char or char is not True):
+            for x, y in coords:
+                if y is not None:
+                    self.point(x, y,
+                        color=color,
+                        char=char)
+
+    def draw(self, row, *,
+            color=False):
+        # scale if needed
+        if self.braille:
+            xscale, yscale = 2, 4
+        elif self.dots:
+            xscale, yscale = 1, 2
+        else:
+            xscale, yscale = 1, 1
+
+        y = self.height//yscale-1 - row
+        row_ = []
+        for x in range(self.width//xscale):
+            best_f = ''
+            best_c = False
+
+            # encode into a byte
+            b = 0
+            for i in range(xscale*yscale):
+                f, c = self.grid[x*xscale+(xscale-1-(i%xscale))
+                        + (y*yscale+(i//xscale))*self.width]
+                if c:
+                    b |= 1 << i
+
+                if f:
+                    best_f = f
+                if c and c is not True:
+                    best_c = c
+
+            # use byte to lookup character
+            if b:
+                if best_c:
+                    c = best_c
+                elif self.braille:
+                    c = CHARS_BRAILLE[b]
+                else:
+                    c = CHARS_DOTS[b]
+            else:
+                c = ' '
+
+            # color?
+            if b and color and best_f:
+                c = '\x1b[%sm%s\x1b[m' % (best_f, c)
+
+            # draw axis in blank spaces
+            if not b:
+                if x == 0 and y == 0:
+                    c = '+'
+                elif x == 0 and y == self.height//yscale-1:
+                    c = '^'
+                elif x == self.width//xscale-1 and y == 0:
+                    c = '>'
+                elif x == 0:
+                    c = '|'
+                elif y == 0:
+                    c = '-'
+
+            row_.append(c)
+
+        return ''.join(row_)
+
+
+def collect(csv_paths, renames=[]):
+    # collect results from CSV files
+    results = []
+    for path in csv_paths:
+        try:
+            with openio(path) as f:
+                reader = csv.DictReader(f, restval='')
+                for r in reader:
+                    results.append(r)
+        except FileNotFoundError:
+            pass
+
+    if renames:
+        for r in results:
+            # make a copy so renames can overlap
+            r_ = {}
+            for new_k, old_k in renames:
+                if old_k in r:
+                    r_[new_k] = r[old_k]
+            r.update(r_)
+
+    return results
+
+def dataset(results, x=None, y=None, define=[]):
+    # organize by 'by', x, and y
+    dataset = {}
+    i = 0
+    for r in results:
+        # filter results by matching defines
+        if not all(k in r and r[k] in vs for k, vs in define):
+            continue
+
+        # find xs
+        if x is not None:
+            if x not in r:
+                continue
+            try:
+                x_ = dat(r[x])
+            except ValueError:
+                continue
+        else:
+            x_ = i
+            i += 1
+
+        # find ys
+        if y is not None:
+            if y not in r:
+                continue
+            try:
+                y_ = dat(r[y])
+            except ValueError:
+                continue
+        else:
+            y_ = None
+
+        if y_ is not None:
+            dataset[x_] = y_ + dataset.get(x_, 0)
+        else:
+            dataset[x_] = y_ or dataset.get(x_, None)
+
+    return dataset
+
+def datasets(results, by=None, x=None, y=None, define=[]):
+    # filter results by matching defines
+    results_ = []
+    for r in results:
+        if all(k in r and r[k] in vs for k, vs in define):
+            results_.append(r)
+    results = results_
+
+    # if y not specified, try to guess from data
+    if y is None:
+        y = co.OrderedDict()
+        for r in results:
+            for k, v in r.items():
+                if (by is None or k not in by) and v.strip():
+                    try:
+                        dat(v)
+                        y[k] = True
+                    except ValueError:
+                        y[k] = False
+        y = list(k for k,v in y.items() if v)
+
+    if by is not None:
+        # find all 'by' values
+        ks = set()
+        for r in results:
+            ks.add(tuple(r.get(k, '') for k in by))
+        ks = sorted(ks)
+
+    # collect all datasets
+    datasets = co.OrderedDict()
+    for ks_ in (ks if by is not None else [()]):
+        for x_ in (x if x is not None else [None]):
+            for y_ in y:
+                # hide x/y if there is only one field
+                k_x = x_ if len(x or []) > 1 else ''
+                k_y = y_ if len(y or []) > 1 or (not ks_ and not k_x) else ''
+
+                datasets[ks_ + (k_x, k_y)] = dataset(
+                    results,
+                    x_,
+                    y_,
+                    [(by_, {k_}) for by_, k_ in zip(by, ks_)]
+                        if by is not None else [])
+
+    return datasets
+
+
+# some classes for organizing subplots into a grid
+class Subplot:
+    def __init__(self, **args):
+        self.x = 0
+        self.y = 0
+        self.xspan = 1
+        self.yspan = 1
+        self.args = args
+
+class Grid:
+    def __init__(self, subplot, width=1.0, height=1.0):
+        self.xweights = [width]
+        self.yweights = [height]
+        self.map = {(0,0): subplot}
+        self.subplots = [subplot]
+
+    def __repr__(self):
+        return 'Grid(%r, %r)' % (self.xweights, self.yweights)
+
+    @property
+    def width(self):
+        return len(self.xweights)
+
+    @property
+    def height(self):
+        return len(self.yweights)
+
+    def __iter__(self):
+        return iter(self.subplots)
+
+    def __getitem__(self, i):
+        x, y = i
+        if x < 0:
+            x += len(self.xweights)
+        if y < 0:
+            y += len(self.yweights)
+
+        return self.map[(x,y)]
+
+    def merge(self, other, dir):
+        if dir in ['above', 'below']:
+            # first scale the two grids so they line up
+            self_xweights = self.xweights
+            other_xweights = other.xweights
+            self_w = sum(self_xweights)
+            other_w = sum(other_xweights)
+            ratio = self_w / other_w
+            other_xweights = [s*ratio for s in other_xweights]
+
+            # now interleave xweights as needed
+            new_xweights = []
+            self_map = {}
+            other_map = {}
+            self_i = 0
+            other_i = 0
+            self_xweight = (self_xweights[self_i]
+                if self_i < len(self_xweights) else m.inf)
+            other_xweight = (other_xweights[other_i]
+                if other_i < len(other_xweights) else m.inf)
+            while self_i < len(self_xweights) and other_i < len(other_xweights):
+                if other_xweight - self_xweight > 0.0000001:
+                    new_xweights.append(self_xweight)
+                    other_xweight -= self_xweight
+
+                    new_i = len(new_xweights)-1
+                    for j in range(len(self.yweights)):
+                        self_map[(new_i, j)] = self.map[(self_i, j)]
+                    for j in range(len(other.yweights)):
+                        other_map[(new_i, j)] = other.map[(other_i, j)]
+                    for s in other.subplots:
+                        if s.x+s.xspan-1 == new_i:
+                            s.xspan += 1
+                        elif s.x > new_i:
+                            s.x += 1
+
+                    self_i += 1
+                    self_xweight = (self_xweights[self_i]
+                        if self_i < len(self_xweights) else m.inf)
+                elif self_xweight - other_xweight > 0.0000001:
+                    new_xweights.append(other_xweight)
+                    self_xweight -= other_xweight
+
+                    new_i = len(new_xweights)-1
+                    for j in range(len(other.yweights)):
+                        other_map[(new_i, j)] = other.map[(other_i, j)]
+                    for j in range(len(self.yweights)):
+                        self_map[(new_i, j)] = self.map[(self_i, j)]
+                    for s in self.subplots:
+                        if s.x+s.xspan-1 == new_i:
+                            s.xspan += 1
+                        elif s.x > new_i:
+                            s.x += 1
+
+                    other_i += 1
+                    other_xweight = (other_xweights[other_i]
+                        if other_i < len(other_xweights) else m.inf)
+                else:
+                    new_xweights.append(self_xweight)
+
+                    new_i = len(new_xweights)-1
+                    for j in range(len(self.yweights)):
+                        self_map[(new_i, j)] = self.map[(self_i, j)]
+                    for j in range(len(other.yweights)):
+                        other_map[(new_i, j)] = other.map[(other_i, j)]
+
+                    self_i += 1
+                    self_xweight = (self_xweights[self_i]
+                        if self_i < len(self_xweights) else m.inf)
+                    other_i += 1
+                    other_xweight = (other_xweights[other_i]
+                        if other_i < len(other_xweights) else m.inf)
+
+            # squish so ratios are preserved
+            self_h = sum(self.yweights)
+            other_h = sum(other.yweights)
+            ratio = (self_h-other_h) / self_h
+            self_yweights = [s*ratio for s in self.yweights]
+
+            # finally concatenate the two grids
+            if dir == 'above':
+                for s in other.subplots:
+                    s.y += len(self_yweights)
+                self.subplots.extend(other.subplots)
+
+                self.xweights = new_xweights
+                self.yweights = self_yweights + other.yweights
+                self.map = self_map | {(x, y+len(self_yweights)): s
+                    for (x, y), s in other_map.items()}
+            else:
+                for s in self.subplots:
+                    s.y += len(other.yweights)
+                self.subplots.extend(other.subplots)
+
+                self.xweights = new_xweights
+                self.yweights = other.yweights + self_yweights
+                self.map = other_map | {(x, y+len(other.yweights)): s
+                    for (x, y), s in self_map.items()}
+
+        if dir in ['right', 'left']:
+            # first scale the two grids so they line up
+            self_yweights = self.yweights
+            other_yweights = other.yweights
+            self_h = sum(self_yweights)
+            other_h = sum(other_yweights)
+            ratio = self_h / other_h
+            other_yweights = [s*ratio for s in other_yweights]
+
+            # now interleave yweights as needed
+            new_yweights = []
+            self_map = {}
+            other_map = {}
+            self_i = 0
+            other_i = 0
+            self_yweight = (self_yweights[self_i]
+                if self_i < len(self_yweights) else m.inf)
+            other_yweight = (other_yweights[other_i]
+                if other_i < len(other_yweights) else m.inf)
+            while self_i < len(self_yweights) and other_i < len(other_yweights):
+                if other_yweight - self_yweight > 0.0000001:
+                    new_yweights.append(self_yweight)
+                    other_yweight -= self_yweight
+
+                    new_i = len(new_yweights)-1
+                    for j in range(len(self.xweights)):
+                        self_map[(j, new_i)] = self.map[(j, self_i)]
+                    for j in range(len(other.xweights)):
+                        other_map[(j, new_i)] = other.map[(j, other_i)]
+                    for s in other.subplots:
+                        if s.y+s.yspan-1 == new_i:
+                            s.yspan += 1
+                        elif s.y > new_i:
+                            s.y += 1
+
+                    self_i += 1
+                    self_yweight = (self_yweights[self_i]
+                        if self_i < len(self_yweights) else m.inf)
+                elif self_yweight - other_yweight > 0.0000001:
+                    new_yweights.append(other_yweight)
+                    self_yweight -= other_yweight
+
+                    new_i = len(new_yweights)-1
+                    for j in range(len(other.xweights)):
+                        other_map[(j, new_i)] = other.map[(j, other_i)]
+                    for j in range(len(self.xweights)):
+                        self_map[(j, new_i)] = self.map[(j, self_i)]
+                    for s in self.subplots:
+                        if s.y+s.yspan-1 == new_i:
+                            s.yspan += 1
+                        elif s.y > new_i:
+                            s.y += 1
+
+                    other_i += 1
+                    other_yweight = (other_yweights[other_i]
+                        if other_i < len(other_yweights) else m.inf)
+                else:
+                    new_yweights.append(self_yweight)
+
+                    new_i = len(new_yweights)-1
+                    for j in range(len(self.xweights)):
+                        self_map[(j, new_i)] = self.map[(j, self_i)]
+                    for j in range(len(other.xweights)):
+                        other_map[(j, new_i)] = other.map[(j, other_i)]
+
+                    self_i += 1
+                    self_yweight = (self_yweights[self_i]
+                        if self_i < len(self_yweights) else m.inf)
+                    other_i += 1
+                    other_yweight = (other_yweights[other_i]
+                        if other_i < len(other_yweights) else m.inf)
+
+            # squish so ratios are preserved
+            self_w = sum(self.xweights)
+            other_w = sum(other.xweights)
+            ratio = (self_w-other_w) / self_w
+            self_xweights = [s*ratio for s in self.xweights]
+
+            # finally concatenate the two grids
+            if dir == 'right':
+                for s in other.subplots:
+                    s.x += len(self_xweights)
+                self.subplots.extend(other.subplots)
+
+                self.xweights = self_xweights + other.xweights
+                self.yweights = new_yweights
+                self.map = self_map | {(x+len(self_xweights), y): s
+                    for (x, y), s in other_map.items()}
+            else:
+                for s in self.subplots:
+                    s.x += len(other.xweights)
+                self.subplots.extend(other.subplots)
+
+                self.xweights = other.xweights + self_xweights
+                self.yweights = new_yweights
+                self.map = other_map | {(x+len(other.xweights), y): s
+                    for (x, y), s in self_map.items()}
+                    
+
+    def scale(self, width, height):
+        self.xweights = [s*width for s in self.xweights]
+        self.yweights = [s*height for s in self.yweights]
+
+    @classmethod
+    def fromargs(cls, width=1.0, height=1.0, *,
+            subplots=[],
+            **args):
+        grid = cls(Subplot(**args))
+
+        for dir, subargs in subplots:
+            subgrid = cls.fromargs(
+                width=subargs.pop('width',
+                    0.5 if dir in ['right', 'left'] else width),
+                height=subargs.pop('height',
+                    0.5 if dir in ['above', 'below'] else height),
+                **subargs)
+            grid.merge(subgrid, dir)
+
+        grid.scale(width, height)
+        return grid
+    
+
+def main(csv_paths, *,
+        by=None,
+        x=None,
+        y=None,
+        define=[],
+        color=False,
+        braille=False,
+        colors=None,
+        chars=None,
+        line_chars=None,
+        points=False,
+        points_and_lines=False,
+        width=None,
+        height=None,
+        xlim=(None,None),
+        ylim=(None,None),
+        xlog=False,
+        ylog=False,
+        x2=False,
+        y2=False,
+        xunits='',
+        yunits='',
+        xlabel=None,
+        ylabel=None,
+        xticklabels=None,
+        yticklabels=None,
+        title=None,
+        legend_right=False,
+        legend_above=False,
+        legend_below=False,
+        subplot={},
+        subplots=[],
+        cat=False,
+        keep_open=False,
+        sleep=None,
+        **args):
+    # figure out what color should be
+    if color == 'auto':
+        color = sys.stdout.isatty()
+    elif color == 'always':
+        color = True
+    else:
+        color = False
+
+    # what colors to use?
+    if colors is not None:
+        colors_ = colors
+    else:
+        colors_ = COLORS
+
+    if chars is not None:
+        chars_ = chars
+    elif points_and_lines:
+        chars_ = CHARS_POINTS_AND_LINES
+    else:
+        chars_ = [True]
+
+    if line_chars is not None:
+        line_chars_ = line_chars
+    elif points_and_lines or not points:
+        line_chars_ = [True]
+    else:
+        line_chars_ = [False]
+
+    # allow escape codes in labels/titles
+    title = escape(title).splitlines() if title is not None else []
+    xlabel = escape(xlabel).splitlines() if xlabel is not None else []
+    ylabel = escape(ylabel).splitlines() if ylabel is not None else []
+
+    # separate out renames
+    renames = list(it.chain.from_iterable(
+        ((k, v) for v in vs)
+        for k, vs in it.chain(by or [], x or [], y or [])))
+    if by is not None:
+        by = [k for k, _ in by]
+    if x is not None:
+        x = [k for k, _ in x]
+    if y is not None:
+        y = [k for k, _ in y]
+
+    # create a grid of subplots
+    grid = Grid.fromargs(
+        subplots=subplots + subplot.pop('subplots', []),
+        **subplot)
+
+    for s in grid:
+        # allow subplot params to override global params
+        x2_ = s.args.get('x2', False) or x2
+        y2_ = s.args.get('y2', False) or y2
+        xunits_ = s.args.get('xunits', xunits)
+        yunits_ = s.args.get('yunits', yunits)
+        xticklabels_ = s.args.get('xticklabels', xticklabels)
+        yticklabels_ = s.args.get('yticklabels', yticklabels)
+
+        # label/titles are handled a bit differently in subplots
+        subtitle = s.args.get('title')
+        xsublabel = s.args.get('xlabel')
+        ysublabel = s.args.get('ylabel')
+
+        # allow escape codes in sublabels/subtitles
+        subtitle = (escape(subtitle).splitlines()
+            if subtitle is not None else [])
+        xsublabel = (escape(xsublabel).splitlines()
+            if xsublabel is not None else [])
+        ysublabel = (escape(ysublabel).splitlines()
+            if ysublabel is not None else [])
+
+        # don't allow >2 ticklabels and render single ticklabels only once
+        if xticklabels_ is not None:
+            if len(xticklabels_) == 1:
+                xticklabels_ = ["", xticklabels_[0]]
+            elif len(xticklabels_) > 2:
+                xticklabels_ = [xticklabels_[0], xticklabels_[-1]]
+        if yticklabels_ is not None:
+            if len(yticklabels_) == 1:
+                yticklabels_ = ["", yticklabels_[0]]
+            elif len(yticklabels_) > 2:
+                yticklabels_ = [yticklabels_[0], yticklabels_[-1]]
+
+        s.x2 = x2_
+        s.y2 = y2_
+        s.xunits = xunits_
+        s.yunits = yunits_
+        s.xticklabels = xticklabels_
+        s.yticklabels = yticklabels_
+        s.title = subtitle
+        s.xlabel = xsublabel
+        s.ylabel = ysublabel
+
+    # preprocess margins so they can be shared
+    for s in grid:
+        s.xmargin = (
+            len(s.ylabel) + (1 if s.ylabel else 0) # fit ysublabel
+                + (1 if s.x > 0 else 0),           # space between
+            ((5 if s.y2 else 4) + len(s.yunits)    # fit yticklabels
+                if s.yticklabels is None
+                else max((len(t) for t in s.yticklabels), default=0))
+                + (1 if s.yticklabels != [] else 0),
+        )
+        s.ymargin = (
+            len(s.xlabel),                   # fit xsublabel
+            1 if s.xticklabels != [] else 0, # fit xticklabels
+            len(s.title),                    # fit subtitle
+        )
+
+    for s in grid:
+        # share margins so everything aligns nicely
+        s.xmargin = (
+            max(s_.xmargin[0] for s_ in grid if s_.x == s.x),
+            max(s_.xmargin[1] for s_ in grid if s_.x == s.x),
+        )
+        s.ymargin = (
+            max(s_.ymargin[0] for s_ in grid if s_.y == s.y),
+            max(s_.ymargin[1] for s_ in grid if s_.y == s.y),
+            max(s_.ymargin[-1] for s_ in grid if s_.y+s_.yspan == s.y+s.yspan),
+        )
+
+
+    def draw(f):
+        def writeln(s=''):
+            f.write(s)
+            f.write('\n')
+        f.writeln = writeln
+
+        # first collect results from CSV files
+        results = collect(csv_paths, renames)
+
+        # then extract the requested datasets
+        datasets_ = datasets(results, by, x, y, define)
+
+        # figure out colors/chars here so that subplot defines
+        # don't change them later, that'd be bad
+        datacolors_ = {
+            name: colors_[i % len(colors_)]
+            for i, name in enumerate(datasets_.keys())}
+        datachars_ = {
+            name: chars_[i % len(chars_)]
+            for i, name in enumerate(datasets_.keys())}
+        dataline_chars_ = {
+            name: line_chars_[i % len(line_chars_)]
+            for i, name in enumerate(datasets_.keys())}
+
+        # build legend?
+        legend_width = 0
+        if legend_right or legend_above or legend_below:
+            legend_ = []
+            for i, k in enumerate(datasets_.keys()):
+                label = '%s%s' % (
+                    '%s ' % chars_[i % len(chars_)]
+                        if chars is not None
+                        else '%s ' % line_chars_[i % len(line_chars_)]
+                        if line_chars is not None
+                        else '',
+                    ','.join(k_ for k_ in k if k_))
+
+                if label:
+                    legend_.append(label)
+                    legend_width = max(legend_width, len(label)+1)
+
+        # figure out our canvas size
+        if width is None:
+            width_ = min(80, shutil.get_terminal_size((80, None))[0])
+        elif width:
+            width_ = width
+        else:
+            width_ = shutil.get_terminal_size((80, None))[0]
+
+        if height is None:
+            height_ = 17 + len(title) + len(xlabel)
+        elif height:
+            height_ = height
+        else:
+            height_ = shutil.get_terminal_size((None,
+                17 + len(title) + len(xlabel)))[1]
+            # make space for shell prompt
+            if not keep_open:
+                height_ -= 1
+
+        # carve out space for the xlabel
+        height_ -= len(xlabel)
+        # carve out space for the ylabel
+        width_ -= len(ylabel) + (1 if ylabel else 0)
+        # carve out space for title
+        height_ -= len(title)
+
+        # carve out space for the legend
+        if legend_right and legend_:
+            width_ -= legend_width
+        if legend_above and legend_:
+            legend_cols = len(legend_)
+            while True:
+                legend_widths = [
+                    max(len(l) for l in legend_[i::legend_cols])
+                    for i in range(legend_cols)]
+                if (legend_cols <= 1
+                        or sum(legend_widths)+2*(legend_cols-1)
+                            + max(sum(s.xmargin[:2]) for s in grid if s.x == 0)
+                            <= width_):
+                    break
+                legend_cols -= 1
+            height_ -= (len(legend_)+legend_cols-1) // legend_cols
+        if legend_below and legend_:
+            legend_cols = len(legend_)
+            while True:
+                legend_widths = [
+                    max(len(l) for l in legend_[i::legend_cols])
+                    for i in range(legend_cols)]
+                if (legend_cols <= 1
+                        or sum(legend_widths)+2*(legend_cols-1)
+                            + max(sum(s.xmargin[:2]) for s in grid if s.x == 0)
+                            <= width_):
+                    break
+                legend_cols -= 1
+            height_ -= (len(legend_)+legend_cols-1) // legend_cols
+
+        # figure out the grid dimensions
+        #
+        # note we floor to give the dimension tweaks the best chance of not
+        # exceeding the requested dimensions, this means we usually are less
+        # than the requested dimensions by quite a bit when we have many
+        # subplots, but it's a tradeoff for a relatively simple implementation
+        widths = [m.floor(w*width_) for w in grid.xweights]
+        heights = [m.floor(w*height_) for w in grid.yweights]
+
+        # tweak dimensions to allow all plots to have a minimum width,
+        # this may force the plot to be larger than the requested dimensions,
+        # but that's the best we can do
+        for s in grid:
+            # fit xunits
+            minwidth = sum(s.xmargin) + max(2,
+                2*((5 if s.x2 else 4)+len(s.xunits))
+                    if s.xticklabels is None
+                    else sum(len(t) for t in s.xticklabels))
+            # fit yunits
+            minheight = sum(s.ymargin) + 2
+
+            i = 0
+            while minwidth > sum(widths[s.x:s.x+s.xspan]):
+                widths[s.x+i] += 1
+                i = (i + 1) % s.xspan
+
+            i = 0
+            while minheight > sum(heights[s.y:s.y+s.yspan]):
+                heights[s.y+i] += 1
+                i = (i + 1) % s.yspan
+
+        width_ = sum(widths)
+        height_ = sum(heights)
+
+        # create a plot for each subplot
+        for s in grid:
+            # allow subplot params to override global params
+            define_ = define + s.args.get('define', [])
+            xlim_ = s.args.get('xlim', xlim)
+            ylim_ = s.args.get('ylim', ylim)
+            xlog_ = s.args.get('xlog', False) or xlog
+            ylog_ = s.args.get('ylog', False) or ylog
+
+            # allow shortened ranges
+            if len(xlim_) == 1:
+                xlim_ = (0, xlim_[0])
+            if len(ylim_) == 1:
+                ylim_ = (0, ylim_[0])
+
+            # data can be constrained by subplot-specific defines,
+            # so re-extract for each plot
+            subdatasets = datasets(results, by, x, y, define_)
+
+            # find actual xlim/ylim
+            xlim_ = (
+                xlim_[0] if xlim_[0] is not None
+                    else min(it.chain([0], (k
+                        for r in subdatasets.values()
+                        for k, v in r.items()
+                        if v is not None))),
+                xlim_[1] if xlim_[1] is not None
+                    else max(it.chain([0], (k
+                        for r in subdatasets.values()
+                        for k, v in r.items()
+                        if v is not None))))
+
+            ylim_ = (
+                ylim_[0] if ylim_[0] is not None
+                    else min(it.chain([0], (v
+                        for r in subdatasets.values()
+                        for _, v in r.items()
+                        if v is not None))),
+                ylim_[1] if ylim_[1] is not None
+                    else max(it.chain([0], (v
+                        for r in subdatasets.values()
+                        for _, v in r.items()
+                        if v is not None))))
+
+            # find actual width/height
+            subwidth = sum(widths[s.x:s.x+s.xspan]) - sum(s.xmargin)
+            subheight = sum(heights[s.y:s.y+s.yspan]) - sum(s.ymargin)
+
+            # plot!
+            plot = Plot(
+                subwidth,
+                subheight,
+                xlim=xlim_,
+                ylim=ylim_,
+                xlog=xlog_,
+                ylog=ylog_,
+                braille=line_chars is None and braille,
+                dots=line_chars is None and not braille)
+
+            for name, dataset in subdatasets.items():
+                plot.plot(
+                    sorted((x,y) for x,y in dataset.items()),
+                    color=datacolors_[name],
+                    char=datachars_[name],
+                    line_char=dataline_chars_[name])
+
+            s.plot = plot
+            s.width = subwidth
+            s.height = subheight
+            s.xlim = xlim_
+            s.ylim = ylim_
+
+
+        # now that everything's plotted, let's render things to the terminal
+
+        # figure out margin
+        xmargin = (
+            len(ylabel) + (1 if ylabel else 0),
+            sum(grid[0,0].xmargin[:2]),
+        )
+        ymargin = (
+            sum(grid[0,0].ymargin[:2]),
+            grid[-1,-1].ymargin[-1],
+        )
+
+        # draw title?
+        for line in title:
+            f.writeln('%*s%s' % (
+                sum(xmargin[:2]), '',
+                line.center(width_-xmargin[1])))
+
+        # draw legend_above?
+        if legend_above and legend_:
+            for i in range(0, len(legend_), legend_cols):
+                f.writeln('%*s%s' % (
+                    max(sum(xmargin[:2])
+                        + (width_-xmargin[1]
+                            - (sum(legend_widths)+2*(legend_cols-1)))
+                            // 2,
+                        0), '',
+                    '  '.join('%s%s%s' % (
+                        '\x1b[%sm' % colors_[(i+j) % len(colors_)]
+                            if color else '',
+                        '%-*s' % (legend_widths[j], legend_[i+j]),
+                        '\x1b[m'
+                            if color else '')
+                        for j in range(min(legend_cols, len(legend_)-i)))))
+
+        for row in range(height_):
+            # draw ylabel?
+            f.write(
+                '%s ' % ''.join(
+                    ('%*s%s%*s' % (
+                        ymargin[-1], '',
+                        line.center(height_-sum(ymargin)),
+                        ymargin[0], ''))[row]
+                    for line in ylabel)
+                if ylabel else '')
+
+            for x_ in range(grid.width):
+                # figure out the grid x/y position
+                subrow = row
+                y_ = len(heights)-1
+                while subrow >= heights[y_]:
+                    subrow -= heights[y_]
+                    y_ -= 1
+
+                s = grid[x_, y_]
+                subrow = row - sum(heights[s.y+s.yspan:])
+
+                # header
+                if subrow < s.ymargin[-1]:
+                    # draw subtitle?
+                    if subrow < len(s.title):
+                        f.write('%*s%s' % (
+                            sum(s.xmargin[:2]), '',
+                            s.title[subrow].center(s.width)))
+                    else:
+                        f.write('%*s%*s' % (
+                            sum(s.xmargin[:2]), '',
+                            s.width, ''))
+                # draw plot?
+                elif subrow-s.ymargin[-1] < s.height:
+                    subrow = subrow-s.ymargin[-1]
+
+                    # draw ysublabel?
+                    f.write('%-*s' % (
+                        s.xmargin[0],
+                        '%s ' % ''.join(
+                                line.center(s.height)[subrow]
+                                for line in s.ylabel)
+                            if s.ylabel else ''))
+
+                    # draw yunits?
+                    if subrow == 0 and s.yticklabels != []:
+                        f.write('%*s' % (
+                            s.xmargin[1],
+                            ((si2 if s.y2 else si)(s.ylim[1]) + s.yunits
+                                if s.yticklabels is None
+                                else s.yticklabels[1])
+                                + ' '))
+                    elif subrow == s.height-1 and s.yticklabels != []:
+                        f.write('%*s' % (
+                            s.xmargin[1],
+                            ((si2 if s.y2 else si)(s.ylim[0]) + s.yunits
+                                if s.yticklabels is None
+                                else s.yticklabels[0])
+                                + ' '))
+                    else:
+                        f.write('%*s' % (
+                            s.xmargin[1], ''))
+
+                    # draw plot!
+                    f.write(s.plot.draw(subrow, color=color))
+
+                # footer
+                else:
+                    subrow = subrow-s.ymargin[-1]-s.height
+
+                    # draw xunits?
+                    if subrow < (1 if s.xticklabels != [] else 0):
+                        f.write('%*s%-*s%*s%*s' % (
+                            sum(s.xmargin[:2]), '',
+                            (5 if s.x2 else 4) + len(s.xunits)
+                                if s.xticklabels is None
+                                else len(s.xticklabels[0]),
+                            (si2 if s.x2 else si)(s.xlim[0]) + s.xunits
+                                if s.xticklabels is None
+                                else s.xticklabels[0],
+                            s.width - (2*((5 if s.x2 else 4)+len(s.xunits))
+                                if s.xticklabels is None
+                                else sum(len(t) for t in s.xticklabels)), '',
+                            (5 if s.x2 else 4) + len(s.xunits)
+                                if s.xticklabels is None
+                                else len(s.xticklabels[1]),
+                            (si2 if s.x2 else si)(s.xlim[1]) + s.xunits
+                                if s.xticklabels is None
+                                else s.xticklabels[1]))
+                    # draw xsublabel?
+                    elif (subrow < s.ymargin[1]
+                            or subrow-s.ymargin[1] >= len(s.xlabel)):
+                        f.write('%*s%*s' % (
+                            sum(s.xmargin[:2]), '',
+                            s.width, ''))
+                    else:
+                        f.write('%*s%s' % (
+                            sum(s.xmargin[:2]), '',
+                            s.xlabel[subrow-s.ymargin[1]].center(s.width)))
+
+            # draw legend_right?
+            if (legend_right and legend_
+                    and row >= ymargin[-1]
+                    and row-ymargin[-1] < len(legend_)):
+                j = row-ymargin[-1]
+                f.write(' %s%s%s' % (
+                    '\x1b[%sm' % colors_[j % len(colors_)] if color else '',
+                    legend_[j],
+                    '\x1b[m' if color else ''))
+
+            f.writeln()
+
+        # draw xlabel?
+        for line in xlabel:
+            f.writeln('%*s%s' % (
+                sum(xmargin[:2]), '',
+                line.center(width_-xmargin[1])))
+
+        # draw legend below?
+        if legend_below and legend_:
+            for i in range(0, len(legend_), legend_cols):
+                f.writeln('%*s%s' % (
+                    max(sum(xmargin[:2])
+                        + (width_-xmargin[1]
+                            - (sum(legend_widths)+2*(legend_cols-1)))
+                            // 2,
+                        0), '',
+                    '  '.join('%s%s%s' % (
+                        '\x1b[%sm' % colors_[(i+j) % len(colors_)]
+                            if color else '',
+                        '%-*s' % (legend_widths[j], legend_[i+j]),
+                        '\x1b[m'
+                            if color else '')
+                        for j in range(min(legend_cols, len(legend_)-i)))))
+
+
+    if keep_open:
+        try:
+            while True:
+                if cat:
+                    draw(sys.stdout)
+                else:
+                    ring = LinesIO()
+                    draw(ring)
+                    ring.draw()
+
+                # try to inotifywait
+                if inotify_simple is not None:
+                    ptime = time.time()
+                    inotifywait(csv_paths)
+                    # sleep for a minimum amount of time, this helps issues
+                    # around rapidly updating files
+                    time.sleep(max(0, (sleep or 0.01) - (time.time()-ptime)))
+                else:
+                    time.sleep(sleep or 0.1)
+        except KeyboardInterrupt:
+            pass
+
+        if cat:
+            draw(sys.stdout)
+        else:
+            ring = LinesIO()
+            draw(ring)
+            ring.draw()
+        sys.stdout.write('\n')
+    else:
+        draw(sys.stdout)
+
+
+if __name__ == "__main__":
+    import sys
+    import argparse
+    parser = argparse.ArgumentParser(
+        description="Plot CSV files in terminal.",
+        allow_abbrev=False)
+    parser.add_argument(
+        'csv_paths',
+        nargs='*',
+        help="Input *.csv files.")
+    parser.add_argument(
+        '-b', '--by',
+        action='append',
+        type=lambda x: (
+            lambda k,v=None: (k, v.split(',') if v is not None else ())
+            )(*x.split('=', 1)),
+        help="Group by this field. Can rename fields with new_name=old_name.")
+    parser.add_argument(
+        '-x',
+        action='append',
+        type=lambda x: (
+            lambda k,v=None: (k, v.split(',') if v is not None else ())
+            )(*x.split('=', 1)),
+        help="Field to use for the x-axis. Can rename fields with "
+            "new_name=old_name.")
+    parser.add_argument(
+        '-y',
+        action='append',
+        type=lambda x: (
+            lambda k,v=None: (k, v.split(',') if v is not None else ())
+            )(*x.split('=', 1)),
+        help="Field to use for the y-axis. Can rename fields with "
+            "new_name=old_name.")
+    parser.add_argument(
+        '-D', '--define',
+        type=lambda x: (lambda k,v: (k, set(v.split(','))))(*x.split('=', 1)),
+        action='append',
+        help="Only include results where this field is this value. May include "
+            "comma-separated options.")
+    parser.add_argument(
+        '--color',
+        choices=['never', 'always', 'auto'],
+        default='auto',
+        help="When to use terminal colors. Defaults to 'auto'.")
+    parser.add_argument(
+        '-⣿', '--braille',
+        action='store_true',
+        help="Use 2x4 unicode braille characters. Note that braille characters "
+            "sometimes suffer from inconsistent widths.")
+    parser.add_argument(
+        '-.', '--points',
+        action='store_true',
+        help="Only draw data points.")
+    parser.add_argument(
+        '-!', '--points-and-lines',
+        action='store_true',
+        help="Draw data points and lines.")
+    parser.add_argument(
+        '--colors',
+        type=lambda x: [x.strip() for x in x.split(',')],
+        help="Comma-separated colors to use.")
+    parser.add_argument(
+        '--chars',
+        help="Characters to use for points.")
+    parser.add_argument(
+        '--line-chars',
+        help="Characters to use for lines.")
+    parser.add_argument(
+        '-W', '--width',
+        nargs='?',
+        type=lambda x: int(x, 0),
+        const=0,
+        help="Width in columns. 0 uses the terminal width. Defaults to "
+            "min(terminal, 80).")
+    parser.add_argument(
+        '-H', '--height',
+        nargs='?',
+        type=lambda x: int(x, 0),
+        const=0,
+        help="Height in rows. 0 uses the terminal height. Defaults to 17.")
+    parser.add_argument(
+        '-X', '--xlim',
+        type=lambda x: tuple(
+            dat(x) if x.strip() else None
+            for x in x.split(',')),
+        help="Range for the x-axis.")
+    parser.add_argument(
+        '-Y', '--ylim',
+        type=lambda x: tuple(
+            dat(x) if x.strip() else None
+            for x in x.split(',')),
+        help="Range for the y-axis.")
+    parser.add_argument(
+        '--xlog',
+        action='store_true',
+        help="Use a logarithmic x-axis.")
+    parser.add_argument(
+        '--ylog',
+        action='store_true',
+        help="Use a logarithmic y-axis.")
+    parser.add_argument(
+        '--x2',
+        action='store_true',
+        help="Use base-2 prefixes for the x-axis.")
+    parser.add_argument(
+        '--y2',
+        action='store_true',
+        help="Use base-2 prefixes for the y-axis.")
+    parser.add_argument(
+        '--xunits',
+        help="Units for the x-axis.")
+    parser.add_argument(
+        '--yunits',
+        help="Units for the y-axis.")
+    parser.add_argument(
+        '--xlabel',
+        help="Add a label to the x-axis.")
+    parser.add_argument(
+        '--ylabel',
+        help="Add a label to the y-axis.")
+    parser.add_argument(
+        '--xticklabels',
+        type=lambda x:
+            [x.strip() for x in x.split(',')]
+            if x.strip() else [],
+        help="Comma separated xticklabels.")
+    parser.add_argument(
+        '--yticklabels',
+        type=lambda x:
+            [x.strip() for x in x.split(',')]
+            if x.strip() else [],
+        help="Comma separated yticklabels.")
+    parser.add_argument(
+        '-t', '--title',
+        help="Add a title.")
+    parser.add_argument(
+        '-l', '--legend-right',
+        action='store_true',
+        help="Place a legend to the right.")
+    parser.add_argument(
+        '--legend-above',
+        action='store_true',
+        help="Place a legend above.")
+    parser.add_argument(
+        '--legend-below',
+        action='store_true',
+        help="Place a legend below.")
+    class AppendSubplot(argparse.Action):
+        @staticmethod
+        def parse(value):
+            import copy
+            subparser = copy.deepcopy(parser)
+            next(a for a in subparser._actions
+                if '--width' in a.option_strings).type = float
+            next(a for a in subparser._actions
+                if '--height' in a.option_strings).type = float
+            return subparser.parse_intermixed_args(shlex.split(value or ""))
+        def __call__(self, parser, namespace, value, option):
+            if not hasattr(namespace, 'subplots'):
+                namespace.subplots = []
+            namespace.subplots.append((
+                option.split('-')[-1],
+                self.__class__.parse(value)))
+    parser.add_argument(
+        '--subplot-above',
+        action=AppendSubplot,
+        help="Add subplot above with the same dataset. Takes an arg string to "
+            "control the subplot which supports most (but not all) of the "
+            "parameters listed here. The relative dimensions of the subplot "
+            "can be controlled with -W/-H which now take a percentage.")
+    parser.add_argument(
+        '--subplot-below',
+        action=AppendSubplot,
+        help="Add subplot below with the same dataset.")
+    parser.add_argument(
+        '--subplot-left',
+        action=AppendSubplot,
+        help="Add subplot left with the same dataset.")
+    parser.add_argument(
+        '--subplot-right',
+        action=AppendSubplot,
+        help="Add subplot right with the same dataset.")
+    parser.add_argument(
+        '--subplot',
+        type=AppendSubplot.parse,
+        help="Add subplot-specific arguments to the main plot.")
+    parser.add_argument(
+        '-z', '--cat',
+        action='store_true',
+        help="Pipe directly to stdout.")
+    parser.add_argument(
+        '-k', '--keep-open',
+        action='store_true',
+        help="Continue to open and redraw the CSV files in a loop.")
+    parser.add_argument(
+        '-s', '--sleep',
+        type=float,
+        help="Time in seconds to sleep between redraws when running with -k. "
+            "Defaults to 0.01.")
+
+    def dictify(ns):
+        if hasattr(ns, 'subplots'):
+            ns.subplots = [(dir, dictify(subplot_ns))
+                for dir, subplot_ns in ns.subplots]
+        if ns.subplot is not None:
+            ns.subplot = dictify(ns.subplot)
+        return {k: v
+            for k, v in vars(ns).items()
+            if v is not None}
+
+    sys.exit(main(**dictify(parser.parse_intermixed_args())))
diff --git a/scripts/plotmpl.py b/scripts/plotmpl.py
new file mode 100644
index 0000000..1bc1158
--- /dev/null
+++ b/scripts/plotmpl.py
@@ -0,0 +1,1262 @@
+#!/usr/bin/env python3
+#
+# Plot CSV files with matplotlib.
+#
+# Example:
+# ./scripts/plotmpl.py bench.csv -xSIZE -ybench_read -obench.svg
+#
+# Copyright (c) 2022, The littlefs authors.
+# SPDX-License-Identifier: BSD-3-Clause
+#
+
+import codecs
+import collections as co
+import csv
+import io
+import itertools as it
+import logging
+import math as m
+import numpy as np
+import os
+import shlex
+import shutil
+import time
+
+import matplotlib as mpl
+import matplotlib.pyplot as plt
+
+# some nicer colors borrowed from Seaborn
+# note these include a non-opaque alpha
+COLORS = [
+    '#4c72b0bf', # blue
+    '#dd8452bf', # orange
+    '#55a868bf', # green
+    '#c44e52bf', # red
+    '#8172b3bf', # purple
+    '#937860bf', # brown
+    '#da8bc3bf', # pink
+    '#8c8c8cbf', # gray
+    '#ccb974bf', # yellow
+    '#64b5cdbf', # cyan
+]
+COLORS_DARK = [
+    '#a1c9f4bf', # blue
+    '#ffb482bf', # orange
+    '#8de5a1bf', # green
+    '#ff9f9bbf', # red
+    '#d0bbffbf', # purple
+    '#debb9bbf', # brown
+    '#fab0e4bf', # pink
+    '#cfcfcfbf', # gray
+    '#fffea3bf', # yellow
+    '#b9f2f0bf', # cyan
+]
+ALPHAS = [0.75]
+FORMATS = ['-']
+FORMATS_POINTS = ['.']
+FORMATS_POINTS_AND_LINES = ['.-']
+
+WIDTH = 750
+HEIGHT = 350
+FONT_SIZE = 11
+
+SI_PREFIXES = {
+    18:  'E',
+    15:  'P',
+    12:  'T',
+    9:   'G',
+    6:   'M',
+    3:   'K',
+    0:   '',
+    -3:  'm',
+    -6:  'u',
+    -9:  'n',
+    -12: 'p',
+    -15: 'f',
+    -18: 'a',
+}
+
+SI2_PREFIXES = {
+    60:  'Ei',
+    50:  'Pi',
+    40:  'Ti',
+    30:  'Gi',
+    20:  'Mi',
+    10:  'Ki',
+    0:   '',
+    -10: 'mi',
+    -20: 'ui',
+    -30: 'ni',
+    -40: 'pi',
+    -50: 'fi',
+    -60: 'ai',
+}
+
+
+# formatter for matplotlib
+def si(x):
+    if x == 0:
+        return '0'
+    # figure out prefix and scale
+    p = 3*int(m.log(abs(x), 10**3))
+    p = min(18, max(-18, p))
+    # format with 3 digits of precision
+    s = '%.3f' % (abs(x) / (10.0**p))
+    s = s[:3+1]
+    # truncate but only digits that follow the dot
+    if '.' in s:
+        s = s.rstrip('0')
+        s = s.rstrip('.')
+    return '%s%s%s' % ('-' if x < 0 else '', s, SI_PREFIXES[p])
+
+# formatter for matplotlib
+def si2(x):
+    if x == 0:
+        return '0'
+    # figure out prefix and scale
+    p = 10*int(m.log(abs(x), 2**10))
+    p = min(30, max(-30, p))
+    # format with 3 digits of precision
+    s = '%.3f' % (abs(x) / (2.0**p))
+    s = s[:3+1]
+    # truncate but only digits that follow the dot
+    if '.' in s:
+        s = s.rstrip('0')
+        s = s.rstrip('.')
+    return '%s%s%s' % ('-' if x < 0 else '', s, SI2_PREFIXES[p])
+
+# parse escape strings
+def escape(s):
+    return codecs.escape_decode(s.encode('utf8'))[0].decode('utf8')
+
+# we want to use MaxNLocator, but since MaxNLocator forces multiples of 10
+# to be an option, we can't really...
+class AutoMultipleLocator(mpl.ticker.MultipleLocator):
+    def __init__(self, base, nbins=None):
+        # note base needs to be floats to avoid integer pow issues
+        self.base = float(base)
+        self.nbins = nbins
+        super().__init__(self.base)
+
+    def __call__(self):
+        # find best tick count, conveniently matplotlib has a function for this
+        vmin, vmax = self.axis.get_view_interval()
+        vmin, vmax = mpl.transforms.nonsingular(vmin, vmax, 1e-12, 1e-13)
+        if self.nbins is not None:
+            nbins = self.nbins
+        else:
+            nbins = np.clip(self.axis.get_tick_space(), 1, 9)
+
+        # find the best power, use this as our locator's actual base
+        scale = self.base ** (m.ceil(m.log((vmax-vmin) / (nbins+1), self.base)))
+        self.set_params(scale)
+
+        return super().__call__()
+
+
+def openio(path, mode='r', buffering=-1):
+    # allow '-' for stdin/stdout
+    if path == '-':
+        if mode == 'r':
+            return os.fdopen(os.dup(sys.stdin.fileno()), mode, buffering)
+        else:
+            return os.fdopen(os.dup(sys.stdout.fileno()), mode, buffering)
+    else:
+        return open(path, mode, buffering)
+
+
+# parse different data representations
+def dat(x):
+    # allow the first part of an a/b fraction
+    if '/' in x:
+        x, _ = x.split('/', 1)
+
+    # first try as int
+    try:
+        return int(x, 0)
+    except ValueError:
+        pass
+
+    # then try as float
+    try:
+        return float(x)
+        # just don't allow infinity or nan
+        if m.isinf(x) or m.isnan(x):
+            raise ValueError("invalid dat %r" % x)
+    except ValueError:
+        pass
+
+    # else give up
+    raise ValueError("invalid dat %r" % x)
+
+def collect(csv_paths, renames=[]):
+    # collect results from CSV files
+    results = []
+    for path in csv_paths:
+        try:
+            with openio(path) as f:
+                reader = csv.DictReader(f, restval='')
+                for r in reader:
+                    results.append(r)
+        except FileNotFoundError:
+            pass
+
+    if renames:
+        for r in results:
+            # make a copy so renames can overlap
+            r_ = {}
+            for new_k, old_k in renames:
+                if old_k in r:
+                    r_[new_k] = r[old_k]
+            r.update(r_)
+
+    return results
+
+def dataset(results, x=None, y=None, define=[]):
+    # organize by 'by', x, and y
+    dataset = {}
+    i = 0
+    for r in results:
+        # filter results by matching defines
+        if not all(k in r and r[k] in vs for k, vs in define):
+            continue
+
+        # find xs
+        if x is not None:
+            if x not in r:
+                continue
+            try:
+                x_ = dat(r[x])
+            except ValueError:
+                continue
+        else:
+            x_ = i
+            i += 1
+
+        # find ys
+        if y is not None:
+            if y not in r:
+                continue
+            try:
+                y_ = dat(r[y])
+            except ValueError:
+                continue
+        else:
+            y_ = None
+
+        if y_ is not None:
+            dataset[x_] = y_ + dataset.get(x_, 0)
+        else:
+            dataset[x_] = y_ or dataset.get(x_, None)
+
+    return dataset
+
+def datasets(results, by=None, x=None, y=None, define=[]):
+    # filter results by matching defines
+    results_ = []
+    for r in results:
+        if all(k in r and r[k] in vs for k, vs in define):
+            results_.append(r)
+    results = results_
+
+    # if y not specified, try to guess from data
+    if y is None:
+        y = co.OrderedDict()
+        for r in results:
+            for k, v in r.items():
+                if (by is None or k not in by) and v.strip():
+                    try:
+                        dat(v)
+                        y[k] = True
+                    except ValueError:
+                        y[k] = False
+        y = list(k for k,v in y.items() if v)
+
+    if by is not None:
+        # find all 'by' values
+        ks = set()
+        for r in results:
+            ks.add(tuple(r.get(k, '') for k in by))
+        ks = sorted(ks)
+
+    # collect all datasets
+    datasets = co.OrderedDict()
+    for ks_ in (ks if by is not None else [()]):
+        for x_ in (x if x is not None else [None]):
+            for y_ in y:
+                # hide x/y if there is only one field
+                k_x = x_ if len(x or []) > 1 else ''
+                k_y = y_ if len(y or []) > 1 or (not ks_ and not k_x) else ''
+
+                datasets[ks_ + (k_x, k_y)] = dataset(
+                    results,
+                    x_,
+                    y_,
+                    [(by_, {k_}) for by_, k_ in zip(by, ks_)]
+                        if by is not None else [])
+
+    return datasets
+
+
+# some classes for organizing subplots into a grid
+class Subplot:
+    def __init__(self, **args):
+        self.x = 0
+        self.y = 0
+        self.xspan = 1
+        self.yspan = 1
+        self.args = args
+
+class Grid:
+    def __init__(self, subplot, width=1.0, height=1.0):
+        self.xweights = [width]
+        self.yweights = [height]
+        self.map = {(0,0): subplot}
+        self.subplots = [subplot]
+
+    def __repr__(self):
+        return 'Grid(%r, %r)' % (self.xweights, self.yweights)
+
+    @property
+    def width(self):
+        return len(self.xweights)
+
+    @property
+    def height(self):
+        return len(self.yweights)
+
+    def __iter__(self):
+        return iter(self.subplots)
+
+    def __getitem__(self, i):
+        x, y = i
+        if x < 0:
+            x += len(self.xweights)
+        if y < 0:
+            y += len(self.yweights)
+
+        return self.map[(x,y)]
+
+    def merge(self, other, dir):
+        if dir in ['above', 'below']:
+            # first scale the two grids so they line up
+            self_xweights = self.xweights
+            other_xweights = other.xweights
+            self_w = sum(self_xweights)
+            other_w = sum(other_xweights)
+            ratio = self_w / other_w
+            other_xweights = [s*ratio for s in other_xweights]
+
+            # now interleave xweights as needed
+            new_xweights = []
+            self_map = {}
+            other_map = {}
+            self_i = 0
+            other_i = 0
+            self_xweight = (self_xweights[self_i]
+                if self_i < len(self_xweights) else m.inf)
+            other_xweight = (other_xweights[other_i]
+                if other_i < len(other_xweights) else m.inf)
+            while self_i < len(self_xweights) and other_i < len(other_xweights):
+                if other_xweight - self_xweight > 0.0000001:
+                    new_xweights.append(self_xweight)
+                    other_xweight -= self_xweight
+
+                    new_i = len(new_xweights)-1
+                    for j in range(len(self.yweights)):
+                        self_map[(new_i, j)] = self.map[(self_i, j)]
+                    for j in range(len(other.yweights)):
+                        other_map[(new_i, j)] = other.map[(other_i, j)]
+                    for s in other.subplots:
+                        if s.x+s.xspan-1 == new_i:
+                            s.xspan += 1
+                        elif s.x > new_i:
+                            s.x += 1
+
+                    self_i += 1
+                    self_xweight = (self_xweights[self_i]
+                        if self_i < len(self_xweights) else m.inf)
+                elif self_xweight - other_xweight > 0.0000001:
+                    new_xweights.append(other_xweight)
+                    self_xweight -= other_xweight
+
+                    new_i = len(new_xweights)-1
+                    for j in range(len(other.yweights)):
+                        other_map[(new_i, j)] = other.map[(other_i, j)]
+                    for j in range(len(self.yweights)):
+                        self_map[(new_i, j)] = self.map[(self_i, j)]
+                    for s in self.subplots:
+                        if s.x+s.xspan-1 == new_i:
+                            s.xspan += 1
+                        elif s.x > new_i:
+                            s.x += 1
+
+                    other_i += 1
+                    other_xweight = (other_xweights[other_i]
+                        if other_i < len(other_xweights) else m.inf)
+                else:
+                    new_xweights.append(self_xweight)
+
+                    new_i = len(new_xweights)-1
+                    for j in range(len(self.yweights)):
+                        self_map[(new_i, j)] = self.map[(self_i, j)]
+                    for j in range(len(other.yweights)):
+                        other_map[(new_i, j)] = other.map[(other_i, j)]
+
+                    self_i += 1
+                    self_xweight = (self_xweights[self_i]
+                        if self_i < len(self_xweights) else m.inf)
+                    other_i += 1
+                    other_xweight = (other_xweights[other_i]
+                        if other_i < len(other_xweights) else m.inf)
+
+            # squish so ratios are preserved
+            self_h = sum(self.yweights)
+            other_h = sum(other.yweights)
+            ratio = (self_h-other_h) / self_h
+            self_yweights = [s*ratio for s in self.yweights]
+
+            # finally concatenate the two grids
+            if dir == 'above':
+                for s in other.subplots:
+                    s.y += len(self_yweights)
+                self.subplots.extend(other.subplots)
+
+                self.xweights = new_xweights
+                self.yweights = self_yweights + other.yweights
+                self.map = self_map | {(x, y+len(self_yweights)): s
+                    for (x, y), s in other_map.items()}
+            else:
+                for s in self.subplots:
+                    s.y += len(other.yweights)
+                self.subplots.extend(other.subplots)
+
+                self.xweights = new_xweights
+                self.yweights = other.yweights + self_yweights
+                self.map = other_map | {(x, y+len(other.yweights)): s
+                    for (x, y), s in self_map.items()}
+
+        if dir in ['right', 'left']:
+            # first scale the two grids so they line up
+            self_yweights = self.yweights
+            other_yweights = other.yweights
+            self_h = sum(self_yweights)
+            other_h = sum(other_yweights)
+            ratio = self_h / other_h
+            other_yweights = [s*ratio for s in other_yweights]
+
+            # now interleave yweights as needed
+            new_yweights = []
+            self_map = {}
+            other_map = {}
+            self_i = 0
+            other_i = 0
+            self_yweight = (self_yweights[self_i]
+                if self_i < len(self_yweights) else m.inf)
+            other_yweight = (other_yweights[other_i]
+                if other_i < len(other_yweights) else m.inf)
+            while self_i < len(self_yweights) and other_i < len(other_yweights):
+                if other_yweight - self_yweight > 0.0000001:
+                    new_yweights.append(self_yweight)
+                    other_yweight -= self_yweight
+
+                    new_i = len(new_yweights)-1
+                    for j in range(len(self.xweights)):
+                        self_map[(j, new_i)] = self.map[(j, self_i)]
+                    for j in range(len(other.xweights)):
+                        other_map[(j, new_i)] = other.map[(j, other_i)]
+                    for s in other.subplots:
+                        if s.y+s.yspan-1 == new_i:
+                            s.yspan += 1
+                        elif s.y > new_i:
+                            s.y += 1
+
+                    self_i += 1
+                    self_yweight = (self_yweights[self_i]
+                        if self_i < len(self_yweights) else m.inf)
+                elif self_yweight - other_yweight > 0.0000001:
+                    new_yweights.append(other_yweight)
+                    self_yweight -= other_yweight
+
+                    new_i = len(new_yweights)-1
+                    for j in range(len(other.xweights)):
+                        other_map[(j, new_i)] = other.map[(j, other_i)]
+                    for j in range(len(self.xweights)):
+                        self_map[(j, new_i)] = self.map[(j, self_i)]
+                    for s in self.subplots:
+                        if s.y+s.yspan-1 == new_i:
+                            s.yspan += 1
+                        elif s.y > new_i:
+                            s.y += 1
+
+                    other_i += 1
+                    other_yweight = (other_yweights[other_i]
+                        if other_i < len(other_yweights) else m.inf)
+                else:
+                    new_yweights.append(self_yweight)
+
+                    new_i = len(new_yweights)-1
+                    for j in range(len(self.xweights)):
+                        self_map[(j, new_i)] = self.map[(j, self_i)]
+                    for j in range(len(other.xweights)):
+                        other_map[(j, new_i)] = other.map[(j, other_i)]
+
+                    self_i += 1
+                    self_yweight = (self_yweights[self_i]
+                        if self_i < len(self_yweights) else m.inf)
+                    other_i += 1
+                    other_yweight = (other_yweights[other_i]
+                        if other_i < len(other_yweights) else m.inf)
+
+            # squish so ratios are preserved
+            self_w = sum(self.xweights)
+            other_w = sum(other.xweights)
+            ratio = (self_w-other_w) / self_w
+            self_xweights = [s*ratio for s in self.xweights]
+
+            # finally concatenate the two grids
+            if dir == 'right':
+                for s in other.subplots:
+                    s.x += len(self_xweights)
+                self.subplots.extend(other.subplots)
+
+                self.xweights = self_xweights + other.xweights
+                self.yweights = new_yweights
+                self.map = self_map | {(x+len(self_xweights), y): s
+                    for (x, y), s in other_map.items()}
+            else:
+                for s in self.subplots:
+                    s.x += len(other.xweights)
+                self.subplots.extend(other.subplots)
+
+                self.xweights = other.xweights + self_xweights
+                self.yweights = new_yweights
+                self.map = other_map | {(x+len(other.xweights), y): s
+                    for (x, y), s in self_map.items()}
+                    
+
+    def scale(self, width, height):
+        self.xweights = [s*width for s in self.xweights]
+        self.yweights = [s*height for s in self.yweights]
+
+    @classmethod
+    def fromargs(cls, width=1.0, height=1.0, *,
+            subplots=[],
+            **args):
+        grid = cls(Subplot(**args))
+
+        for dir, subargs in subplots:
+            subgrid = cls.fromargs(
+                width=subargs.pop('width',
+                    0.5 if dir in ['right', 'left'] else width),
+                height=subargs.pop('height',
+                    0.5 if dir in ['above', 'below'] else height),
+                **subargs)
+            grid.merge(subgrid, dir)
+
+        grid.scale(width, height)
+        return grid
+
+
+def main(csv_paths, output, *,
+        svg=False,
+        png=False,
+        quiet=False,
+        by=None,
+        x=None,
+        y=None,
+        define=[],
+        points=False,
+        points_and_lines=False,
+        colors=None,
+        formats=None,
+        width=WIDTH,
+        height=HEIGHT,
+        xlim=(None,None),
+        ylim=(None,None),
+        xlog=False,
+        ylog=False,
+        x2=False,
+        y2=False,
+        xticks=None,
+        yticks=None,
+        xunits=None,
+        yunits=None,
+        xlabel=None,
+        ylabel=None,
+        xticklabels=None,
+        yticklabels=None,
+        title=None,
+        legend_right=False,
+        legend_above=False,
+        legend_below=False,
+        dark=False,
+        ggplot=False,
+        xkcd=False,
+        github=False,
+        font=None,
+        font_size=FONT_SIZE,
+        font_color=None,
+        foreground=None,
+        background=None,
+        subplot={},
+        subplots=[],
+        **args):
+    # guess the output format
+    if not png and not svg:
+        if output.endswith('.png'):
+            png = True
+        else:
+            svg = True
+
+    # some shortcuts for color schemes
+    if github:
+        ggplot = True
+        if font_color is None:
+            if dark:
+                font_color = '#c9d1d9'
+            else:
+                font_color = '#24292f'
+        if foreground is None:
+            if dark:
+                foreground = '#343942'
+            else:
+                foreground = '#eff1f3'
+        if background is None:
+            if dark:
+                background = '#0d1117'
+            else:
+                background = '#ffffff'
+
+    # what colors/alphas/formats to use?
+    if colors is not None:
+        colors_ = colors
+    elif dark:
+        colors_ = COLORS_DARK
+    else:
+        colors_ = COLORS
+
+    if formats is not None:
+        formats_ = formats
+    elif points_and_lines:
+        formats_ = FORMATS_POINTS_AND_LINES
+    elif points:
+        formats_ = FORMATS_POINTS
+    else:
+        formats_ = FORMATS
+
+    if font_color is not None:
+        font_color_ = font_color
+    elif dark:
+        font_color_ = '#ffffff'
+    else:
+        font_color_ = '#000000'
+
+    if foreground is not None:
+        foreground_ = foreground
+    elif dark:
+        foreground_ = '#333333'
+    else:
+        foreground_ = '#e5e5e5'
+
+    if background is not None:
+        background_ = background
+    elif dark:
+        background_ = '#000000'
+    else:
+        background_ = '#ffffff'
+
+    # configure some matplotlib settings
+    if xkcd:
+        # the font search here prints a bunch of unhelpful warnings
+        logging.getLogger('matplotlib.font_manager').setLevel(logging.ERROR)
+        plt.xkcd()
+        # turn off the white outline, this breaks some things
+        plt.rc('path', effects=[])
+    if ggplot:
+        plt.style.use('ggplot')
+        plt.rc('patch', linewidth=0)
+        plt.rc('axes', facecolor=foreground_, edgecolor=background_)
+        plt.rc('grid', color=background_)
+        # fix the the gridlines when ggplot+xkcd
+        if xkcd:
+            plt.rc('grid', linewidth=1)
+            plt.rc('axes.spines', bottom=False, left=False)
+    if dark:
+        plt.style.use('dark_background')
+        plt.rc('savefig', facecolor='auto', edgecolor='auto')
+        # fix ggplot when dark
+        if ggplot:
+            plt.rc('axes',
+                facecolor=foreground_,
+                edgecolor=background_)
+            plt.rc('grid', color=background_)
+
+    if font is not None:
+        plt.rc('font', family=font)
+    plt.rc('font', size=font_size)
+    plt.rc('text', color=font_color_)
+    plt.rc('figure',
+        titlesize='medium',
+        labelsize='small')
+    plt.rc('axes',
+        titlesize='small',
+        labelsize='small',
+        labelcolor=font_color_)
+    if not ggplot:
+        plt.rc('axes', edgecolor=font_color_)
+    plt.rc('xtick', labelsize='small', color=font_color_)
+    plt.rc('ytick', labelsize='small', color=font_color_)
+    plt.rc('legend',
+        fontsize='small',
+        fancybox=False,
+        framealpha=None,
+        edgecolor=foreground_,
+        borderaxespad=0)
+    plt.rc('axes.spines', top=False, right=False)
+
+    plt.rc('figure', facecolor=background_, edgecolor=background_)
+    if not ggplot:
+        plt.rc('axes', facecolor='#00000000')
+
+    # I think the svg backend just ignores DPI, but seems to use something
+    # equivalent to 96, maybe this is the default for SVG rendering?
+    plt.rc('figure', dpi=96)
+
+    # separate out renames
+    renames = list(it.chain.from_iterable(
+        ((k, v) for v in vs)
+        for k, vs in it.chain(by or [], x or [], y or [])))
+    if by is not None:
+        by = [k for k, _ in by]
+    if x is not None:
+        x = [k for k, _ in x]
+    if y is not None:
+        y = [k for k, _ in y]
+
+    # first collect results from CSV files
+    results = collect(csv_paths, renames)
+
+    # then extract the requested datasets
+    datasets_ = datasets(results, by, x, y, define)
+
+    # figure out formats/colors here so that subplot defines
+    # don't change them later, that'd be bad
+    dataformats_ = {
+        name: formats_[i % len(formats_)]
+        for i, name in enumerate(datasets_.keys())}
+    datacolors_ = {
+        name: colors_[i % len(colors_)]
+        for i, name in enumerate(datasets_.keys())}
+
+    # create a grid of subplots
+    grid = Grid.fromargs(
+        subplots=subplots + subplot.pop('subplots', []),
+        **subplot)
+
+    # create a matplotlib plot
+    fig = plt.figure(figsize=(
+        width/plt.rcParams['figure.dpi'],
+        height/plt.rcParams['figure.dpi']),
+        layout='constrained',
+        # we need a linewidth to keep xkcd mode happy
+        linewidth=8 if xkcd else 0)
+
+    gs = fig.add_gridspec(
+        grid.height
+            + (1 if legend_above else 0)
+            + (1 if legend_below else 0),
+        grid.width
+            + (1 if legend_right else 0),
+        height_ratios=([0.001] if legend_above else [])
+            + [max(s, 0.01) for s in reversed(grid.yweights)]
+            + ([0.001] if legend_below else []),
+        width_ratios=[max(s, 0.01) for s in grid.xweights]
+            + ([0.001] if legend_right else []))
+
+    # first create axes so that plots can interact with each other
+    for s in grid:
+        s.ax = fig.add_subplot(gs[
+            grid.height-(s.y+s.yspan) + (1 if legend_above else 0)
+                : grid.height-s.y + (1 if legend_above else 0),
+            s.x
+                : s.x+s.xspan])
+
+    # now plot each subplot
+    for s in grid:
+        # allow subplot params to override global params
+        define_ = define + s.args.get('define', [])
+        xlim_ = s.args.get('xlim', xlim)
+        ylim_ = s.args.get('ylim', ylim)
+        xlog_ = s.args.get('xlog', False) or xlog
+        ylog_ = s.args.get('ylog', False) or ylog
+        x2_ = s.args.get('x2', False) or x2
+        y2_ = s.args.get('y2', False) or y2
+        xticks_ = s.args.get('xticks', xticks)
+        yticks_ = s.args.get('yticks', yticks)
+        xunits_ = s.args.get('xunits', xunits)
+        yunits_ = s.args.get('yunits', yunits)
+        xticklabels_ = s.args.get('xticklabels', xticklabels)
+        yticklabels_ = s.args.get('yticklabels', yticklabels)
+
+        # label/titles are handled a bit differently in subplots
+        subtitle = s.args.get('title')
+        xsublabel = s.args.get('xlabel')
+        ysublabel = s.args.get('ylabel')
+
+        # allow shortened ranges
+        if len(xlim_) == 1:
+            xlim_ = (0, xlim_[0])
+        if len(ylim_) == 1:
+            ylim_ = (0, ylim_[0])
+
+        # data can be constrained by subplot-specific defines,
+        # so re-extract for each plot
+        subdatasets = datasets(results, by, x, y, define_)
+
+        # plot!
+        ax = s.ax
+        for name, dataset in subdatasets.items():
+            dats = sorted((x,y) for x,y in dataset.items())
+            ax.plot([x for x,_ in dats], [y for _,y in dats],
+                dataformats_[name],
+                color=datacolors_[name],
+                label=','.join(k for k in name if k))
+
+        # axes scaling
+        if xlog_:
+            ax.set_xscale('symlog')
+            ax.xaxis.set_minor_locator(mpl.ticker.NullLocator())
+        if ylog_:
+            ax.set_yscale('symlog')
+            ax.yaxis.set_minor_locator(mpl.ticker.NullLocator())
+        # axes limits
+        ax.set_xlim(
+            xlim_[0] if xlim_[0] is not None
+                else min(it.chain([0], (k
+                    for r in subdatasets.values()
+                    for k, v in r.items()
+                    if v is not None))),
+            xlim_[1] if xlim_[1] is not None
+                else max(it.chain([0], (k
+                    for r in subdatasets.values()
+                    for k, v in r.items()
+                    if v is not None))))
+        ax.set_ylim(
+            ylim_[0] if ylim_[0] is not None
+                else min(it.chain([0], (v
+                    for r in subdatasets.values()
+                    for _, v in r.items()
+                    if v is not None))),
+            ylim_[1] if ylim_[1] is not None
+                else max(it.chain([0], (v
+                    for r in subdatasets.values()
+                    for _, v in r.items()
+                    if v is not None))))
+        # axes ticks
+        if x2_:
+            ax.xaxis.set_major_formatter(lambda x, pos:
+                si2(x)+(xunits_ if xunits_ else ''))
+            if xticklabels_ is not None:
+                ax.xaxis.set_ticklabels(xticklabels_)
+            if xticks_ is None:
+                ax.xaxis.set_major_locator(AutoMultipleLocator(2))
+            elif isinstance(xticks_, list):
+                ax.xaxis.set_major_locator(mpl.ticker.FixedLocator(xticks_))
+            elif xticks_ != 0:
+                ax.xaxis.set_major_locator(AutoMultipleLocator(2, xticks_-1))
+            else:
+                ax.xaxis.set_major_locator(mpl.ticker.NullLocator())
+        else:
+            ax.xaxis.set_major_formatter(lambda x, pos:
+                si(x)+(xunits_ if xunits_ else ''))
+            if xticklabels_ is not None:
+                ax.xaxis.set_ticklabels(xticklabels_)
+            if xticks_ is None:
+                ax.xaxis.set_major_locator(mpl.ticker.AutoLocator())
+            elif isinstance(xticks_, list):
+                ax.xaxis.set_major_locator(mpl.ticker.FixedLocator(xticks_))
+            elif xticks_ != 0:
+                ax.xaxis.set_major_locator(mpl.ticker.MaxNLocator(xticks_-1))
+            else:
+                ax.xaxis.set_major_locator(mpl.ticker.NullLocator())
+        if y2_:
+            ax.yaxis.set_major_formatter(lambda x, pos:
+                si2(x)+(yunits_ if yunits_ else ''))
+            if yticklabels_ is not None:
+                ax.yaxis.set_ticklabels(yticklabels_)
+            if yticks_ is None:
+                ax.yaxis.set_major_locator(AutoMultipleLocator(2))
+            elif isinstance(yticks_, list):
+                ax.yaxis.set_major_locator(mpl.ticker.FixedLocator(yticks_))
+            elif yticks_ != 0:
+                ax.yaxis.set_major_locator(AutoMultipleLocator(2, yticks_-1))
+            else:
+                ax.yaxis.set_major_locator(mpl.ticker.NullLocator())
+        else:
+            ax.yaxis.set_major_formatter(lambda x, pos:
+                si(x)+(yunits_ if yunits_ else ''))
+            if yticklabels_ is not None:
+                ax.yaxis.set_ticklabels(yticklabels_)
+            if yticks_ is None:
+                ax.yaxis.set_major_locator(mpl.ticker.AutoLocator())
+            elif isinstance(yticks_, list):
+                ax.yaxis.set_major_locator(mpl.ticker.FixedLocator(yticks_))
+            elif yticks_ != 0:
+                ax.yaxis.set_major_locator(mpl.ticker.MaxNLocator(yticks_-1))
+            else:
+                ax.yaxis.set_major_locator(mpl.ticker.NullLocator())
+        if ggplot:
+            ax.grid(sketch_params=None)
+
+        # axes subplot labels
+        if xsublabel is not None:
+            ax.set_xlabel(escape(xsublabel))
+        if ysublabel is not None:
+            ax.set_ylabel(escape(ysublabel))
+        if subtitle is not None:
+            ax.set_title(escape(subtitle))
+
+    # add a legend? a bit tricky with matplotlib
+    #
+    # the best solution I've found is a dedicated, invisible axes for the
+    # legend, hacky, but it works.
+    #
+    # note this was written before constrained_layout supported legend
+    # collisions, hopefully this is added in the future
+    labels = co.OrderedDict()
+    for s in grid:
+        for h, l in zip(*s.ax.get_legend_handles_labels()):
+            labels[l] = h
+
+    if legend_right:
+        ax = fig.add_subplot(gs[(1 if legend_above else 0):,-1])
+        ax.set_axis_off()
+        ax.legend(
+            labels.values(),
+            labels.keys(),
+            loc='upper left',
+            fancybox=False,
+            borderaxespad=0)
+
+    if legend_above:
+        ax = fig.add_subplot(gs[0, :grid.width])
+        ax.set_axis_off()
+
+        # try different column counts until we fit in the axes
+        for ncol in reversed(range(1, len(labels)+1)):
+            legend_ = ax.legend(
+                labels.values(),
+                labels.keys(),
+                loc='upper center',
+                ncol=ncol,
+                fancybox=False,
+                borderaxespad=0)
+
+            if (legend_.get_window_extent().width
+                    <= ax.get_window_extent().width):
+                break
+
+    if legend_below:
+        ax = fig.add_subplot(gs[-1, :grid.width])
+        ax.set_axis_off()
+
+        # big hack to get xlabel above the legend! but hey this
+        # works really well actually
+        if xlabel:
+            ax.set_title(escape(xlabel),
+                size=plt.rcParams['axes.labelsize'],
+                weight=plt.rcParams['axes.labelweight'])
+
+        # try different column counts until we fit in the axes
+        for ncol in reversed(range(1, len(labels)+1)):
+            legend_ = ax.legend(
+                labels.values(),
+                labels.keys(),
+                loc='upper center',
+                ncol=ncol,
+                fancybox=False,
+                borderaxespad=0)
+
+            if (legend_.get_window_extent().width
+                    <= ax.get_window_extent().width):
+                break
+
+
+    # axes labels, NOTE we reposition these below
+    if xlabel is not None and not legend_below:
+        fig.supxlabel(escape(xlabel))
+    if ylabel is not None:
+        fig.supylabel(escape(ylabel))
+    if title is not None:
+        fig.suptitle(escape(title))
+
+    # precompute constrained layout and find midpoints to adjust things
+    # that should be centered so they are actually centered
+    fig.canvas.draw()
+    xmid = (grid[0,0].ax.get_position().x0 + grid[-1,0].ax.get_position().x1)/2
+    ymid = (grid[0,0].ax.get_position().y0 + grid[0,-1].ax.get_position().y1)/2
+
+    if xlabel is not None and not legend_below:
+        fig.supxlabel(escape(xlabel), x=xmid)
+    if ylabel is not None:
+        fig.supylabel(escape(ylabel), y=ymid)
+    if title is not None:
+        fig.suptitle(escape(title), x=xmid)
+
+
+    # write the figure!
+    plt.savefig(output, format='png' if png else 'svg')
+
+    # some stats
+    if not quiet:
+        print('updated %s, %s datasets, %s points' % (
+            output,
+            len(datasets_),
+            sum(len(dataset) for dataset in datasets_.values())))
+
+
+if __name__ == "__main__":
+    import sys
+    import argparse
+    parser = argparse.ArgumentParser(
+        description="Plot CSV files with matplotlib.",
+        allow_abbrev=False)
+    parser.add_argument(
+        'csv_paths',
+        nargs='*',
+        help="Input *.csv files.")
+    output_rule = parser.add_argument(
+        '-o', '--output',
+        required=True,
+        help="Output *.svg/*.png file.")
+    parser.add_argument(
+        '--svg',
+        action='store_true',
+        help="Output an svg file. By default this is infered.")
+    parser.add_argument(
+        '--png',
+        action='store_true',
+        help="Output a png file. By default this is infered.")
+    parser.add_argument(
+        '-q', '--quiet',
+        action='store_true',
+        help="Don't print info.")
+    parser.add_argument(
+        '-b', '--by',
+        action='append',
+        type=lambda x: (
+            lambda k,v=None: (k, v.split(',') if v is not None else ())
+            )(*x.split('=', 1)),
+        help="Group by this field. Can rename fields with new_name=old_name.")
+    parser.add_argument(
+        '-x',
+        action='append',
+        type=lambda x: (
+            lambda k,v=None: (k, v.split(',') if v is not None else ())
+            )(*x.split('=', 1)),
+        help="Field to use for the x-axis. Can rename fields with "
+            "new_name=old_name.")
+    parser.add_argument(
+        '-y',
+        action='append',
+        type=lambda x: (
+            lambda k,v=None: (k, v.split(',') if v is not None else ())
+            )(*x.split('=', 1)),
+        help="Field to use for the y-axis. Can rename fields with "
+            "new_name=old_name.")
+    parser.add_argument(
+        '-D', '--define',
+        type=lambda x: (lambda k,v: (k, set(v.split(','))))(*x.split('=', 1)),
+        action='append',
+        help="Only include results where this field is this value. May include "
+            "comma-separated options.")
+    parser.add_argument(
+        '-.', '--points',
+        action='store_true',
+        help="Only draw data points.")
+    parser.add_argument(
+        '-!', '--points-and-lines',
+        action='store_true',
+        help="Draw data points and lines.")
+    parser.add_argument(
+        '--colors',
+        type=lambda x: [x.strip() for x in x.split(',')],
+        help="Comma-separated hex colors to use.")
+    parser.add_argument(
+        '--formats',
+        type=lambda x: [x.strip().replace('0',',') for x in x.split(',')],
+        help="Comma-separated matplotlib formats to use. Allows '0' as an "
+            "alternative for ','.")
+    parser.add_argument(
+        '-W', '--width',
+        type=lambda x: int(x, 0),
+        help="Width in pixels. Defaults to %r." % WIDTH)
+    parser.add_argument(
+        '-H', '--height',
+        type=lambda x: int(x, 0),
+        help="Height in pixels. Defaults to %r." % HEIGHT)
+    parser.add_argument(
+        '-X', '--xlim',
+        type=lambda x: tuple(
+            dat(x) if x.strip() else None
+            for x in x.split(',')),
+        help="Range for the x-axis.")
+    parser.add_argument(
+        '-Y', '--ylim',
+        type=lambda x: tuple(
+            dat(x) if x.strip() else None
+            for x in x.split(',')),
+        help="Range for the y-axis.")
+    parser.add_argument(
+        '--xlog',
+        action='store_true',
+        help="Use a logarithmic x-axis.")
+    parser.add_argument(
+        '--ylog',
+        action='store_true',
+        help="Use a logarithmic y-axis.")
+    parser.add_argument(
+        '--x2',
+        action='store_true',
+        help="Use base-2 prefixes for the x-axis.")
+    parser.add_argument(
+        '--y2',
+        action='store_true',
+        help="Use base-2 prefixes for the y-axis.")
+    parser.add_argument(
+        '--xticks',
+        type=lambda x: int(x, 0) if ',' not in x
+            else [dat(x) for x in x.split(',')],
+        help="Ticks for the x-axis. This can be explicit comma-separated "
+            "ticks, the number of ticks, or 0 to disable.")
+    parser.add_argument(
+        '--yticks',
+        type=lambda x: int(x, 0) if ',' not in x
+            else [dat(x) for x in x.split(',')],
+        help="Ticks for the y-axis. This can be explicit comma-separated "
+            "ticks, the number of ticks, or 0 to disable.")
+    parser.add_argument(
+        '--xunits',
+        help="Units for the x-axis.")
+    parser.add_argument(
+        '--yunits',
+        help="Units for the y-axis.")
+    parser.add_argument(
+        '--xlabel',
+        help="Add a label to the x-axis.")
+    parser.add_argument(
+        '--ylabel',
+        help="Add a label to the y-axis.")
+    parser.add_argument(
+        '--xticklabels',
+        type=lambda x:
+            [x.strip() for x in x.split(',')]
+            if x.strip() else [],
+        help="Comma separated xticklabels.")
+    parser.add_argument(
+        '--yticklabels',
+        type=lambda x:
+            [x.strip() for x in x.split(',')]
+            if x.strip() else [],
+        help="Comma separated yticklabels.")
+    parser.add_argument(
+        '-t', '--title',
+        help="Add a title.")
+    parser.add_argument(
+        '-l', '--legend-right',
+        action='store_true',
+        help="Place a legend to the right.")
+    parser.add_argument(
+        '--legend-above',
+        action='store_true',
+        help="Place a legend above.")
+    parser.add_argument(
+        '--legend-below',
+        action='store_true',
+        help="Place a legend below.")
+    parser.add_argument(
+        '--dark',
+        action='store_true',
+        help="Use the dark style.")
+    parser.add_argument(
+        '--ggplot',
+        action='store_true',
+        help="Use the ggplot style.")
+    parser.add_argument(
+        '--xkcd',
+        action='store_true',
+        help="Use the xkcd style.")
+    parser.add_argument(
+        '--github',
+        action='store_true',
+        help="Use the ggplot style with GitHub colors.")
+    parser.add_argument(
+        '--font',
+        type=lambda x: [x.strip() for x in x.split(',')],
+        help="Font family for matplotlib.")
+    parser.add_argument(
+        '--font-size',
+        help="Font size for matplotlib. Defaults to %r." % FONT_SIZE)
+    parser.add_argument(
+        '--font-color',
+        help="Color for the font and other line elements.")
+    parser.add_argument(
+        '--foreground',
+        help="Foreground color to use.")
+    parser.add_argument(
+        '--background',
+        help="Background color to use.")
+    class AppendSubplot(argparse.Action):
+        @staticmethod
+        def parse(value):
+            import copy
+            subparser = copy.deepcopy(parser)
+            next(a for a in subparser._actions
+                if '--output' in a.option_strings).required = False
+            next(a for a in subparser._actions
+                if '--width' in a.option_strings).type = float
+            next(a for a in subparser._actions
+                if '--height' in a.option_strings).type = float
+            return subparser.parse_intermixed_args(shlex.split(value or ""))
+        def __call__(self, parser, namespace, value, option):
+            if not hasattr(namespace, 'subplots'):
+                namespace.subplots = []
+            namespace.subplots.append((
+                option.split('-')[-1],
+                self.__class__.parse(value)))
+    parser.add_argument(
+        '--subplot-above',
+        action=AppendSubplot,
+        help="Add subplot above with the same dataset. Takes an arg string to "
+            "control the subplot which supports most (but not all) of the "
+            "parameters listed here. The relative dimensions of the subplot "
+            "can be controlled with -W/-H which now take a percentage.")
+    parser.add_argument(
+        '--subplot-below',
+        action=AppendSubplot,
+        help="Add subplot below with the same dataset.")
+    parser.add_argument(
+        '--subplot-left',
+        action=AppendSubplot,
+        help="Add subplot left with the same dataset.")
+    parser.add_argument(
+        '--subplot-right',
+        action=AppendSubplot,
+        help="Add subplot right with the same dataset.")
+    parser.add_argument(
+        '--subplot',
+        type=AppendSubplot.parse,
+        help="Add subplot-specific arguments to the main plot.")
+
+    def dictify(ns):
+        if hasattr(ns, 'subplots'):
+            ns.subplots = [(dir, dictify(subplot_ns))
+                for dir, subplot_ns in ns.subplots]
+        if ns.subplot is not None:
+            ns.subplot = dictify(ns.subplot)
+        return {k: v
+            for k, v in vars(ns).items()
+            if v is not None}
+
+    sys.exit(main(**dictify(parser.parse_intermixed_args())))
diff --git a/scripts/prettyasserts.py b/scripts/prettyasserts.py
new file mode 100644
index 0000000..3a62d36
--- /dev/null
+++ b/scripts/prettyasserts.py
@@ -0,0 +1,452 @@
+#!/usr/bin/env python3
+#
+# Preprocessor that makes asserts easier to debug.
+#
+# Example:
+# ./scripts/prettyasserts.py -p LFS_ASSERT lfs.c -o lfs.a.c
+#
+# Copyright (c) 2022, The littlefs authors.
+# Copyright (c) 2020, Arm Limited. All rights reserved.
+# SPDX-License-Identifier: BSD-3-Clause
+#
+
+import re
+import sys
+
+# NOTE the use of macros here helps keep a consistent stack depth which
+# tools may rely on.
+#
+# If compilation errors are noisy consider using -ftrack-macro-expansion=0.
+#
+
+LIMIT = 16
+
+CMP = {
+    '==': 'eq',
+    '!=': 'ne',
+    '<=': 'le',
+    '>=': 'ge',
+    '<':  'lt',
+    '>':  'gt',
+}
+
+LEXEMES = {
+    'ws':       [r'(?:\s|\n|#.*?\n|//.*?\n|/\*.*?\*/)+'],
+    'assert':   ['assert'],
+    'arrow':    ['=>'],
+    'string':   [r'"(?:\\.|[^"])*"', r"'(?:\\.|[^'])\'"],
+    'paren':    ['\(', '\)'],
+    'cmp':      CMP.keys(),
+    'logic':    ['\&\&', '\|\|'],
+    'sep':      [':', ';', '\{', '\}', ','],
+    'op':       ['->'], # specifically ops that conflict with cmp
+}
+
+
+def openio(path, mode='r', buffering=-1):
+    # allow '-' for stdin/stdout
+    if path == '-':
+        if mode == 'r':
+            return os.fdopen(os.dup(sys.stdin.fileno()), mode, buffering)
+        else:
+            return os.fdopen(os.dup(sys.stdout.fileno()), mode, buffering)
+    else:
+        return open(path, mode, buffering)
+
+def write_header(f, limit=LIMIT):
+    f.writeln("// Generated by %s:" % sys.argv[0])
+    f.writeln("//")
+    f.writeln("// %s" % ' '.join(sys.argv))
+    f.writeln("//")
+    f.writeln()
+
+    f.writeln("#include <stdbool.h>")
+    f.writeln("#include <stdint.h>")
+    f.writeln("#include <inttypes.h>")
+    f.writeln("#include <stdio.h>")
+    f.writeln("#include <string.h>")
+    f.writeln("#include <signal.h>")
+    # give source a chance to define feature macros
+    f.writeln("#undef _FEATURES_H")
+    f.writeln()
+
+    # write print macros
+    f.writeln("__attribute__((unused))")
+    f.writeln("static void __pretty_assert_print_bool(")
+    f.writeln("        const void *v, size_t size) {")
+    f.writeln("    (void)size;")
+    f.writeln("    printf(\"%s\", *(const bool*)v ? \"true\" : \"false\");")
+    f.writeln("}")
+    f.writeln()
+    f.writeln("__attribute__((unused))")
+    f.writeln("static void __pretty_assert_print_int(")
+    f.writeln("        const void *v, size_t size) {")
+    f.writeln("    (void)size;")
+    f.writeln("    printf(\"%\"PRIiMAX, *(const intmax_t*)v);")
+    f.writeln("}")
+    f.writeln()
+    f.writeln("__attribute__((unused))")
+    f.writeln("static void __pretty_assert_print_mem(")
+    f.writeln("        const void *v, size_t size) {")
+    f.writeln("    const uint8_t *v_ = v;")
+    f.writeln("    printf(\"\\\"\");")
+    f.writeln("    for (size_t i = 0; i < size && i < %d; i++) {" % limit)
+    f.writeln("        if (v_[i] >= ' ' && v_[i] <= '~') {")
+    f.writeln("            printf(\"%c\", v_[i]);")
+    f.writeln("        } else {")
+    f.writeln("            printf(\"\\\\x%02x\", v_[i]);")
+    f.writeln("        }")
+    f.writeln("    }")
+    f.writeln("    if (size > %d) {" % limit)
+    f.writeln("        printf(\"...\");")
+    f.writeln("    }")
+    f.writeln("    printf(\"\\\"\");")
+    f.writeln("}")
+    f.writeln()
+    f.writeln("__attribute__((unused))")
+    f.writeln("static void __pretty_assert_print_str(")
+    f.writeln("        const void *v, size_t size) {")
+    f.writeln("    __pretty_assert_print_mem(v, size);")
+    f.writeln("}")
+    f.writeln()
+    f.writeln("__attribute__((unused, noinline))")
+    f.writeln("static void __pretty_assert_fail(")
+    f.writeln("        const char *file, int line,")
+    f.writeln("        void (*type_print_cb)(const void*, size_t),")
+    f.writeln("        const char *cmp,")
+    f.writeln("        const void *lh, size_t lsize,")
+    f.writeln("        const void *rh, size_t rsize) {")
+    f.writeln("    printf(\"%s:%d:assert: assert failed with \", file, line);")
+    f.writeln("    type_print_cb(lh, lsize);")
+    f.writeln("    printf(\", expected %s \", cmp);")
+    f.writeln("    type_print_cb(rh, rsize);")
+    f.writeln("    printf(\"\\n\");")
+    f.writeln("    fflush(NULL);")
+    f.writeln("    raise(SIGABRT);")
+    f.writeln("}")
+    f.writeln()
+
+    # write assert macros
+    for op, cmp in sorted(CMP.items()):
+        f.writeln("#define __PRETTY_ASSERT_BOOL_%s(lh, rh) do { \\"
+            % cmp.upper())
+        f.writeln("    bool _lh = !!(lh); \\")
+        f.writeln("    bool _rh = !!(rh); \\")
+        f.writeln("    if (!(_lh %s _rh)) { \\" % op)
+        f.writeln("        __pretty_assert_fail( \\")
+        f.writeln("                __FILE__, __LINE__, \\")
+        f.writeln("                __pretty_assert_print_bool, \"%s\", \\"
+            % cmp)
+        f.writeln("                &_lh, 0, \\")
+        f.writeln("                &_rh, 0); \\")
+        f.writeln("    } \\")
+        f.writeln("} while (0)")
+    for op, cmp in sorted(CMP.items()):
+        f.writeln("#define __PRETTY_ASSERT_INT_%s(lh, rh) do { \\"
+            % cmp.upper())
+        f.writeln("    __typeof__(lh) _lh = lh; \\")
+        f.writeln("    __typeof__(lh) _rh = rh; \\")
+        f.writeln("    if (!(_lh %s _rh)) { \\" % op)
+        f.writeln("        __pretty_assert_fail( \\")
+        f.writeln("                __FILE__, __LINE__, \\")
+        f.writeln("                __pretty_assert_print_int, \"%s\", \\"
+            % cmp)
+        f.writeln("                &(intmax_t){_lh}, 0, \\")
+        f.writeln("                &(intmax_t){_rh}, 0); \\")
+        f.writeln("    } \\")
+        f.writeln("} while (0)")
+    for op, cmp in sorted(CMP.items()):
+        f.writeln("#define __PRETTY_ASSERT_MEM_%s(lh, rh, size) do { \\"
+            % cmp.upper())
+        f.writeln("    const void *_lh = lh; \\")
+        f.writeln("    const void *_rh = rh; \\")
+        f.writeln("    if (!(memcmp(_lh, _rh, size) %s 0)) { \\" % op)
+        f.writeln("        __pretty_assert_fail( \\")
+        f.writeln("                __FILE__, __LINE__, \\")
+        f.writeln("                __pretty_assert_print_mem, \"%s\", \\"
+            % cmp)
+        f.writeln("                _lh, size, \\")
+        f.writeln("                _rh, size); \\")
+        f.writeln("    } \\")
+        f.writeln("} while (0)")
+    for op, cmp in sorted(CMP.items()):
+        f.writeln("#define __PRETTY_ASSERT_STR_%s(lh, rh) do { \\"
+            % cmp.upper())
+        f.writeln("    const char *_lh = lh; \\")
+        f.writeln("    const char *_rh = rh; \\")
+        f.writeln("    if (!(strcmp(_lh, _rh) %s 0)) { \\" % op)
+        f.writeln("        __pretty_assert_fail( \\")
+        f.writeln("                __FILE__, __LINE__, \\")
+        f.writeln("                __pretty_assert_print_str, \"%s\", \\"
+            % cmp)
+        f.writeln("                _lh, strlen(_lh), \\")
+        f.writeln("                _rh, strlen(_rh)); \\")
+        f.writeln("    } \\")
+        f.writeln("} while (0)")
+    f.writeln()
+    f.writeln()
+
+def mkassert(type, cmp, lh, rh, size=None):
+    if size is not None:
+        return ("__PRETTY_ASSERT_%s_%s(%s, %s, %s)"
+            % (type.upper(), cmp.upper(), lh, rh, size))
+    else:
+        return ("__PRETTY_ASSERT_%s_%s(%s, %s)"
+            % (type.upper(), cmp.upper(), lh, rh))
+
+
+# simple recursive descent parser
+class ParseFailure(Exception):
+    def __init__(self, expected, found):
+        self.expected = expected
+        self.found = found
+
+    def __str__(self):
+        return "expected %r, found %s..." % (
+            self.expected, repr(self.found)[:70])
+
+class Parser:
+    def __init__(self, in_f, lexemes=LEXEMES):
+        p = '|'.join('(?P<%s>%s)' % (n, '|'.join(l))
+            for n, l in lexemes.items())
+        p = re.compile(p, re.DOTALL)
+        data = in_f.read()
+        tokens = []
+        line = 1
+        col = 0
+        while True:
+            m = p.search(data)
+            if m:
+                if m.start() > 0:
+                    tokens.append((None, data[:m.start()], line, col))
+                tokens.append((m.lastgroup, m.group(), line, col))
+                data = data[m.end():]
+            else:
+                tokens.append((None, data, line, col))
+                break
+        self.tokens = tokens
+        self.off = 0
+
+    def lookahead(self, *pattern):
+        if self.off < len(self.tokens):
+            token = self.tokens[self.off]
+            if token[0] in pattern or token[1] in pattern:
+                self.m = token[1]
+                return self.m
+        self.m = None
+        return self.m
+
+    def accept(self, *patterns):
+        m = self.lookahead(*patterns)
+        if m is not None:
+            self.off += 1
+        return m
+
+    def expect(self, *patterns):
+        m = self.accept(*patterns)
+        if not m:
+            raise ParseFailure(patterns, self.tokens[self.off:])
+        return m
+
+    def push(self):
+        return self.off
+
+    def pop(self, state):
+        self.off = state
+
+def p_assert(p):
+    state = p.push()
+
+    # assert(memcmp(a,b,size) cmp 0)?
+    try:
+        p.expect('assert') ; p.accept('ws')
+        p.expect('(') ; p.accept('ws')
+        p.expect('memcmp') ; p.accept('ws')
+        p.expect('(') ; p.accept('ws')
+        lh = p_expr(p) ; p.accept('ws')
+        p.expect(',') ; p.accept('ws')
+        rh = p_expr(p) ; p.accept('ws')
+        p.expect(',') ; p.accept('ws')
+        size = p_expr(p) ; p.accept('ws')
+        p.expect(')') ; p.accept('ws')
+        cmp = p.expect('cmp') ; p.accept('ws')
+        p.expect('0') ; p.accept('ws')
+        p.expect(')')
+        return mkassert('mem', CMP[cmp], lh, rh, size)
+    except ParseFailure:
+        p.pop(state)
+
+    # assert(strcmp(a,b) cmp 0)?
+    try:
+        p.expect('assert') ; p.accept('ws')
+        p.expect('(') ; p.accept('ws')
+        p.expect('strcmp') ; p.accept('ws')
+        p.expect('(') ; p.accept('ws')
+        lh = p_expr(p) ; p.accept('ws')
+        p.expect(',') ; p.accept('ws')
+        rh = p_expr(p) ; p.accept('ws')
+        p.expect(')') ; p.accept('ws')
+        cmp = p.expect('cmp') ; p.accept('ws')
+        p.expect('0') ; p.accept('ws')
+        p.expect(')')
+        return mkassert('str', CMP[cmp], lh, rh)
+    except ParseFailure:
+        p.pop(state)
+
+    # assert(a cmp b)?
+    try:
+        p.expect('assert') ; p.accept('ws')
+        p.expect('(') ; p.accept('ws')
+        lh = p_expr(p) ; p.accept('ws')
+        cmp = p.expect('cmp') ; p.accept('ws')
+        rh = p_expr(p) ; p.accept('ws')
+        p.expect(')')
+        return mkassert('int', CMP[cmp], lh, rh)
+    except ParseFailure:
+        p.pop(state)
+
+    # assert(a)?
+    p.expect('assert') ; p.accept('ws')
+    p.expect('(') ; p.accept('ws')
+    lh = p_exprs(p) ; p.accept('ws')
+    p.expect(')')
+    return mkassert('bool', 'eq', lh, 'true')
+
+def p_expr(p):
+    res = []
+    while True:
+        if p.accept('('):
+            res.append(p.m)
+            while True:
+                res.append(p_exprs(p))
+                if p.accept('sep'):
+                    res.append(p.m)
+                else:
+                    break
+            res.append(p.expect(')'))
+        elif p.lookahead('assert'):
+            state = p.push()
+            try:
+                res.append(p_assert(p))
+            except ParseFailure:
+                p.pop(state)
+                res.append(p.expect('assert'))
+        elif p.accept('string', 'op', 'ws', None):
+            res.append(p.m)
+        else:
+            return ''.join(res)
+
+def p_exprs(p):
+    res = []
+    while True:
+        res.append(p_expr(p))
+        if p.accept('cmp', 'logic', ','):
+            res.append(p.m)
+        else:
+            return ''.join(res)
+
+def p_stmt(p):
+    ws = p.accept('ws') or ''
+
+    # memcmp(lh,rh,size) => 0?
+    if p.lookahead('memcmp'):
+        state = p.push()
+        try:
+            p.expect('memcmp') ; p.accept('ws')
+            p.expect('(') ; p.accept('ws')
+            lh = p_expr(p) ; p.accept('ws')
+            p.expect(',') ; p.accept('ws')
+            rh = p_expr(p) ; p.accept('ws')
+            p.expect(',') ; p.accept('ws')
+            size = p_expr(p) ; p.accept('ws')
+            p.expect(')') ; p.accept('ws')
+            p.expect('=>') ; p.accept('ws')
+            p.expect('0') ; p.accept('ws')
+            return ws + mkassert('mem', 'eq', lh, rh, size)
+        except ParseFailure:
+            p.pop(state)
+
+    # strcmp(lh,rh) => 0?
+    if p.lookahead('strcmp'):
+        state = p.push()
+        try:
+            p.expect('strcmp') ; p.accept('ws') ; p.expect('(') ; p.accept('ws')
+            lh = p_expr(p) ; p.accept('ws')
+            p.expect(',') ; p.accept('ws')
+            rh = p_expr(p) ; p.accept('ws')
+            p.expect(')') ; p.accept('ws')
+            p.expect('=>') ; p.accept('ws')
+            p.expect('0') ; p.accept('ws')
+            return ws + mkassert('str', 'eq', lh, rh)
+        except ParseFailure:
+            p.pop(state)
+
+    # lh => rh?
+    lh = p_exprs(p)
+    if p.accept('=>'):
+        rh = p_exprs(p)
+        return ws + mkassert('int', 'eq', lh, rh)
+    else:
+        return ws + lh
+
+def main(input=None, output=None, pattern=[], limit=LIMIT):
+    with openio(input or '-', 'r') as in_f:
+        # create parser
+        lexemes = LEXEMES.copy()
+        lexemes['assert'] += pattern
+        p = Parser(in_f, lexemes)
+
+        with openio(output or '-', 'w') as f:
+            def writeln(s=''):
+                f.write(s)
+                f.write('\n')
+            f.writeln = writeln
+
+            # write extra verbose asserts
+            write_header(f, limit=limit)
+            if input is not None:
+                f.writeln("#line %d \"%s\"" % (1, input))
+
+            # parse and write out stmt at a time
+            try:
+                while True:
+                    f.write(p_stmt(p))
+                    if p.accept('sep'):
+                        f.write(p.m)
+                    else:
+                        break
+            except ParseFailure as e:
+                print('warning: %s' % e)
+                pass
+
+            for i in range(p.off, len(p.tokens)):
+                f.write(p.tokens[i][1])
+
+
+if __name__ == "__main__":
+    import argparse
+    import sys
+    parser = argparse.ArgumentParser(
+        description="Preprocessor that makes asserts easier to debug.",
+        allow_abbrev=False)
+    parser.add_argument(
+        'input',
+        help="Input C file.")
+    parser.add_argument(
+        '-o', '--output',
+        required=True,
+        help="Output C file.")
+    parser.add_argument(
+        '-p', '--pattern',
+        action='append',
+        help="Regex patterns to search for starting an assert statement. This"
+            " implicitly includes \"assert\" and \"=>\".")
+    parser.add_argument(
+        '-l', '--limit',
+        type=lambda x: int(x, 0),
+        default=LIMIT,
+        help="Maximum number of characters to display in strcmp and memcmp. "
+            "Defaults to %r." % LIMIT)
+    sys.exit(main(**{k: v
+        for k, v in vars(parser.parse_intermixed_args()).items()
+        if v is not None}))
diff --git a/scripts/readmdir.py b/scripts/readmdir.py
index b6c3dcc..98816df 100755
--- a/scripts/readmdir.py
+++ b/scripts/readmdir.py
@@ -24,6 +24,8 @@ TAG_TYPES = {
     'gstate':       (0x700, 0x700),
     'movestate':    (0x7ff, 0x7ff),
     'crc':          (0x700, 0x500),
+    'ccrc':         (0x780, 0x500),
+    'fcrc':         (0x7ff, 0x5ff),
 }
 
 class Tag:
@@ -99,7 +101,16 @@ class Tag:
         return struct.unpack('b', struct.pack('B', self.chunk))[0]
 
     def is_(self, type):
-        return (self.type & TAG_TYPES[type][0]) == TAG_TYPES[type][1]
+        try:
+            if ' ' in type:
+                type1, type3 = type.split()
+                return (self.is_(type1) and
+                    (self.type & ~TAG_TYPES[type1][0]) == int(type3, 0))
+
+            return self.type == int(type, 0)
+
+        except (ValueError, KeyError):
+            return (self.type & TAG_TYPES[type][0]) == TAG_TYPES[type][1]
 
     def mkmask(self):
         return Tag(
@@ -109,14 +120,20 @@ class Tag:
 
     def chid(self, nid):
         ntag = Tag(self.type, nid, self.size)
-        if hasattr(self, 'off'):  ntag.off  = self.off
-        if hasattr(self, 'data'): ntag.data = self.data
-        if hasattr(self, 'crc'):  ntag.crc  = self.crc
+        if hasattr(self, 'off'):    ntag.off    = self.off
+        if hasattr(self, 'data'):   ntag.data   = self.data
+        if hasattr(self, 'ccrc'):   ntag.crc    = self.crc
+        if hasattr(self, 'erased'): ntag.erased = self.erased
         return ntag
 
     def typerepr(self):
-        if self.is_('crc') and getattr(self, 'crc', 0xffffffff) != 0xffffffff:
-            return 'crc (bad)'
+        if (self.is_('ccrc')
+                and getattr(self, 'ccrc', 0xffffffff) != 0xffffffff):
+            crc_status = ' (bad)'
+        elif self.is_('fcrc') and getattr(self, 'erased', False):
+            crc_status = ' (era)'
+        else:
+            crc_status = ''
 
         reverse_types = {v: k for k, v in TAG_TYPES.items()}
         for prefix in range(12):
@@ -124,12 +141,12 @@ class Tag:
             if (mask, self.type & mask) in reverse_types:
                 type = reverse_types[mask, self.type & mask]
                 if prefix > 0:
-                    return '%s %#0*x' % (
-                        type, prefix//4, self.type & ((1 << prefix)-1))
+                    return '%s %#x%s' % (
+                        type, self.type & ((1 << prefix)-1), crc_status)
                 else:
-                    return type
+                    return '%s%s' % (type, crc_status)
         else:
-            return '%02x' % self.type
+            return '%02x%s' % (self.type, crc_status)
 
     def idrepr(self):
         return repr(self.id) if self.id != 0x3ff else '.'
@@ -172,6 +189,8 @@ class MetadataPair:
 
         self.rev, = struct.unpack('<I', block[0:4])
         crc = binascii.crc32(block[0:4])
+        fcrctag = None
+        fcrcdata = None
 
         # parse tags
         corrupt = False
@@ -182,11 +201,11 @@ class MetadataPair:
         while len(block) - off >= 4:
             ntag, = struct.unpack('>I', block[off:off+4])
 
-            tag = Tag(int(tag) ^ ntag)
+            tag = Tag((int(tag) ^ ntag) & 0x7fffffff)
             tag.off = off + 4
             tag.data = block[off+4:off+tag.dsize]
-            if tag.is_('crc'):
-                crc = binascii.crc32(block[off:off+4+4], crc)
+            if tag.is_('ccrc'):
+                crc = binascii.crc32(block[off:off+2*4], crc)
             else:
                 crc = binascii.crc32(block[off:off+tag.dsize], crc)
             tag.crc = crc
@@ -194,16 +213,29 @@ class MetadataPair:
 
             self.all_.append(tag)
 
-            if tag.is_('crc'):
+            if tag.is_('fcrc') and len(tag.data) == 8:
+                fcrctag = tag
+                fcrcdata = struct.unpack('<II', tag.data)
+            elif tag.is_('ccrc'):
                 # is valid commit?
                 if crc != 0xffffffff:
                     corrupt = True
                 if not corrupt:
                     self.log = self.all_.copy()
+                    # end of commit?
+                    if fcrcdata:
+                        fcrcsize, fcrc = fcrcdata
+                        fcrc_ = 0xffffffff ^ binascii.crc32(
+                            block[off:off+fcrcsize])
+                        if fcrc_ == fcrc:
+                            fcrctag.erased = True
+                            corrupt = True
 
                 # reset tag parsing
                 crc = 0
                 tag = Tag(int(tag) ^ ((tag.type & 1) << 31))
+                fcrctag = None
+                fcrcdata = None
 
         # find active ids
         self.ids = list(it.takewhile(
@@ -280,7 +312,7 @@ class MetadataPair:
         f.write('\n')
 
         for tag in tags:
-            f.write("%08x: %08x  %-13s %4s %4s" % (
+            f.write("%08x: %08x  %-14s %3s %4s" % (
                 tag.off, tag,
                 tag.typerepr(), tag.idrepr(), tag.sizerepr()))
             if truncate:
diff --git a/scripts/stack.py b/scripts/stack.py
index 0c652d8..ba95433 100644
--- a/scripts/stack.py
+++ b/scripts/stack.py
@@ -3,19 +3,136 @@
 # Script to find stack usage at the function level. Will detect recursion and
 # report as infinite stack usage.
 #
+# Example:
+# ./scripts/stack.py lfs.ci lfs_util.ci -Slimit
+#
+# Copyright (c) 2022, The littlefs authors.
+# SPDX-License-Identifier: BSD-3-Clause
+#
 
-import os
-import glob
-import itertools as it
-import re
-import csv
 import collections as co
+import csv
+import itertools as it
 import math as m
+import os
+import re
 
 
-CI_PATHS = ['*.ci']
 
-def collect(paths, **args):
+# integer fields
+class Int(co.namedtuple('Int', 'x')):
+    __slots__ = ()
+    def __new__(cls, x=0):
+        if isinstance(x, Int):
+            return x
+        if isinstance(x, str):
+            try:
+                x = int(x, 0)
+            except ValueError:
+                # also accept +-∞ and +-inf
+                if re.match('^\s*\+?\s*(?:∞|inf)\s*$', x):
+                    x = m.inf
+                elif re.match('^\s*-\s*(?:∞|inf)\s*$', x):
+                    x = -m.inf
+                else:
+                    raise
+        assert isinstance(x, int) or m.isinf(x), x
+        return super().__new__(cls, x)
+
+    def __str__(self):
+        if self.x == m.inf:
+            return '∞'
+        elif self.x == -m.inf:
+            return '-∞'
+        else:
+            return str(self.x)
+
+    def __int__(self):
+        assert not m.isinf(self.x)
+        return self.x
+
+    def __float__(self):
+        return float(self.x)
+
+    none = '%7s' % '-'
+    def table(self):
+        return '%7s' % (self,)
+
+    diff_none = '%7s' % '-'
+    diff_table = table
+
+    def diff_diff(self, other):
+        new = self.x if self else 0
+        old = other.x if other else 0
+        diff = new - old
+        if diff == +m.inf:
+            return '%7s' % '+∞'
+        elif diff == -m.inf:
+            return '%7s' % '-∞'
+        else:
+            return '%+7d' % diff
+
+    def ratio(self, other):
+        new = self.x if self else 0
+        old = other.x if other else 0
+        if m.isinf(new) and m.isinf(old):
+            return 0.0
+        elif m.isinf(new):
+            return +m.inf
+        elif m.isinf(old):
+            return -m.inf
+        elif not old and not new:
+            return 0.0
+        elif not old:
+            return 1.0
+        else:
+            return (new-old) / old
+
+    def __add__(self, other):
+        return self.__class__(self.x + other.x)
+
+    def __sub__(self, other):
+        return self.__class__(self.x - other.x)
+
+    def __mul__(self, other):
+        return self.__class__(self.x * other.x)
+
+# size results
+class StackResult(co.namedtuple('StackResult', [
+        'file', 'function', 'frame', 'limit', 'children'])):
+    _by = ['file', 'function']
+    _fields = ['frame', 'limit']
+    _sort = ['limit', 'frame']
+    _types = {'frame': Int, 'limit': Int}
+
+    __slots__ = ()
+    def __new__(cls, file='', function='',
+            frame=0, limit=0, children=set()):
+        return super().__new__(cls, file, function,
+            Int(frame), Int(limit),
+            children)
+
+    def __add__(self, other):
+        return StackResult(self.file, self.function,
+            self.frame + other.frame,
+            max(self.limit, other.limit),
+            self.children | other.children)
+
+
+def openio(path, mode='r', buffering=-1):
+    # allow '-' for stdin/stdout
+    if path == '-':
+        if mode == 'r':
+            return os.fdopen(os.dup(sys.stdin.fileno()), mode, buffering)
+        else:
+            return os.fdopen(os.dup(sys.stdout.fileno()), mode, buffering)
+    else:
+        return open(path, mode, buffering)
+
+def collect(ci_paths, *,
+        sources=None,
+        everything=False,
+        **args):
     # parse the vcg format
     k_pattern = re.compile('([a-z]+)\s*:', re.DOTALL)
     v_pattern = re.compile('(?:"(.*?)"|([a-z]+))', re.DOTALL)
@@ -24,10 +141,10 @@ def collect(paths, **args):
             node = []
             while True:
                 rest = rest.lstrip()
-                m = k_pattern.match(rest)
-                if not m:
+                m_ = k_pattern.match(rest)
+                if not m_:
                     return (node, rest)
-                k, rest = m.group(1), rest[m.end(0):]
+                k, rest = m_.group(1), rest[m_.end(0):]
 
                 rest = rest.lstrip()
                 if rest.startswith('{'):
@@ -36,9 +153,9 @@ def collect(paths, **args):
                     rest = rest[1:]
                     node.append((k, v))
                 else:
-                    m = v_pattern.match(rest)
-                    assert m, "unexpected %r" % rest[0:1]
-                    v, rest = m.group(1) or m.group(2), rest[m.end(0):]
+                    m_ = v_pattern.match(rest)
+                    assert m_, "unexpected %r" % rest[0:1]
+                    v, rest = m_.group(1) or m_.group(2), rest[m_.end(0):]
                     node.append((k, v))
 
         node, rest = parse_vcg(rest)
@@ -46,10 +163,10 @@ def collect(paths, **args):
         return node
 
     # collect into functions
-    results = co.defaultdict(lambda: (None, None, 0, set()))
+    callgraph = co.defaultdict(lambda: (None, None, 0, set()))
     f_pattern = re.compile(
         r'([^\\]*)\\n([^:]*)[^\\]*\\n([0-9]+) bytes \((.*)\)')
-    for path in paths:
+    for path in ci_paths:
         with open(path) as f:
             vcg = parse_vcg(f.read())
         for k, graph in vcg:
@@ -58,373 +175,561 @@ def collect(paths, **args):
             for k, info in graph:
                 if k == 'node':
                     info = dict(info)
-                    m = f_pattern.match(info['label'])
-                    if m:
-                        function, file, size, type = m.groups()
-                        if not args.get('quiet') and type != 'static':
-                            print('warning: found non-static stack for %s (%s)'
-                                % (function, type))
-                        _, _, _, targets = results[info['title']]
-                        results[info['title']] = (
+                    m_ = f_pattern.match(info['label'])
+                    if m_:
+                        function, file, size, type = m_.groups()
+                        if (not args.get('quiet')
+                                and 'static' not in type
+                                and 'bounded' not in type):
+                            print("warning: "
+                                "found non-static stack for %s (%s, %s)" % (
+                                function, type, size))
+                        _, _, _, targets = callgraph[info['title']]
+                        callgraph[info['title']] = (
                             file, function, int(size), targets)
                 elif k == 'edge':
                     info = dict(info)
-                    _, _, _, targets = results[info['sourcename']]
+                    _, _, _, targets = callgraph[info['sourcename']]
                     targets.add(info['targetname'])
                 else:
                     continue
 
-    if not args.get('everything'):
-        for source, (s_file, s_function, _, _) in list(results.items()):
+    callgraph_ = co.defaultdict(lambda: (None, None, 0, set()))
+    for source, (s_file, s_function, frame, targets) in callgraph.items():
+        # discard internal functions
+        if not everything and s_function.startswith('__'):
+            continue
+        # ignore filtered sources
+        if sources is not None:
+            if not any(
+                    os.path.abspath(s_file) == os.path.abspath(s)
+                    for s in sources):
+                continue
+        else:
+            # default to only cwd
+            if not everything and not os.path.commonpath([
+                    os.getcwd(),
+                    os.path.abspath(s_file)]) == os.getcwd():
+                continue
+
+        # smiplify path
+        if os.path.commonpath([
+                os.getcwd(),
+                os.path.abspath(s_file)]) == os.getcwd():
+            s_file = os.path.relpath(s_file)
+        else:
+            s_file = os.path.abspath(s_file)
+
+        callgraph_[source] = (s_file, s_function, frame, targets)
+    callgraph = callgraph_
+
+    if not everything:
+        callgraph_ = co.defaultdict(lambda: (None, None, 0, set()))
+        for source, (s_file, s_function, frame, targets) in callgraph.items():
+            # discard filtered sources
+            if sources is not None and not any(
+                    os.path.abspath(s_file) == os.path.abspath(s)
+                    for s in sources):
+                continue
             # discard internal functions
-            if s_file.startswith('<') or s_file.startswith('/usr/include'):
-                del results[source]
+            if s_function.startswith('__'):
+                continue
+            callgraph_[source] = (s_file, s_function, frame, targets)
+        callgraph = callgraph_
 
     # find maximum stack size recursively, this requires also detecting cycles
     # (in case of recursion)
     def find_limit(source, seen=None):
         seen = seen or set()
-        if source not in results:
+        if source not in callgraph:
             return 0
-        _, _, frame, targets = results[source]
+        _, _, frame, targets = callgraph[source]
 
         limit = 0
         for target in targets:
             if target in seen:
                 # found a cycle
-                return float('inf')
+                return m.inf
             limit_ = find_limit(target, seen | {target})
             limit = max(limit, limit_)
 
         return frame + limit
 
-    def find_deps(targets):
-        deps = set()
+    def find_children(targets):
+        children = set()
         for target in targets:
-            if target in results:
-                t_file, t_function, _, _ = results[target]
-                deps.add((t_file, t_function))
-        return deps
+            if target in callgraph:
+                t_file, t_function, _, _ = callgraph[target]
+                children.add((t_file, t_function))
+        return children
 
-    # flatten into a list
-    flat_results = []
-    for source, (s_file, s_function, frame, targets) in results.items():
+    # build results
+    results = []
+    for source, (s_file, s_function, frame, targets) in callgraph.items():
         limit = find_limit(source)
-        deps = find_deps(targets)
-        flat_results.append((s_file, s_function, frame, limit, deps))
+        children = find_children(targets)
+        results.append(StackResult(s_file, s_function, frame, limit, children))
 
-    return flat_results
+    return results
 
-def main(**args):
-    def openio(path, mode='r'):
-        if path == '-':
-            if 'r' in mode:
-                return os.fdopen(os.dup(sys.stdin.fileno()), 'r')
-            else:
-                return os.fdopen(os.dup(sys.stdout.fileno()), 'w')
+
+def fold(Result, results, *,
+        by=None,
+        defines=None,
+        **_):
+    if by is None:
+        by = Result._by
+
+    for k in it.chain(by or [], (k for k, _ in defines or [])):
+        if k not in Result._by and k not in Result._fields:
+            print("error: could not find field %r?" % k)
+            sys.exit(-1)
+
+    # filter by matching defines
+    if defines is not None:
+        results_ = []
+        for r in results:
+            if all(getattr(r, k) in vs for k, vs in defines):
+                results_.append(r)
+        results = results_
+
+    # organize results into conflicts
+    folding = co.OrderedDict()
+    for r in results:
+        name = tuple(getattr(r, k) for k in by)
+        if name not in folding:
+            folding[name] = []
+        folding[name].append(r)
+
+    # merge conflicts
+    folded = []
+    for name, rs in folding.items():
+        folded.append(sum(rs[1:], start=rs[0]))
+
+    return folded
+
+def table(Result, results, diff_results=None, *,
+        by=None,
+        fields=None,
+        sort=None,
+        summary=False,
+        all=False,
+        percent=False,
+        tree=False,
+        depth=1,
+        **_):
+    all_, all = all, __builtins__.all
+
+    if by is None:
+        by = Result._by
+    if fields is None:
+        fields = Result._fields
+    types = Result._types
+
+    # fold again
+    results = fold(Result, results, by=by)
+    if diff_results is not None:
+        diff_results = fold(Result, diff_results, by=by)
+
+    # organize by name
+    table = {
+        ','.join(str(getattr(r, k) or '') for k in by): r
+        for r in results}
+    diff_table = {
+        ','.join(str(getattr(r, k) or '') for k in by): r
+        for r in diff_results or []}
+    names = list(table.keys() | diff_table.keys())
+
+    # sort again, now with diff info, note that python's sort is stable
+    names.sort()
+    if diff_results is not None:
+        names.sort(key=lambda n: tuple(
+            types[k].ratio(
+                getattr(table.get(n), k, None),
+                getattr(diff_table.get(n), k, None))
+            for k in fields),
+            reverse=True)
+    if sort:
+        for k, reverse in reversed(sort):
+            names.sort(
+                key=lambda n: tuple(
+                    (getattr(table[n], k),)
+                    if getattr(table.get(n), k, None) is not None else ()
+                    for k in ([k] if k else [
+                        k for k in Result._sort if k in fields])),
+                reverse=reverse ^ (not k or k in Result._fields))
+
+
+    # build up our lines
+    lines = []
+
+    # header
+    header = []
+    header.append('%s%s' % (
+        ','.join(by),
+        ' (%d added, %d removed)' % (
+            sum(1 for n in table if n not in diff_table),
+            sum(1 for n in diff_table if n not in table))
+            if diff_results is not None and not percent else '')
+        if not summary else '')
+    if diff_results is None:
+        for k in fields:
+            header.append(k)
+    elif percent:
+        for k in fields:
+            header.append(k)
+    else:
+        for k in fields:
+            header.append('o'+k)
+        for k in fields:
+            header.append('n'+k)
+        for k in fields:
+            header.append('d'+k)
+    header.append('')
+    lines.append(header)
+
+    def table_entry(name, r, diff_r=None, ratios=[]):
+        entry = []
+        entry.append(name)
+        if diff_results is None:
+            for k in fields:
+                entry.append(getattr(r, k).table()
+                    if getattr(r, k, None) is not None
+                    else types[k].none)
+        elif percent:
+            for k in fields:
+                entry.append(getattr(r, k).diff_table()
+                    if getattr(r, k, None) is not None
+                    else types[k].diff_none)
         else:
-            return open(path, mode)
+            for k in fields:
+                entry.append(getattr(diff_r, k).diff_table()
+                    if getattr(diff_r, k, None) is not None
+                    else types[k].diff_none)
+            for k in fields:
+                entry.append(getattr(r, k).diff_table()
+                    if getattr(r, k, None) is not None
+                    else types[k].diff_none)
+            for k in fields:
+                entry.append(types[k].diff_diff(
+                        getattr(r, k, None),
+                        getattr(diff_r, k, None)))
+        if diff_results is None:
+            entry.append('')
+        elif percent:
+            entry.append(' (%s)' % ', '.join(
+                '+∞%' if t == +m.inf
+                else '-∞%' if t == -m.inf
+                else '%+.1f%%' % (100*t)
+                for t in ratios))
+        else:
+            entry.append(' (%s)' % ', '.join(
+                    '+∞%' if t == +m.inf
+                    else '-∞%' if t == -m.inf
+                    else '%+.1f%%' % (100*t)
+                    for t in ratios
+                    if t)
+                if any(ratios) else '')
+        return entry
+
+    # entries
+    if not summary:
+        for name in names:
+            r = table.get(name)
+            if diff_results is None:
+                diff_r = None
+                ratios = None
+            else:
+                diff_r = diff_table.get(name)
+                ratios = [
+                    types[k].ratio(
+                        getattr(r, k, None),
+                        getattr(diff_r, k, None))
+                    for k in fields]
+                if not all_ and not any(ratios):
+                    continue
+            lines.append(table_entry(name, r, diff_r, ratios))
+
+    # total
+    r = next(iter(fold(Result, results, by=[])), None)
+    if diff_results is None:
+        diff_r = None
+        ratios = None
+    else:
+        diff_r = next(iter(fold(Result, diff_results, by=[])), None)
+        ratios = [
+            types[k].ratio(
+                getattr(r, k, None),
+                getattr(diff_r, k, None))
+            for k in fields]
+    lines.append(table_entry('TOTAL', r, diff_r, ratios))
+
+    # find the best widths, note that column 0 contains the names and column -1
+    # the ratios, so those are handled a bit differently
+    widths = [
+        ((max(it.chain([w], (len(l[i]) for l in lines)))+1+4-1)//4)*4-1
+        for w, i in zip(
+            it.chain([23], it.repeat(7)),
+            range(len(lines[0])-1))]
+
+    # adjust the name width based on the expected call depth, though
+    # note this doesn't really work with unbounded recursion
+    if not summary and not m.isinf(depth):
+        widths[0] += 4*(depth-1)
+
+    # print the tree recursively
+    if not tree:
+        print('%-*s  %s%s' % (
+            widths[0], lines[0][0],
+            ' '.join('%*s' % (w, x)
+                for w, x in zip(widths[1:], lines[0][1:-1])),
+            lines[0][-1]))
+
+    if not summary:
+        line_table = {n: l for n, l in zip(names, lines[1:-1])}
+
+        def recurse(names_, depth_, prefixes=('', '', '', '')):
+            for i, name in enumerate(names_):
+                if name not in line_table:
+                    continue
+                line = line_table[name]
+                is_last = (i == len(names_)-1)
+
+                print('%s%-*s ' % (
+                    prefixes[0+is_last],
+                    widths[0] - (
+                        len(prefixes[0+is_last])
+                        if not m.isinf(depth) else 0),
+                    line[0]),
+                    end='')
+                if not tree:
+                    print(' %s%s' % (
+                        ' '.join('%*s' % (w, x)
+                            for w, x in zip(widths[1:], line[1:-1])),
+                        line[-1]),
+                        end='')
+                print() 
+
+                # recurse?
+                if name in table and depth_ > 1:
+                    children = {
+                        ','.join(str(getattr(Result(*c), k) or '') for k in by)
+                        for c in table[name].children}
+                    recurse(
+                        # note we're maintaining sort order
+                        [n for n in names if n in children],
+                        depth_-1,
+                        (prefixes[2+is_last] + "|-> ",
+                         prefixes[2+is_last] + "'-> ",
+                         prefixes[2+is_last] + "|   ",
+                         prefixes[2+is_last] + "    "))
+
+        recurse(names, depth)
+
+    if not tree:
+        print('%-*s  %s%s' % (
+            widths[0], lines[-1][0],
+            ' '.join('%*s' % (w, x)
+                for w, x in zip(widths[1:], lines[-1][1:-1])),
+            lines[-1][-1]))
+
+
+def main(ci_paths,
+        by=None,
+        fields=None,
+        defines=None,
+        sort=None,
+        **args):
+    # it doesn't really make sense to not have a depth with tree,
+    # so assume depth=inf if tree by default
+    if args.get('depth') is None:
+        args['depth'] = m.inf if args['tree'] else 1
+    elif args.get('depth') == 0:
+        args['depth'] = m.inf
 
     # find sizes
     if not args.get('use', None):
-        # find .ci files
-        paths = []
-        for path in args['ci_paths']:
-            if os.path.isdir(path):
-                path = path + '/*.ci'
-
-            for path in glob.glob(path):
-                paths.append(path)
-
-        if not paths:
-            print('no .ci files found in %r?' % args['ci_paths'])
-            sys.exit(-1)
-
-        results = collect(paths, **args)
+        results = collect(ci_paths, **args)
     else:
+        results = []
         with openio(args['use']) as f:
-            r = csv.DictReader(f)
-            results = [
-                (   result['file'],
-                    result['name'],
-                    int(result['stack_frame']),
-                    float(result['stack_limit']), # note limit can be inf
-                    set())
-                for result in r
-                if result.get('stack_frame') not in {None, ''}
-                if result.get('stack_limit') not in {None, ''}]
+            reader = csv.DictReader(f, restval='')
+            for r in reader:
+                if not any('stack_'+k in r and r['stack_'+k].strip()
+                        for k in StackResult._fields):
+                    continue
+                try:
+                    results.append(StackResult(
+                        **{k: r[k] for k in StackResult._by
+                            if k in r and r[k].strip()},
+                        **{k: r['stack_'+k] for k in StackResult._fields
+                            if 'stack_'+k in r and r['stack_'+k].strip()}))
+                except TypeError:
+                    pass
 
-    total_frame = 0
-    total_limit = 0
-    for _, _, frame, limit, _ in results:
-        total_frame += frame
-        total_limit = max(total_limit, limit)
+    # fold
+    results = fold(StackResult, results, by=by, defines=defines)
 
-    # find previous results?
-    if args.get('diff'):
-        try:
-            with openio(args['diff']) as f:
-                r = csv.DictReader(f)
-                prev_results = [
-                    (   result['file'],
-                        result['name'],
-                        int(result['stack_frame']),
-                        float(result['stack_limit']),
-                        set())
-                    for result in r
-                    if result.get('stack_frame') not in {None, ''}
-                    if result.get('stack_limit') not in {None, ''}]
-        except FileNotFoundError:
-            prev_results = []
-
-        prev_total_frame = 0
-        prev_total_limit = 0
-        for _, _, frame, limit, _ in prev_results:
-            prev_total_frame += frame
-            prev_total_limit = max(prev_total_limit, limit)
+    # sort, note that python's sort is stable
+    results.sort()
+    if sort:
+        for k, reverse in reversed(sort):
+            results.sort(
+                key=lambda r: tuple(
+                    (getattr(r, k),) if getattr(r, k) is not None else ()
+                    for k in ([k] if k else StackResult._sort)),
+                reverse=reverse ^ (not k or k in StackResult._fields))
 
     # write results to CSV
     if args.get('output'):
-        merged_results = co.defaultdict(lambda: {})
-        other_fields = []
-
-        # merge?
-        if args.get('merge'):
-            try:
-                with openio(args['merge']) as f:
-                    r = csv.DictReader(f)
-                    for result in r:
-                        file = result.pop('file', '')
-                        func = result.pop('name', '')
-                        result.pop('stack_frame', None)
-                        result.pop('stack_limit', None)
-                        merged_results[(file, func)] = result
-                        other_fields = result.keys()
-            except FileNotFoundError:
-                pass
-
-        for file, func, frame, limit, _ in results:
-            merged_results[(file, func)]['stack_frame'] = frame
-            merged_results[(file, func)]['stack_limit'] = limit
-
         with openio(args['output'], 'w') as f:
-            w = csv.DictWriter(f, ['file', 'name', *other_fields, 'stack_frame', 'stack_limit'])
-            w.writeheader()
-            for (file, func), result in sorted(merged_results.items()):
-                w.writerow({'file': file, 'name': func, **result})
+            writer = csv.DictWriter(f,
+                (by if by is not None else StackResult._by)
+                + ['stack_'+k for k in (
+                    fields if fields is not None else StackResult._fields)])
+            writer.writeheader()
+            for r in results:
+                writer.writerow(
+                    {k: getattr(r, k) for k in (
+                        by if by is not None else StackResult._by)}
+                    | {'stack_'+k: getattr(r, k) for k in (
+                        fields if fields is not None else StackResult._fields)})
 
-    # print results
-    def dedup_entries(results, by='name'):
-        entries = co.defaultdict(lambda: (0, 0, set()))
-        for file, func, frame, limit, deps in results:
-            entry = (file if by == 'file' else func)
-            entry_frame, entry_limit, entry_deps = entries[entry]
-            entries[entry] = (
-                entry_frame + frame,
-                max(entry_limit, limit),
-                entry_deps | {file if by == 'file' else func
-                    for file, func in deps})
-        return entries
+    # find previous results?
+    if args.get('diff'):
+        diff_results = []
+        try:
+            with openio(args['diff']) as f:
+                reader = csv.DictReader(f, restval='')
+                for r in reader:
+                    if not any('stack_'+k in r and r['stack_'+k].strip()
+                            for k in StackResult._fields):
+                        continue
+                    try:
+                        diff_results.append(StackResult(
+                            **{k: r[k] for k in StackResult._by
+                                if k in r and r[k].strip()},
+                            **{k: r['stack_'+k] for k in StackResult._fields
+                                if 'stack_'+k in r and r['stack_'+k].strip()}))
+                    except TypeError:
+                        raise
+        except FileNotFoundError:
+            pass
 
-    def diff_entries(olds, news):
-        diff = co.defaultdict(lambda: (None, None, None, None, 0, 0, 0, set()))
-        for name, (new_frame, new_limit, deps) in news.items():
-            diff[name] = (
-                None, None,
-                new_frame, new_limit,
-                new_frame, new_limit,
-                1.0,
-                deps)
-        for name, (old_frame, old_limit, _) in olds.items():
-            _, _, new_frame, new_limit, _, _, _, deps = diff[name]
-            diff[name] = (
-                old_frame, old_limit,
-                new_frame, new_limit,
-                (new_frame or 0) - (old_frame or 0),
-                0 if m.isinf(new_limit or 0) and m.isinf(old_limit or 0)
-                    else (new_limit or 0) - (old_limit or 0),
-                0.0 if m.isinf(new_limit or 0) and m.isinf(old_limit or 0)
-                    else +float('inf') if m.isinf(new_limit or 0)
-                    else -float('inf') if m.isinf(old_limit or 0)
-                    else +0.0 if not old_limit and not new_limit
-                    else +1.0 if not old_limit
-                    else ((new_limit or 0) - (old_limit or 0))/(old_limit or 0),
-                deps)
-        return diff
+        # fold
+        diff_results = fold(StackResult, diff_results, by=by, defines=defines)
 
-    def sorted_entries(entries):
-        if args.get('limit_sort'):
-            return sorted(entries, key=lambda x: (-x[1][1], x))
-        elif args.get('reverse_limit_sort'):
-            return sorted(entries, key=lambda x: (+x[1][1], x))
-        elif args.get('frame_sort'):
-            return sorted(entries, key=lambda x: (-x[1][0], x))
-        elif args.get('reverse_frame_sort'):
-            return sorted(entries, key=lambda x: (+x[1][0], x))
-        else:
-            return sorted(entries)
+    # print table
+    if not args.get('quiet'):
+        table(StackResult, results,
+            diff_results if args.get('diff') else None,
+            by=by if by is not None else ['function'],
+            fields=fields,
+            sort=sort,
+            **args)
 
-    def sorted_diff_entries(entries):
-        if args.get('limit_sort'):
-            return sorted(entries, key=lambda x: (-(x[1][3] or 0), x))
-        elif args.get('reverse_limit_sort'):
-            return sorted(entries, key=lambda x: (+(x[1][3] or 0), x))
-        elif args.get('frame_sort'):
-            return sorted(entries, key=lambda x: (-(x[1][2] or 0), x))
-        elif args.get('reverse_frame_sort'):
-            return sorted(entries, key=lambda x: (+(x[1][2] or 0), x))
-        else:
-            return sorted(entries, key=lambda x: (-x[1][6], x))
-
-    def print_header(by=''):
-        if not args.get('diff'):
-            print('%-36s %7s %7s' % (by, 'frame', 'limit'))
-        else:
-            print('%-36s %15s %15s %15s' % (by, 'old', 'new', 'diff'))
-
-    def print_entry(name, frame, limit):
-        print("%-36s %7d %7s" % (name,
-            frame, '∞' if m.isinf(limit) else int(limit)))
-
-    def print_diff_entry(name,
-            old_frame, old_limit,
-            new_frame, new_limit,
-            diff_frame, diff_limit,
-            ratio):
-        print('%-36s %7s %7s %7s %7s %+7d %7s%s' % (name,
-            old_frame if old_frame is not None else "-",
-            ('∞' if m.isinf(old_limit) else int(old_limit))
-                if old_limit is not None else "-",
-            new_frame if new_frame is not None else "-",
-            ('∞' if m.isinf(new_limit) else int(new_limit))
-                if new_limit is not None else "-",
-            diff_frame,
-            ('+∞' if diff_limit > 0 and m.isinf(diff_limit)
-                else '-∞' if diff_limit < 0 and m.isinf(diff_limit)
-                else '%+d' % diff_limit),
-            '' if not ratio
-                else ' (+∞%)' if ratio > 0 and m.isinf(ratio)
-                else ' (-∞%)' if ratio < 0 and m.isinf(ratio)
-                else ' (%+.1f%%)' % (100*ratio)))
-
-    def print_entries(by='name'):
-        # build optional tree of dependencies
-        def print_deps(entries, depth, print,
-                filter=lambda _: True,
-                prefixes=('', '', '', '')):
-            entries = entries if isinstance(entries, list) else list(entries)
-            filtered_entries = [(name, entry)
-                for name, entry in entries
-                if filter(name)]
-            for i, (name, entry) in enumerate(filtered_entries):
-                last = (i == len(filtered_entries)-1)
-                print(prefixes[0+last] + name, entry)
-
-                if depth > 0:
-                    deps = entry[-1]
-                    print_deps(entries, depth-1, print,
-                        lambda name: name in deps,
-                        (   prefixes[2+last] + "|-> ",
-                            prefixes[2+last] + "'-> ",
-                            prefixes[2+last] + "|   ",
-                            prefixes[2+last] + "    "))
-
-        entries = dedup_entries(results, by=by)
-
-        if not args.get('diff'):
-            print_header(by=by)
-            print_deps(
-                sorted_entries(entries.items()),
-                args.get('depth') or 0,
-                lambda name, entry: print_entry(name, *entry[:-1]))
-        else:
-            prev_entries = dedup_entries(prev_results, by=by)
-            diff = diff_entries(prev_entries, entries)
-
-            print_header(by='%s (%d added, %d removed)' % (by,
-                sum(1 for _, old, _, _, _, _, _, _ in diff.values() if old is None),
-                sum(1 for _, _, _, new, _, _, _, _ in diff.values() if new is None)))
-            print_deps(
-                filter(
-                    lambda x: x[1][6] or args.get('all'),
-                    sorted_diff_entries(diff.items())),
-                args.get('depth') or 0,
-                lambda name, entry: print_diff_entry(name, *entry[:-1]))
-
-    def print_totals():
-        if not args.get('diff'):
-            print_entry('TOTAL', total_frame, total_limit)
-        else:
-            diff_frame = total_frame - prev_total_frame
-            diff_limit = (
-                0 if m.isinf(total_limit or 0) and m.isinf(prev_total_limit or 0)
-                    else (total_limit or 0) - (prev_total_limit or 0))
-            ratio = (
-                0.0 if m.isinf(total_limit or 0) and m.isinf(prev_total_limit or 0)
-                    else +float('inf') if m.isinf(total_limit or 0)
-                    else -float('inf') if m.isinf(prev_total_limit or 0)
-                    else 0.0 if not prev_total_limit and not total_limit
-                    else 1.0 if not prev_total_limit
-                    else ((total_limit or 0) - (prev_total_limit or 0))/(prev_total_limit or 0))
-            print_diff_entry('TOTAL',
-                prev_total_frame, prev_total_limit,
-                total_frame, total_limit,
-                diff_frame, diff_limit,
-                ratio)
-
-    if args.get('quiet'):
-        pass
-    elif args.get('summary'):
-        print_header()
-        print_totals()
-    elif args.get('files'):
-        print_entries(by='file')
-        print_totals()
-    else:
-        print_entries(by='name')
-        print_totals()
+    # error on recursion
+    if args.get('error_on_recursion') and any(
+            m.isinf(float(r.limit)) for r in results):
+        sys.exit(2)
 
 
 if __name__ == "__main__":
     import argparse
     import sys
     parser = argparse.ArgumentParser(
-        description="Find stack usage at the function level.")
-    parser.add_argument('ci_paths', nargs='*', default=CI_PATHS,
-        help="Description of where to find *.ci files. May be a directory \
-            or a list of paths. Defaults to %r." % CI_PATHS)
-    parser.add_argument('-v', '--verbose', action='store_true',
+        description="Find stack usage at the function level.",
+        allow_abbrev=False)
+    parser.add_argument(
+        'ci_paths',
+        nargs='*',
+        help="Input *.ci files.")
+    parser.add_argument(
+        '-v', '--verbose',
+        action='store_true',
         help="Output commands that run behind the scenes.")
-    parser.add_argument('-q', '--quiet', action='store_true',
+    parser.add_argument(
+        '-q', '--quiet',
+        action='store_true',
         help="Don't show anything, useful with -o.")
-    parser.add_argument('-o', '--output',
+    parser.add_argument(
+        '-o', '--output',
         help="Specify CSV file to store results.")
-    parser.add_argument('-u', '--use',
-        help="Don't parse callgraph files, instead use this CSV file.")
-    parser.add_argument('-d', '--diff',
+    parser.add_argument(
+        '-u', '--use',
+        help="Don't parse anything, use this CSV file.")
+    parser.add_argument(
+        '-d', '--diff',
         help="Specify CSV file to diff against.")
-    parser.add_argument('-m', '--merge',
-        help="Merge with an existing CSV file when writing to output.")
-    parser.add_argument('-a', '--all', action='store_true',
-        help="Show all functions, not just the ones that changed.")
-    parser.add_argument('-A', '--everything', action='store_true',
+    parser.add_argument(
+        '-a', '--all',
+        action='store_true',
+        help="Show all, not just the ones that changed.")
+    parser.add_argument(
+        '-p', '--percent',
+        action='store_true',
+        help="Only show percentage change, not a full diff.")
+    parser.add_argument(
+        '-b', '--by',
+        action='append',
+        choices=StackResult._by,
+        help="Group by this field.")
+    parser.add_argument(
+        '-f', '--field',
+        dest='fields',
+        action='append',
+        choices=StackResult._fields,
+        help="Show this field.")
+    parser.add_argument(
+        '-D', '--define',
+        dest='defines',
+        action='append',
+        type=lambda x: (lambda k,v: (k, set(v.split(','))))(*x.split('=', 1)),
+        help="Only include results where this field is this value.")
+    class AppendSort(argparse.Action):
+        def __call__(self, parser, namespace, value, option):
+            if namespace.sort is None:
+                namespace.sort = []
+            namespace.sort.append((value, True if option == '-S' else False))
+    parser.add_argument(
+        '-s', '--sort',
+        nargs='?',
+        action=AppendSort,
+        help="Sort by this field.")
+    parser.add_argument(
+        '-S', '--reverse-sort',
+        nargs='?',
+        action=AppendSort,
+        help="Sort by this field, but backwards.")
+    parser.add_argument(
+        '-Y', '--summary',
+        action='store_true',
+        help="Only show the total.")
+    parser.add_argument(
+        '-F', '--source',
+        dest='sources',
+        action='append',
+        help="Only consider definitions in this file. Defaults to anything "
+            "in the current directory.")
+    parser.add_argument(
+        '--everything',
+        action='store_true',
         help="Include builtin and libc specific symbols.")
-    parser.add_argument('-s', '--limit-sort', action='store_true',
-        help="Sort by stack limit.")
-    parser.add_argument('-S', '--reverse-limit-sort', action='store_true',
-        help="Sort by stack limit, but backwards.")
-    parser.add_argument('--frame-sort', action='store_true',
-        help="Sort by stack frame size.")
-    parser.add_argument('--reverse-frame-sort', action='store_true',
-        help="Sort by stack frame size, but backwards.")
-    parser.add_argument('-L', '--depth', default=0, type=lambda x: int(x, 0),
-        nargs='?', const=float('inf'),
-        help="Depth of dependencies to show.")
-    parser.add_argument('-F', '--files', action='store_true',
-        help="Show file-level calls.")
-    parser.add_argument('-Y', '--summary', action='store_true',
-        help="Only show the total stack size.")
-    parser.add_argument('--build-dir',
-        help="Specify the relative build directory. Used to map object files \
-            to the correct source files.")
-    sys.exit(main(**vars(parser.parse_args())))
+    parser.add_argument(
+        '--tree',
+        action='store_true',
+        help="Only show the function call tree.")
+    parser.add_argument(
+        '-Z', '--depth',
+        nargs='?',
+        type=lambda x: int(x, 0),
+        const=0,
+        help="Depth of function calls to show. 0 shows all calls but may not "
+            "terminate!")
+    parser.add_argument(
+        '-e', '--error-on-recursion',
+        action='store_true',
+        help="Error if any functions are recursive.")
+    sys.exit(main(**{k: v
+        for k, v in vars(parser.parse_intermixed_args()).items()
+        if v is not None}))
diff --git a/scripts/structs.py b/scripts/structs.py
index e8d7193..e9b97e7 100644
--- a/scripts/structs.py
+++ b/scripts/structs.py
@@ -2,49 +2,183 @@
 #
 # Script to find struct sizes.
 #
+# Example:
+# ./scripts/structs.py lfs.o lfs_util.o -Ssize
+#
+# Copyright (c) 2022, The littlefs authors.
+# SPDX-License-Identifier: BSD-3-Clause
+#
 
-import os
-import glob
-import itertools as it
-import subprocess as sp
-import shlex
-import re
-import csv
 import collections as co
+import csv
+import difflib
+import itertools as it
+import math as m
+import os
+import re
+import shlex
+import subprocess as sp
 
 
-OBJ_PATHS = ['*.o']
+OBJDUMP_PATH = ['objdump']
 
-def collect(paths, **args):
-    decl_pattern = re.compile(
+
+
+# integer fields
+class Int(co.namedtuple('Int', 'x')):
+    __slots__ = ()
+    def __new__(cls, x=0):
+        if isinstance(x, Int):
+            return x
+        if isinstance(x, str):
+            try:
+                x = int(x, 0)
+            except ValueError:
+                # also accept +-∞ and +-inf
+                if re.match('^\s*\+?\s*(?:∞|inf)\s*$', x):
+                    x = m.inf
+                elif re.match('^\s*-\s*(?:∞|inf)\s*$', x):
+                    x = -m.inf
+                else:
+                    raise
+        assert isinstance(x, int) or m.isinf(x), x
+        return super().__new__(cls, x)
+
+    def __str__(self):
+        if self.x == m.inf:
+            return '∞'
+        elif self.x == -m.inf:
+            return '-∞'
+        else:
+            return str(self.x)
+
+    def __int__(self):
+        assert not m.isinf(self.x)
+        return self.x
+
+    def __float__(self):
+        return float(self.x)
+
+    none = '%7s' % '-'
+    def table(self):
+        return '%7s' % (self,)
+
+    diff_none = '%7s' % '-'
+    diff_table = table
+
+    def diff_diff(self, other):
+        new = self.x if self else 0
+        old = other.x if other else 0
+        diff = new - old
+        if diff == +m.inf:
+            return '%7s' % '+∞'
+        elif diff == -m.inf:
+            return '%7s' % '-∞'
+        else:
+            return '%+7d' % diff
+
+    def ratio(self, other):
+        new = self.x if self else 0
+        old = other.x if other else 0
+        if m.isinf(new) and m.isinf(old):
+            return 0.0
+        elif m.isinf(new):
+            return +m.inf
+        elif m.isinf(old):
+            return -m.inf
+        elif not old and not new:
+            return 0.0
+        elif not old:
+            return 1.0
+        else:
+            return (new-old) / old
+
+    def __add__(self, other):
+        return self.__class__(self.x + other.x)
+
+    def __sub__(self, other):
+        return self.__class__(self.x - other.x)
+
+    def __mul__(self, other):
+        return self.__class__(self.x * other.x)
+
+# struct size results
+class StructResult(co.namedtuple('StructResult', ['file', 'struct', 'size'])):
+    _by = ['file', 'struct']
+    _fields = ['size']
+    _sort = ['size']
+    _types = {'size': Int}
+
+    __slots__ = ()
+    def __new__(cls, file='', struct='', size=0):
+        return super().__new__(cls, file, struct,
+            Int(size))
+
+    def __add__(self, other):
+        return StructResult(self.file, self.struct,
+            self.size + other.size)
+
+
+def openio(path, mode='r', buffering=-1):
+    # allow '-' for stdin/stdout
+    if path == '-':
+        if mode == 'r':
+            return os.fdopen(os.dup(sys.stdin.fileno()), mode, buffering)
+        else:
+            return os.fdopen(os.dup(sys.stdout.fileno()), mode, buffering)
+    else:
+        return open(path, mode, buffering)
+
+def collect(obj_paths, *,
+        objdump_path=OBJDUMP_PATH,
+        sources=None,
+        everything=False,
+        internal=False,
+        **args):
+    line_pattern = re.compile(
         '^\s+(?P<no>[0-9]+)'
-            '\s+(?P<dir>[0-9]+)'
+            '(?:\s+(?P<dir>[0-9]+))?'
             '\s+.*'
-            '\s+(?P<file>[^\s]+)$')
-    struct_pattern = re.compile(
-        '^(?:.*DW_TAG_(?P<tag>[a-z_]+).*'
-            '|^.*DW_AT_name.*:\s*(?P<name>[^:\s]+)\s*'
-            '|^.*DW_AT_decl_file.*:\s*(?P<decl>[0-9]+)\s*'
-            '|^.*DW_AT_byte_size.*:\s*(?P<size>[0-9]+)\s*)$')
+            '\s+(?P<path>[^\s]+)$')
+    info_pattern = re.compile(
+        '^(?:.*(?P<tag>DW_TAG_[a-z_]+).*'
+            '|.*DW_AT_name.*:\s*(?P<name>[^:\s]+)\s*'
+            '|.*DW_AT_decl_file.*:\s*(?P<file>[0-9]+)\s*'
+            '|.*DW_AT_byte_size.*:\s*(?P<size>[0-9]+)\s*)$')
 
-    results = co.defaultdict(lambda: 0)
-    for path in paths:
-        # find decl, we want to filter by structs in .h files
-        decls = {}
-        # note objdump-tool may contain extra args
-        cmd = args['objdump_tool'] + ['--dwarf=rawline', path]
+    results = []
+    for path in obj_paths:
+        # find files, we want to filter by structs in .h files
+        dirs = {}
+        files = {}
+        # note objdump-path may contain extra args
+        cmd = objdump_path + ['--dwarf=rawline', path]
         if args.get('verbose'):
             print(' '.join(shlex.quote(c) for c in cmd))
         proc = sp.Popen(cmd,
             stdout=sp.PIPE,
             stderr=sp.PIPE if not args.get('verbose') else None,
             universal_newlines=True,
-            errors='replace')
+            errors='replace',
+            close_fds=False)
         for line in proc.stdout:
-            # find file numbers
-            m = decl_pattern.match(line)
+            # note that files contain references to dirs, which we
+            # dereference as soon as we see them as each file table follows a
+            # dir table
+            m = line_pattern.match(line)
             if m:
-                decls[int(m.group('no'))] = m.group('file')
+                if not m.group('dir'):
+                    # found a directory entry
+                    dirs[int(m.group('no'))] = m.group('path')
+                else:
+                    # found a file entry
+                    dir = int(m.group('dir'))
+                    if dir in dirs:
+                        files[int(m.group('no'))] = os.path.join(
+                            dirs[dir],
+                            m.group('path'))
+                    else:
+                        files[int(m.group('no'))] = m.group('path')
         proc.wait()
         if proc.returncode != 0:
             if not args.get('verbose'):
@@ -53,40 +187,39 @@ def collect(paths, **args):
             sys.exit(-1)
 
         # collect structs as we parse dwarf info
-        found = False
-        name = None
-        decl = None
-        size = None
-
-        # note objdump-tool may contain extra args
-        cmd = args['objdump_tool'] + ['--dwarf=info', path]
+        results_ = []
+        is_struct = False
+        s_name = None
+        s_file = None
+        s_size = None
+        # note objdump-path may contain extra args
+        cmd = objdump_path + ['--dwarf=info', path]
         if args.get('verbose'):
             print(' '.join(shlex.quote(c) for c in cmd))
         proc = sp.Popen(cmd,
             stdout=sp.PIPE,
             stderr=sp.PIPE if not args.get('verbose') else None,
             universal_newlines=True,
-            errors='replace')
+            errors='replace',
+            close_fds=False)
         for line in proc.stdout:
             # state machine here to find structs
-            m = struct_pattern.match(line)
+            m = info_pattern.match(line)
             if m:
                 if m.group('tag'):
-                    if (name is not None
-                            and decl is not None
-                            and size is not None):
-                        decl = decls.get(decl, '?')
-                        results[(decl, name)] = size
-                    found = (m.group('tag') == 'structure_type')
-                    name = None
-                    decl = None
-                    size = None
-                elif found and m.group('name'):
-                    name = m.group('name')
-                elif found and name and m.group('decl'):
-                    decl = int(m.group('decl'))
-                elif found and name and m.group('size'):
-                    size = int(m.group('size'))
+                    if is_struct:
+                        file = files.get(s_file, '?')
+                        results_.append(StructResult(file, s_name, s_size))
+                    is_struct = (m.group('tag') == 'DW_TAG_structure_type')
+                elif m.group('name'):
+                    s_name = m.group('name')
+                elif m.group('file'):
+                    s_file = int(m.group('file'))
+                elif m.group('size'):
+                    s_size = int(m.group('size'))
+        if is_struct:
+            file = files.get(s_file, '?')
+            results_.append(StructResult(file, s_name, s_size))
         proc.wait()
         if proc.returncode != 0:
             if not args.get('verbose'):
@@ -94,238 +227,426 @@ def collect(paths, **args):
                     sys.stdout.write(line)
             sys.exit(-1)
 
-    flat_results = []
-    for (file, struct), size in results.items():
-        # map to source files
-        if args.get('build_dir'):
-            file = re.sub('%s/*' % re.escape(args['build_dir']), '', file)
-        # only include structs declared in header files in the current
-        # directory, ignore internal-only # structs (these are represented
-        # in other measurements)
-        if not args.get('everything'):
-            if not file.endswith('.h'):
-                continue
-        # replace .o with .c, different scripts report .o/.c, we need to
-        # choose one if we want to deduplicate csv files
-        file = re.sub('\.o$', '.c', file)
-
-        flat_results.append((file, struct, size))
-
-    return flat_results
-
-
-def main(**args):
-    def openio(path, mode='r'):
-        if path == '-':
-            if 'r' in mode:
-                return os.fdopen(os.dup(sys.stdin.fileno()), 'r')
+        for r in results_:
+            # ignore filtered sources
+            if sources is not None:
+                if not any(
+                        os.path.abspath(r.file) == os.path.abspath(s)
+                        for s in sources):
+                    continue
             else:
-                return os.fdopen(os.dup(sys.stdout.fileno()), 'w')
-        else:
-            return open(path, mode)
+                # default to only cwd
+                if not everything and not os.path.commonpath([
+                        os.getcwd(),
+                        os.path.abspath(r.file)]) == os.getcwd():
+                    continue
 
-    # find sizes
-    if not args.get('use', None):
-        # find .o files
-        paths = []
-        for path in args['obj_paths']:
-            if os.path.isdir(path):
-                path = path + '/*.o'
+                # limit to .h files unless --internal
+                if not internal and not r.file.endswith('.h'):
+                    continue
 
-            for path in glob.glob(path):
-                paths.append(path)
+            # simplify path
+            if os.path.commonpath([
+                    os.getcwd(),
+                    os.path.abspath(r.file)]) == os.getcwd():
+                file = os.path.relpath(r.file)
+            else:
+                file = os.path.abspath(r.file)
 
-        if not paths:
-            print('no .obj files found in %r?' % args['obj_paths'])
+            results.append(r._replace(file=file))
+
+    return results
+
+
+def fold(Result, results, *,
+        by=None,
+        defines=None,
+        **_):
+    if by is None:
+        by = Result._by
+
+    for k in it.chain(by or [], (k for k, _ in defines or [])):
+        if k not in Result._by and k not in Result._fields:
+            print("error: could not find field %r?" % k)
             sys.exit(-1)
 
-        results = collect(paths, **args)
+    # filter by matching defines
+    if defines is not None:
+        results_ = []
+        for r in results:
+            if all(getattr(r, k) in vs for k, vs in defines):
+                results_.append(r)
+        results = results_
+
+    # organize results into conflicts
+    folding = co.OrderedDict()
+    for r in results:
+        name = tuple(getattr(r, k) for k in by)
+        if name not in folding:
+            folding[name] = []
+        folding[name].append(r)
+
+    # merge conflicts
+    folded = []
+    for name, rs in folding.items():
+        folded.append(sum(rs[1:], start=rs[0]))
+
+    return folded
+
+def table(Result, results, diff_results=None, *,
+        by=None,
+        fields=None,
+        sort=None,
+        summary=False,
+        all=False,
+        percent=False,
+        **_):
+    all_, all = all, __builtins__.all
+
+    if by is None:
+        by = Result._by
+    if fields is None:
+        fields = Result._fields
+    types = Result._types
+
+    # fold again
+    results = fold(Result, results, by=by)
+    if diff_results is not None:
+        diff_results = fold(Result, diff_results, by=by)
+
+    # organize by name
+    table = {
+        ','.join(str(getattr(r, k) or '') for k in by): r
+        for r in results}
+    diff_table = {
+        ','.join(str(getattr(r, k) or '') for k in by): r
+        for r in diff_results or []}
+    names = list(table.keys() | diff_table.keys())
+
+    # sort again, now with diff info, note that python's sort is stable
+    names.sort()
+    if diff_results is not None:
+        names.sort(key=lambda n: tuple(
+            types[k].ratio(
+                getattr(table.get(n), k, None),
+                getattr(diff_table.get(n), k, None))
+            for k in fields),
+            reverse=True)
+    if sort:
+        for k, reverse in reversed(sort):
+            names.sort(
+                key=lambda n: tuple(
+                    (getattr(table[n], k),)
+                    if getattr(table.get(n), k, None) is not None else ()
+                    for k in ([k] if k else [
+                        k for k in Result._sort if k in fields])),
+                reverse=reverse ^ (not k or k in Result._fields))
+
+
+    # build up our lines
+    lines = []
+
+    # header
+    header = []
+    header.append('%s%s' % (
+        ','.join(by),
+        ' (%d added, %d removed)' % (
+            sum(1 for n in table if n not in diff_table),
+            sum(1 for n in diff_table if n not in table))
+            if diff_results is not None and not percent else '')
+        if not summary else '')
+    if diff_results is None:
+        for k in fields:
+            header.append(k)
+    elif percent:
+        for k in fields:
+            header.append(k)
     else:
+        for k in fields:
+            header.append('o'+k)
+        for k in fields:
+            header.append('n'+k)
+        for k in fields:
+            header.append('d'+k)
+    header.append('')
+    lines.append(header)
+
+    def table_entry(name, r, diff_r=None, ratios=[]):
+        entry = []
+        entry.append(name)
+        if diff_results is None:
+            for k in fields:
+                entry.append(getattr(r, k).table()
+                    if getattr(r, k, None) is not None
+                    else types[k].none)
+        elif percent:
+            for k in fields:
+                entry.append(getattr(r, k).diff_table()
+                    if getattr(r, k, None) is not None
+                    else types[k].diff_none)
+        else:
+            for k in fields:
+                entry.append(getattr(diff_r, k).diff_table()
+                    if getattr(diff_r, k, None) is not None
+                    else types[k].diff_none)
+            for k in fields:
+                entry.append(getattr(r, k).diff_table()
+                    if getattr(r, k, None) is not None
+                    else types[k].diff_none)
+            for k in fields:
+                entry.append(types[k].diff_diff(
+                        getattr(r, k, None),
+                        getattr(diff_r, k, None)))
+        if diff_results is None:
+            entry.append('')
+        elif percent:
+            entry.append(' (%s)' % ', '.join(
+                '+∞%' if t == +m.inf
+                else '-∞%' if t == -m.inf
+                else '%+.1f%%' % (100*t)
+                for t in ratios))
+        else:
+            entry.append(' (%s)' % ', '.join(
+                    '+∞%' if t == +m.inf
+                    else '-∞%' if t == -m.inf
+                    else '%+.1f%%' % (100*t)
+                    for t in ratios
+                    if t)
+                if any(ratios) else '')
+        return entry
+
+    # entries
+    if not summary:
+        for name in names:
+            r = table.get(name)
+            if diff_results is None:
+                diff_r = None
+                ratios = None
+            else:
+                diff_r = diff_table.get(name)
+                ratios = [
+                    types[k].ratio(
+                        getattr(r, k, None),
+                        getattr(diff_r, k, None))
+                    for k in fields]
+                if not all_ and not any(ratios):
+                    continue
+            lines.append(table_entry(name, r, diff_r, ratios))
+
+    # total
+    r = next(iter(fold(Result, results, by=[])), None)
+    if diff_results is None:
+        diff_r = None
+        ratios = None
+    else:
+        diff_r = next(iter(fold(Result, diff_results, by=[])), None)
+        ratios = [
+            types[k].ratio(
+                getattr(r, k, None),
+                getattr(diff_r, k, None))
+            for k in fields]
+    lines.append(table_entry('TOTAL', r, diff_r, ratios))
+
+    # find the best widths, note that column 0 contains the names and column -1
+    # the ratios, so those are handled a bit differently
+    widths = [
+        ((max(it.chain([w], (len(l[i]) for l in lines)))+1+4-1)//4)*4-1
+        for w, i in zip(
+            it.chain([23], it.repeat(7)),
+            range(len(lines[0])-1))]
+
+    # print our table
+    for line in lines:
+        print('%-*s  %s%s' % (
+            widths[0], line[0],
+            ' '.join('%*s' % (w, x)
+                for w, x in zip(widths[1:], line[1:-1])),
+            line[-1]))
+
+
+def main(obj_paths, *,
+        by=None,
+        fields=None,
+        defines=None,
+        sort=None,
+        **args):
+    # find sizes
+    if not args.get('use', None):
+        results = collect(obj_paths, **args)
+    else:
+        results = []
         with openio(args['use']) as f:
-            r = csv.DictReader(f)
-            results = [
-                (   result['file'],
-                    result['name'],
-                    int(result['struct_size']))
-                for result in r
-                if result.get('struct_size') not in {None, ''}]
+            reader = csv.DictReader(f, restval='')
+            for r in reader:
+                if not any('struct_'+k in r and r['struct_'+k].strip()
+                        for k in StructResult._fields):
+                    continue
+                try:
+                    results.append(StructResult(
+                        **{k: r[k] for k in StructResult._by
+                            if k in r and r[k].strip()},
+                        **{k: r['struct_'+k]
+                            for k in StructResult._fields
+                            if 'struct_'+k in r
+                                and r['struct_'+k].strip()}))
+                except TypeError:
+                    pass
 
-    total = 0
-    for _, _, size in results:
-        total += size
+    # fold
+    results = fold(StructResult, results, by=by, defines=defines)
 
-    # find previous results?
-    if args.get('diff'):
-        try:
-            with openio(args['diff']) as f:
-                r = csv.DictReader(f)
-                prev_results = [
-                    (   result['file'],
-                        result['name'],
-                        int(result['struct_size']))
-                    for result in r
-                    if result.get('struct_size') not in {None, ''}]
-        except FileNotFoundError:
-            prev_results = []
-
-        prev_total = 0
-        for _, _, size in prev_results:
-            prev_total += size
+    # sort, note that python's sort is stable
+    results.sort()
+    if sort:
+        for k, reverse in reversed(sort):
+            results.sort(
+                key=lambda r: tuple(
+                    (getattr(r, k),) if getattr(r, k) is not None else ()
+                    for k in ([k] if k else StructResult._sort)),
+                reverse=reverse ^ (not k or k in StructResult._fields))
 
     # write results to CSV
     if args.get('output'):
-        merged_results = co.defaultdict(lambda: {})
-        other_fields = []
-
-        # merge?
-        if args.get('merge'):
-            try:
-                with openio(args['merge']) as f:
-                    r = csv.DictReader(f)
-                    for result in r:
-                        file = result.pop('file', '')
-                        struct = result.pop('name', '')
-                        result.pop('struct_size', None)
-                        merged_results[(file, struct)] = result
-                        other_fields = result.keys()
-            except FileNotFoundError:
-                pass
-
-        for file, struct, size in results:
-            merged_results[(file, struct)]['struct_size'] = size
-
         with openio(args['output'], 'w') as f:
-            w = csv.DictWriter(f, ['file', 'name', *other_fields, 'struct_size'])
-            w.writeheader()
-            for (file, struct), result in sorted(merged_results.items()):
-                w.writerow({'file': file, 'name': struct, **result})
+            writer = csv.DictWriter(f,
+                (by if by is not None else StructResult._by)
+                + ['struct_'+k for k in (
+                    fields if fields is not None else StructResult._fields)])
+            writer.writeheader()
+            for r in results:
+                writer.writerow(
+                    {k: getattr(r, k) for k in (
+                        by if by is not None else StructResult._by)}
+                    | {'struct_'+k: getattr(r, k) for k in (
+                        fields if fields is not None else StructResult._fields)})
 
-    # print results
-    def dedup_entries(results, by='name'):
-        entries = co.defaultdict(lambda: 0)
-        for file, struct, size in results:
-            entry = (file if by == 'file' else struct)
-            entries[entry] += size
-        return entries
+    # find previous results?
+    if args.get('diff'):
+        diff_results = []
+        try:
+            with openio(args['diff']) as f:
+                reader = csv.DictReader(f, restval='')
+                for r in reader:
+                    if not any('struct_'+k in r and r['struct_'+k].strip()
+                            for k in StructResult._fields):
+                        continue
+                    try:
+                        diff_results.append(StructResult(
+                            **{k: r[k] for k in StructResult._by
+                                if k in r and r[k].strip()},
+                            **{k: r['struct_'+k]
+                                for k in StructResult._fields
+                                if 'struct_'+k in r
+                                    and r['struct_'+k].strip()}))
+                    except TypeError:
+                        pass
+        except FileNotFoundError:
+            pass
 
-    def diff_entries(olds, news):
-        diff = co.defaultdict(lambda: (0, 0, 0, 0))
-        for name, new in news.items():
-            diff[name] = (0, new, new, 1.0)
-        for name, old in olds.items():
-            _, new, _, _ = diff[name]
-            diff[name] = (old, new, new-old, (new-old)/old if old else 1.0)
-        return diff
+        # fold
+        diff_results = fold(StructResult, diff_results, by=by, defines=defines)
 
-    def sorted_entries(entries):
-        if args.get('size_sort'):
-            return sorted(entries, key=lambda x: (-x[1], x))
-        elif args.get('reverse_size_sort'):
-            return sorted(entries, key=lambda x: (+x[1], x))
-        else:
-            return sorted(entries)
+    # print table
+    if not args.get('quiet'):
+        table(StructResult, results,
+            diff_results if args.get('diff') else None,
+            by=by if by is not None else ['struct'],
+            fields=fields,
+            sort=sort,
+            **args)
 
-    def sorted_diff_entries(entries):
-        if args.get('size_sort'):
-            return sorted(entries, key=lambda x: (-x[1][1], x))
-        elif args.get('reverse_size_sort'):
-            return sorted(entries, key=lambda x: (+x[1][1], x))
-        else:
-            return sorted(entries, key=lambda x: (-x[1][3], x))
-
-    def print_header(by=''):
-        if not args.get('diff'):
-            print('%-36s %7s' % (by, 'size'))
-        else:
-            print('%-36s %7s %7s %7s' % (by, 'old', 'new', 'diff'))
-
-    def print_entry(name, size):
-        print("%-36s %7d" % (name, size))
-
-    def print_diff_entry(name, old, new, diff, ratio):
-        print("%-36s %7s %7s %+7d%s" % (name,
-            old or "-",
-            new or "-",
-            diff,
-            ' (%+.1f%%)' % (100*ratio) if ratio else ''))
-
-    def print_entries(by='name'):
-        entries = dedup_entries(results, by=by)
-
-        if not args.get('diff'):
-            print_header(by=by)
-            for name, size in sorted_entries(entries.items()):
-                print_entry(name, size)
-        else:
-            prev_entries = dedup_entries(prev_results, by=by)
-            diff = diff_entries(prev_entries, entries)
-            print_header(by='%s (%d added, %d removed)' % (by,
-                sum(1 for old, _, _, _ in diff.values() if not old),
-                sum(1 for _, new, _, _ in diff.values() if not new)))
-            for name, (old, new, diff, ratio) in sorted_diff_entries(
-                    diff.items()):
-                if ratio or args.get('all'):
-                    print_diff_entry(name, old, new, diff, ratio)
-
-    def print_totals():
-        if not args.get('diff'):
-            print_entry('TOTAL', total)
-        else:
-            ratio = (0.0 if not prev_total and not total
-                else 1.0 if not prev_total
-                else (total-prev_total)/prev_total)
-            print_diff_entry('TOTAL',
-                prev_total, total,
-                total-prev_total,
-                ratio)
-
-    if args.get('quiet'):
-        pass
-    elif args.get('summary'):
-        print_header()
-        print_totals()
-    elif args.get('files'):
-        print_entries(by='file')
-        print_totals()
-    else:
-        print_entries(by='name')
-        print_totals()
 
 if __name__ == "__main__":
     import argparse
     import sys
     parser = argparse.ArgumentParser(
-        description="Find struct sizes.")
-    parser.add_argument('obj_paths', nargs='*', default=OBJ_PATHS,
-        help="Description of where to find *.o files. May be a directory \
-            or a list of paths. Defaults to %r." % OBJ_PATHS)
-    parser.add_argument('-v', '--verbose', action='store_true',
+        description="Find struct sizes.",
+        allow_abbrev=False)
+    parser.add_argument(
+        'obj_paths',
+        nargs='*',
+        help="Input *.o files.")
+    parser.add_argument(
+        '-v', '--verbose',
+        action='store_true',
         help="Output commands that run behind the scenes.")
-    parser.add_argument('-q', '--quiet', action='store_true',
+    parser.add_argument(
+        '-q', '--quiet',
+        action='store_true',
         help="Don't show anything, useful with -o.")
-    parser.add_argument('-o', '--output',
+    parser.add_argument(
+        '-o', '--output',
         help="Specify CSV file to store results.")
-    parser.add_argument('-u', '--use',
-        help="Don't compile and find struct sizes, instead use this CSV file.")
-    parser.add_argument('-d', '--diff',
-        help="Specify CSV file to diff struct size against.")
-    parser.add_argument('-m', '--merge',
-        help="Merge with an existing CSV file when writing to output.")
-    parser.add_argument('-a', '--all', action='store_true',
-        help="Show all functions, not just the ones that changed.")
-    parser.add_argument('-A', '--everything', action='store_true',
+    parser.add_argument(
+        '-u', '--use',
+        help="Don't parse anything, use this CSV file.")
+    parser.add_argument(
+        '-d', '--diff',
+        help="Specify CSV file to diff against.")
+    parser.add_argument(
+        '-a', '--all',
+        action='store_true',
+        help="Show all, not just the ones that changed.")
+    parser.add_argument(
+        '-p', '--percent',
+        action='store_true',
+        help="Only show percentage change, not a full diff.")
+    parser.add_argument(
+        '-b', '--by',
+        action='append',
+        choices=StructResult._by,
+        help="Group by this field.")
+    parser.add_argument(
+        '-f', '--field',
+        dest='fields',
+        action='append',
+        choices=StructResult._fields,
+        help="Show this field.")
+    parser.add_argument(
+        '-D', '--define',
+        dest='defines',
+        action='append',
+        type=lambda x: (lambda k,v: (k, set(v.split(','))))(*x.split('=', 1)),
+        help="Only include results where this field is this value.")
+    class AppendSort(argparse.Action):
+        def __call__(self, parser, namespace, value, option):
+            if namespace.sort is None:
+                namespace.sort = []
+            namespace.sort.append((value, True if option == '-S' else False))
+    parser.add_argument(
+        '-s', '--sort',
+        nargs='?',
+        action=AppendSort,
+        help="Sort by this field.")
+    parser.add_argument(
+        '-S', '--reverse-sort',
+        nargs='?',
+        action=AppendSort,
+        help="Sort by this field, but backwards.")
+    parser.add_argument(
+        '-Y', '--summary',
+        action='store_true',
+        help="Only show the total.")
+    parser.add_argument(
+        '-F', '--source',
+        dest='sources',
+        action='append',
+        help="Only consider definitions in this file. Defaults to anything "
+            "in the current directory.")
+    parser.add_argument(
+        '--everything',
+        action='store_true',
         help="Include builtin and libc specific symbols.")
-    parser.add_argument('-s', '--size-sort', action='store_true',
-        help="Sort by size.")
-    parser.add_argument('-S', '--reverse-size-sort', action='store_true',
-        help="Sort by size, but backwards.")
-    parser.add_argument('-F', '--files', action='store_true',
-        help="Show file-level struct sizes.")
-    parser.add_argument('-Y', '--summary', action='store_true',
-        help="Only show the total struct size.")
-    parser.add_argument('--objdump-tool', default=['objdump'], type=lambda x: x.split(),
-        help="Path to the objdump tool to use.")
-    parser.add_argument('--build-dir',
-        help="Specify the relative build directory. Used to map object files \
-            to the correct source files.")
-    sys.exit(main(**vars(parser.parse_args())))
+    parser.add_argument(
+        '--internal',
+        action='store_true',
+        help="Also show structs in .c files.")
+    parser.add_argument(
+        '--objdump-path',
+        type=lambda x: x.split(),
+        default=OBJDUMP_PATH,
+        help="Path to the objdump executable, may include flags. "
+            "Defaults to %r." % OBJDUMP_PATH)
+    sys.exit(main(**{k: v
+        for k, v in vars(parser.parse_intermixed_args()).items()
+        if v is not None}))
diff --git a/scripts/summary.py b/scripts/summary.py
index 7ce769b..445368e 100644
--- a/scripts/summary.py
+++ b/scripts/summary.py
@@ -2,278 +2,828 @@
 #
 # Script to summarize the outputs of other scripts. Operates on CSV files.
 #
+# Example:
+# ./scripts/code.py lfs.o lfs_util.o -q -o lfs.code.csv
+# ./scripts/data.py lfs.o lfs_util.o -q -o lfs.data.csv
+# ./scripts/summary.py lfs.code.csv lfs.data.csv -q -o lfs.csv
+# ./scripts/summary.py -Y lfs.csv -f code=code_size,data=data_size
+#
+# Copyright (c) 2022, The littlefs authors.
+# SPDX-License-Identifier: BSD-3-Clause
+#
 
-import functools as ft
 import collections as co
-import os
 import csv
-import re
+import functools as ft
+import itertools as it
 import math as m
-
-# displayable fields
-Field = co.namedtuple('Field', 'name,parse,acc,key,fmt,repr,null,ratio')
-FIELDS = [
-    # name, parse, accumulate, fmt, print, null
-    Field('code',
-        lambda r: int(r['code_size']),
-        sum,
-        lambda r: r,
-        '%7s',
-        lambda r: r,
-        '-',
-        lambda old, new: (new-old)/old),
-    Field('data',
-        lambda r: int(r['data_size']),
-        sum,
-        lambda r: r,
-        '%7s',
-        lambda r: r,
-        '-',
-        lambda old, new: (new-old)/old),
-    Field('stack',
-        lambda r: float(r['stack_limit']),
-        max,
-        lambda r: r,
-        '%7s',
-        lambda r: '∞' if m.isinf(r) else int(r),
-        '-',
-        lambda old, new: (new-old)/old),
-    Field('structs',
-        lambda r: int(r['struct_size']),
-        sum,
-        lambda r: r,
-        '%8s',
-        lambda r: r,
-        '-',
-        lambda old, new: (new-old)/old),
-    Field('coverage',
-        lambda r: (int(r['coverage_hits']), int(r['coverage_count'])),
-        lambda rs: ft.reduce(lambda a, b: (a[0]+b[0], a[1]+b[1]), rs),
-        lambda r: r[0]/r[1],
-        '%19s',
-        lambda r: '%11s %7s' % ('%d/%d' % (r[0], r[1]), '%.1f%%' % (100*r[0]/r[1])),
-        '%11s %7s' % ('-', '-'),
-        lambda old, new: ((new[0]/new[1]) - (old[0]/old[1])))
-]
+import os
+import re
 
 
-def main(**args):
-    def openio(path, mode='r'):
-        if path == '-':
-            if 'r' in mode:
-                return os.fdopen(os.dup(sys.stdin.fileno()), 'r')
-            else:
-                return os.fdopen(os.dup(sys.stdout.fileno()), 'w')
+# supported merge operations
+#
+# this is a terrible way to express these
+#
+OPS = {
+    'sum':     lambda xs: sum(xs[1:], start=xs[0]),
+    'prod':    lambda xs: m.prod(xs[1:], start=xs[0]),
+    'min':     min,
+    'max':     max,
+    'mean':    lambda xs: Float(sum(float(x) for x in xs) / len(xs)),
+    'stddev':  lambda xs: (
+        lambda mean: Float(
+            m.sqrt(sum((float(x) - mean)**2 for x in xs) / len(xs)))
+        )(sum(float(x) for x in xs) / len(xs)),
+    'gmean':   lambda xs: Float(m.prod(float(x) for x in xs)**(1/len(xs))),
+    'gstddev': lambda xs: (
+        lambda gmean: Float(
+            m.exp(m.sqrt(sum(m.log(float(x)/gmean)**2 for x in xs) / len(xs)))
+            if gmean else m.inf)
+        )(m.prod(float(x) for x in xs)**(1/len(xs))),
+}
+
+
+# integer fields
+class Int(co.namedtuple('Int', 'x')):
+    __slots__ = ()
+    def __new__(cls, x=0):
+        if isinstance(x, Int):
+            return x
+        if isinstance(x, str):
+            try:
+                x = int(x, 0)
+            except ValueError:
+                # also accept +-∞ and +-inf
+                if re.match('^\s*\+?\s*(?:∞|inf)\s*$', x):
+                    x = m.inf
+                elif re.match('^\s*-\s*(?:∞|inf)\s*$', x):
+                    x = -m.inf
+                else:
+                    raise
+        assert isinstance(x, int) or m.isinf(x), x
+        return super().__new__(cls, x)
+
+    def __str__(self):
+        if self.x == m.inf:
+            return '∞'
+        elif self.x == -m.inf:
+            return '-∞'
         else:
-            return open(path, mode)
+            return str(self.x)
 
-    # find results
-    results = co.defaultdict(lambda: {})
-    for path in args.get('csv_paths', '-'):
+    def __int__(self):
+        assert not m.isinf(self.x)
+        return self.x
+
+    def __float__(self):
+        return float(self.x)
+
+    none = '%7s' % '-'
+    def table(self):
+        return '%7s' % (self,)
+
+    diff_none = '%7s' % '-'
+    diff_table = table
+
+    def diff_diff(self, other):
+        new = self.x if self else 0
+        old = other.x if other else 0
+        diff = new - old
+        if diff == +m.inf:
+            return '%7s' % '+∞'
+        elif diff == -m.inf:
+            return '%7s' % '-∞'
+        else:
+            return '%+7d' % diff
+
+    def ratio(self, other):
+        new = self.x if self else 0
+        old = other.x if other else 0
+        if m.isinf(new) and m.isinf(old):
+            return 0.0
+        elif m.isinf(new):
+            return +m.inf
+        elif m.isinf(old):
+            return -m.inf
+        elif not old and not new:
+            return 0.0
+        elif not old:
+            return 1.0
+        else:
+            return (new-old) / old
+
+    def __add__(self, other):
+        return self.__class__(self.x + other.x)
+
+    def __sub__(self, other):
+        return self.__class__(self.x - other.x)
+
+    def __mul__(self, other):
+        return self.__class__(self.x * other.x)
+
+# float fields
+class Float(co.namedtuple('Float', 'x')):
+    __slots__ = ()
+    def __new__(cls, x=0.0):
+        if isinstance(x, Float):
+            return x
+        if isinstance(x, str):
+            try:
+                x = float(x)
+            except ValueError:
+                # also accept +-∞ and +-inf
+                if re.match('^\s*\+?\s*(?:∞|inf)\s*$', x):
+                    x = m.inf
+                elif re.match('^\s*-\s*(?:∞|inf)\s*$', x):
+                    x = -m.inf
+                else:
+                    raise
+        assert isinstance(x, float), x
+        return super().__new__(cls, x)
+
+    def __str__(self):
+        if self.x == m.inf:
+            return '∞'
+        elif self.x == -m.inf:
+            return '-∞'
+        else:
+            return '%.1f' % self.x
+
+    def __float__(self):
+        return float(self.x)
+
+    none = Int.none
+    table = Int.table
+    diff_none = Int.diff_none
+    diff_table = Int.diff_table
+    diff_diff = Int.diff_diff
+    ratio = Int.ratio
+    __add__ = Int.__add__
+    __sub__ = Int.__sub__
+    __mul__ = Int.__mul__
+
+# fractional fields, a/b
+class Frac(co.namedtuple('Frac', 'a,b')):
+    __slots__ = ()
+    def __new__(cls, a=0, b=None):
+        if isinstance(a, Frac) and b is None:
+            return a
+        if isinstance(a, str) and b is None:
+            a, b = a.split('/', 1)
+        if b is None:
+            b = a
+        return super().__new__(cls, Int(a), Int(b))
+
+    def __str__(self):
+        return '%s/%s' % (self.a, self.b)
+
+    def __float__(self):
+        return float(self.a)
+
+    none = '%11s %7s' % ('-', '-')
+    def table(self):
+        t = self.a.x/self.b.x if self.b.x else 1.0
+        return '%11s %7s' % (
+            self,
+            '∞%' if t == +m.inf
+            else '-∞%' if t == -m.inf
+            else '%.1f%%' % (100*t))
+
+    diff_none = '%11s' % '-'
+    def diff_table(self):
+        return '%11s' % (self,)
+
+    def diff_diff(self, other):
+        new_a, new_b = self if self else (Int(0), Int(0))
+        old_a, old_b = other if other else (Int(0), Int(0))
+        return '%11s' % ('%s/%s' % (
+            new_a.diff_diff(old_a).strip(),
+            new_b.diff_diff(old_b).strip()))
+
+    def ratio(self, other):
+        new_a, new_b = self if self else (Int(0), Int(0))
+        old_a, old_b = other if other else (Int(0), Int(0))
+        new = new_a.x/new_b.x if new_b.x else 1.0
+        old = old_a.x/old_b.x if old_b.x else 1.0
+        return new - old
+
+    def __add__(self, other):
+        return self.__class__(self.a + other.a, self.b + other.b)
+
+    def __sub__(self, other):
+        return self.__class__(self.a - other.a, self.b - other.b)
+
+    def __mul__(self, other):
+        return self.__class__(self.a * other.a, self.b + other.b)
+
+    def __lt__(self, other):
+        self_t = self.a.x/self.b.x if self.b.x else 1.0
+        other_t = other.a.x/other.b.x if other.b.x else 1.0
+        return (self_t, self.a.x) < (other_t, other.a.x)
+
+    def __gt__(self, other):
+        return self.__class__.__lt__(other, self)
+
+    def __le__(self, other):
+        return not self.__gt__(other)
+
+    def __ge__(self, other):
+        return not self.__lt__(other)
+
+# available types
+TYPES = co.OrderedDict([
+    ('int',   Int),
+    ('float', Float),
+    ('frac',  Frac)
+])
+
+
+def infer(results, *,
+        by=None,
+        fields=None,
+        types={},
+        ops={},
+        renames=[],
+        **_):
+    # if fields not specified, try to guess from data
+    if fields is None:
+        fields = co.OrderedDict()
+        for r in results:
+            for k, v in r.items():
+                if (by is None or k not in by) and v.strip():
+                    types_ = []
+                    for t in fields.get(k, TYPES.values()):
+                        try:
+                            t(v)
+                            types_.append(t)
+                        except ValueError:
+                            pass
+                    fields[k] = types_
+        fields = list(k for k, v in fields.items() if v)
+
+    # deduplicate fields
+    fields = list(co.OrderedDict.fromkeys(fields).keys())
+
+    # if by not specified, guess it's anything not in fields and not a
+    # source of a rename
+    if by is None:
+        by = co.OrderedDict()
+        for r in results:
+            # also ignore None keys, these are introduced by csv.DictReader
+            # when header + row mismatch
+            by.update((k, True) for k in r.keys()
+                if k is not None
+                    and k not in fields
+                    and not any(k == old_k for _, old_k in renames))
+        by = list(by.keys())
+
+    # deduplicate fields
+    by = list(co.OrderedDict.fromkeys(by).keys())
+
+    # find best type for all fields
+    types_ = {}
+    for k in fields:
+        if k in types:
+            types_[k] = types[k]
+        else:
+            for t in TYPES.values():
+                for r in results:
+                    if k in r and r[k].strip():
+                        try:
+                            t(r[k])
+                        except ValueError:
+                            break
+                else:
+                    types_[k] = t
+                    break
+            else:
+                print("error: no type matches field %r?" % k)
+                sys.exit(-1)
+    types = types_
+
+    # does folding change the type?
+    types_ = {}
+    for k, t in types.items():
+        types_[k] = ops.get(k, OPS['sum'])([t()]).__class__
+
+
+    # create result class
+    def __new__(cls, **r):
+        return cls.__mro__[1].__new__(cls,
+            **{k: r.get(k, '') for k in by},
+            **{k: r[k] if k in r and isinstance(r[k], list)
+                else [types[k](r[k])] if k in r
+                else []
+                for k in fields})
+
+    def __add__(self, other):
+        return self.__class__(
+            **{k: getattr(self, k) for k in by},
+            **{k: object.__getattribute__(self, k)
+                + object.__getattribute__(other, k)
+                for k in fields})
+
+    def __getattribute__(self, k):
+        if k in fields:
+            if object.__getattribute__(self, k):
+                return ops.get(k, OPS['sum'])(object.__getattribute__(self, k))
+            else:
+                return None
+        return object.__getattribute__(self, k)
+
+    return type('Result', (co.namedtuple('Result', by + fields),), {
+        '__slots__': (),
+        '__new__': __new__,
+        '__add__': __add__,
+        '__getattribute__': __getattribute__,
+        '_by': by,
+        '_fields': fields,
+        '_sort': fields,
+        '_types': types_,
+    })
+
+
+def fold(Result, results, *,
+        by=None,
+        defines=None,
+        **_):
+    if by is None:
+        by = Result._by
+
+    for k in it.chain(by or [], (k for k, _ in defines or [])):
+        if k not in Result._by and k not in Result._fields:
+            print("error: could not find field %r?" % k)
+            sys.exit(-1)
+
+    # filter by matching defines
+    if defines is not None:
+        results_ = []
+        for r in results:
+            if all(getattr(r, k) in vs for k, vs in defines):
+                results_.append(r)
+        results = results_
+
+    # organize results into conflicts
+    folding = co.OrderedDict()
+    for r in results:
+        name = tuple(getattr(r, k) for k in by)
+        if name not in folding:
+            folding[name] = []
+        folding[name].append(r)
+
+    # merge conflicts
+    folded = []
+    for name, rs in folding.items():
+        folded.append(sum(rs[1:], start=rs[0]))
+
+    return folded
+
+def table(Result, results, diff_results=None, *,
+        by=None,
+        fields=None,
+        sort=None,
+        summary=False,
+        all=False,
+        percent=False,
+        **_):
+    all_, all = all, __builtins__.all
+
+    if by is None:
+        by = Result._by
+    if fields is None:
+        fields = Result._fields
+    types = Result._types
+
+    # fold again
+    results = fold(Result, results, by=by)
+    if diff_results is not None:
+        diff_results = fold(Result, diff_results, by=by)
+
+    # organize by name
+    table = {
+        ','.join(str(getattr(r, k) or '') for k in by): r
+        for r in results}
+    diff_table = {
+        ','.join(str(getattr(r, k) or '') for k in by): r
+        for r in diff_results or []}
+    names = list(table.keys() | diff_table.keys())
+
+    # sort again, now with diff info, note that python's sort is stable
+    names.sort()
+    if diff_results is not None:
+        names.sort(key=lambda n: tuple(
+            types[k].ratio(
+                getattr(table.get(n), k, None),
+                getattr(diff_table.get(n), k, None))
+            for k in fields),
+            reverse=True)
+    if sort:
+        for k, reverse in reversed(sort):
+            names.sort(
+                key=lambda n: tuple(
+                    (getattr(table[n], k),)
+                    if getattr(table.get(n), k, None) is not None else ()
+                    for k in ([k] if k else [
+                        k for k in Result._sort if k in fields])),
+                reverse=reverse ^ (not k or k in Result._fields))
+
+
+    # build up our lines
+    lines = []
+
+    # header
+    header = []
+    header.append('%s%s' % (
+        ','.join(by),
+        ' (%d added, %d removed)' % (
+            sum(1 for n in table if n not in diff_table),
+            sum(1 for n in diff_table if n not in table))
+            if diff_results is not None and not percent else '')
+        if not summary else '')
+    if diff_results is None:
+        for k in fields:
+            header.append(k)
+    elif percent:
+        for k in fields:
+            header.append(k)
+    else:
+        for k in fields:
+            header.append('o'+k)
+        for k in fields:
+            header.append('n'+k)
+        for k in fields:
+            header.append('d'+k)
+    header.append('')
+    lines.append(header)
+
+    def table_entry(name, r, diff_r=None, ratios=[]):
+        entry = []
+        entry.append(name)
+        if diff_results is None:
+            for k in fields:
+                entry.append(getattr(r, k).table()
+                    if getattr(r, k, None) is not None
+                    else types[k].none)
+        elif percent:
+            for k in fields:
+                entry.append(getattr(r, k).diff_table()
+                    if getattr(r, k, None) is not None
+                    else types[k].diff_none)
+        else:
+            for k in fields:
+                entry.append(getattr(diff_r, k).diff_table()
+                    if getattr(diff_r, k, None) is not None
+                    else types[k].diff_none)
+            for k in fields:
+                entry.append(getattr(r, k).diff_table()
+                    if getattr(r, k, None) is not None
+                    else types[k].diff_none)
+            for k in fields:
+                entry.append(types[k].diff_diff(
+                        getattr(r, k, None),
+                        getattr(diff_r, k, None)))
+        if diff_results is None:
+            entry.append('')
+        elif percent:
+            entry.append(' (%s)' % ', '.join(
+                '+∞%' if t == +m.inf
+                else '-∞%' if t == -m.inf
+                else '%+.1f%%' % (100*t)
+                for t in ratios))
+        else:
+            entry.append(' (%s)' % ', '.join(
+                    '+∞%' if t == +m.inf
+                    else '-∞%' if t == -m.inf
+                    else '%+.1f%%' % (100*t)
+                    for t in ratios
+                    if t)
+                if any(ratios) else '')
+        return entry
+
+    # entries
+    if not summary:
+        for name in names:
+            r = table.get(name)
+            if diff_results is None:
+                diff_r = None
+                ratios = None
+            else:
+                diff_r = diff_table.get(name)
+                ratios = [
+                    types[k].ratio(
+                        getattr(r, k, None),
+                        getattr(diff_r, k, None))
+                    for k in fields]
+                if not all_ and not any(ratios):
+                    continue
+            lines.append(table_entry(name, r, diff_r, ratios))
+
+    # total
+    r = next(iter(fold(Result, results, by=[])), None)
+    if diff_results is None:
+        diff_r = None
+        ratios = None
+    else:
+        diff_r = next(iter(fold(Result, diff_results, by=[])), None)
+        ratios = [
+            types[k].ratio(
+                getattr(r, k, None),
+                getattr(diff_r, k, None))
+            for k in fields]
+    lines.append(table_entry('TOTAL', r, diff_r, ratios))
+
+    # find the best widths, note that column 0 contains the names and column -1
+    # the ratios, so those are handled a bit differently
+    widths = [
+        ((max(it.chain([w], (len(l[i]) for l in lines)))+1+4-1)//4)*4-1
+        for w, i in zip(
+            it.chain([23], it.repeat(7)),
+            range(len(lines[0])-1))]
+
+    # print our table
+    for line in lines:
+        print('%-*s  %s%s' % (
+            widths[0], line[0],
+            ' '.join('%*s' % (w, x)
+                for w, x in zip(widths[1:], line[1:-1])),
+            line[-1]))
+
+
+def openio(path, mode='r', buffering=-1):
+    # allow '-' for stdin/stdout
+    if path == '-':
+        if mode == 'r':
+            return os.fdopen(os.dup(sys.stdin.fileno()), mode, buffering)
+        else:
+            return os.fdopen(os.dup(sys.stdout.fileno()), mode, buffering)
+    else:
+        return open(path, mode, buffering)
+
+def main(csv_paths, *,
+        by=None,
+        fields=None,
+        defines=None,
+        sort=None,
+        **args):
+    # separate out renames
+    renames = list(it.chain.from_iterable(
+        ((k, v) for v in vs)
+        for k, vs in it.chain(by or [], fields or [])))
+    if by is not None:
+        by = [k for k, _ in by]
+    if fields is not None:
+        fields = [k for k, _ in fields]
+
+    # figure out types
+    types = {}
+    for t in TYPES.keys():
+        for k in args.get(t, []):
+            if k in types:
+                print("error: conflicting type for field %r?" % k)
+                sys.exit(-1)
+            types[k] = TYPES[t]
+    # rename types?
+    if renames:
+        types_ = {}
+        for new_k, old_k in renames:
+            if old_k in types:
+                types_[new_k] = types[old_k]
+        types.update(types_)
+
+    # figure out merge operations
+    ops = {}
+    for o in OPS.keys():
+        for k in args.get(o, []):
+            if k in ops:
+                print("error: conflicting op for field %r?" % k)
+                sys.exit(-1)
+            ops[k] = OPS[o]
+    # rename ops?
+    if renames:
+        ops_ = {}
+        for new_k, old_k in renames:
+            if old_k in ops:
+                ops_[new_k] = ops[old_k]
+        ops.update(ops_)
+
+    # find CSV files
+    results = []
+    for path in csv_paths:
         try:
             with openio(path) as f:
-                r = csv.DictReader(f)
-                for result in r:
-                    file = result.pop('file', '')
-                    name = result.pop('name', '')
-                    prev = results[(file, name)]
-                    for field in FIELDS:
-                        try:
-                            r = field.parse(result)
-                            if field.name in prev:
-                                results[(file, name)][field.name] = field.acc(
-                                    [prev[field.name], r])
-                            else:
-                                results[(file, name)][field.name] = r
-                        except (KeyError, ValueError):
-                            pass
+                reader = csv.DictReader(f, restval='')
+                for r in reader:
+                    # rename fields?
+                    if renames:
+                        # make a copy so renames can overlap
+                        r_ = {}
+                        for new_k, old_k in renames:
+                            if old_k in r:
+                                r_[new_k] = r[old_k]
+                        r.update(r_)
+
+                    results.append(r)
         except FileNotFoundError:
             pass
 
-    # find fields
-    if args.get('all_fields'):
-        fields = FIELDS
-    elif args.get('fields') is not None:
-        fields_dict = {field.name: field for field in FIELDS}
-        fields = [fields_dict[f] for f in args['fields']]
-    else:
-        fields = []
-        for field in FIELDS:
-            if any(field.name in result for result in results.values()):
-                fields.append(field)
+    # homogenize
+    Result = infer(results,
+        by=by,
+        fields=fields,
+        types=types,
+        ops=ops,
+        renames=renames)
+    results_ = []
+    for r in results:
+        if not any(k in r and r[k].strip()
+                for k in Result._fields):
+            continue
+        try:
+            results_.append(Result(**{
+                k: r[k] for k in Result._by + Result._fields
+                if k in r and r[k].strip()}))
+        except TypeError:
+            pass
+    results = results_
 
-    # find total for every field
-    total = {}
-    for result in results.values():
-        for field in fields:
-            if field.name in result and field.name in total:
-                total[field.name] = field.acc(
-                    [total[field.name], result[field.name]])
-            elif field.name in result:
-                total[field.name] = result[field.name]
+    # fold
+    results = fold(Result, results, by=by, defines=defines)
+
+    # sort, note that python's sort is stable
+    results.sort()
+    if sort:
+        for k, reverse in reversed(sort):
+            results.sort(
+                key=lambda r: tuple(
+                    (getattr(r, k),) if getattr(r, k) is not None else ()
+                    for k in ([k] if k else Result._sort)),
+                reverse=reverse ^ (not k or k in Result._fields))
+
+    # write results to CSV
+    if args.get('output'):
+        with openio(args['output'], 'w') as f:
+            writer = csv.DictWriter(f, Result._by + Result._fields)
+            writer.writeheader()
+            for r in results:
+                # note we need to go through getattr to resolve lazy fields
+                writer.writerow({
+                    k: getattr(r, k) for k in Result._by + Result._fields})
 
     # find previous results?
     if args.get('diff'):
-        prev_results = co.defaultdict(lambda: {})
+        diff_results = []
         try:
             with openio(args['diff']) as f:
-                r = csv.DictReader(f)
-                for result in r:
-                    file = result.pop('file', '')
-                    name = result.pop('name', '')
-                    prev = prev_results[(file, name)]
-                    for field in FIELDS:
-                        try:
-                            r = field.parse(result)
-                            if field.name in prev:
-                                prev_results[(file, name)][field.name] = field.acc(
-                                    [prev[field.name], r])
-                            else:
-                                prev_results[(file, name)][field.name] = r
-                        except (KeyError, ValueError):
-                            pass
+                reader = csv.DictReader(f, restval='')
+                for r in reader:
+                    # rename fields?
+                    if renames:
+                        # make a copy so renames can overlap
+                        r_ = {}
+                        for new_k, old_k in renames:
+                            if old_k in r:
+                                r_[new_k] = r[old_k]
+                        r.update(r_)
+
+                    if not any(k in r and r[k].strip()
+                            for k in Result._fields):
+                        continue
+                    try:
+                        diff_results.append(Result(**{
+                            k: r[k] for k in Result._by + Result._fields
+                            if k in r and r[k].strip()}))
+                    except TypeError:
+                        pass
         except FileNotFoundError:
             pass
 
-        prev_total = {}
-        for result in prev_results.values():
-            for field in fields:
-                if field.name in result and field.name in prev_total:
-                    prev_total[field.name] = field.acc(
-                        [prev_total[field.name], result[field.name]])
-                elif field.name in result:
-                    prev_total[field.name] = result[field.name]
+        # fold
+        diff_results = fold(Result, diff_results, by=by, defines=defines)
 
-    # print results
-    def dedup_entries(results, by='name'):
-        entries = co.defaultdict(lambda: {})
-        for (file, func), result in results.items():
-            entry = (file if by == 'file' else func)
-            prev = entries[entry]
-            for field in fields:
-                if field.name in result and field.name in prev:
-                    entries[entry][field.name] = field.acc(
-                        [prev[field.name], result[field.name]])
-                elif field.name in result:
-                    entries[entry][field.name] = result[field.name]
-        return entries
-
-    def sorted_entries(entries):
-        if args.get('sort') is not None:
-            field = {field.name: field for field in FIELDS}[args['sort']]
-            return sorted(entries, key=lambda x: (
-                -(field.key(x[1][field.name])) if field.name in x[1] else -1, x))
-        elif args.get('reverse_sort') is not None:
-            field = {field.name: field for field in FIELDS}[args['reverse_sort']]
-            return sorted(entries, key=lambda x: (
-                +(field.key(x[1][field.name])) if field.name in x[1] else -1, x))
-        else:
-            return sorted(entries)
-
-    def print_header(by=''):
-        if not args.get('diff'):
-            print('%-36s' % by, end='')
-            for field in fields:
-                print((' '+field.fmt) % field.name, end='')
-            print()
-        else:
-            print('%-36s' % by, end='')
-            for field in fields:
-                print((' '+field.fmt) % field.name, end='')
-                print(' %-9s' % '', end='')
-            print()
-
-    def print_entry(name, result):
-        print('%-36s' % name, end='')
-        for field in fields:
-            r = result.get(field.name)
-            if r is not None:
-                print((' '+field.fmt) % field.repr(r), end='')
-            else:
-                print((' '+field.fmt) % '-', end='')
-        print()
-
-    def print_diff_entry(name, old, new):
-        print('%-36s' % name, end='')
-        for field in fields:
-            n = new.get(field.name)
-            if n is not None:
-                print((' '+field.fmt) % field.repr(n), end='')
-            else:
-                print((' '+field.fmt) % '-', end='')
-            o = old.get(field.name)
-            ratio = (
-                0.0 if m.isinf(o or 0) and m.isinf(n or 0)
-                    else +float('inf') if m.isinf(n or 0)
-                    else -float('inf') if m.isinf(o or 0)
-                    else 0.0 if not o and not n
-                    else +1.0 if not o
-                    else -1.0 if not n
-                    else field.ratio(o, n))
-            print(' %-9s' % (
-                '' if not ratio
-                    else '(+∞%)' if ratio > 0 and m.isinf(ratio)
-                    else '(-∞%)' if ratio < 0 and m.isinf(ratio)
-                    else '(%+.1f%%)' % (100*ratio)), end='')
-        print()
-
-    def print_entries(by='name'):
-        entries = dedup_entries(results, by=by)
-
-        if not args.get('diff'):
-            print_header(by=by)
-            for name, result in sorted_entries(entries.items()):
-                print_entry(name, result)
-        else:
-            prev_entries = dedup_entries(prev_results, by=by)
-            print_header(by='%s (%d added, %d removed)' % (by,
-                sum(1 for name in entries if name not in prev_entries),
-                sum(1 for name in prev_entries if name not in entries)))
-            for name, result in sorted_entries(entries.items()):
-                if args.get('all') or result != prev_entries.get(name, {}):
-                    print_diff_entry(name, prev_entries.get(name, {}), result)
-
-    def print_totals():
-        if not args.get('diff'):
-            print_entry('TOTAL', total)
-        else:
-            print_diff_entry('TOTAL', prev_total, total)
-
-    if args.get('summary'):
-        print_header()
-        print_totals()
-    elif args.get('files'):
-        print_entries(by='file')
-        print_totals()
-    else:
-        print_entries(by='name')
-        print_totals()
+    # print table
+    if not args.get('quiet'):
+        table(Result, results,
+            diff_results if args.get('diff') else None,
+            by=by,
+            fields=fields,
+            sort=sort,
+            **args)
 
 
 if __name__ == "__main__":
     import argparse
     import sys
     parser = argparse.ArgumentParser(
-        description="Summarize measurements")
-    parser.add_argument('csv_paths', nargs='*', default='-',
-        help="Description of where to find *.csv files. May be a directory \
-            or list of paths. *.csv files will be merged to show the total \
-            coverage.")
-    parser.add_argument('-d', '--diff',
+        description="Summarize measurements in CSV files.",
+        allow_abbrev=False)
+    parser.add_argument(
+        'csv_paths',
+        nargs='*',
+        help="Input *.csv files.")
+    parser.add_argument(
+        '-q', '--quiet',
+        action='store_true',
+        help="Don't show anything, useful with -o.")
+    parser.add_argument(
+        '-o', '--output',
+        help="Specify CSV file to store results.")
+    parser.add_argument(
+        '-d', '--diff',
         help="Specify CSV file to diff against.")
-    parser.add_argument('-a', '--all', action='store_true',
-        help="Show all objects, not just the ones that changed.")
-    parser.add_argument('-e', '--all-fields', action='store_true',
-        help="Show all fields, even those with no results.")
-    parser.add_argument('-f', '--fields', type=lambda x: re.split('\s*,\s*', x),
-        help="Comma separated list of fields to print, by default all fields \
-            that are found in the CSV files are printed.")
-    parser.add_argument('-s', '--sort',
+    parser.add_argument(
+        '-a', '--all',
+        action='store_true',
+        help="Show all, not just the ones that changed.")
+    parser.add_argument(
+        '-p', '--percent',
+        action='store_true',
+        help="Only show percentage change, not a full diff.")
+    parser.add_argument(
+        '-b', '--by',
+        action='append',
+        type=lambda x: (
+            lambda k,v=None: (k, v.split(',') if v is not None else ())
+            )(*x.split('=', 1)),
+        help="Group by this field. Can rename fields with new_name=old_name.")
+    parser.add_argument(
+        '-f', '--field',
+        dest='fields',
+        action='append',
+        type=lambda x: (
+            lambda k,v=None: (k, v.split(',') if v is not None else ())
+            )(*x.split('=', 1)),
+        help="Show this field. Can rename fields with new_name=old_name.")
+    parser.add_argument(
+        '-D', '--define',
+        dest='defines',
+        action='append',
+        type=lambda x: (lambda k,v: (k, set(v.split(','))))(*x.split('=', 1)),
+        help="Only include results where this field is this value. May include "
+            "comma-separated options.")
+    class AppendSort(argparse.Action):
+        def __call__(self, parser, namespace, value, option):
+            if namespace.sort is None:
+                namespace.sort = []
+            namespace.sort.append((value, True if option == '-S' else False))
+    parser.add_argument(
+        '-s', '--sort',
+        nargs='?',
+        action=AppendSort,
         help="Sort by this field.")
-    parser.add_argument('-S', '--reverse-sort',
+    parser.add_argument(
+        '-S', '--reverse-sort',
+        nargs='?',
+        action=AppendSort,
         help="Sort by this field, but backwards.")
-    parser.add_argument('-F', '--files', action='store_true',
-        help="Show file-level calls.")
-    parser.add_argument('-Y', '--summary', action='store_true',
-        help="Only show the totals.")
-    sys.exit(main(**vars(parser.parse_args())))
+    parser.add_argument(
+        '-Y', '--summary',
+        action='store_true',
+        help="Only show the total.")
+    parser.add_argument(
+        '--int',
+        action='append',
+        help="Treat these fields as ints.")
+    parser.add_argument(
+        '--float',
+        action='append',
+        help="Treat these fields as floats.")
+    parser.add_argument(
+        '--frac',
+        action='append',
+        help="Treat these fields as fractions.")
+    parser.add_argument(
+        '--sum',
+        action='append',
+        help="Add these fields (the default).")
+    parser.add_argument(
+        '--prod',
+        action='append',
+        help="Multiply these fields.")
+    parser.add_argument(
+        '--min',
+        action='append',
+        help="Take the minimum of these fields.")
+    parser.add_argument(
+        '--max',
+        action='append',
+        help="Take the maximum of these fields.")
+    parser.add_argument(
+        '--mean',
+        action='append',
+        help="Average these fields.")
+    parser.add_argument(
+        '--stddev',
+        action='append',
+        help="Find the standard deviation of these fields.")
+    parser.add_argument(
+        '--gmean',
+        action='append',
+        help="Find the geometric mean of these fields.")
+    parser.add_argument(
+        '--gstddev',
+        action='append',
+        help="Find the geometric standard deviation of these fields.")
+    sys.exit(main(**{k: v
+        for k, v in vars(parser.parse_intermixed_args()).items()
+        if v is not None}))
diff --git a/scripts/tailpipe.py b/scripts/tailpipe.py
new file mode 100644
index 0000000..802f74d
--- /dev/null
+++ b/scripts/tailpipe.py
@@ -0,0 +1,177 @@
+#!/usr/bin/env python3
+#
+# Efficiently displays the last n lines of a file/pipe.
+#
+# Example:
+# ./scripts/tailpipe.py trace -n5
+#
+# Copyright (c) 2022, The littlefs authors.
+# SPDX-License-Identifier: BSD-3-Clause
+#
+
+import collections as co
+import io
+import os
+import select
+import shutil
+import sys
+import threading as th
+import time
+
+
+def openio(path, mode='r', buffering=-1):
+    # allow '-' for stdin/stdout
+    if path == '-':
+        if mode == 'r':
+            return os.fdopen(os.dup(sys.stdin.fileno()), mode, buffering)
+        else:
+            return os.fdopen(os.dup(sys.stdout.fileno()), mode, buffering)
+    else:
+        return open(path, mode, buffering)
+
+class LinesIO:
+    def __init__(self, maxlen=None):
+        self.maxlen = maxlen
+        self.lines = co.deque(maxlen=maxlen)
+        self.tail = io.StringIO()
+
+        # trigger automatic sizing
+        if maxlen == 0:
+            self.resize(0)
+
+    def write(self, s):
+        # note using split here ensures the trailing string has no newline
+        lines = s.split('\n')
+
+        if len(lines) > 1 and self.tail.getvalue():
+            self.tail.write(lines[0])
+            lines[0] = self.tail.getvalue()
+            self.tail = io.StringIO()
+
+        self.lines.extend(lines[:-1])
+
+        if lines[-1]:
+            self.tail.write(lines[-1])
+
+    def resize(self, maxlen):
+        self.maxlen = maxlen
+        if maxlen == 0:
+            maxlen = shutil.get_terminal_size((80, 5))[1]
+        if maxlen != self.lines.maxlen:
+            self.lines = co.deque(self.lines, maxlen=maxlen)
+
+    canvas_lines = 1
+    def draw(self):
+        # did terminal size change?
+        if self.maxlen == 0:
+            self.resize(0)
+
+        # first thing first, give ourself a canvas
+        while LinesIO.canvas_lines < len(self.lines):
+            sys.stdout.write('\n')
+            LinesIO.canvas_lines += 1
+
+        # clear the bottom of the canvas if we shrink
+        shrink = LinesIO.canvas_lines - len(self.lines)
+        if shrink > 0:
+            for i in range(shrink):
+                sys.stdout.write('\r')
+                if shrink-1-i > 0:
+                    sys.stdout.write('\x1b[%dA' % (shrink-1-i))
+                sys.stdout.write('\x1b[K')
+                if shrink-1-i > 0:
+                    sys.stdout.write('\x1b[%dB' % (shrink-1-i))
+            sys.stdout.write('\x1b[%dA' % shrink)
+            LinesIO.canvas_lines = len(self.lines)
+
+        for i, line in enumerate(self.lines):
+            # move cursor, clear line, disable/reenable line wrapping
+            sys.stdout.write('\r')
+            if len(self.lines)-1-i > 0:
+                sys.stdout.write('\x1b[%dA' % (len(self.lines)-1-i))
+            sys.stdout.write('\x1b[K')
+            sys.stdout.write('\x1b[?7l')
+            sys.stdout.write(line)
+            sys.stdout.write('\x1b[?7h')
+            if len(self.lines)-1-i > 0:
+                sys.stdout.write('\x1b[%dB' % (len(self.lines)-1-i))
+        sys.stdout.flush()
+
+
+def main(path='-', *, lines=5, cat=False, sleep=None, keep_open=False):
+    if cat:
+        ring = sys.stdout
+    else:
+        ring = LinesIO(lines)
+
+    # if sleep print in background thread to avoid getting stuck in a read call
+    event = th.Event()
+    lock = th.Lock()
+    if not cat:
+        done = False
+        def background():
+            while not done:
+                event.wait()
+                event.clear()
+                with lock:
+                    ring.draw()
+                time.sleep(sleep or 0.01)
+        th.Thread(target=background, daemon=True).start()
+
+    try:
+        while True:
+            with openio(path) as f:
+                for line in f:
+                    with lock:
+                        ring.write(line)
+                        event.set()
+
+            if not keep_open:
+                break
+            # don't just flood open calls
+            time.sleep(sleep or 0.1)
+    except FileNotFoundError as e:
+        print("error: file not found %r" % path)
+        sys.exit(-1)
+    except KeyboardInterrupt:
+        pass
+
+    if not cat:
+        done = True
+        lock.acquire() # avoids https://bugs.python.org/issue42717
+        sys.stdout.write('\n')
+
+
+if __name__ == "__main__":
+    import sys
+    import argparse
+    parser = argparse.ArgumentParser(
+        description="Efficiently displays the last n lines of a file/pipe.",
+        allow_abbrev=False)
+    parser.add_argument(
+        'path',
+        nargs='?',
+        help="Path to read from.")
+    parser.add_argument(
+        '-n', '--lines',
+        nargs='?',
+        type=lambda x: int(x, 0),
+        const=0,
+        help="Show this many lines of history. 0 uses the terminal height. "
+            "Defaults to 5.")
+    parser.add_argument(
+        '-z', '--cat',
+        action='store_true',
+        help="Pipe directly to stdout.")
+    parser.add_argument(
+        '-s', '--sleep',
+        type=float,
+        help="Seconds to sleep between reads. Defaults to 0.01.")
+    parser.add_argument(
+        '-k', '--keep-open',
+        action='store_true',
+        help="Reopen the pipe on EOF, useful when multiple "
+            "processes are writing.")
+    sys.exit(main(**{k: v
+        for k, v in vars(parser.parse_intermixed_args()).items()
+        if v is not None}))
diff --git a/scripts/teepipe.py b/scripts/teepipe.py
new file mode 100644
index 0000000..ee32e44
--- /dev/null
+++ b/scripts/teepipe.py
@@ -0,0 +1,73 @@
+#!/usr/bin/env python3
+#
+# tee, but for pipes
+#
+# Example:
+# ./scripts/tee.py in_pipe out_pipe1 out_pipe2
+#
+# Copyright (c) 2022, The littlefs authors.
+# SPDX-License-Identifier: BSD-3-Clause
+#
+
+import os
+import io
+import time
+import sys
+
+
+def openio(path, mode='r', buffering=-1):
+    # allow '-' for stdin/stdout
+    if path == '-':
+        if mode == 'r':
+            return os.fdopen(os.dup(sys.stdin.fileno()), mode, buffering)
+        else:
+            return os.fdopen(os.dup(sys.stdout.fileno()), mode, buffering)
+    else:
+        return open(path, mode, buffering)
+
+def main(in_path, out_paths, *, keep_open=False):
+    out_pipes = [openio(p, 'wb', 0) for p in out_paths]
+    try:
+        with openio(in_path, 'rb', 0) as f:
+            while True:
+                buf = f.read(io.DEFAULT_BUFFER_SIZE)
+                if not buf:
+                    if not keep_open:
+                        break
+                    # don't just flood reads
+                    time.sleep(0.1)
+                    continue
+
+                for p in out_pipes:
+                    try:
+                        p.write(buf)
+                    except BrokenPipeError:
+                        pass
+    except FileNotFoundError as e:
+        print("error: file not found %r" % in_path)
+        sys.exit(-1)
+    except KeyboardInterrupt:
+        pass
+
+
+if __name__ == "__main__":
+    import sys
+    import argparse
+    parser = argparse.ArgumentParser(
+        description="tee, but for pipes.",
+        allow_abbrev=False)
+    parser.add_argument(
+        'in_path',
+        help="Path to read from.")
+    parser.add_argument(
+        'out_paths',
+        nargs='+',
+        help="Path to write to.")
+    parser.add_argument(
+        '-k', '--keep-open',
+        action='store_true',
+        help="Reopen the pipe on EOF, useful when multiple "
+            "processes are writing.")
+    sys.exit(main(**{k: v
+        for k, v in vars(parser.parse_intermixed_args()).items()
+        if v is not None}))
diff --git a/scripts/test.py b/scripts/test.py
index 92a13b1..6e8a201 100755
--- a/scripts/test.py
+++ b/scripts/test.py
@@ -1,860 +1,1484 @@
 #!/usr/bin/env python3
-
-# This script manages littlefs tests, which are configured with
-# .toml files stored in the tests directory.
+#
+# Script to compile and runs tests.
+#
+# Example:
+# ./scripts/test.py runners/test_runner -b
+#
+# Copyright (c) 2022, The littlefs authors.
+# SPDX-License-Identifier: BSD-3-Clause
 #
 
-import toml
-import glob
-import re
-import os
-import io
-import itertools as it
-import collections.abc as abc
-import subprocess as sp
-import base64
-import sys
-import copy
-import shlex
-import pty
+import collections as co
+import csv
 import errno
+import glob
+import itertools as it
+import math as m
+import os
+import pty
+import re
+import shlex
+import shutil
 import signal
+import subprocess as sp
+import threading as th
+import time
+import toml
 
-TEST_PATHS = 'tests'
-RULES = """
-# add block devices to sources
-TESTSRC ?= $(SRC) $(wildcard bd/*.c)
 
-define FLATTEN
-%(path)s%%$(subst /,.,$(target)): $(target)
-    ./scripts/explode_asserts.py $$< -o $$@
-endef
-$(foreach target,$(TESTSRC),$(eval $(FLATTEN)))
+RUNNER_PATH = './runners/test_runner'
+HEADER_PATH = 'runners/test_runner.h'
 
--include %(path)s*.d
-.SECONDARY:
+GDB_PATH = ['gdb']
+VALGRIND_PATH = ['valgrind']
+PERF_SCRIPT = ['./scripts/perf.py']
 
-%(path)s.test: %(path)s.test.o \\
-        $(foreach t,$(subst /,.,$(TESTSRC:.c=.o)),%(path)s.$t)
-    $(CC) $(CFLAGS) $^ $(LFLAGS) -o $@
 
-# needed in case builddir is different
-%(path)s%%.o: %(path)s%%.c
-    $(CC) -c -MMD $(CFLAGS) $< -o $@
-"""
-COVERAGE_RULES = """
-%(path)s.test: override CFLAGS += -fprofile-arcs -ftest-coverage
-
-# delete lingering coverage
-%(path)s.test: | %(path)s.info.clean
-.PHONY: %(path)s.info.clean
-%(path)s.info.clean:
-    rm -f %(path)s*.gcda
-
-# accumulate coverage info
-.PHONY: %(path)s.info
-%(path)s.info:
-    $(strip $(LCOV) -c \\
-        $(addprefix -d ,$(wildcard %(path)s*.gcda)) \\
-        --rc 'geninfo_adjust_src_path=$(shell pwd)' \\
-        -o $@)
-    $(LCOV) -e $@ $(addprefix /,$(SRC)) -o $@
-ifdef COVERAGETARGET
-    $(strip $(LCOV) -a $@ \\
-        $(addprefix -a ,$(wildcard $(COVERAGETARGET))) \\
-        -o $(COVERAGETARGET))
-endif
-"""
-GLOBALS = """
-//////////////// AUTOGENERATED TEST ////////////////
-#include "lfs.h"
-#include "bd/lfs_testbd.h"
-#include <stdio.h>
-extern const char *lfs_testbd_path;
-extern uint32_t lfs_testbd_cycles;
-"""
-DEFINES = {
-    'LFS_READ_SIZE': 16,
-    'LFS_PROG_SIZE': 'LFS_READ_SIZE',
-    'LFS_BLOCK_SIZE': 512,
-    'LFS_BLOCK_COUNT': 1024,
-    'LFS_BLOCK_CYCLES': -1,
-    'LFS_CACHE_SIZE': '(64 % LFS_PROG_SIZE == 0 ? 64 : LFS_PROG_SIZE)',
-    'LFS_LOOKAHEAD_SIZE': 16,
-    'LFS_ERASE_VALUE': 0xff,
-    'LFS_ERASE_CYCLES': 0,
-    'LFS_BADBLOCK_BEHAVIOR': 'LFS_TESTBD_BADBLOCK_PROGERROR',
-}
-PROLOGUE = """
-    // prologue
-    __attribute__((unused)) lfs_t lfs;
-    __attribute__((unused)) lfs_testbd_t bd;
-    __attribute__((unused)) lfs_file_t file;
-    __attribute__((unused)) lfs_dir_t dir;
-    __attribute__((unused)) struct lfs_info info;
-    __attribute__((unused)) char path[1024];
-    __attribute__((unused)) uint8_t buffer[1024];
-    __attribute__((unused)) lfs_size_t size;
-    __attribute__((unused)) int err;
-    
-    __attribute__((unused)) const struct lfs_config cfg = {
-        .context        = &bd,
-        .read           = lfs_testbd_read,
-        .prog           = lfs_testbd_prog,
-        .erase          = lfs_testbd_erase,
-        .sync           = lfs_testbd_sync,
-        .read_size      = LFS_READ_SIZE,
-        .prog_size      = LFS_PROG_SIZE,
-        .block_size     = LFS_BLOCK_SIZE,
-        .block_count    = LFS_BLOCK_COUNT,
-        .block_cycles   = LFS_BLOCK_CYCLES,
-        .cache_size     = LFS_CACHE_SIZE,
-        .lookahead_size = LFS_LOOKAHEAD_SIZE,
-    };
-
-    __attribute__((unused)) const struct lfs_testbd_config bdcfg = {
-        .erase_value        = LFS_ERASE_VALUE,
-        .erase_cycles       = LFS_ERASE_CYCLES,
-        .badblock_behavior  = LFS_BADBLOCK_BEHAVIOR,
-        .power_cycles       = lfs_testbd_cycles,
-    };
-
-    lfs_testbd_createcfg(&cfg, lfs_testbd_path, &bdcfg) => 0;
-"""
-EPILOGUE = """
-    // epilogue
-    lfs_testbd_destroy(&cfg) => 0;
-"""
-PASS = '\033[32m✓\033[0m'
-FAIL = '\033[31m✗\033[0m'
-
-class TestFailure(Exception):
-    def __init__(self, case, returncode=None, stdout=None, assert_=None):
-        self.case = case
-        self.returncode = returncode
-        self.stdout = stdout
-        self.assert_ = assert_
+def openio(path, mode='r', buffering=-1):
+    # allow '-' for stdin/stdout
+    if path == '-':
+        if mode == 'r':
+            return os.fdopen(os.dup(sys.stdin.fileno()), mode, buffering)
+        else:
+            return os.fdopen(os.dup(sys.stdout.fileno()), mode, buffering)
+    else:
+        return open(path, mode, buffering)
 
 class TestCase:
-    def __init__(self, config, filter=filter,
-            suite=None, caseno=None, lineno=None, **_):
-        self.config = config
-        self.filter = filter
-        self.suite = suite
-        self.caseno = caseno
-        self.lineno = lineno
+    # create a TestCase object from a config
+    def __init__(self, config, args={}):
+        self.name = config.pop('name')
+        self.path = config.pop('path')
+        self.suite = config.pop('suite')
+        self.lineno = config.pop('lineno', None)
+        self.if_ = config.pop('if', None)
+        if isinstance(self.if_, bool):
+            self.if_ = 'true' if self.if_ else 'false'
+        self.code = config.pop('code')
+        self.code_lineno = config.pop('code_lineno', None)
+        self.in_ = config.pop('in',
+            config.pop('suite_in', None))
 
-        self.code = config['code']
-        self.code_lineno = config['code_lineno']
-        self.defines = config.get('define', {})
-        self.if_ = config.get('if', None)
-        self.in_ = config.get('in', None)
+        self.reentrant = config.pop('reentrant',
+            config.pop('suite_reentrant', False))
 
-        self.result = None
+        # figure out defines and build possible permutations
+        self.defines = set()
+        self.permutations = []
 
-    def __str__(self):
-        if hasattr(self, 'permno'):
-            if any(k not in self.case.defines for k in self.defines):
-                return '%s#%d#%d (%s)' % (
-                    self.suite.name, self.caseno, self.permno, ', '.join(
-                        '%s=%s' % (k, v) for k, v in self.defines.items()
-                        if k not in self.case.defines))
+        # defines can be a dict or a list or dicts
+        suite_defines = config.pop('suite_defines', {})
+        if not isinstance(suite_defines, list):
+            suite_defines = [suite_defines]
+        defines = config.pop('defines', {})
+        if not isinstance(defines, list):
+            defines = [defines]
+
+        def csplit(v):
+            # split commas but only outside of parens
+            parens = 0
+            i_ = 0
+            for i in range(len(v)):
+                if v[i] == ',' and parens == 0:
+                    yield v[i_:i]
+                    i_ = i+1
+                elif v[i] in '([{':
+                    parens += 1
+                elif v[i] in '}])':
+                    parens -= 1
+            if v[i_:].strip():
+                yield v[i_:]
+
+        def parse_define(v):
+            # a define entry can be a list
+            if isinstance(v, list):
+                for v_ in v:
+                    yield from parse_define(v_)
+            # or a string
+            elif isinstance(v, str):
+                # which can be comma-separated values, with optional
+                # range statements. This matches the runtime define parser in
+                # the runner itself.
+                for v_ in csplit(v):
+                    m = re.search(r'\brange\b\s*\('
+                        '(?P<start>[^,\s]*)'
+                        '\s*(?:,\s*(?P<stop>[^,\s]*)'
+                        '\s*(?:,\s*(?P<step>[^,\s]*)\s*)?)?\)',
+                        v_)
+                    if m:
+                        start = (int(m.group('start'), 0)
+                            if m.group('start') else 0)
+                        stop = (int(m.group('stop'), 0)
+                            if m.group('stop') else None)
+                        step = (int(m.group('step'), 0)
+                            if m.group('step') else 1)
+                        if m.lastindex <= 1:
+                            start, stop = 0, start
+                        for x in range(start, stop, step):
+                            yield from parse_define('%s(%d)%s' % (
+                                v_[:m.start()], x, v_[m.end():]))
+                    else:
+                        yield v_
+            # or a literal value
+            elif isinstance(v, bool):
+                yield 'true' if v else 'false'
             else:
-                return '%s#%d#%d' % (
-                    self.suite.name, self.caseno, self.permno)
-        else:
-            return '%s#%d' % (
-                self.suite.name, self.caseno)
+                yield v
 
-    def permute(self, class_=None, defines={}, permno=None, **_):
-        ncase = (class_ or type(self))(self.config)
-        for k, v in self.__dict__.items():
-            setattr(ncase, k, v)
-        ncase.case = self
-        ncase.perms = [ncase]
-        ncase.permno = permno
-        ncase.defines = defines
-        return ncase
+        # build possible permutations
+        for suite_defines_ in suite_defines:
+            self.defines |= suite_defines_.keys()
+            for defines_ in defines:
+                self.defines |= defines_.keys()
+                self.permutations.extend(dict(perm) for perm in it.product(*(
+                    [(k, v) for v in parse_define(vs)]
+                    for k, vs in sorted((suite_defines_ | defines_).items()))))
 
-    def build(self, f, **_):
-        # prologue
-        for k, v in sorted(self.defines.items()):
-            if k not in self.suite.defines:
-                f.write('#define %s %s\n' % (k, v))
+        for k in config.keys():
+            print('%swarning:%s in %s, found unused key %r' % (
+                '\x1b[01;33m' if args['color'] else '',
+                '\x1b[m' if args['color'] else '',
+                self.name,
+                k),
+                file=sys.stderr)
 
-        f.write('void test_case%d(%s) {' % (self.caseno, ','.join(
-            '\n'+8*' '+'__attribute__((unused)) intmax_t %s' % k
-            for k in sorted(self.perms[0].defines)
-            if k not in self.defines)))
-
-        f.write(PROLOGUE)
-        f.write('\n')
-        f.write(4*' '+'// test case %d\n' % self.caseno)
-        f.write(4*' '+'#line %d "%s"\n' % (self.code_lineno, self.suite.path))
-
-        # test case goes here
-        f.write(self.code)
-
-        # epilogue
-        f.write(EPILOGUE)
-        f.write('}\n')
-
-        for k, v in sorted(self.defines.items()):
-            if k not in self.suite.defines:
-                f.write('#undef %s\n' % k)
-
-    def shouldtest(self, **args):
-        if (self.filter is not None and
-                len(self.filter) >= 1 and
-                self.filter[0] != self.caseno):
-            return False
-        elif (self.filter is not None and
-                len(self.filter) >= 2 and
-                self.filter[1] != self.permno):
-            return False
-        elif args.get('no_internal') and self.in_ is not None:
-            return False
-        elif self.if_ is not None:
-            if_ = self.if_
-            while True:
-                for k, v in sorted(self.defines.items(),
-                        key=lambda x: len(x[0]), reverse=True):
-                    if k in if_:
-                        if_ = if_.replace(k, '(%s)' % v)
-                        break
-                else:
-                    break
-            if_ = (
-                re.sub('(\&\&|\?)', ' and ',
-                re.sub('(\|\||:)', ' or ',
-                re.sub('!(?!=)', ' not ', if_))))
-            return eval(if_)
-        else:
-            return True
-
-    def test(self, exec=[], persist=False, cycles=None,
-            gdb=False, failure=None, disk=None, **args):
-        # build command
-        cmd = exec + ['./%s.test' % self.suite.path,
-            repr(self.caseno), repr(self.permno)]
-
-        # persist disk or keep in RAM for speed?
-        if persist:
-            if not disk:
-                disk = self.suite.path + '.disk'
-            if persist != 'noerase':
-                try:
-                    with open(disk, 'w') as f:
-                        f.truncate(0)
-                    if args.get('verbose'):
-                        print('truncate --size=0', disk)
-                except FileNotFoundError:
-                    pass
-
-            cmd.append(disk)
-
-        # simulate power-loss after n cycles?
-        if cycles:
-            cmd.append(str(cycles))
-
-        # failed? drop into debugger?
-        if gdb and failure:
-            ncmd = ['gdb']
-            if gdb == 'assert':
-                ncmd.extend(['-ex', 'r'])
-                if failure.assert_:
-                    ncmd.extend(['-ex', 'up 2'])
-            elif gdb == 'main':
-                ncmd.extend([
-                    '-ex', 'b %s:%d' % (self.suite.path, self.code_lineno),
-                    '-ex', 'r'])
-            ncmd.extend(['--args'] + cmd)
-
-            if args.get('verbose'):
-                print(' '.join(shlex.quote(c) for c in ncmd))
-            signal.signal(signal.SIGINT, signal.SIG_IGN)
-            sys.exit(sp.call(ncmd))
-
-        # run test case!
-        mpty, spty = pty.openpty()
-        if args.get('verbose'):
-            print(' '.join(shlex.quote(c) for c in cmd))
-        proc = sp.Popen(cmd, stdout=spty, stderr=spty)
-        os.close(spty)
-        mpty = os.fdopen(mpty, 'r', 1)
-        stdout = []
-        assert_ = None
-        try:
-            while True:
-                try:
-                    line = mpty.readline()
-                except OSError as e:
-                    if e.errno == errno.EIO:
-                        break
-                    raise
-                if not line:
-                    break;
-                stdout.append(line)
-                if args.get('verbose'):
-                    sys.stdout.write(line)
-                # intercept asserts
-                m = re.match(
-                    '^{0}([^:]+):(\d+):(?:\d+:)?{0}{1}:{0}(.*)$'
-                    .format('(?:\033\[[\d;]*.| )*', 'assert'),
-                    line)
-                if m and assert_ is None:
-                    try:
-                        with open(m.group(1)) as f:
-                            lineno = int(m.group(2))
-                            line = (next(it.islice(f, lineno-1, None))
-                                .strip('\n'))
-                        assert_ = {
-                            'path': m.group(1),
-                            'line': line,
-                            'lineno': lineno,
-                            'message': m.group(3)}
-                    except:
-                        pass
-        except KeyboardInterrupt:
-            raise TestFailure(self, 1, stdout, None)
-        proc.wait()
-
-        # did we pass?
-        if proc.returncode != 0:
-            raise TestFailure(self, proc.returncode, stdout, assert_)
-        else:
-            return PASS
-
-class ValgrindTestCase(TestCase):
-    def __init__(self, config, **args):
-        self.leaky = config.get('leaky', False)
-        super().__init__(config, **args)
-
-    def shouldtest(self, **args):
-        return not self.leaky and super().shouldtest(**args)
-
-    def test(self, exec=[], **args):
-        verbose = args.get('verbose')
-        uninit = (self.defines.get('LFS_ERASE_VALUE', None) == -1)
-        exec = [
-            'valgrind',
-            '--leak-check=full',
-            ] + (['--undef-value-errors=no'] if uninit else []) + [
-            ] + (['--track-origins=yes'] if not uninit else []) + [
-            '--error-exitcode=4',
-            '--error-limit=no',
-            ] + (['--num-callers=1'] if not verbose else []) + [
-            '-q'] + exec
-        return super().test(exec=exec, **args)
-
-class ReentrantTestCase(TestCase):
-    def __init__(self, config, **args):
-        self.reentrant = config.get('reentrant', False)
-        super().__init__(config, **args)
-
-    def shouldtest(self, **args):
-        return self.reentrant and super().shouldtest(**args)
-
-    def test(self, persist=False, gdb=False, failure=None, **args):
-        for cycles in it.count(1):
-            # clear disk first?
-            if cycles == 1 and persist != 'noerase':
-                persist = 'erase'
-            else:
-                persist = 'noerase'
-
-            # exact cycle we should drop into debugger?
-            if gdb and failure and failure.cycleno == cycles:
-                return super().test(gdb=gdb, persist=persist, cycles=cycles,
-                    failure=failure, **args)
-
-            # run tests, but kill the program after prog/erase has
-            # been hit n cycles. We exit with a special return code if the
-            # program has not finished, since this isn't a test failure.
-            try:
-                return super().test(persist=persist, cycles=cycles, **args)
-            except TestFailure as nfailure:
-                if nfailure.returncode == 33:
-                    continue
-                else:
-                    nfailure.cycleno = cycles
-                    raise
 
 class TestSuite:
-    def __init__(self, path, classes=[TestCase], defines={},
-            filter=None, **args):
+    # create a TestSuite object from a toml file
+    def __init__(self, path, args={}):
+        self.path = path
         self.name = os.path.basename(path)
         if self.name.endswith('.toml'):
             self.name = self.name[:-len('.toml')]
-        if args.get('build_dir'):
-            self.toml = path
-            self.path = args['build_dir'] + '/' + path
-        else:
-            self.toml = path
-            self.path = path
-        self.classes = classes
-        self.defines = defines.copy()
-        self.filter = filter
 
-        with open(self.toml) as f:
+        # load toml file and parse test cases
+        with open(self.path) as f:
             # load tests
             config = toml.load(f)
 
             # find line numbers
             f.seek(0)
-            linenos = []
+            case_linenos = []
             code_linenos = []
             for i, line in enumerate(f):
-                if re.match(r'\[\[\s*case\s*\]\]', line):
-                    linenos.append(i+1)
-                if re.match(r'code\s*=\s*(\'\'\'|""")', line):
+                match = re.match(
+                    '(?P<case>\[\s*cases\s*\.\s*(?P<name>\w+)\s*\])'
+                        '|' '(?P<code>code\s*=)',
+                    line)
+                if match and match.group('case'):
+                    case_linenos.append((i+1, match.group('name')))
+                elif match and match.group('code'):
                     code_linenos.append(i+2)
 
-            code_linenos.reverse()
+            # sort in case toml parsing did not retain order
+            case_linenos.sort()
 
-        # grab global config
-        for k, v in config.get('define', {}).items():
-            if k not in self.defines:
-                self.defines[k] = v
-        self.code = config.get('code', None)
-        if self.code is not None:
-            self.code_lineno = code_linenos.pop()
+            cases = config.pop('cases')
+            for (lineno, name), (nlineno, _) in it.zip_longest(
+                    case_linenos, case_linenos[1:],
+                    fillvalue=(float('inf'), None)):
+                code_lineno = min(
+                    (l for l in code_linenos if l >= lineno and l < nlineno),
+                    default=None)
+                cases[name]['lineno'] = lineno
+                cases[name]['code_lineno'] = code_lineno
 
-        # create initial test cases
-        self.cases = []
-        for i, (case, lineno) in enumerate(zip(config['case'], linenos)):
-            # code lineno?
-            if 'code' in case:
-                case['code_lineno'] = code_linenos.pop()
-            # merge conditions if necessary
-            if 'if' in config and 'if' in case:
-                case['if'] = '(%s) && (%s)' % (config['if'], case['if'])
-            elif 'if' in config:
-                case['if'] = config['if']
-            # initialize test case
-            self.cases.append(TestCase(case, filter=filter,
-                suite=self, caseno=i+1, lineno=lineno, **args))
+            self.if_ = config.pop('if', None)
+            if isinstance(self.if_, bool):
+                self.if_ = 'true' if self.if_ else 'false'
 
-    def __str__(self):
-        return self.name
+            self.code = config.pop('code', None)
+            self.code_lineno = min(
+                (l for l in code_linenos
+                    if not case_linenos or l < case_linenos[0][0]),
+                default=None)
 
-    def __lt__(self, other):
-        return self.name < other.name
+            # a couple of these we just forward to all cases
+            defines = config.pop('defines', {})
+            in_ = config.pop('in', None)
+            reentrant = config.pop('reentrant', False)
 
-    def permute(self, **args):
-        for case in self.cases:
-            # lets find all parameterized definitions, in one of [args.D,
-            # suite.defines, case.defines, DEFINES]. Note that each of these
-            # can be either a dict of defines, or a list of dicts, expressing
-            # an initial set of permutations.
-            pending = [{}]
-            for inits in [self.defines, case.defines, DEFINES]:
-                if not isinstance(inits, list):
-                    inits = [inits]
+            self.cases = []
+            for name, case in sorted(cases.items(),
+                    key=lambda c: c[1].get('lineno')):
+                self.cases.append(TestCase(config={
+                    'name': name,
+                    'path': path + (':%d' % case['lineno']
+                        if 'lineno' in case else ''),
+                    'suite': self.name,
+                    'suite_defines': defines,
+                    'suite_in': in_,
+                    'suite_reentrant': reentrant,
+                    **case},
+                    args=args))
 
-                npending = []
-                for init, pinit in it.product(inits, pending):
-                    ninit = pinit.copy()
-                    for k, v in init.items():
-                        if k not in ninit:
-                            try:
-                                ninit[k] = eval(v)
-                            except:
-                                ninit[k] = v
-                    npending.append(ninit)
+            # combine per-case defines
+            self.defines = set.union(*(
+                set(case.defines) for case in self.cases))
 
-                pending = npending
+            # combine other per-case things
+            self.reentrant = any(case.reentrant for case in self.cases)
 
-            # expand permutations
-            pending = list(reversed(pending))
-            expanded = []
-            while pending:
-                perm = pending.pop()
-                for k, v in sorted(perm.items()):
-                    if not isinstance(v, str) and isinstance(v, abc.Iterable):
-                        for nv in reversed(v):
-                            nperm = perm.copy()
-                            nperm[k] = nv
-                            pending.append(nperm)
-                        break
-                else:
-                    expanded.append(perm)
+        for k in config.keys():
+            print('%swarning:%s in %s, found unused key %r' % (
+                '\x1b[01;33m' if args['color'] else '',
+                '\x1b[m' if args['color'] else '',
+                self.name,
+                k),
+                file=sys.stderr)
 
-            # generate permutations
-            case.perms = []
-            for i, (class_, defines) in enumerate(
-                    it.product(self.classes, expanded)):
-                case.perms.append(case.permute(
-                    class_, defines, permno=i+1, **args))
 
-            # also track non-unique defines
-            case.defines = {}
-            for k, v in case.perms[0].defines.items():
-                if all(perm.defines[k] == v for perm in case.perms):
-                    case.defines[k] = v
 
-        # track all perms and non-unique defines
-        self.perms = []
-        for case in self.cases:
-            self.perms.extend(case.perms)
+def compile(test_paths, **args):
+    # find .toml files
+    paths = []
+    for path in test_paths:
+        if os.path.isdir(path):
+            path = path + '/*.toml'
 
-        self.defines = {}
-        for k, v in self.perms[0].defines.items():
-            if all(perm.defines.get(k, None) == v for perm in self.perms):
-                self.defines[k] = v
+        for path in glob.glob(path):
+            paths.append(path)
 
-        return self.perms
+    if not paths:
+        print('no test suites found in %r?' % test_paths)
+        sys.exit(-1)
 
-    def build(self, **args):
-        # build test files
-        tf = open(self.path + '.test.tc', 'w')
-        tf.write(GLOBALS)
-        if self.code is not None:
-            tf.write('#line %d "%s"\n' % (self.code_lineno, self.path))
-            tf.write(self.code)
+    # load the suites
+    suites = [TestSuite(path, args) for path in paths]
+    suites.sort(key=lambda s: s.name)
 
-        tfs = {None: tf}
-        for case in self.cases:
-            if case.in_ not in tfs:
-                tfs[case.in_] = open(self.path+'.'+
-                    re.sub('(\.c)?$', '.tc', case.in_.replace('/', '.')), 'w')
-                tfs[case.in_].write('#line 1 "%s"\n' % case.in_)
-                with open(case.in_) as f:
-                    for line in f:
-                        tfs[case.in_].write(line)
-                tfs[case.in_].write('\n')
-                tfs[case.in_].write(GLOBALS)
+    # check for name conflicts, these will cause ambiguity problems later
+    # when running tests
+    seen = {}
+    for suite in suites:
+        if suite.name in seen:
+            print('%swarning:%s conflicting suite %r, %s and %s' % (
+                '\x1b[01;33m' if args['color'] else '',
+                '\x1b[m' if args['color'] else '',
+                suite.name,
+                suite.path,
+                seen[suite.name].path),
+                file=sys.stderr)
+        seen[suite.name] = suite
 
-            tfs[case.in_].write('\n')
-            case.build(tfs[case.in_], **args)
+        for case in suite.cases:
+            # only allow conflicts if a case and its suite share a name
+            if case.name in seen and not (
+                    isinstance(seen[case.name], TestSuite)
+                    and seen[case.name].cases == [case]):
+                print('%swarning:%s conflicting case %r, %s and %s' % (
+                    '\x1b[01;33m' if args['color'] else '',
+                    '\x1b[m' if args['color'] else '',
+                    case.name,
+                    case.path,
+                    seen[case.name].path),
+                    file=sys.stderr)
+            seen[case.name] = case
 
-        tf.write('\n')
-        tf.write('const char *lfs_testbd_path;\n')
-        tf.write('uint32_t lfs_testbd_cycles;\n')
-        tf.write('int main(int argc, char **argv) {\n')
-        tf.write(4*' '+'int case_         = (argc > 1) ? atoi(argv[1]) : 0;\n')
-        tf.write(4*' '+'int perm          = (argc > 2) ? atoi(argv[2]) : 0;\n')
-        tf.write(4*' '+'lfs_testbd_path   = (argc > 3) ? argv[3] : NULL;\n')
-        tf.write(4*' '+'lfs_testbd_cycles = (argc > 4) ? atoi(argv[4]) : 0;\n')
-        for perm in self.perms:
-            # test declaration
-            tf.write(4*' '+'extern void test_case%d(%s);\n' % (
-                perm.caseno, ', '.join(
-                    'intmax_t %s' % k for k in sorted(perm.defines)
-                    if k not in perm.case.defines)))
-            # test call
-            tf.write(4*' '+
-                'if (argc < 3 || (case_ == %d && perm == %d)) {'
-                ' test_case%d(%s); '
-                '}\n' % (perm.caseno, perm.permno, perm.caseno, ', '.join(
-                    str(v) for k, v in sorted(perm.defines.items())
-                    if k not in perm.case.defines)))
-        tf.write('}\n')
+    # we can only compile one test suite at a time
+    if not args.get('source'):
+        if len(suites) > 1:
+            print('more than one test suite for compilation? (%r)' % test_paths)
+            sys.exit(-1)
 
-        for tf in tfs.values():
-            tf.close()
+        suite = suites[0]
 
-        # write makefiles
-        with open(self.path + '.mk', 'w') as mk:
-            mk.write(RULES.replace(4*' ', '\t') % dict(path=self.path))
-            mk.write('\n')
+    # write generated test source
+    if 'output' in args:
+        with openio(args['output'], 'w') as f:
+            _write = f.write
+            def write(s):
+                f.lineno += s.count('\n')
+                _write(s)
+            def writeln(s=''):
+                f.lineno += s.count('\n') + 1
+                _write(s)
+                _write('\n')
+            f.lineno = 1
+            f.write = write
+            f.writeln = writeln
 
-            # add coverage hooks?
-            if args.get('coverage'):
-                mk.write(COVERAGE_RULES.replace(4*' ', '\t') % dict(
-                    path=self.path))
-                mk.write('\n')
+            f.writeln("// Generated by %s:" % sys.argv[0])
+            f.writeln("//")
+            f.writeln("// %s" % ' '.join(sys.argv))
+            f.writeln("//")
+            f.writeln()
 
-            # add truly global defines globally
-            for k, v in sorted(self.defines.items()):
-                mk.write('%s.test: override CFLAGS += -D%s=%r\n'
-                    % (self.path, k, v))
+            # include test_runner.h in every generated file
+            f.writeln("#include \"%s\"" % args['include'])
+            f.writeln()
 
-            for path in tfs:
-                if path is None:
-                    mk.write('%s: %s | %s\n' % (
-                        self.path+'.test.c',
-                        self.toml,
-                        self.path+'.test.tc'))
-                else:
-                    mk.write('%s: %s %s | %s\n' % (
-                        self.path+'.'+path.replace('/', '.'),
-                        self.toml,
-                        path,
-                        self.path+'.'+re.sub('(\.c)?$', '.tc',
-                            path.replace('/', '.'))))
-                mk.write('\t./scripts/explode_asserts.py $| -o $@\n')
+            # write out generated functions, this can end up in different
+            # files depending on the "in" attribute
+            #
+            # note it's up to the specific generated file to declare
+            # the test defines
+            def write_case_functions(f, suite, case):
+                    # create case define functions
+                    if case.defines:
+                        # deduplicate defines by value to try to reduce the
+                        # number of functions we generate
+                        define_cbs = {}
+                        for i, defines in enumerate(case.permutations):
+                            for k, v in sorted(defines.items()):
+                                if v not in define_cbs:
+                                    name = ('__test__%s__%s__%d'
+                                        % (case.name, k, i))
+                                    define_cbs[v] = name
+                                    f.writeln('intmax_t %s('
+                                        '__attribute__((unused)) '
+                                        'void *data) {' % name)
+                                    f.writeln(4*' '+'return %s;' % v)
+                                    f.writeln('}')
+                                    f.writeln()
+                        f.writeln('const test_define_t '
+                            '__test__%s__defines[]['
+                            'TEST_IMPLICIT_DEFINE_COUNT+%d] = {'
+                            % (case.name, len(suite.defines)))
+                        for defines in case.permutations:
+                            f.writeln(4*' '+'{')
+                            for k, v in sorted(defines.items()):
+                                f.writeln(8*' '+'[%-24s] = {%s, NULL},' % (
+                                    k+'_i', define_cbs[v]))
+                            f.writeln(4*' '+'},')
+                        f.writeln('};')
+                        f.writeln()
 
-        self.makefile = self.path + '.mk'
-        self.target = self.path + '.test'
-        return self.makefile, self.target
+                    # create case filter function
+                    if suite.if_ is not None or case.if_ is not None:
+                        f.writeln('bool __test__%s__filter(void) {'
+                            % (case.name))
+                        f.writeln(4*' '+'return %s;'
+                            % ' && '.join('(%s)' % if_
+                                for if_ in [suite.if_, case.if_]
+                                if if_ is not None))
+                        f.writeln('}')
+                        f.writeln()
 
-    def test(self, **args):
-        # run test suite!
-        if not args.get('verbose', True):
-            sys.stdout.write(self.name + ' ')
-            sys.stdout.flush()
-        for perm in self.perms:
-            if not perm.shouldtest(**args):
-                continue
+                    # create case run function
+                    f.writeln('void __test__%s__run('
+                        '__attribute__((unused)) struct lfs_config *cfg) {'
+                        % (case.name))
+                    f.writeln(4*' '+'// test case %s' % case.name)
+                    if case.code_lineno is not None:
+                        f.writeln(4*' '+'#line %d "%s"'
+                            % (case.code_lineno, suite.path))
+                    f.write(case.code)
+                    if case.code_lineno is not None:
+                        f.writeln(4*' '+'#line %d "%s"'
+                            % (f.lineno+1, args['output']))
+                    f.writeln('}')
+                    f.writeln()
+
+            if not args.get('source'):
+                if suite.code is not None:
+                    if suite.code_lineno is not None:
+                        f.writeln('#line %d "%s"'
+                            % (suite.code_lineno, suite.path))
+                    f.write(suite.code)
+                    if suite.code_lineno is not None:
+                        f.writeln('#line %d "%s"'
+                            % (f.lineno+1, args['output']))
+                    f.writeln()
+
+                if suite.defines:
+                    for i, define in enumerate(sorted(suite.defines)):
+                        f.writeln('#ifndef %s' % define)
+                        f.writeln('#define %-24s '
+                            'TEST_IMPLICIT_DEFINE_COUNT+%d' % (define+'_i', i))
+                        f.writeln('#define %-24s '
+                            'TEST_DEFINE(%s)' % (define, define+'_i'))
+                        f.writeln('#endif')
+                    f.writeln()
+
+                # create case functions
+                for case in suite.cases:
+                    if case.in_ is None:
+                        write_case_functions(f, suite, case)
+                    else:
+                        if case.defines:
+                            f.writeln('extern const test_define_t '
+                                '__test__%s__defines[]['
+                                'TEST_IMPLICIT_DEFINE_COUNT+%d];'
+                                % (case.name, len(suite.defines)))
+                        if suite.if_ is not None or case.if_ is not None:
+                            f.writeln('extern bool __test__%s__filter('
+                                'void);'
+                                % (case.name))
+                        f.writeln('extern void __test__%s__run('
+                            'struct lfs_config *cfg);'
+                            % (case.name))
+                        f.writeln()
+
+                # create suite struct
+                #
+                # note we place this in the custom test_suites section with
+                # minimum alignment, otherwise GCC ups the alignment to
+                # 32-bytes for some reason
+                f.writeln('__attribute__((section("_test_suites"), '
+                    'aligned(1)))')
+                f.writeln('const struct test_suite __test__%s__suite = {'
+                    % suite.name)
+                f.writeln(4*' '+'.name = "%s",' % suite.name)
+                f.writeln(4*' '+'.path = "%s",' % suite.path)
+                f.writeln(4*' '+'.flags = %s,'
+                    % (' | '.join(filter(None, [
+                        'TEST_REENTRANT' if suite.reentrant else None]))
+                        or 0))
+                if suite.defines:
+                    # create suite define names
+                    f.writeln(4*' '+'.define_names = (const char *const['
+                        'TEST_IMPLICIT_DEFINE_COUNT+%d]){' % (
+                        len(suite.defines)))
+                    for k in sorted(suite.defines):
+                        f.writeln(8*' '+'[%-24s] = "%s",' % (k+'_i', k))
+                    f.writeln(4*' '+'},')
+                    f.writeln(4*' '+'.define_count = '
+                        'TEST_IMPLICIT_DEFINE_COUNT+%d,' % len(suite.defines))
+                f.writeln(4*' '+'.cases = (const struct test_case[]){')
+                for case in suite.cases:
+                    # create case structs
+                    f.writeln(8*' '+'{')
+                    f.writeln(12*' '+'.name = "%s",' % case.name)
+                    f.writeln(12*' '+'.path = "%s",' % case.path)
+                    f.writeln(12*' '+'.flags = %s,'
+                        % (' | '.join(filter(None, [
+                            'TEST_REENTRANT' if case.reentrant else None]))
+                            or 0))
+                    f.writeln(12*' '+'.permutations = %d,'
+                        % len(case.permutations))
+                    if case.defines:
+                        f.writeln(12*' '+'.defines '
+                            '= (const test_define_t*)__test__%s__defines,'
+                            % (case.name))
+                    if suite.if_ is not None or case.if_ is not None:
+                        f.writeln(12*' '+'.filter = __test__%s__filter,'
+                            % (case.name))
+                    f.writeln(12*' '+'.run = __test__%s__run,'
+                        % (case.name))
+                    f.writeln(8*' '+'},')
+                f.writeln(4*' '+'},')
+                f.writeln(4*' '+'.case_count = %d,' % len(suite.cases))
+                f.writeln('};')
+                f.writeln()
 
-            try:
-                result = perm.test(**args)
-            except TestFailure as failure:
-                perm.result = failure
-                if not args.get('verbose', True):
-                    sys.stdout.write(FAIL)
-                    sys.stdout.flush()
-                if not args.get('keep_going'):
-                    if not args.get('verbose', True):
-                        sys.stdout.write('\n')
-                    raise
             else:
-                perm.result = PASS
-                if not args.get('verbose', True):
-                    sys.stdout.write(PASS)
-                    sys.stdout.flush()
+                # copy source
+                f.writeln('#line 1 "%s"' % args['source'])
+                with open(args['source']) as sf:
+                    shutil.copyfileobj(sf, f)
+                f.writeln()
 
-        if not args.get('verbose', True):
-            sys.stdout.write('\n')
+                # write any internal tests
+                for suite in suites:
+                    for case in suite.cases:
+                        if (case.in_ is not None
+                                and os.path.normpath(case.in_)
+                                    == os.path.normpath(args['source'])):
+                            # write defines, but note we need to undef any
+                            # new defines since we're in someone else's file
+                            if suite.defines:
+                                for i, define in enumerate(
+                                        sorted(suite.defines)):
+                                    f.writeln('#ifndef %s' % define)
+                                    f.writeln('#define %-24s '
+                                        'TEST_IMPLICIT_DEFINE_COUNT+%d' % (
+                                        define+'_i', i))
+                                    f.writeln('#define %-24s '
+                                        'TEST_DEFINE(%s)' % (
+                                        define, define+'_i'))
+                                    f.writeln('#define '
+                                        '__TEST__%s__NEEDS_UNDEF' % (
+                                        define))
+                                    f.writeln('#endif')
+                                f.writeln()
 
-def main(**args):
-    # figure out explicit defines
-    defines = {}
-    for define in args['D']:
-        k, v, *_ = define.split('=', 2) + ['']
-        defines[k] = v
+                            write_case_functions(f, suite, case)
 
-    # and what class of TestCase to run
-    classes = []
-    if args.get('normal'):
-        classes.append(TestCase)
-    if args.get('reentrant'):
-        classes.append(ReentrantTestCase)
+                            if suite.defines:
+                                for define in sorted(suite.defines):
+                                    f.writeln('#ifdef __TEST__%s__NEEDS_UNDEF'
+                                        % define)
+                                    f.writeln('#undef __TEST__%s__NEEDS_UNDEF'
+                                        % define)
+                                    f.writeln('#undef %s' % define)
+                                    f.writeln('#undef %s' % (define+'_i'))
+                                    f.writeln('#endif')
+                                f.writeln()
+
+def find_runner(runner, **args):
+    cmd = runner.copy()
+
+    # run under some external command?
+    if args.get('exec'):
+        cmd[:0] = args['exec']
+
+    # run under valgrind?
     if args.get('valgrind'):
-        classes.append(ValgrindTestCase)
-    if not classes:
-        classes = [TestCase]
+        cmd[:0] = args['valgrind_path'] + [
+            '--leak-check=full',
+            '--track-origins=yes',
+            '--error-exitcode=4',
+            '-q']
 
-    suites = []
-    for testpath in args['test_paths']:
-        # optionally specified test case/perm
-        testpath, *filter = testpath.split('#')
-        filter = [int(f) for f in filter]
+    # run under perf?
+    if args.get('perf'):
+        cmd[:0] = args['perf_script'] + list(filter(None, [
+            '-R',
+            '--perf-freq=%s' % args['perf_freq']
+                if args.get('perf_freq') else None,
+            '--perf-period=%s' % args['perf_period']
+                if args.get('perf_period') else None,
+            '--perf-events=%s' % args['perf_events']
+                if args.get('perf_events') else None,
+            '--perf-path=%s' % args['perf_path']
+                if args.get('perf_path') else None,
+            '-o%s' % args['perf']]))
 
-        # figure out the suite's toml file
-        if os.path.isdir(testpath):
-            testpath = testpath + '/*.toml'
-        elif os.path.isfile(testpath):
-            testpath = testpath
-        elif testpath.endswith('.toml'):
-            testpath = TEST_PATHS + '/' + testpath
-        else:
-            testpath = TEST_PATHS + '/' + testpath + '.toml'
+    # other context
+    if args.get('geometry'):
+        cmd.append('-G%s' % args['geometry'])
+    if args.get('powerloss'):
+        cmd.append('-P%s' % args['powerloss'])
+    if args.get('disk'):
+        cmd.append('-d%s' % args['disk'])
+    if args.get('trace'):
+        cmd.append('-t%s' % args['trace'])
+    if args.get('trace_backtrace'):
+        cmd.append('--trace-backtrace')
+    if args.get('trace_period'):
+        cmd.append('--trace-period=%s' % args['trace_period'])
+    if args.get('trace_freq'):
+        cmd.append('--trace-freq=%s' % args['trace_freq'])
+    if args.get('read_sleep'):
+        cmd.append('--read-sleep=%s' % args['read_sleep'])
+    if args.get('prog_sleep'):
+        cmd.append('--prog-sleep=%s' % args['prog_sleep'])
+    if args.get('erase_sleep'):
+        cmd.append('--erase-sleep=%s' % args['erase_sleep'])
 
-        # find tests
-        for path in glob.glob(testpath):
-            suites.append(TestSuite(path, classes, defines, filter, **args))
+    # defines?
+    if args.get('define'):
+        for define in args.get('define'):
+            cmd.append('-D%s' % define)
 
-    # sort for reproducibility
-    suites = sorted(suites)
+    return cmd
 
-    # generate permutations
-    for suite in suites:
-        suite.permute(**args)
+def list_(runner, test_ids=[], **args):
+    cmd = find_runner(runner, **args) + test_ids
+    if args.get('summary'):          cmd.append('--summary')
+    if args.get('list_suites'):      cmd.append('--list-suites')
+    if args.get('list_cases'):       cmd.append('--list-cases')
+    if args.get('list_suite_paths'): cmd.append('--list-suite-paths')
+    if args.get('list_case_paths'):  cmd.append('--list-case-paths')
+    if args.get('list_defines'):     cmd.append('--list-defines')
+    if args.get('list_permutation_defines'):
+                                     cmd.append('--list-permutation-defines')
+    if args.get('list_implicit_defines'):
+                                     cmd.append('--list-implicit-defines')
+    if args.get('list_geometries'):  cmd.append('--list-geometries')
+    if args.get('list_powerlosses'): cmd.append('--list-powerlosses')
 
-    # build tests in parallel
-    print('====== building ======')
-    makefiles = []
-    targets = []
-    for suite in suites:
-        makefile, target = suite.build(**args)
-        makefiles.append(makefile)
-        targets.append(target)
-
-    cmd = (['make', '-f', 'Makefile'] +
-        list(it.chain.from_iterable(['-f', m] for m in makefiles)) +
-        [target for target in targets])
-    mpty, spty = pty.openpty()
     if args.get('verbose'):
         print(' '.join(shlex.quote(c) for c in cmd))
-    proc = sp.Popen(cmd, stdout=spty, stderr=spty)
-    os.close(spty)
-    mpty = os.fdopen(mpty, 'r', 1)
-    stdout = []
-    while True:
-        try:
-            line = mpty.readline()
-        except OSError as e:
-            if e.errno == errno.EIO:
-                break
-            raise
-        if not line:
-            break;
-        stdout.append(line)
-        if args.get('verbose'):
-            sys.stdout.write(line)
-        # intercept warnings
-        m = re.match(
-            '^{0}([^:]+):(\d+):(?:\d+:)?{0}{1}:{0}(.*)$'
-            .format('(?:\033\[[\d;]*.| )*', 'warning'),
-            line)
-        if m and not args.get('verbose'):
-            try:
-                with open(m.group(1)) as f:
-                    lineno = int(m.group(2))
-                    line = next(it.islice(f, lineno-1, None)).strip('\n')
-                sys.stdout.write(
-                    "\033[01m{path}:{lineno}:\033[01;35mwarning:\033[m "
-                    "{message}\n{line}\n\n".format(
-                        path=m.group(1), line=line, lineno=lineno,
-                        message=m.group(3)))
-            except:
-                pass
+    return sp.call(cmd)
+
+
+def find_perms(runner_, ids=[], **args):
+    case_suites = {}
+    expected_case_perms = co.defaultdict(lambda: 0)
+    expected_perms = 0
+    total_perms = 0
+
+    # query cases from the runner
+    cmd = runner_ + ['--list-cases'] + ids
+    if args.get('verbose'):
+        print(' '.join(shlex.quote(c) for c in cmd))
+    proc = sp.Popen(cmd,
+        stdout=sp.PIPE,
+        stderr=sp.PIPE if not args.get('verbose') else None,
+        universal_newlines=True,
+        errors='replace',
+        close_fds=False)
+    pattern = re.compile(
+        '^(?P<case>[^\s]+)'
+            '\s+(?P<flags>[^\s]+)'
+            '\s+(?P<filtered>\d+)/(?P<perms>\d+)')
+    # skip the first line
+    for line in it.islice(proc.stdout, 1, None):
+        m = pattern.match(line)
+        if m:
+            filtered = int(m.group('filtered'))
+            perms = int(m.group('perms'))
+            expected_case_perms[m.group('case')] += filtered
+            expected_perms += filtered
+            total_perms += perms
     proc.wait()
     if proc.returncode != 0:
         if not args.get('verbose'):
-            for line in stdout:
+            for line in proc.stderr:
                 sys.stdout.write(line)
         sys.exit(-1)
 
-    print('built %d test suites, %d test cases, %d permutations' % (
-        len(suites),
-        sum(len(suite.cases) for suite in suites),
-        sum(len(suite.perms) for suite in suites)))
+    # get which suite each case belongs to via paths
+    cmd = runner_ + ['--list-case-paths'] + ids
+    if args.get('verbose'):
+        print(' '.join(shlex.quote(c) for c in cmd))
+    proc = sp.Popen(cmd,
+        stdout=sp.PIPE,
+        stderr=sp.PIPE if not args.get('verbose') else None,
+        universal_newlines=True,
+        errors='replace',
+        close_fds=False)
+    pattern = re.compile(
+        '^(?P<case>[^\s]+)'
+            '\s+(?P<path>[^:]+):(?P<lineno>\d+)')
+    # skip the first line
+    for line in it.islice(proc.stdout, 1, None):
+        m = pattern.match(line)
+        if m:
+            path = m.group('path')
+            # strip path/suffix here
+            suite = os.path.basename(path)
+            if suite.endswith('.toml'):
+                suite = suite[:-len('.toml')]
+            case_suites[m.group('case')] = suite
+    proc.wait()
+    if proc.returncode != 0:
+        if not args.get('verbose'):
+            for line in proc.stderr:
+                sys.stdout.write(line)
+        sys.exit(-1)
 
-    total = 0
-    for suite in suites:
-        for perm in suite.perms:
-            total += perm.shouldtest(**args)
-    if total != sum(len(suite.perms) for suite in suites):
-        print('filtered down to %d permutations' % total)
+    # figure out expected suite perms
+    expected_suite_perms = co.defaultdict(lambda: 0)
+    for case, suite in case_suites.items():
+        expected_suite_perms[suite] += expected_case_perms[case]
 
-    # only requested to build?
-    if args.get('build'):
-        return 0
+    return (
+        case_suites,
+        expected_suite_perms,
+        expected_case_perms,
+        expected_perms,
+        total_perms)
 
-    print('====== testing ======')
-    try:
-        for suite in suites:
-            suite.test(**args)
-    except TestFailure:
-        pass
+def find_path(runner_, id, **args):
+    path = None
+    # query from runner
+    cmd = runner_ + ['--list-case-paths', id]
+    if args.get('verbose'):
+        print(' '.join(shlex.quote(c) for c in cmd))
+    proc = sp.Popen(cmd,
+        stdout=sp.PIPE,
+        stderr=sp.PIPE if not args.get('verbose') else None,
+        universal_newlines=True,
+        errors='replace',
+        close_fds=False)
+    pattern = re.compile(
+        '^(?P<case>[^\s]+)'
+            '\s+(?P<path>[^:]+):(?P<lineno>\d+)')
+    # skip the first line
+    for line in it.islice(proc.stdout, 1, None):
+        m = pattern.match(line)
+        if m and path is None:
+            path_ = m.group('path')
+            lineno = int(m.group('lineno'))
+            path = (path_, lineno)
+    proc.wait()
+    if proc.returncode != 0:
+        if not args.get('verbose'):
+            for line in proc.stderr:
+                sys.stdout.write(line)
+        sys.exit(-1)
 
-    print('====== results ======')
-    passed = 0
-    failed = 0
-    for suite in suites:
-        for perm in suite.perms:
-            if perm.result == PASS:
-                passed += 1
-            elif isinstance(perm.result, TestFailure):
-                sys.stdout.write(
-                    "\033[01m{path}:{lineno}:\033[01;31mfailure:\033[m "
-                    "{perm} failed\n".format(
-                        perm=perm, path=perm.suite.path, lineno=perm.lineno,
-                        returncode=perm.result.returncode or 0))
-                if perm.result.stdout:
-                    if perm.result.assert_:
-                        stdout = perm.result.stdout[:-1]
-                    else:
-                        stdout = perm.result.stdout
-                    for line in stdout[-5:]:
-                        sys.stdout.write(line)
-                if perm.result.assert_:
-                    sys.stdout.write(
-                        "\033[01m{path}:{lineno}:\033[01;31massert:\033[m "
-                        "{message}\n{line}\n".format(
-                            **perm.result.assert_))
-                sys.stdout.write('\n')
-                failed += 1
+    return path
 
-    if args.get('coverage'):
-        # collect coverage info
-        # why -j1? lcov doesn't work in parallel because of gcov limitations
-        cmd = (['make', '-j1', '-f', 'Makefile'] +
-            list(it.chain.from_iterable(['-f', m] for m in makefiles)) +
-            (['COVERAGETARGET=%s' % args['coverage']]
-                if isinstance(args['coverage'], str) else []) +
-            [suite.path + '.info' for suite in suites
-                if any(perm.result == PASS for perm in suite.perms)])
+def find_defines(runner_, id, **args):
+    # query permutation defines from runner
+    cmd = runner_ + ['--list-permutation-defines', id]
+    if args.get('verbose'):
+        print(' '.join(shlex.quote(c) for c in cmd))
+    proc = sp.Popen(cmd,
+        stdout=sp.PIPE,
+        stderr=sp.PIPE if not args.get('verbose') else None,
+        universal_newlines=True,
+        errors='replace',
+        close_fds=False)
+    defines = co.OrderedDict()
+    pattern = re.compile('^(?P<define>\w+)=(?P<value>.+)')
+    for line in proc.stdout:
+        m = pattern.match(line)
+        if m:
+            define = m.group('define')
+            value = m.group('value')
+            defines[define] = value
+    proc.wait()
+    if proc.returncode != 0:
+        if not args.get('verbose'):
+            for line in proc.stderr:
+                sys.stdout.write(line)
+        sys.exit(-1)
+
+    return defines
+
+
+# Thread-safe CSV writer
+class TestOutput:
+    def __init__(self, path, head=None, tail=None):
+        self.f = openio(path, 'w+', 1)
+        self.lock = th.Lock()
+        self.head = head or []
+        self.tail = tail or []
+        self.writer = csv.DictWriter(self.f, self.head + self.tail)
+        self.rows = []
+
+    def close(self):
+        self.f.close()
+
+    def __enter__(self):
+        return self
+
+    def __exit__(self, *_):
+        self.f.close()
+
+    def writerow(self, row):
+        with self.lock:
+            self.rows.append(row)
+            if all(k in self.head or k in self.tail for k in row.keys()):
+                # can simply append
+                self.writer.writerow(row)
+            else:
+                # need to rewrite the file
+                self.head.extend(row.keys() - (self.head + self.tail))
+                self.f.seek(0)
+                self.f.truncate()
+                self.writer = csv.DictWriter(self.f, self.head + self.tail)
+                self.writer.writeheader()
+                for row in self.rows:
+                    self.writer.writerow(row)
+
+# A test failure
+class TestFailure(Exception):
+    def __init__(self, id, returncode, stdout, assert_=None):
+        self.id = id
+        self.returncode = returncode
+        self.stdout = stdout
+        self.assert_ = assert_
+
+def run_stage(name, runner_, ids, stdout_, trace_, output_, **args):
+    # get expected suite/case/perm counts
+    (case_suites,
+        expected_suite_perms,
+        expected_case_perms,
+        expected_perms,
+        total_perms) = find_perms(runner_, ids, **args)
+
+    passed_suite_perms = co.defaultdict(lambda: 0)
+    passed_case_perms = co.defaultdict(lambda: 0)
+    passed_perms = 0
+    powerlosses = 0
+    failures = []
+    killed = False
+
+    pattern = re.compile('^(?:'
+            '(?P<op>running|finished|skipped|powerloss) '
+                '(?P<id>(?P<case>[^:]+)[^\s]*)'
+            '|' '(?P<path>[^:]+):(?P<lineno>\d+):(?P<op_>assert):'
+                ' *(?P<message>.*)'
+        ')$')
+    locals = th.local()
+    children = set()
+
+    def run_runner(runner_, ids=[]):
+        nonlocal passed_suite_perms
+        nonlocal passed_case_perms
+        nonlocal passed_perms
+        nonlocal powerlosses
+        nonlocal locals
+
+        # run the tests!
+        cmd = runner_ + ids
         if args.get('verbose'):
             print(' '.join(shlex.quote(c) for c in cmd))
-        proc = sp.Popen(cmd,
-            stdout=sp.PIPE if not args.get('verbose') else None,
-            stderr=sp.STDOUT if not args.get('verbose') else None,
-            universal_newlines=True)
-        stdout = []
-        for line in proc.stdout:
-            stdout.append(line)
+
+        mpty, spty = pty.openpty()
+        proc = sp.Popen(cmd, stdout=spty, stderr=spty, close_fds=False)
+        os.close(spty)
+        children.add(proc)
+        mpty = os.fdopen(mpty, 'r', 1)
+
+        last_id = None
+        last_stdout = co.deque(maxlen=args.get('context', 5) + 1)
+        last_assert = None
+        try:
+            while True:
+                # parse a line for state changes
+                try:
+                    line = mpty.readline()
+                except OSError as e:
+                    if e.errno != errno.EIO:
+                        raise
+                    break
+                if not line:
+                    break
+                last_stdout.append(line)
+                if stdout_:
+                    try:
+                        stdout_.write(line)
+                        stdout_.flush()
+                    except BrokenPipeError:
+                        pass
+
+                m = pattern.match(line)
+                if m:
+                    op = m.group('op') or m.group('op_')
+                    if op == 'running':
+                        locals.seen_perms += 1
+                        last_id = m.group('id')
+                        last_stdout.clear()
+                        last_assert = None
+                    elif op == 'powerloss':
+                        last_id = m.group('id')
+                        powerlosses += 1
+                    elif op == 'finished':
+                        case = m.group('case')
+                        suite = case_suites[case]
+                        passed_suite_perms[suite] += 1
+                        passed_case_perms[case] += 1
+                        passed_perms += 1
+                        if output_:
+                            # get defines and write to csv
+                            defines = find_defines(
+                                runner_, m.group('id'), **args)
+                            output_.writerow({
+                                'suite': suite,
+                                'case': case,
+                                'test_passed': '1/1',
+                                **defines})
+                    elif op == 'skipped':
+                        locals.seen_perms += 1
+                    elif op == 'assert':
+                        last_assert = (
+                            m.group('path'),
+                            int(m.group('lineno')),
+                            m.group('message'))
+                        # go ahead and kill the process, aborting takes a while
+                        if args.get('keep_going'):
+                            proc.kill()
+        except KeyboardInterrupt:
+            raise TestFailure(last_id, 1, list(last_stdout))
+        finally:
+            children.remove(proc)
+            mpty.close()
+
         proc.wait()
         if proc.returncode != 0:
-            if not args.get('verbose'):
-                for line in stdout:
-                    sys.stdout.write(line)
-            sys.exit(-1)
+            raise TestFailure(
+                last_id,
+                proc.returncode,
+                list(last_stdout),
+                last_assert)
 
-    if args.get('gdb'):
-        failure = None
-        for suite in suites:
-            for perm in suite.perms:
-                if isinstance(perm.result, TestFailure):
-                    failure = perm.result
-        if failure is not None:
-            print('======= gdb ======')
-            # drop into gdb
-            failure.case.test(failure=failure, **args)
-            sys.exit(0)
+    def run_job(runner_, ids=[], start=None, step=None):
+        nonlocal failures
+        nonlocal killed
+        nonlocal locals
+
+        start = start or 0
+        step = step or 1
+        while start < total_perms:
+            job_runner = runner_.copy()
+            if args.get('isolate') or args.get('valgrind'):
+                job_runner.append('-s%s,%s,%s' % (start, start+step, step))
+            else:
+                job_runner.append('-s%s,,%s' % (start, step))
+
+            try:
+                # run the tests
+                locals.seen_perms = 0
+                run_runner(job_runner, ids)
+                assert locals.seen_perms > 0
+                start += locals.seen_perms*step
+
+            except TestFailure as failure:
+                # keep track of failures
+                if output_:
+                    case, _ = failure.id.split(':', 1)
+                    suite = case_suites[case]
+                    # get defines and write to csv
+                    defines = find_defines(runner_, failure.id, **args)
+                    output_.writerow({
+                        'suite': suite,
+                        'case': case,
+                        'test_passed': '0/1',
+                        **defines})
+
+                # race condition for multiple failures?
+                if failures and not args.get('keep_going'):
+                    break
+
+                failures.append(failure)
+
+                if args.get('keep_going') and not killed:
+                    # resume after failed test
+                    assert locals.seen_perms > 0
+                    start += locals.seen_perms*step
+                    continue
+                else:
+                    # stop other tests
+                    killed = True
+                    for child in children.copy():
+                        child.kill()
+                    break
+
+
+    # parallel jobs?
+    runners = []
+    if 'jobs' in args:
+        for job in range(args['jobs']):
+            runners.append(th.Thread(
+                target=run_job, args=(runner_, ids, job, args['jobs']),
+                daemon=True))
+    else:
+        runners.append(th.Thread(
+            target=run_job, args=(runner_, ids, None, None),
+            daemon=True))
+
+    def print_update(done):
+        if not args.get('verbose') and (args['color'] or done):
+            sys.stdout.write('%s%srunning %s%s:%s %s%s' % (
+                '\r\x1b[K' if args['color'] else '',
+                '\x1b[?7l' if not done else '',
+                ('\x1b[32m' if not failures else '\x1b[31m')
+                    if args['color'] else '',
+                name,
+                '\x1b[m' if args['color'] else '',
+                ', '.join(filter(None, [
+                    '%d/%d suites' % (
+                        sum(passed_suite_perms[k] == v
+                            for k, v in expected_suite_perms.items()),
+                        len(expected_suite_perms))
+                        if (not args.get('by_suites')
+                            and not args.get('by_cases')) else None,
+                    '%d/%d cases' % (
+                        sum(passed_case_perms[k] == v
+                            for k, v in expected_case_perms.items()),
+                        len(expected_case_perms))
+                        if not args.get('by_cases') else None,
+                    '%d/%d perms' % (passed_perms, expected_perms),
+                    '%dpls!' % powerlosses
+                        if powerlosses else None,
+                    '%s%d/%d failures%s' % (
+                            '\x1b[31m' if args['color'] else '',
+                            len(failures),
+                            expected_perms,
+                            '\x1b[m' if args['color'] else '')
+                        if failures else None])),
+                '\x1b[?7h' if not done else '\n'))
+            sys.stdout.flush()
+
+    for r in runners:
+        r.start()
+
+    try:
+        while any(r.is_alive() for r in runners):
+            time.sleep(0.01)
+            print_update(False)
+    except KeyboardInterrupt:
+        # this is handled by the runner threads, we just
+        # need to not abort here
+        killed = True
+    finally:
+        print_update(True)
+
+    for r in runners:
+        r.join()
+
+    return (
+        expected_perms,
+        passed_perms,
+        powerlosses,
+        failures,
+        killed)
+
+
+def run(runner, test_ids=[], **args):
+    # query runner for tests
+    runner_ = find_runner(runner, **args)
+    print('using runner: %s' % ' '.join(shlex.quote(c) for c in runner_))
+    (_,
+        expected_suite_perms,
+        expected_case_perms,
+        expected_perms,
+        total_perms) = find_perms(runner_, test_ids, **args)
+    print('found %d suites, %d cases, %d/%d permutations' % (
+        len(expected_suite_perms),
+        len(expected_case_perms),
+        expected_perms,
+        total_perms))
+    print()
+
+    # automatic job detection?
+    if args.get('jobs') == 0:
+        args['jobs'] = len(os.sched_getaffinity(0))
+
+    # truncate and open logs here so they aren't disconnected between tests
+    stdout = None
+    if args.get('stdout'):
+        stdout = openio(args['stdout'], 'w', 1)
+    trace = None
+    if args.get('trace'):
+        trace = openio(args['trace'], 'w', 1)
+    output = None
+    if args.get('output'):
+        output = TestOutput(args['output'],
+            ['suite', 'case'],
+            ['test_passed'])
+
+    # measure runtime
+    start = time.time()
+
+    # spawn runners
+    expected = 0
+    passed = 0
+    powerlosses = 0
+    failures = []
+    for by in (test_ids if test_ids
+            else expected_case_perms.keys() if args.get('by_cases')
+            else expected_suite_perms.keys() if args.get('by_suites')
+            else [None]):
+        # spawn jobs for stage
+        (expected_,
+            passed_,
+            powerlosses_,
+            failures_,
+            killed) = run_stage(
+                by or 'tests',
+                runner_,
+                [by] if by is not None else [],
+                stdout,
+                trace,
+                output,
+                **args)
+        # collect passes/failures
+        expected += expected_
+        passed += passed_
+        powerlosses += powerlosses_
+        failures.extend(failures_)
+        if (failures and not args.get('keep_going')) or killed:
+            break
+
+    stop = time.time()
+
+    if stdout:
+        try:
+            stdout.close()
+        except BrokenPipeError:
+            pass
+    if trace:
+        try:
+            trace.close()
+        except BrokenPipeError:
+            pass
+    if output:
+        output.close()
+
+    # show summary
+    print()
+    print('%sdone:%s %s' % (
+        ('\x1b[32m' if not failures else '\x1b[31m')
+            if args['color'] else '',
+        '\x1b[m' if args['color'] else '',
+        ', '.join(filter(None, [
+            '%d/%d passed' % (passed, expected),
+            '%d/%d failed' % (len(failures), expected),
+            '%dpls!' % powerlosses if powerlosses else None,
+            'in %.2fs' % (stop-start)]))))
+    print()
+
+    # print each failure
+    for failure in failures:
+        assert failure.id is not None, '%s broken? %r' % (
+            ' '.join(shlex.quote(c) for c in runner_),
+            failure)
+
+        # get some extra info from runner
+        path, lineno = find_path(runner_, failure.id, **args)
+        defines = find_defines(runner_, failure.id, **args)
+
+        # show summary of failure
+        print('%s%s:%d:%sfailure:%s %s%s failed' % (
+            '\x1b[01m' if args['color'] else '',
+            path, lineno,
+            '\x1b[01;31m' if args['color'] else '',
+            '\x1b[m' if args['color'] else '',
+            failure.id,
+            ' (%s)' % ', '.join('%s=%s' % (k,v) for k,v in defines.items())
+                if defines else ''))
+
+        if failure.stdout:
+            stdout = failure.stdout
+            if failure.assert_ is not None:
+                stdout = stdout[:-1]
+            for line in stdout[-args.get('context', 5):]:
+                sys.stdout.write(line)
+
+        if failure.assert_ is not None:
+            path, lineno, message = failure.assert_
+            print('%s%s:%d:%sassert:%s %s' % (
+                '\x1b[01m' if args['color'] else '',
+                path, lineno,
+                '\x1b[01;31m' if args['color'] else '',
+                '\x1b[m' if args['color'] else '',
+                message))
+            with open(path) as f:
+                line = next(it.islice(f, lineno-1, None)).strip('\n')
+                print(line)
+        print()
+
+    # drop into gdb?
+    if failures and (args.get('gdb')
+            or args.get('gdb_case')
+            or args.get('gdb_main')
+            or args.get('gdb_pl') is not None
+            or args.get('gdb_pl_before')
+            or args.get('gdb_pl_after')):
+        failure = failures[0]
+        cmd = runner_ + [failure.id]
+
+        if args.get('gdb_main'):
+            # we don't really need the case breakpoint here, but it
+            # can be helpful
+            path, lineno = find_path(runner_, failure.id, **args)
+            cmd[:0] = args['gdb_path'] + [
+                '-ex', 'break main',
+                '-ex', 'break %s:%d' % (path, lineno),
+                '-ex', 'run',
+                '--args']
+        elif args.get('gdb_case'):
+            path, lineno = find_path(runner_, failure.id, **args)
+            cmd[:0] = args['gdb_path'] + [
+                '-ex', 'break %s:%d' % (path, lineno),
+                '-ex', 'run',
+                '--args']
+        elif args.get('gdb_pl') is not None:
+            path, lineno = find_path(runner_, failure.id, **args)
+            cmd[:0] = args['gdb_path'] + [
+                '-ex', 'break %s:%d' % (path, lineno),
+                '-ex', 'ignore 1 %d' % args['gdb_pl'],
+                '-ex', 'run',
+                '--args']
+        elif args.get('gdb_pl_before'):
+            # figure out how many powerlosses there were
+            powerlosses = (
+                sum(1 for _ in re.finditer('[0-9a-f]',
+                    failure.id.split(':', 2)[-1]))
+                if failure.id.count(':') >= 2 else 0)
+            path, lineno = find_path(runner_, failure.id, **args)
+            cmd[:0] = args['gdb_path'] + [
+                '-ex', 'break %s:%d' % (path, lineno),
+                '-ex', 'ignore 1 %d' % max(powerlosses-1, 0),
+                '-ex', 'run',
+                '--args']
+        elif args.get('gdb_pl_after'):
+            # figure out how many powerlosses there were
+            powerlosses = (
+                sum(1 for _ in re.finditer('[0-9a-f]',
+                    failure.id.split(':', 2)[-1]))
+                if failure.id.count(':') >= 2 else 0)
+            path, lineno = find_path(runner_, failure.id, **args)
+            cmd[:0] = args['gdb_path'] + [
+                '-ex', 'break %s:%d' % (path, lineno),
+                '-ex', 'ignore 1 %d' % powerlosses,
+                '-ex', 'run',
+                '--args']
+        elif failure.assert_ is not None:
+            cmd[:0] = args['gdb_path'] + [
+                '-ex', 'run',
+                '-ex', 'frame function raise',
+                '-ex', 'up 2',
+                '--args']
+        else:
+            cmd[:0] = args['gdb_path'] + [
+                '-ex', 'run',
+                '--args']
+
+        # exec gdb interactively
+        if args.get('verbose'):
+            print(' '.join(shlex.quote(c) for c in cmd))
+        os.execvp(cmd[0], cmd)
+
+    return 1 if failures else 0
+
+
+def main(**args):
+    # figure out what color should be
+    if args.get('color') == 'auto':
+        args['color'] = sys.stdout.isatty()
+    elif args.get('color') == 'always':
+        args['color'] = True
+    else:
+        args['color'] = False
+
+    if args.get('compile'):
+        return compile(**args)
+    elif (args.get('summary')
+            or args.get('list_suites')
+            or args.get('list_cases')
+            or args.get('list_suite_paths')
+            or args.get('list_case_paths')
+            or args.get('list_defines')
+            or args.get('list_permutation_defines')
+            or args.get('list_implicit_defines')
+            or args.get('list_geometries')
+            or args.get('list_powerlosses')):
+        return list_(**args)
+    else:
+        return run(**args)
 
-    print('tests passed %d/%d (%.1f%%)' % (passed, total,
-        100*(passed/total if total else 1.0)))
-    print('tests failed %d/%d (%.1f%%)' % (failed, total,
-        100*(failed/total if total else 1.0)))
-    return 1 if failed > 0 else 0
 
 if __name__ == "__main__":
     import argparse
+    import sys
+    argparse.ArgumentParser._handle_conflict_ignore = lambda *_: None
+    argparse._ArgumentGroup._handle_conflict_ignore = lambda *_: None
     parser = argparse.ArgumentParser(
-        description="Run parameterized tests in various configurations.")
-    parser.add_argument('test_paths', nargs='*', default=[TEST_PATHS],
-        help="Description of test(s) to run. By default, this is all tests \
-            found in the \"{0}\" directory. Here, you can specify a different \
-            directory of tests, a specific file, a suite by name, and even \
-            specific test cases and permutations. For example \
-            \"test_dirs#1\" or \"{0}/test_dirs.toml#1#1\".".format(TEST_PATHS))
-    parser.add_argument('-D', action='append', default=[],
-        help="Overriding parameter definitions.")
-    parser.add_argument('-v', '--verbose', action='store_true',
-        help="Output everything that is happening.")
-    parser.add_argument('-k', '--keep-going', action='store_true',
-        help="Run all tests instead of stopping on first error. Useful for CI.")
-    parser.add_argument('-p', '--persist', choices=['erase', 'noerase'],
-        nargs='?', const='erase',
-        help="Store disk image in a file.")
-    parser.add_argument('-b', '--build', action='store_true',
-        help="Only build the tests, do not execute.")
-    parser.add_argument('-g', '--gdb', choices=['init', 'main', 'assert'],
-        nargs='?', const='assert',
-        help="Drop into gdb on test failure.")
-    parser.add_argument('--no-internal', action='store_true',
-        help="Don't run tests that require internal knowledge.")
-    parser.add_argument('-n', '--normal', action='store_true',
-        help="Run tests normally.")
-    parser.add_argument('-r', '--reentrant', action='store_true',
-        help="Run reentrant tests with simulated power-loss.")
-    parser.add_argument('--valgrind', action='store_true',
-        help="Run non-leaky tests under valgrind to check for memory leaks.")
-    parser.add_argument('--exec', default=[], type=lambda e: e.split(),
-        help="Run tests with another executable prefixed on the command line.")
-    parser.add_argument('--disk',
-        help="Specify a file to use for persistent/reentrant tests.")
-    parser.add_argument('--coverage', type=lambda x: x if x else True,
-        nargs='?', const='',
-        help="Collect coverage information during testing. This uses lcov/gcov \
-            to accumulate coverage information into *.info files. May also \
-            a path to a *.info file to accumulate coverage info into.")
-    parser.add_argument('--build-dir',
-        help="Build relative to the specified directory instead of the \
-            current directory.")
+        description="Build and run tests.",
+        allow_abbrev=False,
+        conflict_handler='ignore')
+    parser.add_argument(
+        '-v', '--verbose',
+        action='store_true',
+        help="Output commands that run behind the scenes.")
+    parser.add_argument(
+        '--color',
+        choices=['never', 'always', 'auto'],
+        default='auto',
+        help="When to use terminal colors. Defaults to 'auto'.")
 
-    sys.exit(main(**vars(parser.parse_args())))
+    # test flags
+    test_parser = parser.add_argument_group('test options')
+    test_parser.add_argument(
+        'runner',
+        nargs='?',
+        type=lambda x: x.split(),
+        help="Test runner to use for testing. Defaults to %r." % RUNNER_PATH)
+    test_parser.add_argument(
+        'test_ids',
+        nargs='*',
+        help="Description of tests to run.")
+    test_parser.add_argument(
+        '-Y', '--summary',
+        action='store_true',
+        help="Show quick summary.")
+    test_parser.add_argument(
+        '-l', '--list-suites',
+        action='store_true',
+        help="List test suites.")
+    test_parser.add_argument(
+        '-L', '--list-cases',
+        action='store_true',
+        help="List test cases.")
+    test_parser.add_argument(
+        '--list-suite-paths',
+        action='store_true',
+        help="List the path for each test suite.")
+    test_parser.add_argument(
+        '--list-case-paths',
+        action='store_true',
+        help="List the path and line number for each test case.")
+    test_parser.add_argument(
+        '--list-defines',
+        action='store_true',
+        help="List all defines in this test-runner.")
+    test_parser.add_argument(
+        '--list-permutation-defines',
+        action='store_true',
+        help="List explicit defines in this test-runner.")
+    test_parser.add_argument(
+        '--list-implicit-defines',
+        action='store_true',
+        help="List implicit defines in this test-runner.")
+    test_parser.add_argument(
+        '--list-geometries',
+        action='store_true',
+        help="List the available disk geometries.")
+    test_parser.add_argument(
+        '--list-powerlosses',
+        action='store_true',
+        help="List the available power-loss scenarios.")
+    test_parser.add_argument(
+        '-D', '--define',
+        action='append',
+        help="Override a test define.")
+    test_parser.add_argument(
+        '-G', '--geometry',
+        help="Comma-separated list of disk geometries to test.")
+    test_parser.add_argument(
+        '-P', '--powerloss',
+        help="Comma-separated list of power-loss scenarios to test.")
+    test_parser.add_argument(
+        '-d', '--disk',
+        help="Direct block device operations to this file.")
+    test_parser.add_argument(
+        '-t', '--trace',
+        help="Direct trace output to this file.")
+    test_parser.add_argument(
+        '--trace-backtrace',
+        action='store_true',
+        help="Include a backtrace with every trace statement.")
+    test_parser.add_argument(
+        '--trace-period',
+        help="Sample trace output at this period in cycles.")
+    test_parser.add_argument(
+        '--trace-freq',
+        help="Sample trace output at this frequency in hz.")
+    test_parser.add_argument(
+        '-O', '--stdout',
+        help="Direct stdout to this file. Note stderr is already merged here.")
+    test_parser.add_argument(
+        '-o', '--output',
+        help="CSV file to store results.")
+    test_parser.add_argument(
+        '--read-sleep',
+        help="Artificial read delay in seconds.")
+    test_parser.add_argument(
+        '--prog-sleep',
+        help="Artificial prog delay in seconds.")
+    test_parser.add_argument(
+        '--erase-sleep',
+        help="Artificial erase delay in seconds.")
+    test_parser.add_argument(
+        '-j', '--jobs',
+        nargs='?',
+        type=lambda x: int(x, 0),
+        const=0,
+        help="Number of parallel runners to run. 0 runs one runner per core.")
+    test_parser.add_argument(
+        '-k', '--keep-going',
+        action='store_true',
+        help="Don't stop on first error.")
+    test_parser.add_argument(
+        '-i', '--isolate',
+        action='store_true',
+        help="Run each test permutation in a separate process.")
+    test_parser.add_argument(
+        '-b', '--by-suites',
+        action='store_true',
+        help="Step through tests by suite.")
+    test_parser.add_argument(
+        '-B', '--by-cases',
+        action='store_true',
+        help="Step through tests by case.")
+    test_parser.add_argument(
+        '--context',
+        type=lambda x: int(x, 0),
+        default=5,
+        help="Show this many lines of stdout on test failure. "
+            "Defaults to 5.")
+    test_parser.add_argument(
+        '--gdb',
+        action='store_true',
+        help="Drop into gdb on test failure.")
+    test_parser.add_argument(
+        '--gdb-case',
+        action='store_true',
+        help="Drop into gdb on test failure but stop at the beginning "
+            "of the failing test case.")
+    test_parser.add_argument(
+        '--gdb-main',
+        action='store_true',
+        help="Drop into gdb on test failure but stop at the beginning "
+            "of main.")
+    test_parser.add_argument(
+        '--gdb-pl',
+        type=lambda x: int(x, 0),
+        help="Drop into gdb on this specific powerloss.")
+    test_parser.add_argument(
+        '--gdb-pl-before',
+        action='store_true',
+        help="Drop into gdb before the powerloss that caused the failure.")
+    test_parser.add_argument(
+        '--gdb-pl-after',
+        action='store_true',
+        help="Drop into gdb after the powerloss that caused the failure.")
+    test_parser.add_argument(
+        '--gdb-path',
+        type=lambda x: x.split(),
+        default=GDB_PATH,
+        help="Path to the gdb executable, may include flags. "
+            "Defaults to %r." % GDB_PATH)
+    test_parser.add_argument(
+        '--exec',
+        type=lambda e: e.split(),
+        help="Run under another executable.")
+    test_parser.add_argument(
+        '--valgrind',
+        action='store_true',
+        help="Run under Valgrind to find memory errors. Implicitly sets "
+            "--isolate.")
+    test_parser.add_argument(
+        '--valgrind-path',
+        type=lambda x: x.split(),
+        default=VALGRIND_PATH,
+        help="Path to the Valgrind executable, may include flags. "
+            "Defaults to %r." % VALGRIND_PATH)
+    test_parser.add_argument(
+        '-p', '--perf',
+        help="Run under Linux's perf to sample performance counters, writing "
+            "samples to this file.")
+    test_parser.add_argument(
+        '--perf-freq',
+        help="perf sampling frequency. This is passed directly to the perf "
+            "script.")
+    test_parser.add_argument(
+        '--perf-period',
+        help="perf sampling period. This is passed directly to the perf "
+            "script.")
+    test_parser.add_argument(
+        '--perf-events',
+        help="perf events to record. This is passed directly to the perf "
+            "script.")
+    test_parser.add_argument(
+        '--perf-script',
+        type=lambda x: x.split(),
+        default=PERF_SCRIPT,
+        help="Path to the perf script to use. Defaults to %r." % PERF_SCRIPT)
+    test_parser.add_argument(
+        '--perf-path',
+        type=lambda x: x.split(),
+        help="Path to the perf executable, may include flags. This is passed "
+            "directly to the perf script")
+
+    # compilation flags
+    comp_parser = parser.add_argument_group('compilation options')
+    comp_parser.add_argument(
+        'test_paths',
+        nargs='*',
+        help="Description of *.toml files to compile. May be a directory "
+            "or a list of paths.")
+    comp_parser.add_argument(
+        '-c', '--compile',
+        action='store_true',
+        help="Compile a test suite or source file.")
+    comp_parser.add_argument(
+        '-s', '--source',
+        help="Source file to compile, possibly injecting internal tests.")
+    comp_parser.add_argument(
+        '--include',
+        default=HEADER_PATH,
+        help="Inject this header file into every compiled test file. "
+            "Defaults to %r." % HEADER_PATH)
+    comp_parser.add_argument(
+        '-o', '--output',
+        help="Output file.")
+
+    # runner/test_paths overlap, so need to do some munging here
+    args = parser.parse_intermixed_args()
+    args.test_paths = [' '.join(args.runner or [])] + args.test_ids
+    args.runner = args.runner or [RUNNER_PATH]
+
+    sys.exit(main(**{k: v
+        for k, v in vars(args).items()
+        if v is not None}))
diff --git a/scripts/tracebd.py b/scripts/tracebd.py
new file mode 100644
index 0000000..ecf49a7
--- /dev/null
+++ b/scripts/tracebd.py
@@ -0,0 +1,1002 @@
+#!/usr/bin/env python3
+#
+# Display operations on block devices based on trace output
+#
+# Example:
+# ./scripts/tracebd.py trace
+#
+# Copyright (c) 2022, The littlefs authors.
+# SPDX-License-Identifier: BSD-3-Clause
+#
+
+import collections as co
+import functools as ft
+import io
+import itertools as it
+import math as m
+import os
+import re
+import shutil
+import threading as th
+import time
+
+
+CHARS = 'rpe.'
+COLORS = ['42', '45', '44', '']
+
+WEAR_CHARS = '0123456789'
+WEAR_CHARS_SUBSCRIPTS = '.₁₂₃₄₅₆789'
+WEAR_COLORS = ['', '', '', '', '', '', '', '35', '35', '1;31']
+
+CHARS_DOTS = " .':"
+COLORS_DOTS = ['32', '35', '34', '']
+CHARS_BRAILLE = (
+    '⠀⢀⡀⣀⠠⢠⡠⣠⠄⢄⡄⣄⠤⢤⡤⣤' '⠐⢐⡐⣐⠰⢰⡰⣰⠔⢔⡔⣔⠴⢴⡴⣴'
+    '⠂⢂⡂⣂⠢⢢⡢⣢⠆⢆⡆⣆⠦⢦⡦⣦' '⠒⢒⡒⣒⠲⢲⡲⣲⠖⢖⡖⣖⠶⢶⡶⣶'
+    '⠈⢈⡈⣈⠨⢨⡨⣨⠌⢌⡌⣌⠬⢬⡬⣬' '⠘⢘⡘⣘⠸⢸⡸⣸⠜⢜⡜⣜⠼⢼⡼⣼'
+    '⠊⢊⡊⣊⠪⢪⡪⣪⠎⢎⡎⣎⠮⢮⡮⣮' '⠚⢚⡚⣚⠺⢺⡺⣺⠞⢞⡞⣞⠾⢾⡾⣾'
+    '⠁⢁⡁⣁⠡⢡⡡⣡⠅⢅⡅⣅⠥⢥⡥⣥' '⠑⢑⡑⣑⠱⢱⡱⣱⠕⢕⡕⣕⠵⢵⡵⣵'
+    '⠃⢃⡃⣃⠣⢣⡣⣣⠇⢇⡇⣇⠧⢧⡧⣧' '⠓⢓⡓⣓⠳⢳⡳⣳⠗⢗⡗⣗⠷⢷⡷⣷'
+    '⠉⢉⡉⣉⠩⢩⡩⣩⠍⢍⡍⣍⠭⢭⡭⣭' '⠙⢙⡙⣙⠹⢹⡹⣹⠝⢝⡝⣝⠽⢽⡽⣽'
+    '⠋⢋⡋⣋⠫⢫⡫⣫⠏⢏⡏⣏⠯⢯⡯⣯' '⠛⢛⡛⣛⠻⢻⡻⣻⠟⢟⡟⣟⠿⢿⡿⣿')
+
+
+def openio(path, mode='r', buffering=-1):
+    # allow '-' for stdin/stdout
+    if path == '-':
+        if mode == 'r':
+            return os.fdopen(os.dup(sys.stdin.fileno()), mode, buffering)
+        else:
+            return os.fdopen(os.dup(sys.stdout.fileno()), mode, buffering)
+    else:
+        return open(path, mode, buffering)
+
+class LinesIO:
+    def __init__(self, maxlen=None):
+        self.maxlen = maxlen
+        self.lines = co.deque(maxlen=maxlen)
+        self.tail = io.StringIO()
+
+        # trigger automatic sizing
+        if maxlen == 0:
+            self.resize(0)
+
+    def write(self, s):
+        # note using split here ensures the trailing string has no newline
+        lines = s.split('\n')
+
+        if len(lines) > 1 and self.tail.getvalue():
+            self.tail.write(lines[0])
+            lines[0] = self.tail.getvalue()
+            self.tail = io.StringIO()
+
+        self.lines.extend(lines[:-1])
+
+        if lines[-1]:
+            self.tail.write(lines[-1])
+
+    def resize(self, maxlen):
+        self.maxlen = maxlen
+        if maxlen == 0:
+            maxlen = shutil.get_terminal_size((80, 5))[1]
+        if maxlen != self.lines.maxlen:
+            self.lines = co.deque(self.lines, maxlen=maxlen)
+
+    canvas_lines = 1
+    def draw(self):
+        # did terminal size change?
+        if self.maxlen == 0:
+            self.resize(0)
+
+        # first thing first, give ourself a canvas
+        while LinesIO.canvas_lines < len(self.lines):
+            sys.stdout.write('\n')
+            LinesIO.canvas_lines += 1
+
+        # clear the bottom of the canvas if we shrink
+        shrink = LinesIO.canvas_lines - len(self.lines)
+        if shrink > 0:
+            for i in range(shrink):
+                sys.stdout.write('\r')
+                if shrink-1-i > 0:
+                    sys.stdout.write('\x1b[%dA' % (shrink-1-i))
+                sys.stdout.write('\x1b[K')
+                if shrink-1-i > 0:
+                    sys.stdout.write('\x1b[%dB' % (shrink-1-i))
+            sys.stdout.write('\x1b[%dA' % shrink)
+            LinesIO.canvas_lines = len(self.lines)
+
+        for i, line in enumerate(self.lines):
+            # move cursor, clear line, disable/reenable line wrapping
+            sys.stdout.write('\r')
+            if len(self.lines)-1-i > 0:
+                sys.stdout.write('\x1b[%dA' % (len(self.lines)-1-i))
+            sys.stdout.write('\x1b[K')
+            sys.stdout.write('\x1b[?7l')
+            sys.stdout.write(line)
+            sys.stdout.write('\x1b[?7h')
+            if len(self.lines)-1-i > 0:
+                sys.stdout.write('\x1b[%dB' % (len(self.lines)-1-i))
+        sys.stdout.flush()
+
+
+# space filling Hilbert-curve
+#
+# note we memoize the last curve since this is a bit expensive
+#
+@ft.lru_cache(1)
+def hilbert_curve(width, height):
+    # based on generalized Hilbert curves:
+    # https://github.com/jakubcerveny/gilbert
+    #
+    def hilbert_(x, y, a_x, a_y, b_x, b_y):
+        w = abs(a_x+a_y)
+        h = abs(b_x+b_y)
+        a_dx = -1 if a_x < 0 else +1 if a_x > 0 else 0
+        a_dy = -1 if a_y < 0 else +1 if a_y > 0 else 0
+        b_dx = -1 if b_x < 0 else +1 if b_x > 0 else 0
+        b_dy = -1 if b_y < 0 else +1 if b_y > 0 else 0
+
+        # trivial row
+        if h == 1:
+            for _ in range(w):
+                yield (x,y)
+                x, y = x+a_dx, y+a_dy
+            return
+
+        # trivial column
+        if w == 1:
+            for _ in range(h):
+                yield (x,y)
+                x, y = x+b_dx, y+b_dy
+            return
+
+        a_x_, a_y_ = a_x//2, a_y//2
+        b_x_, b_y_ = b_x//2, b_y//2
+        w_ = abs(a_x_+a_y_)
+        h_ = abs(b_x_+b_y_)
+
+        if 2*w > 3*h:
+            # prefer even steps
+            if w_ % 2 != 0 and w > 2:
+                a_x_, a_y_ = a_x_+a_dx, a_y_+a_dy
+
+            # split in two
+            yield from hilbert_(x, y, a_x_, a_y_, b_x, b_y)
+            yield from hilbert_(x+a_x_, y+a_y_, a_x-a_x_, a_y-a_y_, b_x, b_y)
+        else:
+            # prefer even steps
+            if h_ % 2 != 0 and h > 2:
+                b_x_, b_y_ = b_x_+b_dx, b_y_+b_dy
+
+            # split in three
+            yield from hilbert_(x, y, b_x_, b_y_, a_x_, a_y_)
+            yield from hilbert_(x+b_x_, y+b_y_, a_x, a_y, b_x-b_x_, b_y-b_y_)
+            yield from hilbert_(
+                x+(a_x-a_dx)+(b_x_-b_dx), y+(a_y-a_dy)+(b_y_-b_dy),
+                -b_x_, -b_y_, -(a_x-a_x_), -(a_y-a_y_))
+
+    if width >= height:
+        curve = hilbert_(0, 0, +width, 0, 0, +height)
+    else:
+        curve = hilbert_(0, 0, 0, +height, +width, 0)
+
+    return list(curve)
+
+# space filling Z-curve/Lebesgue-curve
+#
+# note we memoize the last curve since this is a bit expensive
+#
+@ft.lru_cache(1)
+def lebesgue_curve(width, height):
+    # we create a truncated Z-curve by simply filtering out the points
+    # that are outside our region
+    curve = []
+    for i in range(2**(2*m.ceil(m.log2(max(width, height))))):
+        # we just operate on binary strings here because it's easier
+        b = '{:0{}b}'.format(i, 2*m.ceil(m.log2(i+1)/2))
+        x = int(b[1::2], 2) if b[1::2] else 0
+        y = int(b[0::2], 2) if b[0::2] else 0
+        if x < width and y < height:
+            curve.append((x, y))
+
+    return curve
+
+
+class Block(int):
+    __slots__ = ()
+    def __new__(cls, state=0, *,
+            wear=0,
+            readed=False,
+            proged=False,
+            erased=False):
+        return super().__new__(cls,
+            state
+            | (wear << 3)
+            | (1 if readed else 0)
+            | (2 if proged else 0)
+            | (4 if erased else 0))
+
+    @property
+    def wear(self):
+        return self >> 3
+
+    @property
+    def readed(self):
+        return (self & 1) != 0
+
+    @property
+    def proged(self):
+        return (self & 2) != 0
+
+    @property
+    def erased(self):
+        return (self & 4) != 0
+
+    def read(self):
+        return Block(int(self) | 1)
+
+    def prog(self):
+        return Block(int(self) | 2)
+
+    def erase(self):
+        return Block((int(self) | 4) + 8)
+
+    def clear(self):
+        return Block(int(self) & ~7)
+
+    def __or__(self, other):
+        return Block(
+            (int(self) | int(other)) & 7,
+            wear=max(self.wear, other.wear))
+
+    def worn(self, max_wear, *,
+            block_cycles=None,
+            wear_chars=None,
+            **_):
+        if wear_chars is None:
+            wear_chars = WEAR_CHARS
+
+        if block_cycles:
+            return self.wear / block_cycles
+        else:
+            return self.wear / max(max_wear, len(wear_chars))
+
+    def draw(self, max_wear, char=None, *,
+            read=True,
+            prog=True,
+            erase=True,
+            wear=False,
+            block_cycles=None,
+            color=True,
+            subscripts=False,
+            dots=False,
+            braille=False,
+            chars=None,
+            wear_chars=None,
+            colors=None,
+            wear_colors=None,
+            **_):
+        # fallback to default chars/colors
+        if chars is None:
+            chars = CHARS
+        if len(chars) < len(CHARS):
+            chars = chars + CHARS[len(chars):]
+
+        if colors is None:
+            if braille or dots:
+                colors = COLORS_DOTS
+            else:
+                colors = COLORS
+        if len(colors) < len(COLORS):
+            colors = colors + COLORS[len(colors):]
+
+        if wear_chars is None:
+            if subscripts:
+                wear_chars = WEAR_CHARS_SUBSCRIPTS
+            else:
+                wear_chars = WEAR_CHARS
+
+        if wear_colors is None:
+            wear_colors = WEAR_COLORS
+
+        # compute char/color
+        c = chars[3]
+        f = [colors[3]]
+
+        if wear:
+            w = min(
+                self.worn(
+                    max_wear,
+                    block_cycles=block_cycles,
+                    wear_chars=wear_chars),
+                1)
+
+            c = wear_chars[int(w * (len(wear_chars)-1))]
+            f.append(wear_colors[int(w * (len(wear_colors)-1))])
+
+        if erase and self.erased:
+            c = chars[2]
+            f.append(colors[2])
+        elif prog and self.proged:
+            c = chars[1]
+            f.append(colors[1])
+        elif read and self.readed:
+            c = chars[0]
+            f.append(colors[0])
+
+        # override char?
+        if char:
+            c = char
+
+        # apply colors
+        if f and color:
+            c = '%s%s\x1b[m' % (
+                ''.join('\x1b[%sm' % f_ for f_ in f),
+                c)
+
+        return c
+
+
+class Bd:
+    def __init__(self, *,
+            size=1,
+            count=1,
+            width=None,
+            height=1,
+            blocks=None):
+        if width is None:
+            width = count
+
+        if blocks is None:
+            self.blocks = [Block() for _ in range(width*height)]
+        else:
+            self.blocks = blocks
+        self.size = size
+        self.count = count
+        self.width = width
+        self.height = height
+
+    def _op(self, f, block=None, off=None, size=None):
+        if block is None:
+            range_ = range(len(self.blocks))
+        else:
+            if off is None:
+                off, size = 0, self.size
+            elif size is None:
+                off, size = 0, off
+
+            # update our geometry? this will do nothing if we haven't changed
+            self.resize(
+                size=max(self.size, off+size),
+                count=max(self.count, block+1))
+
+            # map to our block space
+            start = (block*self.size + off) / (self.size*self.count)
+            stop = (block*self.size + off+size) / (self.size*self.count)
+
+            range_ = range(
+                m.floor(start*len(self.blocks)),
+                m.ceil(stop*len(self.blocks)))
+
+        # apply the op
+        for i in range_:
+            self.blocks[i] = f(self.blocks[i])
+
+    def read(self, block=None, off=None, size=None):
+        self._op(Block.read, block, off, size)
+
+    def prog(self, block=None, off=None, size=None):
+        self._op(Block.prog, block, off, size)
+
+    def erase(self, block=None, off=None, size=None):
+        self._op(Block.erase, block, off, size)
+
+    def clear(self, block=None, off=None, size=None):
+        self._op(Block.clear, block, off, size)
+
+    def copy(self):
+        return Bd(
+            blocks=self.blocks.copy(),
+            size=self.size,
+            count=self.count,
+            width=self.width,
+            height=self.height)
+
+    def resize(self, *,
+            size=None,
+            count=None,
+            width=None,
+            height=None):
+        size = size if size is not None else self.size
+        count = count if count is not None else self.count
+        width = width if width is not None else self.width
+        height = height if height is not None else self.height
+
+        if (size == self.size
+                and count == self.count
+                and width == self.width
+                and height == self.height):
+            return
+
+        # transform our blocks
+        blocks = []
+        for x in range(width*height):
+            # map from new bd space
+            start = m.floor(x * (size*count)/(width*height))
+            stop = m.ceil((x+1) * (size*count)/(width*height))
+            start_block = start // size
+            start_off = start % size
+            stop_block = stop // size
+            stop_off = stop % size
+            # map to old bd space
+            start = start_block*self.size + start_off
+            stop = stop_block*self.size + stop_off
+            start = m.floor(start * len(self.blocks)/(self.size*self.count))
+            stop = m.ceil(stop * len(self.blocks)/(self.size*self.count))
+
+            # aggregate state
+            blocks.append(ft.reduce(
+                Block.__or__,
+                self.blocks[start:stop],
+                Block()))
+            
+        self.size = size
+        self.count = count
+        self.width = width
+        self.height = height
+        self.blocks = blocks
+
+    def draw(self, row, *,
+            read=False,
+            prog=False,
+            erase=False,
+            wear=False,
+            hilbert=False,
+            lebesgue=False,
+            dots=False,
+            braille=False,
+            **args):
+        # find max wear?
+        max_wear = None
+        if wear:
+            max_wear = max(b.wear for b in self.blocks)
+
+        # fold via a curve?
+        if hilbert:
+            grid = [None]*(self.width*self.height)
+            for (x,y), b in zip(
+                    hilbert_curve(self.width, self.height),
+                    self.blocks):
+                grid[x + y*self.width] = b
+        elif lebesgue:
+            grid = [None]*(self.width*self.height)
+            for (x,y), b in zip(
+                    lebesgue_curve(self.width, self.height),
+                    self.blocks):
+                grid[x + y*self.width] = b
+        else:
+            grid = self.blocks
+
+        # need to wait for more trace output before rendering
+        #
+        # this is sort of a hack that knows the output is going to a terminal
+        if (braille and self.height < 4) or (dots and self.height < 2):
+            needed_height = 4 if braille else 2
+
+            self.history = getattr(self, 'history', [])
+            self.history.append(grid)
+
+            if len(self.history)*self.height < needed_height:
+                # skip for now
+                return None
+
+            grid = list(it.chain.from_iterable(
+                # did we resize?
+                it.islice(it.chain(h, it.repeat(Block())),
+                    self.width*self.height)
+                for h in self.history))
+            self.history = []
+
+        line = []
+        if braille:
+            # encode into a byte
+            for x in range(0, self.width, 2):
+                byte_b = 0
+                best_b = Block()
+                for i in range(2*4):
+                    b = grid[x+(2-1-(i%2)) + ((row*4)+(4-1-(i//2)))*self.width]
+                    best_b |= b
+                    if ((read and b.readed)
+                            or (prog and b.proged)
+                            or (erase and b.erased)
+                            or (not read and not prog and not erase
+                                and wear and b.worn(max_wear, **args) >= 0.7)):
+                        byte_b |= 1 << i
+
+                line.append(best_b.draw(
+                    max_wear,
+                    CHARS_BRAILLE[byte_b],
+                    braille=True,
+                    read=read,
+                    prog=prog,
+                    erase=erase,
+                    wear=wear,
+                    **args))
+        elif dots:
+            # encode into a byte
+            for x in range(self.width):
+                byte_b = 0
+                best_b = Block()
+                for i in range(2):
+                    b = grid[x + ((row*2)+(2-1-i))*self.width]
+                    best_b |= b
+                    if ((read and b.readed)
+                            or (prog and b.proged)
+                            or (erase and b.erased)
+                            or (not read and not prog and not erase
+                                and wear and b.worn(max_wear, **args) >= 0.7)):
+                        byte_b |= 1 << i
+
+                line.append(best_b.draw(
+                    max_wear,
+                    CHARS_DOTS[byte_b],
+                    dots=True,
+                    read=read,
+                    prog=prog,
+                    erase=erase,
+                    wear=wear,
+                    **args))
+        else:
+            for x in range(self.width):
+                line.append(grid[x + row*self.width].draw(
+                    max_wear,
+                    read=read,
+                    prog=prog,
+                    erase=erase,
+                    wear=wear,
+                    **args))
+
+        return ''.join(line)
+
+
+
+def main(path='-', *,
+        read=False,
+        prog=False,
+        erase=False,
+        wear=False,
+        block=(None,None),
+        off=(None,None),
+        block_size=None,
+        block_count=None,
+        block_cycles=None,
+        reset=False,
+        color='auto',
+        dots=False,
+        braille=False,
+        width=None,
+        height=None,
+        lines=None,
+        cat=False,
+        hilbert=False,
+        lebesgue=False,
+        coalesce=None,
+        sleep=None,
+        keep_open=False,
+        **args):
+    # figure out what color should be
+    if color == 'auto':
+        color = sys.stdout.isatty()
+    elif color == 'always':
+        color = True
+    else:
+        color = False
+
+    # exclusive wear or read/prog/erase by default
+    if not read and not prog and not erase and not wear:
+        read = True
+        prog = True
+        erase = True
+
+    # assume a reasonable lines/height if not specified
+    #
+    # note that we let height = None if neither hilbert or lebesgue
+    # are specified, this is a bit special as the default may be less
+    # than one character in height.
+    if height is None and (hilbert or lebesgue):
+        if lines is not None:
+            height = lines
+        else:
+            height = 5
+
+    if lines is None:
+        if height is not None:
+            lines = height
+        else:
+            lines = 5
+
+    # allow ranges for blocks/offs
+    block_start = block[0]
+    block_stop = block[1] if len(block) > 1 else block[0]+1
+    off_start = off[0]
+    off_stop = off[1] if len(off) > 1 else off[0]+1
+
+    if block_start is None:
+        block_start = 0
+    if block_stop is None and block_count is not None:
+        block_stop = block_count
+    if off_start is None:
+        off_start = 0
+    if off_stop is None and block_size is not None:
+        off_stop = block_size
+
+    # create a block device representation
+    bd = Bd()
+
+    def resize(*, size=None, count=None):
+        nonlocal bd
+
+        # size may be overriden by cli args
+        if block_size is not None:
+            size = block_size
+        elif off_stop is not None:
+            size = off_stop-off_start
+
+        if block_count is not None:
+            count = block_count
+        elif block_stop is not None:
+            count = block_stop-block_start
+
+        # figure out best width/height
+        if width is None:
+            width_ = min(80, shutil.get_terminal_size((80, 5))[0])
+        elif width:
+            width_ = width
+        else:
+            width_ = shutil.get_terminal_size((80, 5))[0]
+
+        if height is None:
+            height_ = 0
+        elif height:
+            height_ = height
+        else:
+            height_ = shutil.get_terminal_size((80, 5))[1]
+
+        bd.resize(
+            size=size,
+            count=count,
+            # scale if we're printing with dots or braille
+            width=2*width_ if braille else width_,
+            height=max(1,
+                4*height_ if braille
+                else 2*height_ if dots
+                else height_))
+    resize()
+
+    # parse a line of trace output
+    pattern = re.compile(
+        '^(?P<file>[^:]*):(?P<line>[0-9]+):trace:.*?bd_(?:'
+            '(?P<create>create\w*)\('
+                '(?:'
+                    'block_size=(?P<block_size>\w+)'
+                    '|' 'block_count=(?P<block_count>\w+)'
+                    '|' '.*?' ')*' '\)'
+            '|' '(?P<read>read)\('
+                '\s*(?P<read_ctx>\w+)' '\s*,'
+                '\s*(?P<read_block>\w+)' '\s*,'
+                '\s*(?P<read_off>\w+)' '\s*,'
+                '\s*(?P<read_buffer>\w+)' '\s*,'
+                '\s*(?P<read_size>\w+)' '\s*\)'
+            '|' '(?P<prog>prog)\('
+                '\s*(?P<prog_ctx>\w+)' '\s*,'
+                '\s*(?P<prog_block>\w+)' '\s*,'
+                '\s*(?P<prog_off>\w+)' '\s*,'
+                '\s*(?P<prog_buffer>\w+)' '\s*,'
+                '\s*(?P<prog_size>\w+)' '\s*\)'
+            '|' '(?P<erase>erase)\('
+                '\s*(?P<erase_ctx>\w+)' '\s*,'
+                '\s*(?P<erase_block>\w+)'
+                '\s*\(\s*(?P<erase_size>\w+)\s*\)' '\s*\)'
+            '|' '(?P<sync>sync)\('
+                '\s*(?P<sync_ctx>\w+)' '\s*\)' ')\s*$')
+    def parse(line):
+        nonlocal bd
+
+        # string searching is much faster than the regex here, and this
+        # actually has a big impact given how much trace output comes
+        # through here
+        if 'trace' not in line or 'bd' not in line:
+            return False
+        m = pattern.match(line)
+        if not m:
+            return False
+
+        if m.group('create'):
+            # update our block size/count
+            size = int(m.group('block_size'), 0)
+            count = int(m.group('block_count'), 0)
+
+            resize(size=size, count=count)
+            if reset:
+                bd = Bd(
+                    size=bd.size,
+                    count=bd.count,
+                    width=bd.width,
+                    height=bd.height)
+            return True
+
+        elif m.group('read') and read:
+            block = int(m.group('read_block'), 0)
+            off = int(m.group('read_off'), 0)
+            size = int(m.group('read_size'), 0)
+
+            if block_stop is not None and block >= block_stop:
+                return False
+            block -= block_start
+            if off_stop is not None:
+                if off >= off_stop:
+                    return False
+                size = min(size, off_stop-off)
+            off -= off_start
+
+            bd.read(block, off, size)
+            return True
+
+        elif m.group('prog') and prog:
+            block = int(m.group('prog_block'), 0)
+            off = int(m.group('prog_off'), 0)
+            size = int(m.group('prog_size'), 0)
+
+            if block_stop is not None and block >= block_stop:
+                return False
+            block -= block_start
+            if off_stop is not None:
+                if off >= off_stop:
+                    return False
+                size = min(size, off_stop-off)
+            off -= off_start
+
+            bd.prog(block, off, size)
+            return True
+
+        elif m.group('erase') and (erase or wear):
+            block = int(m.group('erase_block'), 0)
+            size = int(m.group('erase_size'), 0)
+
+            if block_stop is not None and block >= block_stop:
+                return False
+            block -= block_start
+            if off_stop is not None:
+                size = min(size, off_stop)
+            off = -off_start
+
+            bd.erase(block, off, size)
+            return True
+
+        else:
+            return False
+
+    # print trace output
+    def draw(f):
+        def writeln(s=''):
+            f.write(s)
+            f.write('\n')
+        f.writeln = writeln
+
+        # don't forget we've scaled this for braille/dots!
+        for row in range(
+                m.ceil(bd.height/4) if braille
+                else m.ceil(bd.height/2) if dots
+                else bd.height):
+            line = bd.draw(row,
+                read=read,
+                prog=prog,
+                erase=erase,
+                wear=wear,
+                block_cycles=block_cycles,
+                color=color,
+                dots=dots,
+                braille=braille,
+                hilbert=hilbert,
+                lebesgue=lebesgue,
+                **args)
+            if line:
+                f.writeln(line)
+
+        bd.clear()
+        resize()
+
+
+    # read/parse/coalesce operations
+    if cat:
+        ring = sys.stdout
+    else:
+        ring = LinesIO(lines)
+
+    # if sleep print in background thread to avoid getting stuck in a read call
+    event = th.Event()
+    lock = th.Lock()
+    if sleep:
+        done = False
+        def background():
+            while not done:
+                event.wait()
+                event.clear()
+                with lock:
+                    draw(ring)
+                    if not cat:
+                        ring.draw()
+                time.sleep(sleep or 0.01)
+        th.Thread(target=background, daemon=True).start()
+
+    try:
+        while True:
+            with openio(path) as f:
+                changed = 0
+                for line in f:
+                    with lock:
+                        changed += parse(line)
+
+                        # need to redraw?
+                        if changed and (not coalesce or changed >= coalesce):
+                            if sleep:
+                                event.set()
+                            else:
+                                draw(ring)
+                                if not cat:
+                                    ring.draw()
+                            changed = 0
+
+            if not keep_open:
+                break
+            # don't just flood open calls
+            time.sleep(sleep or 0.1)
+    except FileNotFoundError as e:
+        print("error: file not found %r" % path)
+        sys.exit(-1)
+    except KeyboardInterrupt:
+        pass
+
+    if sleep:
+        done = True
+        lock.acquire() # avoids https://bugs.python.org/issue42717
+    if not cat:
+        sys.stdout.write('\n')
+
+
+if __name__ == "__main__":
+    import sys
+    import argparse
+    parser = argparse.ArgumentParser(
+        description="Display operations on block devices based on "
+            "trace output.",
+        allow_abbrev=False)
+    parser.add_argument(
+        'path',
+        nargs='?',
+        help="Path to read from.")
+    parser.add_argument(
+        '-r', '--read',
+        action='store_true',
+        help="Render reads.")
+    parser.add_argument(
+        '-p', '--prog',
+        action='store_true',
+        help="Render progs.")
+    parser.add_argument(
+        '-e', '--erase',
+        action='store_true',
+        help="Render erases.")
+    parser.add_argument(
+        '-w', '--wear',
+        action='store_true',
+        help="Render wear.")
+    parser.add_argument(
+        '-b', '--block',
+        type=lambda x: tuple(
+            int(x, 0) if x.strip() else None
+            for x in x.split(',')),
+        help="Show a specific block or range of blocks.")
+    parser.add_argument(
+        '-i', '--off',
+        type=lambda x: tuple(
+            int(x, 0) if x.strip() else None
+            for x in x.split(',')),
+        help="Show a specific offset or range of offsets.")
+    parser.add_argument(
+        '-B', '--block-size',
+        type=lambda x: int(x, 0),
+        help="Assume a specific block size.")
+    parser.add_argument(
+        '--block-count',
+        type=lambda x: int(x, 0),
+        help="Assume a specific block count.")
+    parser.add_argument(
+        '-C', '--block-cycles',
+        type=lambda x: int(x, 0),
+        help="Assumed maximum number of erase cycles when measuring wear.")
+    parser.add_argument(
+        '-R', '--reset',
+        action='store_true',
+        help="Reset wear on block device initialization.")
+    parser.add_argument(
+        '--color',
+        choices=['never', 'always', 'auto'],
+        default='auto',
+        help="When to use terminal colors. Defaults to 'auto'.")
+    parser.add_argument(
+        '--subscripts',
+        action='store_true',
+        help="Use unicode subscripts for showing wear.")
+    parser.add_argument(
+        '-:', '--dots',
+        action='store_true',
+        help="Use 1x2 ascii dot characters.")
+    parser.add_argument(
+        '-⣿', '--braille',
+        action='store_true',
+        help="Use 2x4 unicode braille characters. Note that braille characters "
+            "sometimes suffer from inconsistent widths.")
+    parser.add_argument(
+        '--chars',
+        help="Characters to use for read, prog, erase, noop operations.")
+    parser.add_argument(
+        '--wear-chars',
+        help="Characters to use for showing wear.")
+    parser.add_argument(
+        '--colors',
+        type=lambda x: [x.strip() for x in x.split(',')],
+        help="Colors to use for read, prog, erase, noop operations.")
+    parser.add_argument(
+        '--wear-colors',
+        type=lambda x: [x.strip() for x in x.split(',')],
+        help="Colors to use for showing wear.")
+    parser.add_argument(
+        '-W', '--width',
+        nargs='?',
+        type=lambda x: int(x, 0),
+        const=0,
+        help="Width in columns. 0 uses the terminal width. Defaults to "
+            "min(terminal, 80).")
+    parser.add_argument(
+        '-H', '--height',
+        nargs='?',
+        type=lambda x: int(x, 0),
+        const=0,
+        help="Height in rows. 0 uses the terminal height. Defaults to 1.")
+    parser.add_argument(
+        '-n', '--lines',
+        nargs='?',
+        type=lambda x: int(x, 0),
+        const=0,
+        help="Show this many lines of history. 0 uses the terminal height. "
+            "Defaults to 5.")
+    parser.add_argument(
+        '-z', '--cat',
+        action='store_true',
+        help="Pipe directly to stdout.")
+    parser.add_argument(
+        '-U', '--hilbert',
+        action='store_true',
+        help="Render as a space-filling Hilbert curve.")
+    parser.add_argument(
+        '-Z', '--lebesgue',
+        action='store_true',
+        help="Render as a space-filling Z-curve.")
+    parser.add_argument(
+        '-c', '--coalesce',
+        type=lambda x: int(x, 0),
+        help="Number of operations to coalesce together.")
+    parser.add_argument(
+        '-s', '--sleep',
+        type=float,
+        help="Time in seconds to sleep between reads, coalescing operations.")
+    parser.add_argument(
+        '-k', '--keep-open',
+        action='store_true',
+        help="Reopen the pipe on EOF, useful when multiple "
+            "processes are writing.")
+    sys.exit(main(**{k: v
+        for k, v in vars(parser.parse_intermixed_args()).items()
+        if v is not None}))
diff --git a/scripts/watch.py b/scripts/watch.py
new file mode 100644
index 0000000..dff0601
--- /dev/null
+++ b/scripts/watch.py
@@ -0,0 +1,265 @@
+#!/usr/bin/env python3
+#
+# Traditional watch command, but with higher resolution updates and a bit
+# different options/output format
+#
+# Example:
+# ./scripts/watch.py -s0.1 date
+#
+# Copyright (c) 2022, The littlefs authors.
+# SPDX-License-Identifier: BSD-3-Clause
+#
+
+import collections as co
+import errno
+import fcntl
+import io
+import os
+import pty
+import re
+import shutil
+import struct
+import subprocess as sp
+import sys
+import termios
+import time
+
+try:
+    import inotify_simple
+except ModuleNotFoundError:
+    inotify_simple = None
+
+
+def openio(path, mode='r', buffering=-1):
+    # allow '-' for stdin/stdout
+    if path == '-':
+        if mode == 'r':
+            return os.fdopen(os.dup(sys.stdin.fileno()), mode, buffering)
+        else:
+            return os.fdopen(os.dup(sys.stdout.fileno()), mode, buffering)
+    else:
+        return open(path, mode, buffering)
+
+def inotifywait(paths):
+    # wait for interesting events
+    inotify = inotify_simple.INotify()
+    flags = (inotify_simple.flags.ATTRIB
+        | inotify_simple.flags.CREATE
+        | inotify_simple.flags.DELETE
+        | inotify_simple.flags.DELETE_SELF
+        | inotify_simple.flags.MODIFY
+        | inotify_simple.flags.MOVED_FROM
+        | inotify_simple.flags.MOVED_TO
+        | inotify_simple.flags.MOVE_SELF)
+
+    # recurse into directories
+    for path in paths:
+        if os.path.isdir(path):
+            for dir, _, files in os.walk(path):
+                inotify.add_watch(dir, flags)
+                for f in files:
+                    inotify.add_watch(os.path.join(dir, f), flags)
+        else:
+            inotify.add_watch(path, flags)
+
+    # wait for event
+    inotify.read()
+
+class LinesIO:
+    def __init__(self, maxlen=None):
+        self.maxlen = maxlen
+        self.lines = co.deque(maxlen=maxlen)
+        self.tail = io.StringIO()
+
+        # trigger automatic sizing
+        if maxlen == 0:
+            self.resize(0)
+
+    def write(self, s):
+        # note using split here ensures the trailing string has no newline
+        lines = s.split('\n')
+
+        if len(lines) > 1 and self.tail.getvalue():
+            self.tail.write(lines[0])
+            lines[0] = self.tail.getvalue()
+            self.tail = io.StringIO()
+
+        self.lines.extend(lines[:-1])
+
+        if lines[-1]:
+            self.tail.write(lines[-1])
+
+    def resize(self, maxlen):
+        self.maxlen = maxlen
+        if maxlen == 0:
+            maxlen = shutil.get_terminal_size((80, 5))[1]
+        if maxlen != self.lines.maxlen:
+            self.lines = co.deque(self.lines, maxlen=maxlen)
+
+    canvas_lines = 1
+    def draw(self):
+        # did terminal size change?
+        if self.maxlen == 0:
+            self.resize(0)
+
+        # first thing first, give ourself a canvas
+        while LinesIO.canvas_lines < len(self.lines):
+            sys.stdout.write('\n')
+            LinesIO.canvas_lines += 1
+
+        # clear the bottom of the canvas if we shrink
+        shrink = LinesIO.canvas_lines - len(self.lines)
+        if shrink > 0:
+            for i in range(shrink):
+                sys.stdout.write('\r')
+                if shrink-1-i > 0:
+                    sys.stdout.write('\x1b[%dA' % (shrink-1-i))
+                sys.stdout.write('\x1b[K')
+                if shrink-1-i > 0:
+                    sys.stdout.write('\x1b[%dB' % (shrink-1-i))
+            sys.stdout.write('\x1b[%dA' % shrink)
+            LinesIO.canvas_lines = len(self.lines)
+
+        for i, line in enumerate(self.lines):
+            # move cursor, clear line, disable/reenable line wrapping
+            sys.stdout.write('\r')
+            if len(self.lines)-1-i > 0:
+                sys.stdout.write('\x1b[%dA' % (len(self.lines)-1-i))
+            sys.stdout.write('\x1b[K')
+            sys.stdout.write('\x1b[?7l')
+            sys.stdout.write(line)
+            sys.stdout.write('\x1b[?7h')
+            if len(self.lines)-1-i > 0:
+                sys.stdout.write('\x1b[%dB' % (len(self.lines)-1-i))
+        sys.stdout.flush()
+
+
+def main(command, *,
+        lines=0,
+        cat=False,
+        sleep=None,
+        keep_open=False,
+        keep_open_paths=None,
+        exit_on_error=False):
+    returncode = 0
+    try:
+        while True:
+            # reset ring each run
+            if cat:
+                ring = sys.stdout
+            else:
+                ring = LinesIO(lines)
+
+            try:
+                # run the command under a pseudoterminal 
+                mpty, spty = pty.openpty()
+
+                # forward terminal size
+                w, h = shutil.get_terminal_size((80, 5))
+                if lines:
+                    h = lines
+                fcntl.ioctl(spty, termios.TIOCSWINSZ,
+                    struct.pack('HHHH', h, w, 0, 0))
+
+                proc = sp.Popen(command,
+                    stdout=spty,
+                    stderr=spty,
+                    close_fds=False)
+                os.close(spty)
+                mpty = os.fdopen(mpty, 'r', 1)
+
+                while True:
+                    try:
+                        line = mpty.readline()
+                    except OSError as e:
+                        if e.errno != errno.EIO:
+                            raise
+                        break
+                    if not line:
+                        break
+
+                    ring.write(line)
+                    if not cat:
+                        ring.draw()
+
+                mpty.close()
+                proc.wait()
+                if exit_on_error and proc.returncode != 0:
+                    returncode = proc.returncode
+                    break
+            except OSError as e:
+                if e.errno != errno.ETXTBSY:
+                    raise
+                pass
+
+            # try to inotifywait
+            if keep_open and inotify_simple is not None:
+                if keep_open_paths:
+                    paths = set(keep_paths)
+                else:
+                    # guess inotify paths from command
+                    paths = set()
+                    for p in command:
+                        for p in {
+                                p,
+                                re.sub('^-.', '', p),
+                                re.sub('^--[^=]+=', '', p)}:
+                            if p and os.path.exists(p):
+                                paths.add(p)
+                ptime = time.time()
+                inotifywait(paths)
+                # sleep for a minimum amount of time, this helps issues around
+                # rapidly updating files
+                time.sleep(max(0, (sleep or 0.1) - (time.time()-ptime)))
+            else:
+                time.sleep(sleep or 0.1)
+    except KeyboardInterrupt:
+        pass
+
+    if not cat:
+        sys.stdout.write('\n')
+    sys.exit(returncode)
+
+
+if __name__ == "__main__":
+    import sys
+    import argparse
+    parser = argparse.ArgumentParser(
+        description="Traditional watch command, but with higher resolution "
+            "updates and a bit different options/output format.",
+        allow_abbrev=False)
+    parser.add_argument(
+        'command',
+        nargs=argparse.REMAINDER,
+        help="Command to run.")
+    parser.add_argument(
+        '-n', '--lines',
+        nargs='?',
+        type=lambda x: int(x, 0),
+        const=0,
+        help="Show this many lines of history. 0 uses the terminal height. "
+            "Defaults to 0.")
+    parser.add_argument(
+        '-z', '--cat',
+        action='store_true',
+        help="Pipe directly to stdout.")
+    parser.add_argument(
+        '-s', '--sleep',
+        type=float,
+        help="Seconds to sleep between runs. Defaults to 0.1.")
+    parser.add_argument(
+        '-k', '--keep-open',
+        action='store_true',
+        help="Try to use inotify to wait for changes.")
+    parser.add_argument(
+        '-K', '--keep-open-path',
+        dest='keep_open_paths',
+        action='append',
+        help="Use this path for inotify. Defaults to guessing.")
+    parser.add_argument(
+        '-e', '--exit-on-error',
+        action='store_true',
+        help="Exit on error.")
+    sys.exit(main(**{k: v
+        for k, v in vars(parser.parse_args()).items()
+        if v is not None}))
diff --git a/tests/test_alloc.toml b/tests/test_alloc.toml
index fa92da5..e6fba97 100644
--- a/tests/test_alloc.toml
+++ b/tests/test_alloc.toml
@@ -1,27 +1,34 @@
 # allocator tests
 # note for these to work there are a number constraints on the device geometry
-if = 'LFS_BLOCK_CYCLES == -1'
+if = 'BLOCK_CYCLES == -1'
 
-[[case]] # parallel allocation test
-define.FILES = 3
-define.SIZE = '(((LFS_BLOCK_SIZE-8)*(LFS_BLOCK_COUNT-6)) / FILES)'
+# parallel allocation test
+[cases.test_alloc_parallel]
+defines.FILES = 3
+defines.SIZE = '(((BLOCK_SIZE-8)*(BLOCK_COUNT-6)) / FILES)'
+defines.GC = [false, true]
 code = '''
-    const char *names[FILES] = {"bacon", "eggs", "pancakes"};
+    const char *names[] = {"bacon", "eggs", "pancakes"};
     lfs_file_t files[FILES];
 
-    lfs_format(&lfs, &cfg) => 0;
-    lfs_mount(&lfs, &cfg) => 0;
+    lfs_t lfs;
+    lfs_format(&lfs, cfg) => 0;
+    lfs_mount(&lfs, cfg) => 0;
     lfs_mkdir(&lfs, "breakfast") => 0;
     lfs_unmount(&lfs) => 0;
 
-    lfs_mount(&lfs, &cfg) => 0;
+    lfs_mount(&lfs, cfg) => 0;
     for (int n = 0; n < FILES; n++) {
+        char path[1024];
         sprintf(path, "breakfast/%s", names[n]);
         lfs_file_open(&lfs, &files[n], path,
                 LFS_O_WRONLY | LFS_O_CREAT | LFS_O_APPEND) => 0;
     }
     for (int n = 0; n < FILES; n++) {
-        size = strlen(names[n]);
+        if (GC) {
+            lfs_fs_gc(&lfs) => 0;
+        }
+        size_t size = strlen(names[n]);
         for (lfs_size_t i = 0; i < SIZE; i += size) {
             lfs_file_write(&lfs, &files[n], names[n], size) => size;
         }
@@ -31,12 +38,15 @@ code = '''
     }
     lfs_unmount(&lfs) => 0;
 
-    lfs_mount(&lfs, &cfg) => 0;
+    lfs_mount(&lfs, cfg) => 0;
     for (int n = 0; n < FILES; n++) {
+        char path[1024];
         sprintf(path, "breakfast/%s", names[n]);
+        lfs_file_t file;
         lfs_file_open(&lfs, &file, path, LFS_O_RDONLY) => 0;
-        size = strlen(names[n]);
+        size_t size = strlen(names[n]);
         for (lfs_size_t i = 0; i < SIZE; i += size) {
+            uint8_t buffer[1024];
             lfs_file_read(&lfs, &file, buffer, size) => size;
             assert(memcmp(buffer, names[n], size) == 0);
         }
@@ -45,37 +55,49 @@ code = '''
     lfs_unmount(&lfs) => 0;
 '''
 
-[[case]] # serial allocation test
-define.FILES = 3
-define.SIZE = '(((LFS_BLOCK_SIZE-8)*(LFS_BLOCK_COUNT-6)) / FILES)'
+# serial allocation test
+[cases.test_alloc_serial]
+defines.FILES = 3
+defines.SIZE = '(((BLOCK_SIZE-8)*(BLOCK_COUNT-6)) / FILES)'
+defines.GC = [false, true]
 code = '''
-    const char *names[FILES] = {"bacon", "eggs", "pancakes"};
+    const char *names[] = {"bacon", "eggs", "pancakes"};
 
-    lfs_format(&lfs, &cfg) => 0;
-    lfs_mount(&lfs, &cfg) => 0;
+    lfs_t lfs;
+    lfs_format(&lfs, cfg) => 0;
+    lfs_mount(&lfs, cfg) => 0;
     lfs_mkdir(&lfs, "breakfast") => 0;
     lfs_unmount(&lfs) => 0;
 
     for (int n = 0; n < FILES; n++) {
-        lfs_mount(&lfs, &cfg) => 0;
+        lfs_mount(&lfs, cfg) => 0;
+        char path[1024];
         sprintf(path, "breakfast/%s", names[n]);
+        lfs_file_t file;
         lfs_file_open(&lfs, &file, path,
                 LFS_O_WRONLY | LFS_O_CREAT | LFS_O_APPEND) => 0;
-        size = strlen(names[n]);
+        size_t size = strlen(names[n]);
+        uint8_t buffer[1024];
         memcpy(buffer, names[n], size);
         for (int i = 0; i < SIZE; i += size) {
+            if (GC) {
+                lfs_fs_gc(&lfs) => 0;
+            }
             lfs_file_write(&lfs, &file, buffer, size) => size;
         }
         lfs_file_close(&lfs, &file) => 0;
         lfs_unmount(&lfs) => 0;
     }
 
-    lfs_mount(&lfs, &cfg) => 0;
+    lfs_mount(&lfs, cfg) => 0;
     for (int n = 0; n < FILES; n++) {
+        char path[1024];
         sprintf(path, "breakfast/%s", names[n]);
+        lfs_file_t file;
         lfs_file_open(&lfs, &file, path, LFS_O_RDONLY) => 0;
-        size = strlen(names[n]);
+        size_t size = strlen(names[n]);
         for (int i = 0; i < SIZE; i += size) {
+            uint8_t buffer[1024];
             lfs_file_read(&lfs, &file, buffer, size) => size;
             assert(memcmp(buffer, names[n], size) == 0);
         }
@@ -84,29 +106,32 @@ code = '''
     lfs_unmount(&lfs) => 0;
 '''
 
-[[case]] # parallel allocation reuse test
-define.FILES = 3
-define.SIZE = '(((LFS_BLOCK_SIZE-8)*(LFS_BLOCK_COUNT-6)) / FILES)'
-define.CYCLES = [1, 10]
+# parallel allocation reuse test
+[cases.test_alloc_parallel_reuse]
+defines.FILES = 3
+defines.SIZE = '(((BLOCK_SIZE-8)*(BLOCK_COUNT-6)) / FILES)'
+defines.CYCLES = [1, 10]
 code = '''
-    const char *names[FILES] = {"bacon", "eggs", "pancakes"};
+    const char *names[] = {"bacon", "eggs", "pancakes"};
     lfs_file_t files[FILES];
 
-    lfs_format(&lfs, &cfg) => 0;
+    lfs_t lfs;
+    lfs_format(&lfs, cfg) => 0;
 
     for (int c = 0; c < CYCLES; c++) {
-        lfs_mount(&lfs, &cfg) => 0;
+        lfs_mount(&lfs, cfg) => 0;
         lfs_mkdir(&lfs, "breakfast") => 0;
         lfs_unmount(&lfs) => 0;
 
-        lfs_mount(&lfs, &cfg) => 0;
+        lfs_mount(&lfs, cfg) => 0;
         for (int n = 0; n < FILES; n++) {
+            char path[1024];
             sprintf(path, "breakfast/%s", names[n]);
             lfs_file_open(&lfs, &files[n], path,
                     LFS_O_WRONLY | LFS_O_CREAT | LFS_O_APPEND) => 0;
         }
         for (int n = 0; n < FILES; n++) {
-            size = strlen(names[n]);
+            size_t size = strlen(names[n]);
             for (int i = 0; i < SIZE; i += size) {
                 lfs_file_write(&lfs, &files[n], names[n], size) => size;
             }
@@ -116,12 +141,15 @@ code = '''
         }
         lfs_unmount(&lfs) => 0;
 
-        lfs_mount(&lfs, &cfg) => 0;
+        lfs_mount(&lfs, cfg) => 0;
         for (int n = 0; n < FILES; n++) {
+            char path[1024];
             sprintf(path, "breakfast/%s", names[n]);
+            lfs_file_t file;
             lfs_file_open(&lfs, &file, path, LFS_O_RDONLY) => 0;
-            size = strlen(names[n]);
+            size_t size = strlen(names[n]);
             for (int i = 0; i < SIZE; i += size) {
+                uint8_t buffer[1024];
                 lfs_file_read(&lfs, &file, buffer, size) => size;
                 assert(memcmp(buffer, names[n], size) == 0);
             }
@@ -129,8 +157,9 @@ code = '''
         }
         lfs_unmount(&lfs) => 0;
 
-        lfs_mount(&lfs, &cfg) => 0;
+        lfs_mount(&lfs, cfg) => 0;
         for (int n = 0; n < FILES; n++) {
+            char path[1024];
             sprintf(path, "breakfast/%s", names[n]);
             lfs_remove(&lfs, path) => 0;
         }
@@ -139,26 +168,31 @@ code = '''
     }
 '''
 
-[[case]] # serial allocation reuse test
-define.FILES = 3
-define.SIZE = '(((LFS_BLOCK_SIZE-8)*(LFS_BLOCK_COUNT-6)) / FILES)'
-define.CYCLES = [1, 10]
+# serial allocation reuse test
+[cases.test_alloc_serial_reuse]
+defines.FILES = 3
+defines.SIZE = '(((BLOCK_SIZE-8)*(BLOCK_COUNT-6)) / FILES)'
+defines.CYCLES = [1, 10]
 code = '''
-    const char *names[FILES] = {"bacon", "eggs", "pancakes"};
+    const char *names[] = {"bacon", "eggs", "pancakes"};
 
-    lfs_format(&lfs, &cfg) => 0;
+    lfs_t lfs;
+    lfs_format(&lfs, cfg) => 0;
 
     for (int c = 0; c < CYCLES; c++) {
-        lfs_mount(&lfs, &cfg) => 0;
+        lfs_mount(&lfs, cfg) => 0;
         lfs_mkdir(&lfs, "breakfast") => 0;
         lfs_unmount(&lfs) => 0;
 
         for (int n = 0; n < FILES; n++) {
-            lfs_mount(&lfs, &cfg) => 0;
+            lfs_mount(&lfs, cfg) => 0;
+            char path[1024];
             sprintf(path, "breakfast/%s", names[n]);
+            lfs_file_t file;
             lfs_file_open(&lfs, &file, path,
                     LFS_O_WRONLY | LFS_O_CREAT | LFS_O_APPEND) => 0;
-            size = strlen(names[n]);
+            size_t size = strlen(names[n]);
+            uint8_t buffer[1024];
             memcpy(buffer, names[n], size);
             for (int i = 0; i < SIZE; i += size) {
                 lfs_file_write(&lfs, &file, buffer, size) => size;
@@ -167,12 +201,15 @@ code = '''
             lfs_unmount(&lfs) => 0;
         }
 
-        lfs_mount(&lfs, &cfg) => 0;
+        lfs_mount(&lfs, cfg) => 0;
         for (int n = 0; n < FILES; n++) {
+            char path[1024];
             sprintf(path, "breakfast/%s", names[n]);
+            lfs_file_t file;
             lfs_file_open(&lfs, &file, path, LFS_O_RDONLY) => 0;
-            size = strlen(names[n]);
+            size_t size = strlen(names[n]);
             for (int i = 0; i < SIZE; i += size) {
+                uint8_t buffer[1024];
                 lfs_file_read(&lfs, &file, buffer, size) => size;
                 assert(memcmp(buffer, names[n], size) == 0);
             }
@@ -180,8 +217,9 @@ code = '''
         }
         lfs_unmount(&lfs) => 0;
 
-        lfs_mount(&lfs, &cfg) => 0;
+        lfs_mount(&lfs, cfg) => 0;
         for (int n = 0; n < FILES; n++) {
+            char path[1024];
             sprintf(path, "breakfast/%s", names[n]);
             lfs_remove(&lfs, path) => 0;
         }
@@ -190,12 +228,16 @@ code = '''
     }
 '''
 
-[[case]] # exhaustion test
+# exhaustion test
+[cases.test_alloc_exhaustion]
 code = '''
-    lfs_format(&lfs, &cfg) => 0;
-    lfs_mount(&lfs, &cfg) => 0;
+    lfs_t lfs;
+    lfs_format(&lfs, cfg) => 0;
+    lfs_mount(&lfs, cfg) => 0;
+    lfs_file_t file;
     lfs_file_open(&lfs, &file, "exhaustion", LFS_O_WRONLY | LFS_O_CREAT);
-    size = strlen("exhaustion");
+    size_t size = strlen("exhaustion");
+    uint8_t buffer[1024];
     memcpy(buffer, "exhaustion", size);
     lfs_file_write(&lfs, &file, buffer, size) => size;
     lfs_file_sync(&lfs, &file) => 0;
@@ -213,10 +255,13 @@ code = '''
     }
     res => LFS_ERR_NOSPC;
 
+    // note that lfs_fs_gc should not error here
+    lfs_fs_gc(&lfs) => 0;
+
     lfs_file_close(&lfs, &file) => 0;
     lfs_unmount(&lfs) => 0;
 
-    lfs_mount(&lfs, &cfg) => 0;
+    lfs_mount(&lfs, cfg) => 0;
     lfs_file_open(&lfs, &file, "exhaustion", LFS_O_RDONLY);
     size = strlen("exhaustion");
     lfs_file_size(&lfs, &file) => size;
@@ -226,14 +271,18 @@ code = '''
     lfs_unmount(&lfs) => 0;
 '''
 
-[[case]] # exhaustion wraparound test
-define.SIZE = '(((LFS_BLOCK_SIZE-8)*(LFS_BLOCK_COUNT-4)) / 3)'
+# exhaustion wraparound test
+[cases.test_alloc_exhaustion_wraparound]
+defines.SIZE = '(((BLOCK_SIZE-8)*(BLOCK_COUNT-4)) / 3)'
 code = '''
-    lfs_format(&lfs, &cfg) => 0;
-    lfs_mount(&lfs, &cfg) => 0;
+    lfs_t lfs;
+    lfs_format(&lfs, cfg) => 0;
+    lfs_mount(&lfs, cfg) => 0;
 
+    lfs_file_t file;
     lfs_file_open(&lfs, &file, "padding", LFS_O_WRONLY | LFS_O_CREAT);
-    size = strlen("buffering");
+    size_t size = strlen("buffering");
+    uint8_t buffer[1024];
     memcpy(buffer, "buffering", size);
     for (int i = 0; i < SIZE; i += size) {
         lfs_file_write(&lfs, &file, buffer, size) => size;
@@ -260,10 +309,13 @@ code = '''
     }
     res => LFS_ERR_NOSPC;
 
+    // note that lfs_fs_gc should not error here
+    lfs_fs_gc(&lfs) => 0;
+
     lfs_file_close(&lfs, &file) => 0;
     lfs_unmount(&lfs) => 0;
 
-    lfs_mount(&lfs, &cfg) => 0;
+    lfs_mount(&lfs, cfg) => 0;
     lfs_file_open(&lfs, &file, "exhaustion", LFS_O_RDONLY);
     size = strlen("exhaustion");
     lfs_file_size(&lfs, &file) => size;
@@ -274,17 +326,22 @@ code = '''
     lfs_unmount(&lfs) => 0;
 '''
 
-[[case]] # dir exhaustion test
+# dir exhaustion test
+[cases.test_alloc_dir_exhaustion]
 code = '''
-    lfs_format(&lfs, &cfg) => 0;
-    lfs_mount(&lfs, &cfg) => 0;
+    lfs_t lfs;
+    lfs_format(&lfs, cfg) => 0;
+    lfs_mount(&lfs, cfg) => 0;
 
     // find out max file size
     lfs_mkdir(&lfs, "exhaustiondir") => 0;
-    size = strlen("blahblahblahblah");
+    size_t size = strlen("blahblahblahblah");
+    uint8_t buffer[1024];
     memcpy(buffer, "blahblahblahblah", size);
+    lfs_file_t file;
     lfs_file_open(&lfs, &file, "exhaustion", LFS_O_WRONLY | LFS_O_CREAT);
     int count = 0;
+    int err;
     while (true) {
         err = lfs_file_write(&lfs, &file, buffer, size);
         if (err < 0) {
@@ -294,6 +351,8 @@ code = '''
         count += 1;
     }
     err => LFS_ERR_NOSPC;
+    // note that lfs_fs_gc should not error here
+    lfs_fs_gc(&lfs) => 0;
     lfs_file_close(&lfs, &file) => 0;
 
     lfs_remove(&lfs, "exhaustion") => 0;
@@ -323,17 +382,21 @@ code = '''
     lfs_unmount(&lfs) => 0;
 '''
 
-[[case]] # what if we have a bad block during an allocation scan?
+# what if we have a bad block during an allocation scan?
+[cases.test_alloc_bad_blocks]
 in = "lfs.c"
-define.LFS_ERASE_CYCLES = 0xffffffff
-define.LFS_BADBLOCK_BEHAVIOR = 'LFS_TESTBD_BADBLOCK_READERROR'
+defines.ERASE_CYCLES = 0xffffffff
+defines.BADBLOCK_BEHAVIOR = 'LFS_EMUBD_BADBLOCK_READERROR'
 code = '''
-    lfs_format(&lfs, &cfg) => 0;
-    lfs_mount(&lfs, &cfg) => 0;
+    lfs_t lfs;
+    lfs_format(&lfs, cfg) => 0;
+    lfs_mount(&lfs, cfg) => 0;
     // first fill to exhaustion to find available space
+    lfs_file_t file;
     lfs_file_open(&lfs, &file, "pacman", LFS_O_WRONLY | LFS_O_CREAT) => 0;
+    uint8_t buffer[1024];
     strcpy((char*)buffer, "waka");
-    size = strlen("waka");
+    size_t size = strlen("waka");
     lfs_size_t filesize = 0;
     while (true) {
         lfs_ssize_t res = lfs_file_write(&lfs, &file, buffer, size);
@@ -345,7 +408,7 @@ code = '''
     }
     lfs_file_close(&lfs, &file) => 0;
     // now fill all but a couple of blocks of the filesystem with data
-    filesize -= 3*LFS_BLOCK_SIZE;
+    filesize -= 3*BLOCK_SIZE;
     lfs_file_open(&lfs, &file, "pacman", LFS_O_WRONLY | LFS_O_CREAT) => 0;
     strcpy((char*)buffer, "waka");
     size = strlen("waka");
@@ -358,11 +421,11 @@ code = '''
     lfs_unmount(&lfs) => 0;
 
     // remount to force an alloc scan
-    lfs_mount(&lfs, &cfg) => 0;
+    lfs_mount(&lfs, cfg) => 0;
 
     // but mark the head of our file as a "bad block", this is force our
     // scan to bail early
-    lfs_testbd_setwear(&cfg, fileblock, 0xffffffff) => 0;
+    lfs_emubd_setwear(cfg, fileblock, 0xffffffff) => 0;
     lfs_file_open(&lfs, &file, "ghost", LFS_O_WRONLY | LFS_O_CREAT) => 0;
     strcpy((char*)buffer, "chomp");
     size = strlen("chomp");
@@ -377,7 +440,7 @@ code = '''
 
     // now reverse the "bad block" and try to write the file again until we
     // run out of space
-    lfs_testbd_setwear(&cfg, fileblock, 0) => 0;
+    lfs_emubd_setwear(cfg, fileblock, 0) => 0;
     lfs_file_open(&lfs, &file, "ghost", LFS_O_WRONLY | LFS_O_CREAT) => 0;
     strcpy((char*)buffer, "chomp");
     size = strlen("chomp");
@@ -388,12 +451,14 @@ code = '''
             break;
         }
     }
+    // note that lfs_fs_gc should not error here
+    lfs_fs_gc(&lfs) => 0;
     lfs_file_close(&lfs, &file) => 0;
 
     lfs_unmount(&lfs) => 0;
 
     // check that the disk isn't hurt
-    lfs_mount(&lfs, &cfg) => 0;
+    lfs_mount(&lfs, cfg) => 0;
     lfs_file_open(&lfs, &file, "pacman", LFS_O_RDONLY) => 0;
     strcpy((char*)buffer, "waka");
     size = strlen("waka");
@@ -411,24 +476,29 @@ code = '''
 # on the geometry of the block device. But they are valuable. Eventually they
 # should be removed and replaced with generalized tests.
 
-[[case]] # chained dir exhaustion test
-define.LFS_BLOCK_SIZE = 512
-define.LFS_BLOCK_COUNT = 1024
-if = 'LFS_BLOCK_SIZE == 512 && LFS_BLOCK_COUNT == 1024'
+# chained dir exhaustion test
+[cases.test_alloc_chained_dir_exhaustion]
+if = 'ERASE_SIZE == 512'
+defines.ERASE_COUNT = 1024
 code = '''
-    lfs_format(&lfs, &cfg) => 0;
-    lfs_mount(&lfs, &cfg) => 0;
+    lfs_t lfs;
+    lfs_format(&lfs, cfg) => 0;
+    lfs_mount(&lfs, cfg) => 0;
 
     // find out max file size
     lfs_mkdir(&lfs, "exhaustiondir") => 0;
     for (int i = 0; i < 10; i++) {
+        char path[1024];
         sprintf(path, "dirwithanexhaustivelylongnameforpadding%d", i);
         lfs_mkdir(&lfs, path) => 0;
     }
-    size = strlen("blahblahblahblah");
+    size_t size = strlen("blahblahblahblah");
+    uint8_t buffer[1024];
     memcpy(buffer, "blahblahblahblah", size);
+    lfs_file_t file;
     lfs_file_open(&lfs, &file, "exhaustion", LFS_O_WRONLY | LFS_O_CREAT);
     int count = 0;
+    int err;
     while (true) {
         err = lfs_file_write(&lfs, &file, buffer, size);
         if (err < 0) {
@@ -443,6 +513,7 @@ code = '''
     lfs_remove(&lfs, "exhaustion") => 0;
     lfs_remove(&lfs, "exhaustiondir") => 0;
     for (int i = 0; i < 10; i++) {
+        char path[1024];
         sprintf(path, "dirwithanexhaustivelylongnameforpadding%d", i);
         lfs_remove(&lfs, path) => 0;
     }
@@ -455,6 +526,7 @@ code = '''
     lfs_file_sync(&lfs, &file) => 0;
 
     for (int i = 0; i < 10; i++) {
+        char path[1024];
         sprintf(path, "dirwithanexhaustivelylongnameforpadding%d", i);
         lfs_mkdir(&lfs, path) => 0;
     }
@@ -482,27 +554,31 @@ code = '''
     lfs_unmount(&lfs) => 0;
 '''
 
-[[case]] # split dir test
-define.LFS_BLOCK_SIZE = 512
-define.LFS_BLOCK_COUNT = 1024
-if = 'LFS_BLOCK_SIZE == 512 && LFS_BLOCK_COUNT == 1024'
+# split dir test
+[cases.test_alloc_split_dir]
+if = 'ERASE_SIZE == 512'
+defines.ERASE_COUNT = 1024
 code = '''
-    lfs_format(&lfs, &cfg) => 0;
-    lfs_mount(&lfs, &cfg) => 0;
+    lfs_t lfs;
+    lfs_format(&lfs, cfg) => 0;
+    lfs_mount(&lfs, cfg) => 0;
 
     // create one block hole for half a directory
+    lfs_file_t file;
     lfs_file_open(&lfs, &file, "bump", LFS_O_WRONLY | LFS_O_CREAT) => 0;
-    for (lfs_size_t i = 0; i < cfg.block_size; i += 2) {
+    for (lfs_size_t i = 0; i < cfg->block_size; i += 2) {
+        uint8_t buffer[1024];
         memcpy(&buffer[i], "hi", 2);
     }
-    lfs_file_write(&lfs, &file, buffer, cfg.block_size) => cfg.block_size;
+    uint8_t buffer[1024];
+    lfs_file_write(&lfs, &file, buffer, cfg->block_size) => cfg->block_size;
     lfs_file_close(&lfs, &file) => 0;
 
     lfs_file_open(&lfs, &file, "exhaustion", LFS_O_WRONLY | LFS_O_CREAT);
-    size = strlen("blahblahblahblah");
+    size_t size = strlen("blahblahblahblah");
     memcpy(buffer, "blahblahblahblah", size);
     for (lfs_size_t i = 0;
-            i < (cfg.block_count-4)*(cfg.block_size-8);
+            i < (cfg->block_count-4)*(cfg->block_size-8);
             i += size) {
         lfs_file_write(&lfs, &file, buffer, size) => size;
     }
@@ -510,7 +586,7 @@ code = '''
 
     // remount to force reset of lookahead
     lfs_unmount(&lfs) => 0;
-    lfs_mount(&lfs, &cfg) => 0;
+    lfs_mount(&lfs, cfg) => 0;
 
     // open hole
     lfs_remove(&lfs, "bump") => 0;
@@ -518,30 +594,33 @@ code = '''
     lfs_mkdir(&lfs, "splitdir") => 0;
     lfs_file_open(&lfs, &file, "splitdir/bump",
             LFS_O_WRONLY | LFS_O_CREAT) => 0;
-    for (lfs_size_t i = 0; i < cfg.block_size; i += 2) {
+    for (lfs_size_t i = 0; i < cfg->block_size; i += 2) {
         memcpy(&buffer[i], "hi", 2);
     }
-    lfs_file_write(&lfs, &file, buffer, 2*cfg.block_size) => LFS_ERR_NOSPC;
+    lfs_file_write(&lfs, &file, buffer, 2*cfg->block_size) => LFS_ERR_NOSPC;
     lfs_file_close(&lfs, &file) => 0;
 
     lfs_unmount(&lfs) => 0;
 '''
 
-[[case]] # outdated lookahead test
-define.LFS_BLOCK_SIZE = 512
-define.LFS_BLOCK_COUNT = 1024
-if = 'LFS_BLOCK_SIZE == 512 && LFS_BLOCK_COUNT == 1024'
+# outdated lookahead test
+[cases.test_alloc_outdated_lookahead]
+if = 'ERASE_SIZE == 512'
+defines.ERASE_COUNT = 1024
 code = '''
-    lfs_format(&lfs, &cfg) => 0;
-    lfs_mount(&lfs, &cfg) => 0;
+    lfs_t lfs;
+    lfs_format(&lfs, cfg) => 0;
+    lfs_mount(&lfs, cfg) => 0;
 
     // fill completely with two files
+    lfs_file_t file;
     lfs_file_open(&lfs, &file, "exhaustion1",
             LFS_O_WRONLY | LFS_O_CREAT) => 0;
-    size = strlen("blahblahblahblah");
+    size_t size = strlen("blahblahblahblah");
+    uint8_t buffer[1024];
     memcpy(buffer, "blahblahblahblah", size);
     for (lfs_size_t i = 0;
-            i < ((cfg.block_count-2)/2)*(cfg.block_size-8);
+            i < ((cfg->block_count-2)/2)*(cfg->block_size-8);
             i += size) {
         lfs_file_write(&lfs, &file, buffer, size) => size;
     }
@@ -552,7 +631,7 @@ code = '''
     size = strlen("blahblahblahblah");
     memcpy(buffer, "blahblahblahblah", size);
     for (lfs_size_t i = 0;
-            i < ((cfg.block_count-2+1)/2)*(cfg.block_size-8);
+            i < ((cfg->block_count-2+1)/2)*(cfg->block_size-8);
             i += size) {
         lfs_file_write(&lfs, &file, buffer, size) => size;
     }
@@ -560,7 +639,7 @@ code = '''
 
     // remount to force reset of lookahead
     lfs_unmount(&lfs) => 0;
-    lfs_mount(&lfs, &cfg) => 0;
+    lfs_mount(&lfs, cfg) => 0;
 
     // rewrite one file
     lfs_file_open(&lfs, &file, "exhaustion1",
@@ -569,7 +648,7 @@ code = '''
     size = strlen("blahblahblahblah");
     memcpy(buffer, "blahblahblahblah", size);
     for (lfs_size_t i = 0;
-            i < ((cfg.block_count-2)/2)*(cfg.block_size-8);
+            i < ((cfg->block_count-2)/2)*(cfg->block_size-8);
             i += size) {
         lfs_file_write(&lfs, &file, buffer, size) => size;
     }
@@ -583,7 +662,7 @@ code = '''
     size = strlen("blahblahblahblah");
     memcpy(buffer, "blahblahblahblah", size);
     for (lfs_size_t i = 0;
-            i < ((cfg.block_count-2+1)/2)*(cfg.block_size-8);
+            i < ((cfg->block_count-2+1)/2)*(cfg->block_size-8);
             i += size) {
         lfs_file_write(&lfs, &file, buffer, size) => size;
     }
@@ -592,21 +671,24 @@ code = '''
     lfs_unmount(&lfs) => 0;
 '''
 
-[[case]] # outdated lookahead and split dir test
-define.LFS_BLOCK_SIZE = 512
-define.LFS_BLOCK_COUNT = 1024
-if = 'LFS_BLOCK_SIZE == 512 && LFS_BLOCK_COUNT == 1024'
+# outdated lookahead and split dir test
+[cases.test_alloc_outdated_lookahead_split_dir]
+if = 'ERASE_SIZE == 512'
+defines.ERASE_COUNT = 1024
 code = '''
-    lfs_format(&lfs, &cfg) => 0;
-    lfs_mount(&lfs, &cfg) => 0;
+    lfs_t lfs;
+    lfs_format(&lfs, cfg) => 0;
+    lfs_mount(&lfs, cfg) => 0;
 
     // fill completely with two files
+    lfs_file_t file;
     lfs_file_open(&lfs, &file, "exhaustion1",
             LFS_O_WRONLY | LFS_O_CREAT) => 0;
-    size = strlen("blahblahblahblah");
+    size_t size = strlen("blahblahblahblah");
+    uint8_t buffer[1024];
     memcpy(buffer, "blahblahblahblah", size);
     for (lfs_size_t i = 0;
-            i < ((cfg.block_count-2)/2)*(cfg.block_size-8);
+            i < ((cfg->block_count-2)/2)*(cfg->block_size-8);
             i += size) {
         lfs_file_write(&lfs, &file, buffer, size) => size;
     }
@@ -617,7 +699,7 @@ code = '''
     size = strlen("blahblahblahblah");
     memcpy(buffer, "blahblahblahblah", size);
     for (lfs_size_t i = 0;
-            i < ((cfg.block_count-2+1)/2)*(cfg.block_size-8);
+            i < ((cfg->block_count-2+1)/2)*(cfg->block_size-8);
             i += size) {
         lfs_file_write(&lfs, &file, buffer, size) => size;
     }
@@ -625,7 +707,7 @@ code = '''
 
     // remount to force reset of lookahead
     lfs_unmount(&lfs) => 0;
-    lfs_mount(&lfs, &cfg) => 0;
+    lfs_mount(&lfs, cfg) => 0;
 
     // rewrite one file with a hole of one block
     lfs_file_open(&lfs, &file, "exhaustion1",
@@ -634,7 +716,7 @@ code = '''
     size = strlen("blahblahblahblah");
     memcpy(buffer, "blahblahblahblah", size);
     for (lfs_size_t i = 0;
-            i < ((cfg.block_count-2)/2 - 1)*(cfg.block_size-8);
+            i < ((cfg->block_count-2)/2 - 1)*(cfg->block_size-8);
             i += size) {
         lfs_file_write(&lfs, &file, buffer, size) => size;
     }
diff --git a/tests/test_attrs.toml b/tests/test_attrs.toml
index db8d0c7..3c69001 100644
--- a/tests/test_attrs.toml
+++ b/tests/test_attrs.toml
@@ -1,14 +1,17 @@
-[[case]] # set/get attribute
+[cases.test_attrs_get_set]
 code = '''
-    lfs_format(&lfs, &cfg) => 0;
-    lfs_mount(&lfs, &cfg) => 0;
+    lfs_t lfs;
+    lfs_format(&lfs, cfg) => 0;
+    lfs_mount(&lfs, cfg) => 0;
     lfs_mkdir(&lfs, "hello") => 0;
+    lfs_file_t file;
     lfs_file_open(&lfs, &file, "hello/hello", LFS_O_WRONLY | LFS_O_CREAT) => 0;
     lfs_file_write(&lfs, &file, "hello", strlen("hello")) => strlen("hello");
     lfs_file_close(&lfs, &file);
     lfs_unmount(&lfs) => 0;
 
-    lfs_mount(&lfs, &cfg) => 0;
+    lfs_mount(&lfs, cfg) => 0;
+    uint8_t buffer[1024];
     memset(buffer, 0, sizeof(buffer));
     lfs_setattr(&lfs, "hello", 'A', "aaaa",   4) => 0;
     lfs_setattr(&lfs, "hello", 'B', "bbbbbb", 6) => 0;
@@ -60,7 +63,7 @@ code = '''
 
     lfs_unmount(&lfs) => 0;
 
-    lfs_mount(&lfs, &cfg) => 0;
+    lfs_mount(&lfs, cfg) => 0;
     memset(buffer, 0, sizeof(buffer));
     lfs_getattr(&lfs, "hello", 'A', buffer,    4) => 4;
     lfs_getattr(&lfs, "hello", 'B', buffer+4,  9) => 9;
@@ -76,17 +79,20 @@ code = '''
     lfs_unmount(&lfs) => 0;
 '''
 
-[[case]] # set/get root attribute
+[cases.test_attrs_get_set_root]
 code = '''
-    lfs_format(&lfs, &cfg) => 0;
-    lfs_mount(&lfs, &cfg) => 0;
+    lfs_t lfs;
+    lfs_format(&lfs, cfg) => 0;
+    lfs_mount(&lfs, cfg) => 0;
     lfs_mkdir(&lfs, "hello") => 0;
+    lfs_file_t file;
     lfs_file_open(&lfs, &file, "hello/hello", LFS_O_WRONLY | LFS_O_CREAT) => 0;
     lfs_file_write(&lfs, &file, "hello", strlen("hello")) => strlen("hello");
     lfs_file_close(&lfs, &file);
     lfs_unmount(&lfs) => 0;
 
-    lfs_mount(&lfs, &cfg) => 0;
+    lfs_mount(&lfs, cfg) => 0;
+    uint8_t buffer[1024];
     memset(buffer, 0, sizeof(buffer));
     lfs_setattr(&lfs, "/", 'A', "aaaa",   4) => 0;
     lfs_setattr(&lfs, "/", 'B', "bbbbbb", 6) => 0;
@@ -137,7 +143,7 @@ code = '''
     lfs_getattr(&lfs, "/", 'C', buffer+10, 5) => 5;
     lfs_unmount(&lfs) => 0;
 
-    lfs_mount(&lfs, &cfg) => 0;
+    lfs_mount(&lfs, cfg) => 0;
     memset(buffer, 0, sizeof(buffer));
     lfs_getattr(&lfs, "/", 'A', buffer,    4) => 4;
     lfs_getattr(&lfs, "/", 'B', buffer+4,  9) => 9;
@@ -153,17 +159,20 @@ code = '''
     lfs_unmount(&lfs) => 0;
 '''
 
-[[case]] # set/get file attribute
+[cases.test_attrs_get_set_file]
 code = '''
-    lfs_format(&lfs, &cfg) => 0;
-    lfs_mount(&lfs, &cfg) => 0;
+    lfs_t lfs;
+    lfs_format(&lfs, cfg) => 0;
+    lfs_mount(&lfs, cfg) => 0;
     lfs_mkdir(&lfs, "hello") => 0;
+    lfs_file_t file;
     lfs_file_open(&lfs, &file, "hello/hello", LFS_O_WRONLY | LFS_O_CREAT) => 0;
     lfs_file_write(&lfs, &file, "hello", strlen("hello")) => strlen("hello");
     lfs_file_close(&lfs, &file);
     lfs_unmount(&lfs) => 0;
 
-    lfs_mount(&lfs, &cfg) => 0;
+    lfs_mount(&lfs, cfg) => 0;
+    uint8_t buffer[1024];
     memset(buffer, 0, sizeof(buffer));
     struct lfs_attr attrs1[] = {
         {'A', buffer,    4},
@@ -238,7 +247,7 @@ code = '''
 
     lfs_unmount(&lfs) => 0;
 
-    lfs_mount(&lfs, &cfg) => 0;
+    lfs_mount(&lfs, cfg) => 0;
     memset(buffer, 0, sizeof(buffer));
     struct lfs_attr attrs3[] = {
         {'A', buffer,    4},
@@ -260,20 +269,23 @@ code = '''
     lfs_unmount(&lfs) => 0;
 '''
 
-[[case]] # deferred file attributes
+[cases.test_attrs_deferred_file]
 code = '''
-    lfs_format(&lfs, &cfg) => 0;
-    lfs_mount(&lfs, &cfg) => 0;
+    lfs_t lfs;
+    lfs_format(&lfs, cfg) => 0;
+    lfs_mount(&lfs, cfg) => 0;
     lfs_mkdir(&lfs, "hello") => 0;
+    lfs_file_t file;
     lfs_file_open(&lfs, &file, "hello/hello", LFS_O_WRONLY | LFS_O_CREAT) => 0;
     lfs_file_write(&lfs, &file, "hello", strlen("hello")) => strlen("hello");
     lfs_file_close(&lfs, &file);
     lfs_unmount(&lfs) => 0;
 
-    lfs_mount(&lfs, &cfg) => 0;
+    lfs_mount(&lfs, cfg) => 0;
     lfs_setattr(&lfs, "hello/hello", 'B', "fffffffff",  9) => 0;
     lfs_setattr(&lfs, "hello/hello", 'C', "ccccc",      5) => 0;
 
+    uint8_t buffer[1024];
     memset(buffer, 0, sizeof(buffer));
     struct lfs_attr attrs1[] = {
         {'B', "gggg", 4},
diff --git a/tests/test_badblocks.toml b/tests/test_badblocks.toml
index 06967a6..04ed73f 100644
--- a/tests/test_badblocks.toml
+++ b/tests/test_badblocks.toml
@@ -1,28 +1,30 @@
 # bad blocks with block cycles should be tested in test_relocations
-if = 'LFS_BLOCK_CYCLES == -1'
+if = '(int32_t)BLOCK_CYCLES == -1'
 
-[[case]] # single bad blocks
-define.LFS_BLOCK_COUNT = 256 # small bd so test runs faster
-define.LFS_ERASE_CYCLES = 0xffffffff
-define.LFS_ERASE_VALUE = [0x00, 0xff, -1]
-define.LFS_BADBLOCK_BEHAVIOR = [
-    'LFS_TESTBD_BADBLOCK_PROGERROR',
-    'LFS_TESTBD_BADBLOCK_ERASEERROR',
-    'LFS_TESTBD_BADBLOCK_READERROR',
-    'LFS_TESTBD_BADBLOCK_PROGNOOP',
-    'LFS_TESTBD_BADBLOCK_ERASENOOP',
+[cases.test_badblocks_single]
+defines.ERASE_COUNT = 256 # small bd so test runs faster
+defines.ERASE_CYCLES = 0xffffffff
+defines.ERASE_VALUE = [0x00, 0xff, -1]
+defines.BADBLOCK_BEHAVIOR = [
+    'LFS_EMUBD_BADBLOCK_PROGERROR',
+    'LFS_EMUBD_BADBLOCK_ERASEERROR',
+    'LFS_EMUBD_BADBLOCK_READERROR',
+    'LFS_EMUBD_BADBLOCK_PROGNOOP',
+    'LFS_EMUBD_BADBLOCK_ERASENOOP',
 ]
-define.NAMEMULT = 64
-define.FILEMULT = 1
+defines.NAMEMULT = 64
+defines.FILEMULT = 1
 code = '''
-    for (lfs_block_t badblock = 2; badblock < LFS_BLOCK_COUNT; badblock++) {
-        lfs_testbd_setwear(&cfg, badblock-1, 0) => 0;
-        lfs_testbd_setwear(&cfg, badblock, 0xffffffff) => 0;
-        
-        lfs_format(&lfs, &cfg) => 0;
+    for (lfs_block_t badblock = 2; badblock < BLOCK_COUNT; badblock++) {
+        lfs_emubd_setwear(cfg, badblock-1, 0) => 0;
+        lfs_emubd_setwear(cfg, badblock, 0xffffffff) => 0;
 
-        lfs_mount(&lfs, &cfg) => 0;
+        lfs_t lfs;
+        lfs_format(&lfs, cfg) => 0;
+
+        lfs_mount(&lfs, cfg) => 0;
         for (int i = 1; i < 10; i++) {
+            uint8_t buffer[1024];
             for (int j = 0; j < NAMEMULT; j++) {
                 buffer[j] = '0'+i;
             }
@@ -34,10 +36,11 @@ code = '''
                 buffer[j+NAMEMULT+1] = '0'+i;
             }
             buffer[2*NAMEMULT+1] = '\0';
+            lfs_file_t file;
             lfs_file_open(&lfs, &file, (char*)buffer,
                     LFS_O_WRONLY | LFS_O_CREAT) => 0;
             
-            size = NAMEMULT;
+            lfs_size_t size = NAMEMULT;
             for (int j = 0; j < i*FILEMULT; j++) {
                 lfs_file_write(&lfs, &file, buffer, size) => size;
             }
@@ -46,12 +49,14 @@ code = '''
         }
         lfs_unmount(&lfs) => 0;
 
-        lfs_mount(&lfs, &cfg) => 0;
+        lfs_mount(&lfs, cfg) => 0;
         for (int i = 1; i < 10; i++) {
+            uint8_t buffer[1024];
             for (int j = 0; j < NAMEMULT; j++) {
                 buffer[j] = '0'+i;
             }
             buffer[NAMEMULT] = '\0';
+            struct lfs_info info;
             lfs_stat(&lfs, (char*)buffer, &info) => 0;
             info.type => LFS_TYPE_DIR;
 
@@ -60,9 +65,10 @@ code = '''
                 buffer[j+NAMEMULT+1] = '0'+i;
             }
             buffer[2*NAMEMULT+1] = '\0';
+            lfs_file_t file;
             lfs_file_open(&lfs, &file, (char*)buffer, LFS_O_RDONLY) => 0;
             
-            size = NAMEMULT;
+            int size = NAMEMULT;
             for (int j = 0; j < i*FILEMULT; j++) {
                 uint8_t rbuffer[1024];
                 lfs_file_read(&lfs, &file, rbuffer, size) => size;
@@ -75,28 +81,30 @@ code = '''
     }
 '''
 
-[[case]] # region corruption (causes cascading failures)
-define.LFS_BLOCK_COUNT = 256 # small bd so test runs faster
-define.LFS_ERASE_CYCLES = 0xffffffff
-define.LFS_ERASE_VALUE = [0x00, 0xff, -1]
-define.LFS_BADBLOCK_BEHAVIOR = [
-    'LFS_TESTBD_BADBLOCK_PROGERROR',
-    'LFS_TESTBD_BADBLOCK_ERASEERROR',
-    'LFS_TESTBD_BADBLOCK_READERROR',
-    'LFS_TESTBD_BADBLOCK_PROGNOOP',
-    'LFS_TESTBD_BADBLOCK_ERASENOOP',
+[cases.test_badblocks_region_corruption] # (causes cascading failures)
+defines.ERASE_COUNT = 256 # small bd so test runs faster
+defines.ERASE_CYCLES = 0xffffffff
+defines.ERASE_VALUE = [0x00, 0xff, -1]
+defines.BADBLOCK_BEHAVIOR = [
+    'LFS_EMUBD_BADBLOCK_PROGERROR',
+    'LFS_EMUBD_BADBLOCK_ERASEERROR',
+    'LFS_EMUBD_BADBLOCK_READERROR',
+    'LFS_EMUBD_BADBLOCK_PROGNOOP',
+    'LFS_EMUBD_BADBLOCK_ERASENOOP',
 ]
-define.NAMEMULT = 64
-define.FILEMULT = 1
+defines.NAMEMULT = 64
+defines.FILEMULT = 1
 code = '''
-    for (lfs_block_t i = 0; i < (LFS_BLOCK_COUNT-2)/2; i++) {
-        lfs_testbd_setwear(&cfg, i+2, 0xffffffff) => 0;
+    for (lfs_block_t i = 0; i < (BLOCK_COUNT-2)/2; i++) {
+        lfs_emubd_setwear(cfg, i+2, 0xffffffff) => 0;
     }
-    
-    lfs_format(&lfs, &cfg) => 0;
 
-    lfs_mount(&lfs, &cfg) => 0;
+    lfs_t lfs;
+    lfs_format(&lfs, cfg) => 0;
+
+    lfs_mount(&lfs, cfg) => 0;
     for (int i = 1; i < 10; i++) {
+        uint8_t buffer[1024];
         for (int j = 0; j < NAMEMULT; j++) {
             buffer[j] = '0'+i;
         }
@@ -108,10 +116,11 @@ code = '''
             buffer[j+NAMEMULT+1] = '0'+i;
         }
         buffer[2*NAMEMULT+1] = '\0';
+        lfs_file_t file;
         lfs_file_open(&lfs, &file, (char*)buffer,
                 LFS_O_WRONLY | LFS_O_CREAT) => 0;
         
-        size = NAMEMULT;
+        lfs_size_t size = NAMEMULT;
         for (int j = 0; j < i*FILEMULT; j++) {
             lfs_file_write(&lfs, &file, buffer, size) => size;
         }
@@ -120,12 +129,14 @@ code = '''
     }
     lfs_unmount(&lfs) => 0;
 
-    lfs_mount(&lfs, &cfg) => 0;
+    lfs_mount(&lfs, cfg) => 0;
     for (int i = 1; i < 10; i++) {
+        uint8_t buffer[1024];
         for (int j = 0; j < NAMEMULT; j++) {
             buffer[j] = '0'+i;
         }
         buffer[NAMEMULT] = '\0';
+        struct lfs_info info;
         lfs_stat(&lfs, (char*)buffer, &info) => 0;
         info.type => LFS_TYPE_DIR;
 
@@ -134,9 +145,10 @@ code = '''
             buffer[j+NAMEMULT+1] = '0'+i;
         }
         buffer[2*NAMEMULT+1] = '\0';
+        lfs_file_t file;
         lfs_file_open(&lfs, &file, (char*)buffer, LFS_O_RDONLY) => 0;
         
-        size = NAMEMULT;
+        lfs_size_t size = NAMEMULT;
         for (int j = 0; j < i*FILEMULT; j++) {
             uint8_t rbuffer[1024];
             lfs_file_read(&lfs, &file, rbuffer, size) => size;
@@ -148,28 +160,30 @@ code = '''
     lfs_unmount(&lfs) => 0;
 '''
 
-[[case]] # alternating corruption (causes cascading failures)
-define.LFS_BLOCK_COUNT = 256 # small bd so test runs faster
-define.LFS_ERASE_CYCLES = 0xffffffff
-define.LFS_ERASE_VALUE = [0x00, 0xff, -1]
-define.LFS_BADBLOCK_BEHAVIOR = [
-    'LFS_TESTBD_BADBLOCK_PROGERROR',
-    'LFS_TESTBD_BADBLOCK_ERASEERROR',
-    'LFS_TESTBD_BADBLOCK_READERROR',
-    'LFS_TESTBD_BADBLOCK_PROGNOOP',
-    'LFS_TESTBD_BADBLOCK_ERASENOOP',
+[cases.test_badblocks_alternating_corruption] # (causes cascading failures)
+defines.ERASE_COUNT = 256 # small bd so test runs faster
+defines.ERASE_CYCLES = 0xffffffff
+defines.ERASE_VALUE = [0x00, 0xff, -1]
+defines.BADBLOCK_BEHAVIOR = [
+    'LFS_EMUBD_BADBLOCK_PROGERROR',
+    'LFS_EMUBD_BADBLOCK_ERASEERROR',
+    'LFS_EMUBD_BADBLOCK_READERROR',
+    'LFS_EMUBD_BADBLOCK_PROGNOOP',
+    'LFS_EMUBD_BADBLOCK_ERASENOOP',
 ]
-define.NAMEMULT = 64
-define.FILEMULT = 1
+defines.NAMEMULT = 64
+defines.FILEMULT = 1
 code = '''
-    for (lfs_block_t i = 0; i < (LFS_BLOCK_COUNT-2)/2; i++) {
-        lfs_testbd_setwear(&cfg, (2*i) + 2, 0xffffffff) => 0;
+    for (lfs_block_t i = 0; i < (BLOCK_COUNT-2)/2; i++) {
+        lfs_emubd_setwear(cfg, (2*i) + 2, 0xffffffff) => 0;
     }
-    
-    lfs_format(&lfs, &cfg) => 0;
 
-    lfs_mount(&lfs, &cfg) => 0;
+    lfs_t lfs;
+    lfs_format(&lfs, cfg) => 0;
+
+    lfs_mount(&lfs, cfg) => 0;
     for (int i = 1; i < 10; i++) {
+        uint8_t buffer[1024];
         for (int j = 0; j < NAMEMULT; j++) {
             buffer[j] = '0'+i;
         }
@@ -181,10 +195,11 @@ code = '''
             buffer[j+NAMEMULT+1] = '0'+i;
         }
         buffer[2*NAMEMULT+1] = '\0';
+        lfs_file_t file;
         lfs_file_open(&lfs, &file, (char*)buffer,
                 LFS_O_WRONLY | LFS_O_CREAT) => 0;
         
-        size = NAMEMULT;
+        lfs_size_t size = NAMEMULT;
         for (int j = 0; j < i*FILEMULT; j++) {
             lfs_file_write(&lfs, &file, buffer, size) => size;
         }
@@ -193,12 +208,14 @@ code = '''
     }
     lfs_unmount(&lfs) => 0;
 
-    lfs_mount(&lfs, &cfg) => 0;
+    lfs_mount(&lfs, cfg) => 0;
     for (int i = 1; i < 10; i++) {
+        uint8_t buffer[1024];
         for (int j = 0; j < NAMEMULT; j++) {
             buffer[j] = '0'+i;
         }
         buffer[NAMEMULT] = '\0';
+        struct lfs_info info;
         lfs_stat(&lfs, (char*)buffer, &info) => 0;
         info.type => LFS_TYPE_DIR;
 
@@ -207,9 +224,10 @@ code = '''
             buffer[j+NAMEMULT+1] = '0'+i;
         }
         buffer[2*NAMEMULT+1] = '\0';
+        lfs_file_t file;
         lfs_file_open(&lfs, &file, (char*)buffer, LFS_O_RDONLY) => 0;
         
-        size = NAMEMULT;
+        lfs_size_t size = NAMEMULT;
         for (int j = 0; j < i*FILEMULT; j++) {
             uint8_t rbuffer[1024];
             lfs_file_read(&lfs, &file, rbuffer, size) => size;
@@ -222,20 +240,21 @@ code = '''
 '''
 
 # other corner cases
-[[case]] # bad superblocks (corrupt 1 or 0)
-define.LFS_ERASE_CYCLES = 0xffffffff
-define.LFS_ERASE_VALUE = [0x00, 0xff, -1]
-define.LFS_BADBLOCK_BEHAVIOR = [
-    'LFS_TESTBD_BADBLOCK_PROGERROR',
-    'LFS_TESTBD_BADBLOCK_ERASEERROR',
-    'LFS_TESTBD_BADBLOCK_READERROR',
-    'LFS_TESTBD_BADBLOCK_PROGNOOP',
-    'LFS_TESTBD_BADBLOCK_ERASENOOP',
+[cases.test_badblocks_superblocks] # (corrupt 1 or 0)
+defines.ERASE_CYCLES = 0xffffffff
+defines.ERASE_VALUE = [0x00, 0xff, -1]
+defines.BADBLOCK_BEHAVIOR = [
+    'LFS_EMUBD_BADBLOCK_PROGERROR',
+    'LFS_EMUBD_BADBLOCK_ERASEERROR',
+    'LFS_EMUBD_BADBLOCK_READERROR',
+    'LFS_EMUBD_BADBLOCK_PROGNOOP',
+    'LFS_EMUBD_BADBLOCK_ERASENOOP',
 ]
 code = '''
-    lfs_testbd_setwear(&cfg, 0, 0xffffffff) => 0;
-    lfs_testbd_setwear(&cfg, 1, 0xffffffff) => 0;
+    lfs_emubd_setwear(cfg, 0, 0xffffffff) => 0;
+    lfs_emubd_setwear(cfg, 1, 0xffffffff) => 0;
 
-    lfs_format(&lfs, &cfg) => LFS_ERR_NOSPC;
-    lfs_mount(&lfs, &cfg) => LFS_ERR_CORRUPT;
+    lfs_t lfs;
+    lfs_format(&lfs, cfg) => LFS_ERR_NOSPC;
+    lfs_mount(&lfs, cfg) => LFS_ERR_CORRUPT;
 '''
diff --git a/tests/test_bd.toml b/tests/test_bd.toml
new file mode 100644
index 0000000..8c6510d
--- /dev/null
+++ b/tests/test_bd.toml
@@ -0,0 +1,248 @@
+# These tests don't really test littlefs at all, they are here only to make
+# sure the underlying block device is working.
+#
+# Note we use 251, a prime, in places to avoid aliasing powers of 2.
+# 
+
+[cases.test_bd_one_block]
+defines.READ = ['READ_SIZE', 'BLOCK_SIZE']
+defines.PROG = ['PROG_SIZE', 'BLOCK_SIZE']
+code = '''
+    uint8_t buffer[lfs_max(READ, PROG)];
+
+    // write data
+    cfg->erase(cfg, 0) => 0;
+    for (lfs_off_t i = 0; i < cfg->block_size; i += PROG) {
+        for (lfs_off_t j = 0; j < PROG; j++) {
+            buffer[j] = (i+j) % 251;
+        }
+        cfg->prog(cfg, 0, i, buffer, PROG) => 0;
+    }
+
+    // read data
+    for (lfs_off_t i = 0; i < cfg->block_size; i += READ) {
+        cfg->read(cfg, 0, i, buffer, READ) => 0;
+
+        for (lfs_off_t j = 0; j < READ; j++) {
+            LFS_ASSERT(buffer[j] == (i+j) % 251);
+        }
+    }
+'''
+
+[cases.test_bd_two_block]
+defines.READ = ['READ_SIZE', 'BLOCK_SIZE']
+defines.PROG = ['PROG_SIZE', 'BLOCK_SIZE']
+code = '''
+    uint8_t buffer[lfs_max(READ, PROG)];
+    lfs_block_t block;
+
+    // write block 0
+    block = 0;
+    cfg->erase(cfg, block) => 0;
+    for (lfs_off_t i = 0; i < cfg->block_size; i += PROG) {
+        for (lfs_off_t j = 0; j < PROG; j++) {
+            buffer[j] = (block+i+j) % 251;
+        }
+        cfg->prog(cfg, block, i, buffer, PROG) => 0;
+    }
+
+    // read block 0
+    block = 0;
+    for (lfs_off_t i = 0; i < cfg->block_size; i += READ) {
+        cfg->read(cfg, block, i, buffer, READ) => 0;
+
+        for (lfs_off_t j = 0; j < READ; j++) {
+            LFS_ASSERT(buffer[j] == (block+i+j) % 251);
+        }
+    }
+
+    // write block 1
+    block = 1;
+    cfg->erase(cfg, block) => 0;
+    for (lfs_off_t i = 0; i < cfg->block_size; i += PROG) {
+        for (lfs_off_t j = 0; j < PROG; j++) {
+            buffer[j] = (block+i+j) % 251;
+        }
+        cfg->prog(cfg, block, i, buffer, PROG) => 0;
+    }
+
+    // read block 1
+    block = 1;
+    for (lfs_off_t i = 0; i < cfg->block_size; i += READ) {
+        cfg->read(cfg, block, i, buffer, READ) => 0;
+
+        for (lfs_off_t j = 0; j < READ; j++) {
+            LFS_ASSERT(buffer[j] == (block+i+j) % 251);
+        }
+    }
+
+    // read block 0 again
+    block = 0;
+    for (lfs_off_t i = 0; i < cfg->block_size; i += READ) {
+        cfg->read(cfg, block, i, buffer, READ) => 0;
+
+        for (lfs_off_t j = 0; j < READ; j++) {
+            LFS_ASSERT(buffer[j] == (block+i+j) % 251);
+        }
+    }
+'''
+
+[cases.test_bd_last_block]
+defines.READ = ['READ_SIZE', 'BLOCK_SIZE']
+defines.PROG = ['PROG_SIZE', 'BLOCK_SIZE']
+code = '''
+    uint8_t buffer[lfs_max(READ, PROG)];
+    lfs_block_t block;
+
+    // write block 0
+    block = 0;
+    cfg->erase(cfg, block) => 0;
+    for (lfs_off_t i = 0; i < cfg->block_size; i += PROG) {
+        for (lfs_off_t j = 0; j < PROG; j++) {
+            buffer[j] = (block+i+j) % 251;
+        }
+        cfg->prog(cfg, block, i, buffer, PROG) => 0;
+    }
+
+    // read block 0
+    block = 0;
+    for (lfs_off_t i = 0; i < cfg->block_size; i += READ) {
+        cfg->read(cfg, block, i, buffer, READ) => 0;
+
+        for (lfs_off_t j = 0; j < READ; j++) {
+            LFS_ASSERT(buffer[j] == (block+i+j) % 251);
+        }
+    }
+
+    // write block n-1
+    block = cfg->block_count-1;
+    cfg->erase(cfg, block) => 0;
+    for (lfs_off_t i = 0; i < cfg->block_size; i += PROG) {
+        for (lfs_off_t j = 0; j < PROG; j++) {
+            buffer[j] = (block+i+j) % 251;
+        }
+        cfg->prog(cfg, block, i, buffer, PROG) => 0;
+    }
+
+    // read block n-1
+    block = cfg->block_count-1;
+    for (lfs_off_t i = 0; i < cfg->block_size; i += READ) {
+        cfg->read(cfg, block, i, buffer, READ) => 0;
+
+        for (lfs_off_t j = 0; j < READ; j++) {
+            LFS_ASSERT(buffer[j] == (block+i+j) % 251);
+        }
+    }
+
+    // read block 0 again
+    block = 0;
+    for (lfs_off_t i = 0; i < cfg->block_size; i += READ) {
+        cfg->read(cfg, block, i, buffer, READ) => 0;
+
+        for (lfs_off_t j = 0; j < READ; j++) {
+            LFS_ASSERT(buffer[j] == (block+i+j) % 251);
+        }
+    }
+'''
+
+[cases.test_bd_powers_of_two]
+defines.READ = ['READ_SIZE', 'BLOCK_SIZE']
+defines.PROG = ['PROG_SIZE', 'BLOCK_SIZE']
+code = '''
+    uint8_t buffer[lfs_max(READ, PROG)];
+
+    // write/read every power of 2
+    lfs_block_t block = 1;
+    while (block < cfg->block_count) {
+        // write
+        cfg->erase(cfg, block) => 0;
+        for (lfs_off_t i = 0; i < cfg->block_size; i += PROG) {
+            for (lfs_off_t j = 0; j < PROG; j++) {
+                buffer[j] = (block+i+j) % 251;
+            }
+            cfg->prog(cfg, block, i, buffer, PROG) => 0;
+        }
+
+        // read
+        for (lfs_off_t i = 0; i < cfg->block_size; i += READ) {
+            cfg->read(cfg, block, i, buffer, READ) => 0;
+
+            for (lfs_off_t j = 0; j < READ; j++) {
+                LFS_ASSERT(buffer[j] == (block+i+j) % 251);
+            }
+        }
+
+        block *= 2;
+    }
+
+    // read every power of 2 again
+    block = 1;
+    while (block < cfg->block_count) {
+        // read
+        for (lfs_off_t i = 0; i < cfg->block_size; i += READ) {
+            cfg->read(cfg, block, i, buffer, READ) => 0;
+
+            for (lfs_off_t j = 0; j < READ; j++) {
+                LFS_ASSERT(buffer[j] == (block+i+j) % 251);
+            }
+        }
+
+        block *= 2;
+    }
+'''
+
+[cases.test_bd_fibonacci]
+defines.READ = ['READ_SIZE', 'BLOCK_SIZE']
+defines.PROG = ['PROG_SIZE', 'BLOCK_SIZE']
+code = '''
+    uint8_t buffer[lfs_max(READ, PROG)];
+
+    // write/read every fibonacci number on our device
+    lfs_block_t block = 1;
+    lfs_block_t block_ = 1;
+    while (block < cfg->block_count) {
+        // write
+        cfg->erase(cfg, block) => 0;
+        for (lfs_off_t i = 0; i < cfg->block_size; i += PROG) {
+            for (lfs_off_t j = 0; j < PROG; j++) {
+                buffer[j] = (block+i+j) % 251;
+            }
+            cfg->prog(cfg, block, i, buffer, PROG) => 0;
+        }
+
+        // read
+        for (lfs_off_t i = 0; i < cfg->block_size; i += READ) {
+            cfg->read(cfg, block, i, buffer, READ) => 0;
+
+            for (lfs_off_t j = 0; j < READ; j++) {
+                LFS_ASSERT(buffer[j] == (block+i+j) % 251);
+            }
+        }
+
+        lfs_block_t nblock = block + block_;
+        block_ = block;
+        block = nblock;
+    }
+
+    // read every fibonacci number again
+    block = 1;
+    block_ = 1;
+    while (block < cfg->block_count) {
+        // read
+        for (lfs_off_t i = 0; i < cfg->block_size; i += READ) {
+            cfg->read(cfg, block, i, buffer, READ) => 0;
+
+            for (lfs_off_t j = 0; j < READ; j++) {
+                LFS_ASSERT(buffer[j] == (block+i+j) % 251);
+            }
+        }
+
+        lfs_block_t nblock = block + block_;
+        block_ = block;
+        block = nblock;
+    }
+'''
+
+
+
+
diff --git a/tests/test_compat.toml b/tests/test_compat.toml
new file mode 100644
index 0000000..ba44714
--- /dev/null
+++ b/tests/test_compat.toml
@@ -0,0 +1,1453 @@
+# Test for compatibility between different littlefs versions
+#
+# Note, these tests are a bit special. They expect to be linked against two
+# different versions of littlefs:
+# - lfs  => the new/current version of littlefs
+# - lfsp => the previous version of littlefs
+#
+# If lfsp is not linked, and LFSP is not defined, these tests will alias
+# the relevant lfs types/functions as necessary so at least the tests can
+# themselves be tested locally.
+#
+# But to get value from these tests, it's expected that the previous version
+# of littlefs be linked in during CI, with the help of scripts/changeprefix.py
+#
+
+# alias littlefs symbols as needed
+#
+# there may be a better way to do this, but oh well, explicit aliases works
+code = '''
+#ifdef LFSP
+#define STRINGIZE(x) STRINGIZE_(x)
+#define STRINGIZE_(x) #x
+#include STRINGIZE(LFSP)
+#else
+#define LFSP_DISK_VERSION LFS_DISK_VERSION
+#define LFSP_DISK_VERSION_MAJOR LFS_DISK_VERSION_MAJOR
+#define LFSP_DISK_VERSION_MINOR LFS_DISK_VERSION_MINOR
+#define lfsp_t lfs_t
+#define lfsp_config lfs_config
+#define lfsp_format lfs_format
+#define lfsp_mount lfs_mount
+#define lfsp_unmount lfs_unmount
+#define lfsp_fsinfo lfs_fsinfo
+#define lfsp_fs_stat lfs_fs_stat
+#define lfsp_dir_t lfs_dir_t
+#define lfsp_info lfs_info
+#define LFSP_TYPE_REG LFS_TYPE_REG
+#define LFSP_TYPE_DIR LFS_TYPE_DIR
+#define lfsp_mkdir lfs_mkdir
+#define lfsp_dir_open lfs_dir_open
+#define lfsp_dir_read lfs_dir_read
+#define lfsp_dir_close lfs_dir_close
+#define lfsp_file_t lfs_file_t
+#define LFSP_O_RDONLY LFS_O_RDONLY
+#define LFSP_O_WRONLY LFS_O_WRONLY
+#define LFSP_O_CREAT LFS_O_CREAT
+#define LFSP_O_EXCL LFS_O_EXCL
+#define LFSP_SEEK_SET LFS_SEEK_SET
+#define lfsp_file_open lfs_file_open
+#define lfsp_file_write lfs_file_write
+#define lfsp_file_read lfs_file_read
+#define lfsp_file_seek lfs_file_seek
+#define lfsp_file_close lfs_file_close
+#endif
+'''
+
+
+
+## forward-compatibility tests ##
+
+# test we can mount in a new version
+[cases.test_compat_forward_mount]
+if = '''
+    LFS_DISK_VERSION_MAJOR == LFSP_DISK_VERSION_MAJOR
+        && DISK_VERSION == 0
+'''
+code = '''
+    // create the previous version
+    struct lfsp_config cfgp;
+    memcpy(&cfgp, cfg, sizeof(cfgp));
+    lfsp_t lfsp;
+    lfsp_format(&lfsp, &cfgp) => 0;
+
+    // confirm the previous mount works
+    lfsp_mount(&lfsp, &cfgp) => 0;
+    lfsp_unmount(&lfsp) => 0;
+
+
+    // now test the new mount
+    lfs_t lfs;
+    lfs_mount(&lfs, cfg) => 0;
+
+    // we should be able to read the version using lfs_fs_stat
+    struct lfs_fsinfo fsinfo;
+    lfs_fs_stat(&lfs, &fsinfo) => 0;
+    assert(fsinfo.disk_version == LFSP_DISK_VERSION);
+
+    lfs_unmount(&lfs) => 0;
+'''
+
+# test we can read dirs in a new version
+[cases.test_compat_forward_read_dirs]
+defines.COUNT = 5
+if = '''
+    LFS_DISK_VERSION_MAJOR == LFSP_DISK_VERSION_MAJOR
+        && DISK_VERSION == 0
+'''
+code = '''
+    // create the previous version
+    struct lfsp_config cfgp;
+    memcpy(&cfgp, cfg, sizeof(cfgp));
+    lfsp_t lfsp;
+    lfsp_format(&lfsp, &cfgp) => 0;
+
+    // write COUNT dirs
+    lfsp_mount(&lfsp, &cfgp) => 0;
+    for (lfs_size_t i = 0; i < COUNT; i++) {
+        char name[8];
+        sprintf(name, "dir%03d", i);
+        lfsp_mkdir(&lfsp, name) => 0;
+    }
+    lfsp_unmount(&lfsp) => 0;
+
+
+    // mount the new version
+    lfs_t lfs;
+    lfs_mount(&lfs, cfg) => 0;
+
+    // we should be able to read the version using lfs_fs_stat
+    struct lfs_fsinfo fsinfo;
+    lfs_fs_stat(&lfs, &fsinfo) => 0;
+    assert(fsinfo.disk_version == LFSP_DISK_VERSION);
+
+    // can we list the directories?
+    lfs_dir_t dir;
+    lfs_dir_open(&lfs, &dir, "/") => 0;
+    struct lfs_info info;
+    lfs_dir_read(&lfs, &dir, &info) => 1;
+    assert(info.type == LFS_TYPE_DIR);
+    assert(strcmp(info.name, ".") == 0);
+    lfs_dir_read(&lfs, &dir, &info) => 1;
+    assert(info.type == LFS_TYPE_DIR);
+    assert(strcmp(info.name, "..") == 0);
+
+    for (lfs_size_t i = 0; i < COUNT; i++) {
+        lfs_dir_read(&lfs, &dir, &info) => 1;
+        assert(info.type == LFS_TYPE_DIR);
+        char name[8];
+        sprintf(name, "dir%03d", i);
+        assert(strcmp(info.name, name) == 0);
+    }
+
+    lfs_dir_read(&lfs, &dir, &info) => 0;
+    lfs_dir_close(&lfs, &dir) => 0;
+
+    lfs_unmount(&lfs) => 0;
+'''
+
+# test we can read files in a new version
+[cases.test_compat_forward_read_files]
+defines.COUNT = 5
+defines.SIZE = [4, 32, 512, 8192]
+defines.CHUNK = 4
+if = '''
+    LFS_DISK_VERSION_MAJOR == LFSP_DISK_VERSION_MAJOR
+        && DISK_VERSION == 0
+'''
+code = '''
+    // create the previous version
+    struct lfsp_config cfgp;
+    memcpy(&cfgp, cfg, sizeof(cfgp));
+    lfsp_t lfsp;
+    lfsp_format(&lfsp, &cfgp) => 0;
+
+    // write COUNT files
+    lfsp_mount(&lfsp, &cfgp) => 0;
+    uint32_t prng = 42;
+    for (lfs_size_t i = 0; i < COUNT; i++) {
+        lfsp_file_t file;
+        char name[8];
+        sprintf(name, "file%03d", i);
+        lfsp_file_open(&lfsp, &file, name,
+                LFSP_O_WRONLY | LFSP_O_CREAT | LFSP_O_EXCL) => 0;
+        for (lfs_size_t j = 0; j < SIZE; j += CHUNK) {
+            uint8_t chunk[CHUNK];
+            for (lfs_size_t k = 0; k < CHUNK; k++) {
+                chunk[k] = TEST_PRNG(&prng) & 0xff;
+            }
+
+            lfsp_file_write(&lfsp, &file, chunk, CHUNK) => CHUNK;
+        }
+        lfsp_file_close(&lfsp, &file) => 0;
+    }
+    lfsp_unmount(&lfsp) => 0;
+
+
+    // mount the new version
+    lfs_t lfs;
+    lfs_mount(&lfs, cfg) => 0;
+
+    // we should be able to read the version using lfs_fs_stat
+    struct lfs_fsinfo fsinfo;
+    lfs_fs_stat(&lfs, &fsinfo) => 0;
+    assert(fsinfo.disk_version == LFSP_DISK_VERSION);
+
+    // can we list the files?
+    lfs_dir_t dir;
+    lfs_dir_open(&lfs, &dir, "/") => 0;
+    struct lfs_info info;
+    lfs_dir_read(&lfs, &dir, &info) => 1;
+    assert(info.type == LFS_TYPE_DIR);
+    assert(strcmp(info.name, ".") == 0);
+    lfs_dir_read(&lfs, &dir, &info) => 1;
+    assert(info.type == LFS_TYPE_DIR);
+    assert(strcmp(info.name, "..") == 0);
+
+    for (lfs_size_t i = 0; i < COUNT; i++) {
+        lfs_dir_read(&lfs, &dir, &info) => 1;
+        assert(info.type == LFS_TYPE_REG);
+        char name[8];
+        sprintf(name, "file%03d", i);
+        assert(strcmp(info.name, name) == 0);
+        assert(info.size == SIZE);
+    }
+
+    lfs_dir_read(&lfs, &dir, &info) => 0;
+
+    // now can we read the files?
+    prng = 42;
+    for (lfs_size_t i = 0; i < COUNT; i++) {
+        lfs_file_t file;
+        char name[8];
+        sprintf(name, "file%03d", i);
+        lfs_file_open(&lfs, &file, name, LFS_O_RDONLY) => 0;
+        for (lfs_size_t j = 0; j < SIZE; j += CHUNK) {
+            uint8_t chunk[CHUNK];
+            lfs_file_read(&lfs, &file, chunk, CHUNK) => CHUNK;
+
+            for (lfs_size_t k = 0; k < CHUNK; k++) {
+                assert(chunk[k] == TEST_PRNG(&prng) & 0xff);
+            }
+        }
+        lfs_file_close(&lfs, &file) => 0;
+    }
+
+    lfs_unmount(&lfs) => 0;
+'''
+
+# test we can read files in dirs in a new version
+[cases.test_compat_forward_read_files_in_dirs]
+defines.COUNT = 5
+defines.SIZE = [4, 32, 512, 8192]
+defines.CHUNK = 4
+if = '''
+    LFS_DISK_VERSION_MAJOR == LFSP_DISK_VERSION_MAJOR
+        && DISK_VERSION == 0
+'''
+code = '''
+    // create the previous version
+    struct lfsp_config cfgp;
+    memcpy(&cfgp, cfg, sizeof(cfgp));
+    lfsp_t lfsp;
+    lfsp_format(&lfsp, &cfgp) => 0;
+
+    // write COUNT files+dirs
+    lfsp_mount(&lfsp, &cfgp) => 0;
+    uint32_t prng = 42;
+    for (lfs_size_t i = 0; i < COUNT; i++) {
+        char name[16];
+        sprintf(name, "dir%03d", i);
+        lfsp_mkdir(&lfsp, name) => 0;
+
+        lfsp_file_t file;
+        sprintf(name, "dir%03d/file%03d", i, i);
+        lfsp_file_open(&lfsp, &file, name,
+                LFSP_O_WRONLY | LFSP_O_CREAT | LFSP_O_EXCL) => 0;
+        for (lfs_size_t j = 0; j < SIZE; j += CHUNK) {
+            uint8_t chunk[CHUNK];
+            for (lfs_size_t k = 0; k < CHUNK; k++) {
+                chunk[k] = TEST_PRNG(&prng) & 0xff;
+            }
+
+            lfsp_file_write(&lfsp, &file, chunk, CHUNK) => CHUNK;
+        }
+        lfsp_file_close(&lfsp, &file) => 0;
+    }
+    lfsp_unmount(&lfsp) => 0;
+
+
+    // mount the new version
+    lfs_t lfs;
+    lfs_mount(&lfs, cfg) => 0;
+
+    // we should be able to read the version using lfs_fs_stat
+    struct lfs_fsinfo fsinfo;
+    lfs_fs_stat(&lfs, &fsinfo) => 0;
+    assert(fsinfo.disk_version == LFSP_DISK_VERSION);
+
+    // can we list the directories?
+    lfs_dir_t dir;
+    lfs_dir_open(&lfs, &dir, "/") => 0;
+    struct lfs_info info;
+    lfs_dir_read(&lfs, &dir, &info) => 1;
+    assert(info.type == LFS_TYPE_DIR);
+    assert(strcmp(info.name, ".") == 0);
+    lfs_dir_read(&lfs, &dir, &info) => 1;
+    assert(info.type == LFS_TYPE_DIR);
+    assert(strcmp(info.name, "..") == 0);
+
+    for (lfs_size_t i = 0; i < COUNT; i++) {
+        lfs_dir_read(&lfs, &dir, &info) => 1;
+        assert(info.type == LFS_TYPE_DIR);
+        char name[8];
+        sprintf(name, "dir%03d", i);
+        assert(strcmp(info.name, name) == 0);
+    }
+
+    lfs_dir_read(&lfs, &dir, &info) => 0;
+    lfs_dir_close(&lfs, &dir) => 0;
+
+    // can we list the files?
+    for (lfs_size_t i = 0; i < COUNT; i++) {
+        char name[8];
+        sprintf(name, "dir%03d", i);
+        lfs_dir_t dir;
+        lfs_dir_open(&lfs, &dir, name) => 0;
+        struct lfs_info info;
+        lfs_dir_read(&lfs, &dir, &info) => 1;
+        assert(info.type == LFS_TYPE_DIR);
+        assert(strcmp(info.name, ".") == 0);
+        lfs_dir_read(&lfs, &dir, &info) => 1;
+        assert(info.type == LFS_TYPE_DIR);
+        assert(strcmp(info.name, "..") == 0);
+
+        lfs_dir_read(&lfs, &dir, &info) => 1;
+        assert(info.type == LFS_TYPE_REG);
+        sprintf(name, "file%03d", i);
+        assert(strcmp(info.name, name) == 0);
+        assert(info.size == SIZE);
+
+        lfs_dir_read(&lfs, &dir, &info) => 0;
+        lfs_dir_close(&lfs, &dir) => 0;
+    }
+
+    // now can we read the files?
+    prng = 42;
+    for (lfs_size_t i = 0; i < COUNT; i++) {
+        lfs_file_t file;
+        char name[16];
+        sprintf(name, "dir%03d/file%03d", i, i);
+        lfs_file_open(&lfs, &file, name, LFS_O_RDONLY) => 0;
+        for (lfs_size_t j = 0; j < SIZE; j += CHUNK) {
+            uint8_t chunk[CHUNK];
+            lfs_file_read(&lfs, &file, chunk, CHUNK) => CHUNK;
+
+            for (lfs_size_t k = 0; k < CHUNK; k++) {
+                assert(chunk[k] == TEST_PRNG(&prng) & 0xff);
+            }
+        }
+        lfs_file_close(&lfs, &file) => 0;
+    }
+
+    lfs_unmount(&lfs) => 0;
+'''
+
+# test we can write dirs in a new version
+[cases.test_compat_forward_write_dirs]
+defines.COUNT = 10
+if = '''
+    LFS_DISK_VERSION_MAJOR == LFSP_DISK_VERSION_MAJOR
+        && DISK_VERSION == 0
+'''
+code = '''
+    // create the previous version
+    struct lfsp_config cfgp;
+    memcpy(&cfgp, cfg, sizeof(cfgp));
+    lfsp_t lfsp;
+    lfsp_format(&lfsp, &cfgp) => 0;
+
+    // write COUNT/2 dirs
+    lfsp_mount(&lfsp, &cfgp) => 0;
+    for (lfs_size_t i = 0; i < COUNT/2; i++) {
+        char name[8];
+        sprintf(name, "dir%03d", i);
+        lfsp_mkdir(&lfsp, name) => 0;
+    }
+    lfsp_unmount(&lfsp) => 0;
+
+
+    // mount the new version
+    lfs_t lfs;
+    lfs_mount(&lfs, cfg) => 0;
+
+    // we should be able to read the version using lfs_fs_stat
+    struct lfs_fsinfo fsinfo;
+    lfs_fs_stat(&lfs, &fsinfo) => 0;
+    assert(fsinfo.disk_version == LFSP_DISK_VERSION);
+
+    // write another COUNT/2 dirs
+    for (lfs_size_t i = COUNT/2; i < COUNT; i++) {
+        char name[8];
+        sprintf(name, "dir%03d", i);
+        lfs_mkdir(&lfs, name) => 0;
+    }
+
+    // can we list the directories?
+    lfs_dir_t dir;
+    lfs_dir_open(&lfs, &dir, "/") => 0;
+    struct lfs_info info;
+    lfs_dir_read(&lfs, &dir, &info) => 1;
+    assert(info.type == LFS_TYPE_DIR);
+    assert(strcmp(info.name, ".") == 0);
+    lfs_dir_read(&lfs, &dir, &info) => 1;
+    assert(info.type == LFS_TYPE_DIR);
+    assert(strcmp(info.name, "..") == 0);
+
+    for (lfs_size_t i = 0; i < COUNT; i++) {
+        lfs_dir_read(&lfs, &dir, &info) => 1;
+        assert(info.type == LFS_TYPE_DIR);
+        char name[8];
+        sprintf(name, "dir%03d", i);
+        assert(strcmp(info.name, name) == 0);
+    }
+
+    lfs_dir_read(&lfs, &dir, &info) => 0;
+    lfs_dir_close(&lfs, &dir) => 0;
+
+    lfs_unmount(&lfs) => 0;
+'''
+
+# test we can write files in a new version
+[cases.test_compat_forward_write_files]
+defines.COUNT = 5
+defines.SIZE = [4, 32, 512, 8192]
+defines.CHUNK = 2
+if = '''
+    LFS_DISK_VERSION_MAJOR == LFSP_DISK_VERSION_MAJOR
+        && DISK_VERSION == 0
+'''
+code = '''
+    // create the previous version
+    struct lfsp_config cfgp;
+    memcpy(&cfgp, cfg, sizeof(cfgp));
+    lfsp_t lfsp;
+    lfsp_format(&lfsp, &cfgp) => 0;
+
+    // write half COUNT files
+    lfsp_mount(&lfsp, &cfgp) => 0;
+    uint32_t prng = 42;
+    for (lfs_size_t i = 0; i < COUNT; i++) {
+        // write half
+        lfsp_file_t file;
+        char name[8];
+        sprintf(name, "file%03d", i);
+        lfsp_file_open(&lfsp, &file, name,
+                LFSP_O_WRONLY | LFSP_O_CREAT | LFSP_O_EXCL) => 0;
+        for (lfs_size_t j = 0; j < SIZE/2; j += CHUNK) {
+            uint8_t chunk[CHUNK];
+            for (lfs_size_t k = 0; k < CHUNK; k++) {
+                chunk[k] = TEST_PRNG(&prng) & 0xff;
+            }
+
+            lfsp_file_write(&lfsp, &file, chunk, CHUNK) => CHUNK;
+        }
+        lfsp_file_close(&lfsp, &file) => 0;
+
+        // skip the other half but keep our prng reproducible
+        for (lfs_size_t j = SIZE/2; j < SIZE; j++) {
+            TEST_PRNG(&prng);
+        }
+    }
+    lfsp_unmount(&lfsp) => 0;
+
+
+    // mount the new version
+    lfs_t lfs;
+    lfs_mount(&lfs, cfg) => 0;
+
+    // we should be able to read the version using lfs_fs_stat
+    struct lfs_fsinfo fsinfo;
+    lfs_fs_stat(&lfs, &fsinfo) => 0;
+    assert(fsinfo.disk_version == LFSP_DISK_VERSION);
+
+    // write half COUNT files
+    prng = 42;
+    for (lfs_size_t i = 0; i < COUNT; i++) {
+        // skip half but keep our prng reproducible
+        for (lfs_size_t j = 0; j < SIZE/2; j++) {
+            TEST_PRNG(&prng);
+        }
+
+        // write the other half
+        lfs_file_t file;
+        char name[8];
+        sprintf(name, "file%03d", i);
+        lfs_file_open(&lfs, &file, name, LFS_O_WRONLY) => 0;
+        lfs_file_seek(&lfs, &file, SIZE/2, LFS_SEEK_SET) => SIZE/2;
+
+        for (lfs_size_t j = SIZE/2; j < SIZE; j += CHUNK) {
+            uint8_t chunk[CHUNK];
+            for (lfs_size_t k = 0; k < CHUNK; k++) {
+                chunk[k] = TEST_PRNG(&prng) & 0xff;
+            }
+
+            lfs_file_write(&lfs, &file, chunk, CHUNK) => CHUNK;
+        }
+        lfs_file_close(&lfs, &file) => 0;
+    }
+
+    // can we list the files?
+    lfs_dir_t dir;
+    lfs_dir_open(&lfs, &dir, "/") => 0;
+    struct lfs_info info;
+    lfs_dir_read(&lfs, &dir, &info) => 1;
+    assert(info.type == LFS_TYPE_DIR);
+    assert(strcmp(info.name, ".") == 0);
+    lfs_dir_read(&lfs, &dir, &info) => 1;
+    assert(info.type == LFS_TYPE_DIR);
+    assert(strcmp(info.name, "..") == 0);
+
+    for (lfs_size_t i = 0; i < COUNT; i++) {
+        lfs_dir_read(&lfs, &dir, &info) => 1;
+        assert(info.type == LFS_TYPE_REG);
+        char name[8];
+        sprintf(name, "file%03d", i);
+        assert(strcmp(info.name, name) == 0);
+        assert(info.size == SIZE);
+    }
+
+    lfs_dir_read(&lfs, &dir, &info) => 0;
+
+    // now can we read the files?
+    prng = 42;
+    for (lfs_size_t i = 0; i < COUNT; i++) {
+        lfs_file_t file;
+        char name[8];
+        sprintf(name, "file%03d", i);
+        lfs_file_open(&lfs, &file, name, LFS_O_RDONLY) => 0;
+        for (lfs_size_t j = 0; j < SIZE; j += CHUNK) {
+            uint8_t chunk[CHUNK];
+            lfs_file_read(&lfs, &file, chunk, CHUNK) => CHUNK;
+
+            for (lfs_size_t k = 0; k < CHUNK; k++) {
+                assert(chunk[k] == TEST_PRNG(&prng) & 0xff);
+            }
+        }
+        lfs_file_close(&lfs, &file) => 0;
+    }
+
+    lfs_unmount(&lfs) => 0;
+'''
+
+# test we can write files in dirs in a new version
+[cases.test_compat_forward_write_files_in_dirs]
+defines.COUNT = 5
+defines.SIZE = [4, 32, 512, 8192]
+defines.CHUNK = 2
+if = '''
+    LFS_DISK_VERSION_MAJOR == LFSP_DISK_VERSION_MAJOR
+        && DISK_VERSION == 0
+'''
+code = '''
+    // create the previous version
+    struct lfsp_config cfgp;
+    memcpy(&cfgp, cfg, sizeof(cfgp));
+    lfsp_t lfsp;
+    lfsp_format(&lfsp, &cfgp) => 0;
+
+    // write half COUNT files
+    lfsp_mount(&lfsp, &cfgp) => 0;
+    uint32_t prng = 42;
+    for (lfs_size_t i = 0; i < COUNT; i++) {
+        char name[16];
+        sprintf(name, "dir%03d", i);
+        lfsp_mkdir(&lfsp, name) => 0;
+
+        // write half
+        lfsp_file_t file;
+        sprintf(name, "dir%03d/file%03d", i, i);
+        lfsp_file_open(&lfsp, &file, name,
+                LFSP_O_WRONLY | LFSP_O_CREAT | LFSP_O_EXCL) => 0;
+        for (lfs_size_t j = 0; j < SIZE/2; j += CHUNK) {
+            uint8_t chunk[CHUNK];
+            for (lfs_size_t k = 0; k < CHUNK; k++) {
+                chunk[k] = TEST_PRNG(&prng) & 0xff;
+            }
+
+            lfsp_file_write(&lfsp, &file, chunk, CHUNK) => CHUNK;
+        }
+        lfsp_file_close(&lfsp, &file) => 0;
+
+        // skip the other half but keep our prng reproducible
+        for (lfs_size_t j = SIZE/2; j < SIZE; j++) {
+            TEST_PRNG(&prng);
+        }
+    }
+    lfsp_unmount(&lfsp) => 0;
+
+
+    // mount the new version
+    lfs_t lfs;
+    lfs_mount(&lfs, cfg) => 0;
+
+    // we should be able to read the version using lfs_fs_stat
+    struct lfs_fsinfo fsinfo;
+    lfs_fs_stat(&lfs, &fsinfo) => 0;
+    assert(fsinfo.disk_version == LFSP_DISK_VERSION);
+
+    // write half COUNT files
+    prng = 42;
+    for (lfs_size_t i = 0; i < COUNT; i++) {
+        // skip half but keep our prng reproducible
+        for (lfs_size_t j = 0; j < SIZE/2; j++) {
+            TEST_PRNG(&prng);
+        }
+
+        // write the other half
+        lfs_file_t file;
+        char name[16];
+        sprintf(name, "dir%03d/file%03d", i, i);
+        lfs_file_open(&lfs, &file, name, LFS_O_WRONLY) => 0;
+        lfs_file_seek(&lfs, &file, SIZE/2, LFS_SEEK_SET) => SIZE/2;
+
+        for (lfs_size_t j = SIZE/2; j < SIZE; j += CHUNK) {
+            uint8_t chunk[CHUNK];
+            for (lfs_size_t k = 0; k < CHUNK; k++) {
+                chunk[k] = TEST_PRNG(&prng) & 0xff;
+            }
+
+            lfs_file_write(&lfs, &file, chunk, CHUNK) => CHUNK;
+        }
+        lfs_file_close(&lfs, &file) => 0;
+    }
+
+    // can we list the directories?
+    lfs_dir_t dir;
+    lfs_dir_open(&lfs, &dir, "/") => 0;
+    struct lfs_info info;
+    lfs_dir_read(&lfs, &dir, &info) => 1;
+    assert(info.type == LFS_TYPE_DIR);
+    assert(strcmp(info.name, ".") == 0);
+    lfs_dir_read(&lfs, &dir, &info) => 1;
+    assert(info.type == LFS_TYPE_DIR);
+    assert(strcmp(info.name, "..") == 0);
+
+    for (lfs_size_t i = 0; i < COUNT; i++) {
+        lfs_dir_read(&lfs, &dir, &info) => 1;
+        assert(info.type == LFS_TYPE_DIR);
+        char name[8];
+        sprintf(name, "dir%03d", i);
+        assert(strcmp(info.name, name) == 0);
+    }
+
+    lfs_dir_read(&lfs, &dir, &info) => 0;
+    lfs_dir_close(&lfs, &dir) => 0;
+
+    // can we list the files?
+    for (lfs_size_t i = 0; i < COUNT; i++) {
+        char name[8];
+        sprintf(name, "dir%03d", i);
+        lfs_dir_t dir;
+        lfs_dir_open(&lfs, &dir, name) => 0;
+        struct lfs_info info;
+        lfs_dir_read(&lfs, &dir, &info) => 1;
+        assert(info.type == LFS_TYPE_DIR);
+        assert(strcmp(info.name, ".") == 0);
+        lfs_dir_read(&lfs, &dir, &info) => 1;
+        assert(info.type == LFS_TYPE_DIR);
+        assert(strcmp(info.name, "..") == 0);
+
+        lfs_dir_read(&lfs, &dir, &info) => 1;
+        assert(info.type == LFS_TYPE_REG);
+        sprintf(name, "file%03d", i);
+        assert(strcmp(info.name, name) == 0);
+        assert(info.size == SIZE);
+
+        lfs_dir_read(&lfs, &dir, &info) => 0;
+        lfs_dir_close(&lfs, &dir) => 0;
+    }
+
+    // now can we read the files?
+    prng = 42;
+    for (lfs_size_t i = 0; i < COUNT; i++) {
+        lfs_file_t file;
+        char name[16];
+        sprintf(name, "dir%03d/file%03d", i, i);
+        lfs_file_open(&lfs, &file, name, LFS_O_RDONLY) => 0;
+        for (lfs_size_t j = 0; j < SIZE; j += CHUNK) {
+            uint8_t chunk[CHUNK];
+            lfs_file_read(&lfs, &file, chunk, CHUNK) => CHUNK;
+
+            for (lfs_size_t k = 0; k < CHUNK; k++) {
+                assert(chunk[k] == TEST_PRNG(&prng) & 0xff);
+            }
+        }
+        lfs_file_close(&lfs, &file) => 0;
+    }
+
+    lfs_unmount(&lfs) => 0;
+'''
+
+
+
+## backwards-compatibility tests ##
+
+# test we can mount in an old version
+[cases.test_compat_backward_mount]
+if = '''
+    LFS_DISK_VERSION == LFSP_DISK_VERSION
+        && DISK_VERSION == 0
+'''
+code = '''
+    // create the new version
+    lfs_t lfs;
+    lfs_format(&lfs, cfg) => 0;
+
+    // confirm the new mount works
+    lfs_mount(&lfs, cfg) => 0;
+    lfs_unmount(&lfs) => 0;
+
+    // now test the previous mount
+    struct lfsp_config cfgp;
+    memcpy(&cfgp, cfg, sizeof(cfgp));
+    lfsp_t lfsp;
+    lfsp_mount(&lfsp, &cfgp) => 0;
+
+    lfsp_unmount(&lfsp) => 0;
+'''
+
+# test we can read dirs in an old version
+[cases.test_compat_backward_read_dirs]
+defines.COUNT = 5
+if = '''
+    LFS_DISK_VERSION == LFSP_DISK_VERSION
+        && DISK_VERSION == 0
+'''
+code = '''
+    // create the new version
+    lfs_t lfs;
+    lfs_format(&lfs, cfg) => 0;
+
+    // write COUNT dirs
+    lfs_mount(&lfs, cfg) => 0;
+    for (lfs_size_t i = 0; i < COUNT; i++) {
+        char name[8];
+        sprintf(name, "dir%03d", i);
+        lfs_mkdir(&lfs, name) => 0;
+    }
+    lfs_unmount(&lfs) => 0;
+
+
+    // mount the new version
+    struct lfsp_config cfgp;
+    memcpy(&cfgp, cfg, sizeof(cfgp));
+    lfsp_t lfsp;
+    lfsp_mount(&lfsp, &cfgp) => 0;
+
+    // can we list the directories?
+    lfsp_dir_t dir;
+    lfsp_dir_open(&lfsp, &dir, "/") => 0;
+    struct lfsp_info info;
+    lfsp_dir_read(&lfsp, &dir, &info) => 1;
+    assert(info.type == LFSP_TYPE_DIR);
+    assert(strcmp(info.name, ".") == 0);
+    lfsp_dir_read(&lfsp, &dir, &info) => 1;
+    assert(info.type == LFSP_TYPE_DIR);
+    assert(strcmp(info.name, "..") == 0);
+
+    for (lfs_size_t i = 0; i < COUNT; i++) {
+        lfsp_dir_read(&lfsp, &dir, &info) => 1;
+        assert(info.type == LFSP_TYPE_DIR);
+        char name[8];
+        sprintf(name, "dir%03d", i);
+        assert(strcmp(info.name, name) == 0);
+    }
+
+    lfsp_dir_read(&lfsp, &dir, &info) => 0;
+    lfsp_dir_close(&lfsp, &dir) => 0;
+
+    lfsp_unmount(&lfsp) => 0;
+'''
+
+# test we can read files in an old version
+[cases.test_compat_backward_read_files]
+defines.COUNT = 5
+defines.SIZE = [4, 32, 512, 8192]
+defines.CHUNK = 4
+if = '''
+    LFS_DISK_VERSION == LFSP_DISK_VERSION
+        && DISK_VERSION == 0
+'''
+code = '''
+    // create the new version
+    lfs_t lfs;
+    lfs_format(&lfs, cfg) => 0;
+
+    // write COUNT files
+    lfs_mount(&lfs, cfg) => 0;
+    uint32_t prng = 42;
+    for (lfs_size_t i = 0; i < COUNT; i++) {
+        lfs_file_t file;
+        char name[8];
+        sprintf(name, "file%03d", i);
+        lfs_file_open(&lfs, &file, name,
+                LFS_O_WRONLY | LFS_O_CREAT | LFS_O_EXCL) => 0;
+        for (lfs_size_t j = 0; j < SIZE; j += CHUNK) {
+            uint8_t chunk[CHUNK];
+            for (lfs_size_t k = 0; k < CHUNK; k++) {
+                chunk[k] = TEST_PRNG(&prng) & 0xff;
+            }
+
+            lfs_file_write(&lfs, &file, chunk, CHUNK) => CHUNK;
+        }
+        lfs_file_close(&lfs, &file) => 0;
+    }
+    lfs_unmount(&lfs) => 0;
+
+
+    // mount the previous version
+    struct lfsp_config cfgp;
+    memcpy(&cfgp, cfg, sizeof(cfgp));
+    lfsp_t lfsp;
+    lfsp_mount(&lfsp, &cfgp) => 0;
+
+    // can we list the files?
+    lfsp_dir_t dir;
+    lfsp_dir_open(&lfsp, &dir, "/") => 0;
+    struct lfsp_info info;
+    lfsp_dir_read(&lfsp, &dir, &info) => 1;
+    assert(info.type == LFSP_TYPE_DIR);
+    assert(strcmp(info.name, ".") == 0);
+    lfsp_dir_read(&lfsp, &dir, &info) => 1;
+    assert(info.type == LFSP_TYPE_DIR);
+    assert(strcmp(info.name, "..") == 0);
+
+    for (lfs_size_t i = 0; i < COUNT; i++) {
+        lfsp_dir_read(&lfsp, &dir, &info) => 1;
+        assert(info.type == LFSP_TYPE_REG);
+        char name[8];
+        sprintf(name, "file%03d", i);
+        assert(strcmp(info.name, name) == 0);
+        assert(info.size == SIZE);
+    }
+
+    lfsp_dir_read(&lfsp, &dir, &info) => 0;
+
+    // now can we read the files?
+    prng = 42;
+    for (lfs_size_t i = 0; i < COUNT; i++) {
+        lfsp_file_t file;
+        char name[8];
+        sprintf(name, "file%03d", i);
+        lfsp_file_open(&lfsp, &file, name, LFSP_O_RDONLY) => 0;
+        for (lfs_size_t j = 0; j < SIZE; j += CHUNK) {
+            uint8_t chunk[CHUNK];
+            lfsp_file_read(&lfsp, &file, chunk, CHUNK) => CHUNK;
+
+            for (lfs_size_t k = 0; k < CHUNK; k++) {
+                assert(chunk[k] == TEST_PRNG(&prng) & 0xff);
+            }
+        }
+        lfsp_file_close(&lfsp, &file) => 0;
+    }
+
+    lfsp_unmount(&lfsp) => 0;
+'''
+
+# test we can read files in dirs in an old version
+[cases.test_compat_backward_read_files_in_dirs]
+defines.COUNT = 5
+defines.SIZE = [4, 32, 512, 8192]
+defines.CHUNK = 4
+if = '''
+    LFS_DISK_VERSION == LFSP_DISK_VERSION
+        && DISK_VERSION == 0
+'''
+code = '''
+    // create the new version
+    lfs_t lfs;
+    lfs_format(&lfs, cfg) => 0;
+
+    // write COUNT files+dirs
+    lfs_mount(&lfs, cfg) => 0;
+    uint32_t prng = 42;
+    for (lfs_size_t i = 0; i < COUNT; i++) {
+        char name[16];
+        sprintf(name, "dir%03d", i);
+        lfs_mkdir(&lfs, name) => 0;
+
+        lfs_file_t file;
+        sprintf(name, "dir%03d/file%03d", i, i);
+        lfs_file_open(&lfs, &file, name,
+                LFS_O_WRONLY | LFS_O_CREAT | LFS_O_EXCL) => 0;
+        for (lfs_size_t j = 0; j < SIZE; j += CHUNK) {
+            uint8_t chunk[CHUNK];
+            for (lfs_size_t k = 0; k < CHUNK; k++) {
+                chunk[k] = TEST_PRNG(&prng) & 0xff;
+            }
+
+            lfs_file_write(&lfs, &file, chunk, CHUNK) => CHUNK;
+        }
+        lfs_file_close(&lfs, &file) => 0;
+    }
+    lfs_unmount(&lfs) => 0;
+
+
+    // mount the previous version
+    struct lfsp_config cfgp;
+    memcpy(&cfgp, cfg, sizeof(cfgp));
+    lfsp_t lfsp;
+    lfsp_mount(&lfsp, &cfgp) => 0;
+
+    // can we list the directories?
+    lfsp_dir_t dir;
+    lfsp_dir_open(&lfsp, &dir, "/") => 0;
+    struct lfsp_info info;
+    lfsp_dir_read(&lfsp, &dir, &info) => 1;
+    assert(info.type == LFSP_TYPE_DIR);
+    assert(strcmp(info.name, ".") == 0);
+    lfsp_dir_read(&lfsp, &dir, &info) => 1;
+    assert(info.type == LFSP_TYPE_DIR);
+    assert(strcmp(info.name, "..") == 0);
+
+    for (lfs_size_t i = 0; i < COUNT; i++) {
+        lfsp_dir_read(&lfsp, &dir, &info) => 1;
+        assert(info.type == LFSP_TYPE_DIR);
+        char name[8];
+        sprintf(name, "dir%03d", i);
+        assert(strcmp(info.name, name) == 0);
+    }
+
+    lfsp_dir_read(&lfsp, &dir, &info) => 0;
+    lfsp_dir_close(&lfsp, &dir) => 0;
+
+    // can we list the files?
+    for (lfs_size_t i = 0; i < COUNT; i++) {
+        char name[8];
+        sprintf(name, "dir%03d", i);
+        lfsp_dir_t dir;
+        lfsp_dir_open(&lfsp, &dir, name) => 0;
+        struct lfsp_info info;
+        lfsp_dir_read(&lfsp, &dir, &info) => 1;
+        assert(info.type == LFSP_TYPE_DIR);
+        assert(strcmp(info.name, ".") == 0);
+        lfsp_dir_read(&lfsp, &dir, &info) => 1;
+        assert(info.type == LFSP_TYPE_DIR);
+        assert(strcmp(info.name, "..") == 0);
+
+        lfsp_dir_read(&lfsp, &dir, &info) => 1;
+        assert(info.type == LFSP_TYPE_REG);
+        sprintf(name, "file%03d", i);
+        assert(strcmp(info.name, name) == 0);
+        assert(info.size == SIZE);
+
+        lfsp_dir_read(&lfsp, &dir, &info) => 0;
+        lfsp_dir_close(&lfsp, &dir) => 0;
+    }
+
+    // now can we read the files?
+    prng = 42;
+    for (lfs_size_t i = 0; i < COUNT; i++) {
+        lfsp_file_t file;
+        char name[16];
+        sprintf(name, "dir%03d/file%03d", i, i);
+        lfsp_file_open(&lfsp, &file, name, LFSP_O_RDONLY) => 0;
+        for (lfs_size_t j = 0; j < SIZE; j += CHUNK) {
+            uint8_t chunk[CHUNK];
+            lfsp_file_read(&lfsp, &file, chunk, CHUNK) => CHUNK;
+
+            for (lfs_size_t k = 0; k < CHUNK; k++) {
+                assert(chunk[k] == TEST_PRNG(&prng) & 0xff);
+            }
+        }
+        lfsp_file_close(&lfsp, &file) => 0;
+    }
+
+    lfsp_unmount(&lfsp) => 0;
+'''
+
+# test we can write dirs in an old version
+[cases.test_compat_backward_write_dirs]
+defines.COUNT = 10
+if = '''
+    LFS_DISK_VERSION == LFSP_DISK_VERSION
+        && DISK_VERSION == 0
+'''
+code = '''
+    // create the new version
+    lfs_t lfs;
+    lfs_format(&lfs, cfg) => 0;
+
+    // write COUNT/2 dirs
+    lfs_mount(&lfs, cfg) => 0;
+    for (lfs_size_t i = 0; i < COUNT/2; i++) {
+        char name[8];
+        sprintf(name, "dir%03d", i);
+        lfs_mkdir(&lfs, name) => 0;
+    }
+    lfs_unmount(&lfs) => 0;
+
+
+    // mount the previous version
+    struct lfsp_config cfgp;
+    memcpy(&cfgp, cfg, sizeof(cfgp));
+    lfsp_t lfsp;
+    lfsp_mount(&lfsp, &cfgp) => 0;
+
+    // write another COUNT/2 dirs
+    for (lfs_size_t i = COUNT/2; i < COUNT; i++) {
+        char name[8];
+        sprintf(name, "dir%03d", i);
+        lfsp_mkdir(&lfsp, name) => 0;
+    }
+
+    // can we list the directories?
+    lfsp_dir_t dir;
+    lfsp_dir_open(&lfsp, &dir, "/") => 0;
+    struct lfsp_info info;
+    lfsp_dir_read(&lfsp, &dir, &info) => 1;
+    assert(info.type == LFSP_TYPE_DIR);
+    assert(strcmp(info.name, ".") == 0);
+    lfsp_dir_read(&lfsp, &dir, &info) => 1;
+    assert(info.type == LFSP_TYPE_DIR);
+    assert(strcmp(info.name, "..") == 0);
+
+    for (lfs_size_t i = 0; i < COUNT; i++) {
+        lfsp_dir_read(&lfsp, &dir, &info) => 1;
+        assert(info.type == LFSP_TYPE_DIR);
+        char name[8];
+        sprintf(name, "dir%03d", i);
+        assert(strcmp(info.name, name) == 0);
+    }
+
+    lfsp_dir_read(&lfsp, &dir, &info) => 0;
+    lfsp_dir_close(&lfsp, &dir) => 0;
+
+    lfsp_unmount(&lfsp) => 0;
+'''
+
+# test we can write files in an old version
+[cases.test_compat_backward_write_files]
+defines.COUNT = 5
+defines.SIZE = [4, 32, 512, 8192]
+defines.CHUNK = 2
+if = '''
+    LFS_DISK_VERSION == LFSP_DISK_VERSION
+        && DISK_VERSION == 0
+'''
+code = '''
+    // create the previous version
+    lfs_t lfs;
+    lfs_format(&lfs, cfg) => 0;
+
+    // write half COUNT files
+    lfs_mount(&lfs, cfg) => 0;
+    uint32_t prng = 42;
+    for (lfs_size_t i = 0; i < COUNT; i++) {
+        // write half
+        lfs_file_t file;
+        char name[8];
+        sprintf(name, "file%03d", i);
+        lfs_file_open(&lfs, &file, name,
+                LFS_O_WRONLY | LFS_O_CREAT | LFS_O_EXCL) => 0;
+        for (lfs_size_t j = 0; j < SIZE/2; j += CHUNK) {
+            uint8_t chunk[CHUNK];
+            for (lfs_size_t k = 0; k < CHUNK; k++) {
+                chunk[k] = TEST_PRNG(&prng) & 0xff;
+            }
+
+            lfs_file_write(&lfs, &file, chunk, CHUNK) => CHUNK;
+        }
+        lfs_file_close(&lfs, &file) => 0;
+
+        // skip the other half but keep our prng reproducible
+        for (lfs_size_t j = SIZE/2; j < SIZE; j++) {
+            TEST_PRNG(&prng);
+        }
+    }
+    lfs_unmount(&lfs) => 0;
+
+
+    // mount the new version
+    struct lfsp_config cfgp;
+    memcpy(&cfgp, cfg, sizeof(cfgp));
+    lfsp_t lfsp;
+    lfsp_mount(&lfsp, &cfgp) => 0;
+
+    // write half COUNT files
+    prng = 42;
+    for (lfs_size_t i = 0; i < COUNT; i++) {
+        // skip half but keep our prng reproducible
+        for (lfs_size_t j = 0; j < SIZE/2; j++) {
+            TEST_PRNG(&prng);
+        }
+
+        // write the other half
+        lfsp_file_t file;
+        char name[8];
+        sprintf(name, "file%03d", i);
+        lfsp_file_open(&lfsp, &file, name, LFSP_O_WRONLY) => 0;
+        lfsp_file_seek(&lfsp, &file, SIZE/2, LFSP_SEEK_SET) => SIZE/2;
+
+        for (lfs_size_t j = SIZE/2; j < SIZE; j += CHUNK) {
+            uint8_t chunk[CHUNK];
+            for (lfs_size_t k = 0; k < CHUNK; k++) {
+                chunk[k] = TEST_PRNG(&prng) & 0xff;
+            }
+
+            lfsp_file_write(&lfsp, &file, chunk, CHUNK) => CHUNK;
+        }
+        lfsp_file_close(&lfsp, &file) => 0;
+    }
+
+    // can we list the files?
+    lfsp_dir_t dir;
+    lfsp_dir_open(&lfsp, &dir, "/") => 0;
+    struct lfsp_info info;
+    lfsp_dir_read(&lfsp, &dir, &info) => 1;
+    assert(info.type == LFSP_TYPE_DIR);
+    assert(strcmp(info.name, ".") == 0);
+    lfsp_dir_read(&lfsp, &dir, &info) => 1;
+    assert(info.type == LFSP_TYPE_DIR);
+    assert(strcmp(info.name, "..") == 0);
+
+    for (lfs_size_t i = 0; i < COUNT; i++) {
+        lfsp_dir_read(&lfsp, &dir, &info) => 1;
+        assert(info.type == LFSP_TYPE_REG);
+        char name[8];
+        sprintf(name, "file%03d", i);
+        assert(strcmp(info.name, name) == 0);
+        assert(info.size == SIZE);
+    }
+
+    lfsp_dir_read(&lfsp, &dir, &info) => 0;
+
+    // now can we read the files?
+    prng = 42;
+    for (lfs_size_t i = 0; i < COUNT; i++) {
+        lfsp_file_t file;
+        char name[8];
+        sprintf(name, "file%03d", i);
+        lfsp_file_open(&lfsp, &file, name, LFSP_O_RDONLY) => 0;
+        for (lfs_size_t j = 0; j < SIZE; j += CHUNK) {
+            uint8_t chunk[CHUNK];
+            lfsp_file_read(&lfsp, &file, chunk, CHUNK) => CHUNK;
+
+            for (lfs_size_t k = 0; k < CHUNK; k++) {
+                assert(chunk[k] == TEST_PRNG(&prng) & 0xff);
+            }
+        }
+        lfsp_file_close(&lfsp, &file) => 0;
+    }
+
+    lfsp_unmount(&lfsp) => 0;
+'''
+
+# test we can write files in dirs in an old version
+[cases.test_compat_backward_write_files_in_dirs]
+defines.COUNT = 5
+defines.SIZE = [4, 32, 512, 8192]
+defines.CHUNK = 2
+if = '''
+    LFS_DISK_VERSION == LFSP_DISK_VERSION
+        && DISK_VERSION == 0
+'''
+code = '''
+    // create the previous version
+    lfs_t lfs;
+    lfs_format(&lfs, cfg) => 0;
+
+    // write half COUNT files
+    lfs_mount(&lfs, cfg) => 0;
+    uint32_t prng = 42;
+    for (lfs_size_t i = 0; i < COUNT; i++) {
+        char name[16];
+        sprintf(name, "dir%03d", i);
+        lfs_mkdir(&lfs, name) => 0;
+
+        // write half
+        lfs_file_t file;
+        sprintf(name, "dir%03d/file%03d", i, i);
+        lfs_file_open(&lfs, &file, name,
+                LFS_O_WRONLY | LFS_O_CREAT | LFS_O_EXCL) => 0;
+        for (lfs_size_t j = 0; j < SIZE/2; j += CHUNK) {
+            uint8_t chunk[CHUNK];
+            for (lfs_size_t k = 0; k < CHUNK; k++) {
+                chunk[k] = TEST_PRNG(&prng) & 0xff;
+            }
+
+            lfs_file_write(&lfs, &file, chunk, CHUNK) => CHUNK;
+        }
+        lfs_file_close(&lfs, &file) => 0;
+
+        // skip the other half but keep our prng reproducible
+        for (lfs_size_t j = SIZE/2; j < SIZE; j++) {
+            TEST_PRNG(&prng);
+        }
+    }
+    lfs_unmount(&lfs) => 0;
+
+
+    // mount the new version
+    struct lfsp_config cfgp;
+    memcpy(&cfgp, cfg, sizeof(cfgp));
+    lfsp_t lfsp;
+    lfsp_mount(&lfsp, &cfgp) => 0;
+
+    // write half COUNT files
+    prng = 42;
+    for (lfs_size_t i = 0; i < COUNT; i++) {
+        // skip half but keep our prng reproducible
+        for (lfs_size_t j = 0; j < SIZE/2; j++) {
+            TEST_PRNG(&prng);
+        }
+
+        // write the other half
+        lfsp_file_t file;
+        char name[16];
+        sprintf(name, "dir%03d/file%03d", i, i);
+        lfsp_file_open(&lfsp, &file, name, LFSP_O_WRONLY) => 0;
+        lfsp_file_seek(&lfsp, &file, SIZE/2, LFSP_SEEK_SET) => SIZE/2;
+
+        for (lfs_size_t j = SIZE/2; j < SIZE; j += CHUNK) {
+            uint8_t chunk[CHUNK];
+            for (lfs_size_t k = 0; k < CHUNK; k++) {
+                chunk[k] = TEST_PRNG(&prng) & 0xff;
+            }
+
+            lfsp_file_write(&lfsp, &file, chunk, CHUNK) => CHUNK;
+        }
+        lfsp_file_close(&lfsp, &file) => 0;
+    }
+
+    // can we list the directories?
+    lfsp_dir_t dir;
+    lfsp_dir_open(&lfsp, &dir, "/") => 0;
+    struct lfsp_info info;
+    lfsp_dir_read(&lfsp, &dir, &info) => 1;
+    assert(info.type == LFSP_TYPE_DIR);
+    assert(strcmp(info.name, ".") == 0);
+    lfsp_dir_read(&lfsp, &dir, &info) => 1;
+    assert(info.type == LFSP_TYPE_DIR);
+    assert(strcmp(info.name, "..") == 0);
+
+    for (lfs_size_t i = 0; i < COUNT; i++) {
+        lfsp_dir_read(&lfsp, &dir, &info) => 1;
+        assert(info.type == LFSP_TYPE_DIR);
+        char name[8];
+        sprintf(name, "dir%03d", i);
+        assert(strcmp(info.name, name) == 0);
+    }
+
+    lfsp_dir_read(&lfsp, &dir, &info) => 0;
+    lfsp_dir_close(&lfsp, &dir) => 0;
+
+    // can we list the files?
+    for (lfs_size_t i = 0; i < COUNT; i++) {
+        char name[8];
+        sprintf(name, "dir%03d", i);
+        lfsp_dir_t dir;
+        lfsp_dir_open(&lfsp, &dir, name) => 0;
+        struct lfsp_info info;
+        lfsp_dir_read(&lfsp, &dir, &info) => 1;
+        assert(info.type == LFSP_TYPE_DIR);
+        assert(strcmp(info.name, ".") == 0);
+        lfsp_dir_read(&lfsp, &dir, &info) => 1;
+        assert(info.type == LFSP_TYPE_DIR);
+        assert(strcmp(info.name, "..") == 0);
+
+        lfsp_dir_read(&lfsp, &dir, &info) => 1;
+        assert(info.type == LFSP_TYPE_REG);
+        sprintf(name, "file%03d", i);
+        assert(strcmp(info.name, name) == 0);
+        assert(info.size == SIZE);
+
+        lfsp_dir_read(&lfsp, &dir, &info) => 0;
+        lfsp_dir_close(&lfsp, &dir) => 0;
+    }
+
+    // now can we read the files?
+    prng = 42;
+    for (lfs_size_t i = 0; i < COUNT; i++) {
+        lfsp_file_t file;
+        char name[16];
+        sprintf(name, "dir%03d/file%03d", i, i);
+        lfsp_file_open(&lfsp, &file, name, LFSP_O_RDONLY) => 0;
+        for (lfs_size_t j = 0; j < SIZE; j += CHUNK) {
+            uint8_t chunk[CHUNK];
+            lfsp_file_read(&lfsp, &file, chunk, CHUNK) => CHUNK;
+
+            for (lfs_size_t k = 0; k < CHUNK; k++) {
+                assert(chunk[k] == TEST_PRNG(&prng) & 0xff);
+            }
+        }
+        lfsp_file_close(&lfsp, &file) => 0;
+    }
+
+    lfsp_unmount(&lfsp) => 0;
+'''
+
+
+
+## incompatiblity tests ##
+
+# test that we fail to mount after a major version bump
+[cases.test_compat_major_incompat]
+in = 'lfs.c'
+code = '''
+    // create a superblock
+    lfs_t lfs;
+    lfs_format(&lfs, cfg) => 0;
+
+    // bump the major version
+    //
+    // note we're messing around with internals to do this! this
+    // is not a user API
+    lfs_mount(&lfs, cfg) => 0;
+    lfs_mdir_t mdir;
+    lfs_dir_fetch(&lfs, &mdir, (lfs_block_t[2]){0, 1}) => 0;
+    lfs_superblock_t superblock = {
+        .version     = LFS_DISK_VERSION + 0x00010000,
+        .block_size  = lfs.cfg->block_size,
+        .block_count = lfs.cfg->block_count,
+        .name_max    = lfs.name_max,
+        .file_max    = lfs.file_max,
+        .attr_max    = lfs.attr_max,
+    };
+    lfs_superblock_tole32(&superblock);
+    lfs_dir_commit(&lfs, &mdir, LFS_MKATTRS(
+            {LFS_MKTAG(LFS_TYPE_INLINESTRUCT, 0, sizeof(superblock)),
+                &superblock})) => 0;
+    lfs_unmount(&lfs) => 0;
+
+    // mount should now fail
+    lfs_mount(&lfs, cfg) => LFS_ERR_INVAL;
+'''
+
+# test that we fail to mount after a minor version bump
+[cases.test_compat_minor_incompat]
+in = 'lfs.c'
+code = '''
+    // create a superblock
+    lfs_t lfs;
+    lfs_format(&lfs, cfg) => 0;
+
+    // bump the minor version
+    //
+    // note we're messing around with internals to do this! this
+    // is not a user API
+    lfs_mount(&lfs, cfg) => 0;
+    lfs_mdir_t mdir;
+    lfs_dir_fetch(&lfs, &mdir, (lfs_block_t[2]){0, 1}) => 0;
+    lfs_superblock_t superblock = {
+        .version     = LFS_DISK_VERSION + 0x00000001,
+        .block_size  = lfs.cfg->block_size,
+        .block_count = lfs.cfg->block_count,
+        .name_max    = lfs.name_max,
+        .file_max    = lfs.file_max,
+        .attr_max    = lfs.attr_max,
+    };
+    lfs_superblock_tole32(&superblock);
+    lfs_dir_commit(&lfs, &mdir, LFS_MKATTRS(
+            {LFS_MKTAG(LFS_TYPE_INLINESTRUCT, 0, sizeof(superblock)),
+                &superblock})) => 0;
+    lfs_unmount(&lfs) => 0;
+
+    // mount should now fail
+    lfs_mount(&lfs, cfg) => LFS_ERR_INVAL;
+'''
+
+# test that we correctly bump the minor version
+[cases.test_compat_minor_bump]
+in = 'lfs.c'
+if = '''
+    LFS_DISK_VERSION_MINOR > 0
+        && DISK_VERSION == 0
+'''
+code = '''
+    // create a superblock
+    lfs_t lfs;
+    lfs_format(&lfs, cfg) => 0;
+    lfs_mount(&lfs, cfg) => 0;
+    lfs_file_t file;
+    lfs_file_open(&lfs, &file, "test",
+            LFS_O_WRONLY | LFS_O_CREAT | LFS_O_EXCL) => 0;
+    lfs_file_write(&lfs, &file, "testtest", 8) => 8;
+    lfs_file_close(&lfs, &file) => 0;
+    lfs_unmount(&lfs) => 0;
+
+    // write an old minor version
+    //
+    // note we're messing around with internals to do this! this
+    // is not a user API
+    lfs_mount(&lfs, cfg) => 0;
+    lfs_mdir_t mdir;
+    lfs_dir_fetch(&lfs, &mdir, (lfs_block_t[2]){0, 1}) => 0;
+    lfs_superblock_t superblock = {
+        .version     = LFS_DISK_VERSION - 0x00000001,
+        .block_size  = lfs.cfg->block_size,
+        .block_count = lfs.cfg->block_count,
+        .name_max    = lfs.name_max,
+        .file_max    = lfs.file_max,
+        .attr_max    = lfs.attr_max,
+    };
+    lfs_superblock_tole32(&superblock);
+    lfs_dir_commit(&lfs, &mdir, LFS_MKATTRS(
+            {LFS_MKTAG(LFS_TYPE_INLINESTRUCT, 0, sizeof(superblock)),
+                &superblock})) => 0;
+    lfs_unmount(&lfs) => 0;
+
+    // mount should still work
+    lfs_mount(&lfs, cfg) => 0;
+
+    struct lfs_fsinfo fsinfo;
+    lfs_fs_stat(&lfs, &fsinfo) => 0;
+    assert(fsinfo.disk_version == LFS_DISK_VERSION-1);
+
+    lfs_file_open(&lfs, &file, "test", LFS_O_RDONLY) => 0;
+    uint8_t buffer[8];
+    lfs_file_read(&lfs, &file, buffer, 8) => 8;
+    assert(memcmp(buffer, "testtest", 8) == 0);
+    lfs_file_close(&lfs, &file) => 0;
+
+    // minor version should be unchanged
+    lfs_fs_stat(&lfs, &fsinfo) => 0;
+    assert(fsinfo.disk_version == LFS_DISK_VERSION-1);
+
+    lfs_unmount(&lfs) => 0;
+
+    // if we write, we need to bump the minor version
+    lfs_mount(&lfs, cfg) => 0;
+
+    lfs_fs_stat(&lfs, &fsinfo) => 0;
+    assert(fsinfo.disk_version == LFS_DISK_VERSION-1);
+
+    lfs_file_open(&lfs, &file, "test", LFS_O_WRONLY | LFS_O_TRUNC) => 0;
+    lfs_file_write(&lfs, &file, "teeeeest", 8) => 8;
+    lfs_file_close(&lfs, &file) => 0;
+
+    // minor version should be changed
+    lfs_fs_stat(&lfs, &fsinfo) => 0;
+    assert(fsinfo.disk_version == LFS_DISK_VERSION);
+
+    lfs_unmount(&lfs) => 0;
+
+    // and of course mount should still work
+    lfs_mount(&lfs, cfg) => 0;
+
+    // minor version should have changed
+    lfs_fs_stat(&lfs, &fsinfo) => 0;
+    assert(fsinfo.disk_version == LFS_DISK_VERSION);
+
+    lfs_file_open(&lfs, &file, "test", LFS_O_RDONLY) => 0;
+    lfs_file_read(&lfs, &file, buffer, 8) => 8;
+    assert(memcmp(buffer, "teeeeest", 8) == 0);
+    lfs_file_close(&lfs, &file) => 0;
+
+    // yep, still changed
+    lfs_fs_stat(&lfs, &fsinfo) => 0;
+    assert(fsinfo.disk_version == LFS_DISK_VERSION);
+
+    lfs_unmount(&lfs) => 0;
+'''
diff --git a/tests/test_dirs.toml b/tests/test_dirs.toml
index 270f4f8..4262a1a 100644
--- a/tests/test_dirs.toml
+++ b/tests/test_dirs.toml
@@ -1,8 +1,11 @@
-[[case]] # root
+[cases.test_dirs_root]
 code = '''
-    lfs_format(&lfs, &cfg) => 0;
-    lfs_mount(&lfs, &cfg) => 0;
+    lfs_t lfs;
+    lfs_format(&lfs, cfg) => 0;
+    lfs_mount(&lfs, cfg) => 0;
+    lfs_dir_t dir;
     lfs_dir_open(&lfs, &dir, "/") => 0;
+    struct lfs_info info;
     lfs_dir_read(&lfs, &dir, &info) => 1;
     assert(info.type == LFS_TYPE_DIR);
     assert(strcmp(info.name, ".") == 0);
@@ -14,20 +17,25 @@ code = '''
     lfs_unmount(&lfs) => 0;
 '''
 
-[[case]] # many directory creation
-define.N = 'range(0, 100, 3)'
+[cases.test_dirs_many_creation]
+defines.N = 'range(3, 100, 3)'
+if = 'N < BLOCK_COUNT/2'
 code = '''
-    lfs_format(&lfs, &cfg) => 0;
+    lfs_t lfs;
+    lfs_format(&lfs, cfg) => 0;
 
-    lfs_mount(&lfs, &cfg) => 0;
+    lfs_mount(&lfs, cfg) => 0;
     for (int i = 0; i < N; i++) {
+        char path[1024];
         sprintf(path, "dir%03d", i);
         lfs_mkdir(&lfs, path) => 0;
     }
     lfs_unmount(&lfs) => 0;
 
-    lfs_mount(&lfs, &cfg) => 0;
+    lfs_mount(&lfs, cfg) => 0;
+    lfs_dir_t dir;
     lfs_dir_open(&lfs, &dir, "/") => 0;
+    struct lfs_info info;
     lfs_dir_read(&lfs, &dir, &info) => 1;
     assert(info.type == LFS_TYPE_DIR);
     assert(strcmp(info.name, ".") == 0);
@@ -35,6 +43,7 @@ code = '''
     assert(info.type == LFS_TYPE_DIR);
     assert(strcmp(info.name, "..") == 0);
     for (int i = 0; i < N; i++) {
+        char path[1024];
         sprintf(path, "dir%03d", i);
         lfs_dir_read(&lfs, &dir, &info) => 1;
         assert(info.type == LFS_TYPE_DIR);
@@ -45,20 +54,25 @@ code = '''
     lfs_unmount(&lfs) => 0;
 '''
 
-[[case]] # many directory removal
-define.N = 'range(3, 100, 11)'
+[cases.test_dirs_many_removal]
+defines.N = 'range(3, 100, 11)'
+if = 'N < BLOCK_COUNT/2'
 code = '''
-    lfs_format(&lfs, &cfg) => 0;
+    lfs_t lfs;
+    lfs_format(&lfs, cfg) => 0;
 
-    lfs_mount(&lfs, &cfg) => 0;
+    lfs_mount(&lfs, cfg) => 0;
     for (int i = 0; i < N; i++) {
+        char path[1024];
         sprintf(path, "removeme%03d", i);
         lfs_mkdir(&lfs, path) => 0;
     }
     lfs_unmount(&lfs) => 0;
 
-    lfs_mount(&lfs, &cfg) => 0;
+    lfs_mount(&lfs, cfg) => 0;
+    lfs_dir_t dir;
     lfs_dir_open(&lfs, &dir, "/") => 0;
+    struct lfs_info info;
     lfs_dir_read(&lfs, &dir, &info) => 1;
     assert(info.type == LFS_TYPE_DIR);
     assert(strcmp(info.name, ".") == 0);
@@ -66,6 +80,7 @@ code = '''
     assert(info.type == LFS_TYPE_DIR);
     assert(strcmp(info.name, "..") == 0);
     for (int i = 0; i < N; i++) {
+        char path[1024];
         sprintf(path, "removeme%03d", i);
         lfs_dir_read(&lfs, &dir, &info) => 1;
         assert(info.type == LFS_TYPE_DIR);
@@ -75,14 +90,15 @@ code = '''
     lfs_dir_close(&lfs, &dir) => 0;
     lfs_unmount(&lfs);
 
-    lfs_mount(&lfs, &cfg) => 0;
+    lfs_mount(&lfs, cfg) => 0;
     for (int i = 0; i < N; i++) {
+        char path[1024];
         sprintf(path, "removeme%03d", i);
         lfs_remove(&lfs, path) => 0;
     }
     lfs_unmount(&lfs);
 
-    lfs_mount(&lfs, &cfg) => 0;
+    lfs_mount(&lfs, cfg) => 0;
     lfs_dir_open(&lfs, &dir, "/") => 0;
     lfs_dir_read(&lfs, &dir, &info) => 1;
     assert(info.type == LFS_TYPE_DIR);
@@ -95,20 +111,25 @@ code = '''
     lfs_unmount(&lfs) => 0;
 '''
 
-[[case]] # many directory rename
-define.N = 'range(3, 100, 11)'
+[cases.test_dirs_many_rename]
+defines.N = 'range(3, 100, 11)'
+if = 'N < BLOCK_COUNT/2'
 code = '''
-    lfs_format(&lfs, &cfg) => 0;
+    lfs_t lfs;
+    lfs_format(&lfs, cfg) => 0;
 
-    lfs_mount(&lfs, &cfg) => 0;
+    lfs_mount(&lfs, cfg) => 0;
     for (int i = 0; i < N; i++) {
+        char path[1024];
         sprintf(path, "test%03d", i);
         lfs_mkdir(&lfs, path) => 0;
     }
     lfs_unmount(&lfs) => 0;
 
-    lfs_mount(&lfs, &cfg) => 0;
+    lfs_mount(&lfs, cfg) => 0;
+    lfs_dir_t dir;
     lfs_dir_open(&lfs, &dir, "/") => 0;
+    struct lfs_info info;
     lfs_dir_read(&lfs, &dir, &info) => 1;
     assert(info.type == LFS_TYPE_DIR);
     assert(strcmp(info.name, ".") == 0);
@@ -116,6 +137,7 @@ code = '''
     assert(info.type == LFS_TYPE_DIR);
     assert(strcmp(info.name, "..") == 0);
     for (int i = 0; i < N; i++) {
+        char path[1024];
         sprintf(path, "test%03d", i);
         lfs_dir_read(&lfs, &dir, &info) => 1;
         assert(info.type == LFS_TYPE_DIR);
@@ -125,7 +147,7 @@ code = '''
     lfs_dir_close(&lfs, &dir) => 0;
     lfs_unmount(&lfs);
 
-    lfs_mount(&lfs, &cfg) => 0;
+    lfs_mount(&lfs, cfg) => 0;
     for (int i = 0; i < N; i++) {
         char oldpath[128];
         char newpath[128];
@@ -135,7 +157,7 @@ code = '''
     }
     lfs_unmount(&lfs);
 
-    lfs_mount(&lfs, &cfg) => 0;
+    lfs_mount(&lfs, cfg) => 0;
     lfs_dir_open(&lfs, &dir, "/") => 0;
     lfs_dir_read(&lfs, &dir, &info) => 1;
     assert(info.type == LFS_TYPE_DIR);
@@ -144,6 +166,7 @@ code = '''
     assert(info.type == LFS_TYPE_DIR);
     assert(strcmp(info.name, "..") == 0);
     for (int i = 0; i < N; i++) {
+        char path[1024];
         sprintf(path, "tedd%03d", i);
         lfs_dir_read(&lfs, &dir, &info) => 1;
         assert(info.type == LFS_TYPE_DIR);
@@ -154,29 +177,35 @@ code = '''
     lfs_unmount(&lfs);
 '''
 
-[[case]] # reentrant many directory creation/rename/removal
-define.N = [5, 11]
+[cases.test_dirs_many_reentrant]
+defines.N = [5, 11]
+if = 'BLOCK_COUNT >= 4*N'
 reentrant = true
 code = '''
-    err = lfs_mount(&lfs, &cfg);
+    lfs_t lfs;
+    int err = lfs_mount(&lfs, cfg);
     if (err) {
-        lfs_format(&lfs, &cfg) => 0;
-        lfs_mount(&lfs, &cfg) => 0;
+        lfs_format(&lfs, cfg) => 0;
+        lfs_mount(&lfs, cfg) => 0;
     }
 
     for (int i = 0; i < N; i++) {
+        char path[1024];
         sprintf(path, "hi%03d", i);
         err = lfs_mkdir(&lfs, path);
         assert(err == 0 || err == LFS_ERR_EXIST);
     }
 
     for (int i = 0; i < N; i++) {
+        char path[1024];
         sprintf(path, "hello%03d", i);
         err = lfs_remove(&lfs, path);
         assert(err == 0 || err == LFS_ERR_NOENT);
     }
 
+    lfs_dir_t dir;
     lfs_dir_open(&lfs, &dir, "/") => 0;
+    struct lfs_info info;
     lfs_dir_read(&lfs, &dir, &info) => 1;
     assert(info.type == LFS_TYPE_DIR);
     assert(strcmp(info.name, ".") == 0);
@@ -184,6 +213,7 @@ code = '''
     assert(info.type == LFS_TYPE_DIR);
     assert(strcmp(info.name, "..") == 0);
     for (int i = 0; i < N; i++) {
+        char path[1024];
         sprintf(path, "hi%03d", i);
         lfs_dir_read(&lfs, &dir, &info) => 1;
         assert(info.type == LFS_TYPE_DIR);
@@ -209,6 +239,7 @@ code = '''
     assert(info.type == LFS_TYPE_DIR);
     assert(strcmp(info.name, "..") == 0);
     for (int i = 0; i < N; i++) {
+        char path[1024];
         sprintf(path, "hello%03d", i);
         lfs_dir_read(&lfs, &dir, &info) => 1;
         assert(info.type == LFS_TYPE_DIR);
@@ -218,6 +249,7 @@ code = '''
     lfs_dir_close(&lfs, &dir) => 0;
 
     for (int i = 0; i < N; i++) {
+        char path[1024];
         sprintf(path, "hello%03d", i);
         lfs_remove(&lfs, path) => 0;
     }
@@ -234,22 +266,28 @@ code = '''
     lfs_unmount(&lfs) => 0;
 '''
 
-[[case]] # file creation
-define.N = 'range(3, 100, 11)'
+[cases.test_dirs_file_creation]
+defines.N = 'range(3, 100, 11)'
+if = 'N < BLOCK_COUNT/2'
 code = '''
-    lfs_format(&lfs, &cfg) => 0;
+    lfs_t lfs;
+    lfs_format(&lfs, cfg) => 0;
 
-    lfs_mount(&lfs, &cfg) => 0;
+    lfs_mount(&lfs, cfg) => 0;
     for (int i = 0; i < N; i++) {
+        char path[1024];
         sprintf(path, "file%03d", i);
+        lfs_file_t file;
         lfs_file_open(&lfs, &file, path,
                 LFS_O_WRONLY | LFS_O_CREAT | LFS_O_EXCL) => 0;
         lfs_file_close(&lfs, &file) => 0;
     }
     lfs_unmount(&lfs) => 0;
 
-    lfs_mount(&lfs, &cfg) => 0;
+    lfs_mount(&lfs, cfg) => 0;
+    lfs_dir_t dir;
     lfs_dir_open(&lfs, &dir, "/") => 0;
+    struct lfs_info info;
     lfs_dir_read(&lfs, &dir, &info) => 1;
     assert(info.type == LFS_TYPE_DIR);
     assert(strcmp(info.name, ".") == 0);
@@ -257,6 +295,7 @@ code = '''
     assert(info.type == LFS_TYPE_DIR);
     assert(strcmp(info.name, "..") == 0);
     for (int i = 0; i < N; i++) {
+        char path[1024];
         sprintf(path, "file%03d", i);
         lfs_dir_read(&lfs, &dir, &info) => 1;
         assert(info.type == LFS_TYPE_REG);
@@ -267,22 +306,28 @@ code = '''
     lfs_unmount(&lfs);
 '''
 
-[[case]] # file removal
-define.N = 'range(0, 100, 3)'
+[cases.test_dirs_file_removal]
+defines.N = 'range(3, 100, 11)'
+if = 'N < BLOCK_COUNT/2'
 code = '''
-    lfs_format(&lfs, &cfg) => 0;
+    lfs_t lfs;
+    lfs_format(&lfs, cfg) => 0;
 
-    lfs_mount(&lfs, &cfg) => 0;
+    lfs_mount(&lfs, cfg) => 0;
     for (int i = 0; i < N; i++) {
+        char path[1024];
         sprintf(path, "removeme%03d", i);
+        lfs_file_t file;
         lfs_file_open(&lfs, &file, path,
                 LFS_O_WRONLY | LFS_O_CREAT | LFS_O_EXCL) => 0;
         lfs_file_close(&lfs, &file) => 0;
     }
     lfs_unmount(&lfs) => 0;
 
-    lfs_mount(&lfs, &cfg) => 0;
+    lfs_mount(&lfs, cfg) => 0;
+    lfs_dir_t dir;
     lfs_dir_open(&lfs, &dir, "/") => 0;
+    struct lfs_info info;
     lfs_dir_read(&lfs, &dir, &info) => 1;
     assert(info.type == LFS_TYPE_DIR);
     assert(strcmp(info.name, ".") == 0);
@@ -290,6 +335,7 @@ code = '''
     assert(info.type == LFS_TYPE_DIR);
     assert(strcmp(info.name, "..") == 0);
     for (int i = 0; i < N; i++) {
+        char path[1024];
         sprintf(path, "removeme%03d", i);
         lfs_dir_read(&lfs, &dir, &info) => 1;
         assert(info.type == LFS_TYPE_REG);
@@ -299,14 +345,15 @@ code = '''
     lfs_dir_close(&lfs, &dir) => 0;
     lfs_unmount(&lfs);
 
-    lfs_mount(&lfs, &cfg) => 0;
+    lfs_mount(&lfs, cfg) => 0;
     for (int i = 0; i < N; i++) {
+        char path[1024];
         sprintf(path, "removeme%03d", i);
         lfs_remove(&lfs, path) => 0;
     }
     lfs_unmount(&lfs);
 
-    lfs_mount(&lfs, &cfg) => 0;
+    lfs_mount(&lfs, cfg) => 0;
     lfs_dir_open(&lfs, &dir, "/") => 0;
     lfs_dir_read(&lfs, &dir, &info) => 1;
     assert(info.type == LFS_TYPE_DIR);
@@ -319,22 +366,28 @@ code = '''
     lfs_unmount(&lfs) => 0;
 '''
 
-[[case]] # file rename
-define.N = 'range(0, 100, 3)'
+[cases.test_dirs_file_rename]
+defines.N = 'range(3, 100, 11)'
+if = 'N < BLOCK_COUNT/2'
 code = '''
-    lfs_format(&lfs, &cfg) => 0;
+    lfs_t lfs;
+    lfs_format(&lfs, cfg) => 0;
 
-    lfs_mount(&lfs, &cfg) => 0;
+    lfs_mount(&lfs, cfg) => 0;
     for (int i = 0; i < N; i++) {
+        char path[1024];
         sprintf(path, "test%03d", i);
+        lfs_file_t file;
         lfs_file_open(&lfs, &file, path,
                 LFS_O_WRONLY | LFS_O_CREAT | LFS_O_EXCL) => 0;
         lfs_file_close(&lfs, &file) => 0;
     }
     lfs_unmount(&lfs) => 0;
 
-    lfs_mount(&lfs, &cfg) => 0;
+    lfs_mount(&lfs, cfg) => 0;
+    lfs_dir_t dir;
     lfs_dir_open(&lfs, &dir, "/") => 0;
+    struct lfs_info info;
     lfs_dir_read(&lfs, &dir, &info) => 1;
     assert(info.type == LFS_TYPE_DIR);
     assert(strcmp(info.name, ".") == 0);
@@ -342,6 +395,7 @@ code = '''
     assert(info.type == LFS_TYPE_DIR);
     assert(strcmp(info.name, "..") == 0);
     for (int i = 0; i < N; i++) {
+        char path[1024];
         sprintf(path, "test%03d", i);
         lfs_dir_read(&lfs, &dir, &info) => 1;
         assert(info.type == LFS_TYPE_REG);
@@ -351,7 +405,7 @@ code = '''
     lfs_dir_close(&lfs, &dir) => 0;
     lfs_unmount(&lfs);
 
-    lfs_mount(&lfs, &cfg) => 0;
+    lfs_mount(&lfs, cfg) => 0;
     for (int i = 0; i < N; i++) {
         char oldpath[128];
         char newpath[128];
@@ -361,7 +415,7 @@ code = '''
     }
     lfs_unmount(&lfs);
 
-    lfs_mount(&lfs, &cfg) => 0;
+    lfs_mount(&lfs, cfg) => 0;
     lfs_dir_open(&lfs, &dir, "/") => 0;
     lfs_dir_read(&lfs, &dir, &info) => 1;
     assert(info.type == LFS_TYPE_DIR);
@@ -370,6 +424,7 @@ code = '''
     assert(info.type == LFS_TYPE_DIR);
     assert(strcmp(info.name, "..") == 0);
     for (int i = 0; i < N; i++) {
+        char path[1024];
         sprintf(path, "tedd%03d", i);
         lfs_dir_read(&lfs, &dir, &info) => 1;
         assert(info.type == LFS_TYPE_REG);
@@ -380,29 +435,36 @@ code = '''
     lfs_unmount(&lfs);
 '''
 
-[[case]] # reentrant file creation/rename/removal
-define.N = [5, 25]
+[cases.test_dirs_file_reentrant]
+defines.N = [5, 25]
+if = 'N < BLOCK_COUNT/2'
 reentrant = true
 code = '''
-    err = lfs_mount(&lfs, &cfg);
+    lfs_t lfs;
+    int err = lfs_mount(&lfs, cfg);
     if (err) {
-        lfs_format(&lfs, &cfg) => 0;
-        lfs_mount(&lfs, &cfg) => 0;
+        lfs_format(&lfs, cfg) => 0;
+        lfs_mount(&lfs, cfg) => 0;
     }
 
     for (int i = 0; i < N; i++) {
+        char path[1024];
         sprintf(path, "hi%03d", i);
+        lfs_file_t file;
         lfs_file_open(&lfs, &file, path, LFS_O_CREAT | LFS_O_WRONLY) => 0;
         lfs_file_close(&lfs, &file) => 0;
     }
 
     for (int i = 0; i < N; i++) {
+        char path[1024];
         sprintf(path, "hello%03d", i);
         err = lfs_remove(&lfs, path);
         assert(err == 0 || err == LFS_ERR_NOENT);
     }
 
+    lfs_dir_t dir;
     lfs_dir_open(&lfs, &dir, "/") => 0;
+    struct lfs_info info;
     lfs_dir_read(&lfs, &dir, &info) => 1;
     assert(info.type == LFS_TYPE_DIR);
     assert(strcmp(info.name, ".") == 0);
@@ -410,6 +472,7 @@ code = '''
     assert(info.type == LFS_TYPE_DIR);
     assert(strcmp(info.name, "..") == 0);
     for (int i = 0; i < N; i++) {
+        char path[1024];
         sprintf(path, "hi%03d", i);
         lfs_dir_read(&lfs, &dir, &info) => 1;
         assert(info.type == LFS_TYPE_REG);
@@ -435,6 +498,7 @@ code = '''
     assert(info.type == LFS_TYPE_DIR);
     assert(strcmp(info.name, "..") == 0);
     for (int i = 0; i < N; i++) {
+        char path[1024];
         sprintf(path, "hello%03d", i);
         lfs_dir_read(&lfs, &dir, &info) => 1;
         assert(info.type == LFS_TYPE_REG);
@@ -444,6 +508,7 @@ code = '''
     lfs_dir_close(&lfs, &dir) => 0;
 
     for (int i = 0; i < N; i++) {
+        char path[1024];
         sprintf(path, "hello%03d", i);
         lfs_remove(&lfs, path) => 0;
     }
@@ -460,24 +525,28 @@ code = '''
     lfs_unmount(&lfs) => 0;
 '''
 
-[[case]] # nested directories
+[cases.test_dirs_nested]
 code = '''
-    lfs_format(&lfs, &cfg) => 0;
-    lfs_mount(&lfs, &cfg) => 0;
+    lfs_t lfs;
+    lfs_format(&lfs, cfg) => 0;
+    lfs_mount(&lfs, cfg) => 0;
     lfs_mkdir(&lfs, "potato") => 0;
+    lfs_file_t file;
     lfs_file_open(&lfs, &file, "burito",
             LFS_O_WRONLY | LFS_O_CREAT | LFS_O_EXCL) => 0;
     lfs_file_close(&lfs, &file) => 0;
     lfs_unmount(&lfs) => 0;
 
-    lfs_mount(&lfs, &cfg) => 0;
+    lfs_mount(&lfs, cfg) => 0;
     lfs_mkdir(&lfs, "potato/baked") => 0;
     lfs_mkdir(&lfs, "potato/sweet") => 0;
     lfs_mkdir(&lfs, "potato/fried") => 0;
     lfs_unmount(&lfs) => 0;
 
-    lfs_mount(&lfs, &cfg) => 0;
+    lfs_mount(&lfs, cfg) => 0;
+    lfs_dir_t dir;
     lfs_dir_open(&lfs, &dir, "potato") => 0;
+    struct lfs_info info;
     lfs_dir_read(&lfs, &dir, &info) => 1;
     assert(strcmp(info.name, ".") == 0);
     info.type => LFS_TYPE_DIR;
@@ -498,21 +567,21 @@ code = '''
     lfs_unmount(&lfs) => 0;
 
     // try removing?
-    lfs_mount(&lfs, &cfg) => 0;
+    lfs_mount(&lfs, cfg) => 0;
     lfs_remove(&lfs, "potato") => LFS_ERR_NOTEMPTY;
     lfs_unmount(&lfs) => 0;
 
     // try renaming?
-    lfs_mount(&lfs, &cfg) => 0;
+    lfs_mount(&lfs, cfg) => 0;
     lfs_rename(&lfs, "potato", "coldpotato") => 0;
     lfs_unmount(&lfs) => 0;
 
-    lfs_mount(&lfs, &cfg) => 0;
+    lfs_mount(&lfs, cfg) => 0;
     lfs_rename(&lfs, "coldpotato", "warmpotato") => 0;
     lfs_rename(&lfs, "warmpotato", "hotpotato") => 0;
     lfs_unmount(&lfs) => 0;
 
-    lfs_mount(&lfs, &cfg) => 0;
+    lfs_mount(&lfs, cfg) => 0;
     lfs_remove(&lfs, "potato") => LFS_ERR_NOENT;
     lfs_remove(&lfs, "coldpotato") => LFS_ERR_NOENT;
     lfs_remove(&lfs, "warmpotato") => LFS_ERR_NOENT;
@@ -520,7 +589,7 @@ code = '''
     lfs_unmount(&lfs) => 0;
 
     // try cross-directory renaming
-    lfs_mount(&lfs, &cfg) => 0;
+    lfs_mount(&lfs, cfg) => 0;
     lfs_mkdir(&lfs, "coldpotato") => 0;
     lfs_rename(&lfs, "hotpotato/baked", "coldpotato/baked") => 0;
     lfs_rename(&lfs, "coldpotato", "hotpotato") => LFS_ERR_NOTEMPTY;
@@ -536,7 +605,7 @@ code = '''
     lfs_remove(&lfs, "hotpotato") => LFS_ERR_NOTEMPTY;
     lfs_unmount(&lfs) => 0;
 
-    lfs_mount(&lfs, &cfg) => 0;
+    lfs_mount(&lfs, cfg) => 0;
     lfs_dir_open(&lfs, &dir, "hotpotato") => 0;
     lfs_dir_read(&lfs, &dir, &info) => 1;
     assert(strcmp(info.name, ".") == 0);
@@ -558,7 +627,7 @@ code = '''
     lfs_unmount(&lfs) => 0;
     
     // final remove
-    lfs_mount(&lfs, &cfg) => 0;
+    lfs_mount(&lfs, cfg) => 0;
     lfs_remove(&lfs, "hotpotato") => LFS_ERR_NOTEMPTY;
     lfs_remove(&lfs, "hotpotato/baked") => 0;
     lfs_remove(&lfs, "hotpotato") => LFS_ERR_NOTEMPTY;
@@ -568,7 +637,7 @@ code = '''
     lfs_remove(&lfs, "hotpotato") => 0;
     lfs_unmount(&lfs) => 0;
 
-    lfs_mount(&lfs, &cfg) => 0;
+    lfs_mount(&lfs, cfg) => 0;
     lfs_dir_open(&lfs, &dir, "/") => 0;
     lfs_dir_read(&lfs, &dir, &info) => 1;
     assert(strcmp(info.name, ".") == 0);
@@ -584,17 +653,22 @@ code = '''
     lfs_unmount(&lfs) => 0;
 '''
 
-[[case]] # recursive remove
-define.N = [10, 100]
+[cases.test_dirs_recursive_remove]
+defines.N = [10, 100]
+if = 'N < BLOCK_COUNT/2'
 code = '''
-    lfs_format(&lfs, &cfg) => 0;
-    lfs_mount(&lfs, &cfg) => 0;
+    lfs_t lfs;
+    lfs_format(&lfs, cfg) => 0;
+    lfs_mount(&lfs, cfg) => 0;
     lfs_mkdir(&lfs, "prickly-pear") => 0;
     for (int i = 0; i < N; i++) {
+        char path[1024];
         sprintf(path, "prickly-pear/cactus%03d", i);
         lfs_mkdir(&lfs, path) => 0;
     }
+    lfs_dir_t dir;
     lfs_dir_open(&lfs, &dir, "prickly-pear") => 0;
+    struct lfs_info info;
     lfs_dir_read(&lfs, &dir, &info) => 1;
     assert(info.type == LFS_TYPE_DIR);
     assert(strcmp(info.name, ".") == 0);
@@ -602,6 +676,7 @@ code = '''
     assert(info.type == LFS_TYPE_DIR);
     assert(strcmp(info.name, "..") == 0);
     for (int i = 0; i < N; i++) {
+        char path[1024];
         sprintf(path, "cactus%03d", i);
         lfs_dir_read(&lfs, &dir, &info) => 1;
         assert(info.type == LFS_TYPE_DIR);
@@ -611,7 +686,7 @@ code = '''
     lfs_dir_close(&lfs, &dir) => 0;
     lfs_unmount(&lfs);
 
-    lfs_mount(&lfs, &cfg) => 0;
+    lfs_mount(&lfs, cfg) => 0;
     lfs_remove(&lfs, "prickly-pear") => LFS_ERR_NOTEMPTY;
 
     lfs_dir_open(&lfs, &dir, "prickly-pear") => 0;
@@ -622,6 +697,7 @@ code = '''
     assert(info.type == LFS_TYPE_DIR);
     assert(strcmp(info.name, "..") == 0);
     for (int i = 0; i < N; i++) {
+        char path[1024];
         sprintf(path, "cactus%03d", i);
         lfs_dir_read(&lfs, &dir, &info) => 1;
         assert(info.type == LFS_TYPE_DIR);
@@ -636,22 +712,24 @@ code = '''
     lfs_remove(&lfs, "prickly-pear") => LFS_ERR_NOENT;
     lfs_unmount(&lfs) => 0;
 
-    lfs_mount(&lfs, &cfg) => 0;
+    lfs_mount(&lfs, cfg) => 0;
     lfs_remove(&lfs, "prickly-pear") => LFS_ERR_NOENT;
     lfs_unmount(&lfs) => 0;
 '''
 
-[[case]] # other error cases
+[cases.test_dirs_other_errors]
 code = '''
-    lfs_format(&lfs, &cfg) => 0;
-    lfs_mount(&lfs, &cfg) => 0;
+    lfs_t lfs;
+    lfs_format(&lfs, cfg) => 0;
+    lfs_mount(&lfs, cfg) => 0;
     lfs_mkdir(&lfs, "potato") => 0;
+    lfs_file_t file;
     lfs_file_open(&lfs, &file, "burito",
             LFS_O_WRONLY | LFS_O_CREAT | LFS_O_EXCL) => 0;
     lfs_file_close(&lfs, &file) => 0;
     lfs_unmount(&lfs) => 0;
 
-    lfs_mount(&lfs, &cfg) => 0;
+    lfs_mount(&lfs, cfg) => 0;
 
     lfs_mkdir(&lfs, "potato") => LFS_ERR_EXIST;
     lfs_mkdir(&lfs, "burito") => LFS_ERR_EXIST;
@@ -659,6 +737,7 @@ code = '''
             LFS_O_WRONLY | LFS_O_CREAT | LFS_O_EXCL) => LFS_ERR_EXIST;
     lfs_file_open(&lfs, &file, "potato",
             LFS_O_WRONLY | LFS_O_CREAT | LFS_O_EXCL) => LFS_ERR_EXIST;
+    lfs_dir_t dir;
     lfs_dir_open(&lfs, &dir, "tomato") => LFS_ERR_NOENT;
     lfs_dir_open(&lfs, &dir, "burito") => LFS_ERR_NOTDIR;
     lfs_file_open(&lfs, &file, "tomato", LFS_O_RDONLY) => LFS_ERR_NOENT;
@@ -678,6 +757,7 @@ code = '''
 
     // check that errors did not corrupt directory
     lfs_dir_open(&lfs, &dir, "/") => 0;
+    struct lfs_info info;
     lfs_dir_read(&lfs, &dir, &info) => 1;
     assert(info.type == LFS_TYPE_DIR);
     assert(strcmp(info.name, ".") == 0);
@@ -696,7 +776,7 @@ code = '''
     lfs_unmount(&lfs) => 0;
 
     // or on disk
-    lfs_mount(&lfs, &cfg) => 0;
+    lfs_mount(&lfs, cfg) => 0;
     lfs_dir_open(&lfs, &dir, "/") => 0;
     lfs_dir_read(&lfs, &dir, &info) => 1;
     assert(info.type == LFS_TYPE_DIR);
@@ -715,21 +795,27 @@ code = '''
     lfs_unmount(&lfs) => 0;
 '''
 
-[[case]] # directory seek
-define.COUNT = [4, 128, 132]
+[cases.test_dirs_seek]
+defines.COUNT = [4, 128, 132]
+if = 'COUNT < BLOCK_COUNT/2'
 code = '''
-    lfs_format(&lfs, &cfg) => 0;
-    lfs_mount(&lfs, &cfg) => 0;
+    lfs_t lfs;
+    lfs_format(&lfs, cfg) => 0;
+    lfs_mount(&lfs, cfg) => 0;
     lfs_mkdir(&lfs, "hello") => 0;
     for (int i = 0; i < COUNT; i++) {
+        char path[1024];
         sprintf(path, "hello/kitty%03d", i);
         lfs_mkdir(&lfs, path) => 0;
     }
     lfs_unmount(&lfs) => 0;
 
-    for (int j = 2; j < COUNT; j++) {
-        lfs_mount(&lfs, &cfg) => 0;
+    // try seeking to each dir entry
+    for (int j = 0; j < COUNT; j++) {
+        lfs_mount(&lfs, cfg) => 0;
+        lfs_dir_t dir;
         lfs_dir_open(&lfs, &dir, "hello") => 0;
+        struct lfs_info info;
         lfs_dir_read(&lfs, &dir, &info) => 1;
         assert(strcmp(info.name, ".") == 0);
         assert(info.type == LFS_TYPE_DIR);
@@ -737,24 +823,25 @@ code = '''
         assert(strcmp(info.name, "..") == 0);
         assert(info.type == LFS_TYPE_DIR);
 
-        lfs_soff_t pos;
         for (int i = 0; i < j; i++) {
+            char path[1024];
             sprintf(path, "kitty%03d", i);
             lfs_dir_read(&lfs, &dir, &info) => 1;
             assert(strcmp(info.name, path) == 0);
             assert(info.type == LFS_TYPE_DIR);
-            pos = lfs_dir_tell(&lfs, &dir);
-            assert(pos >= 0);
         }
+        lfs_soff_t pos = lfs_dir_tell(&lfs, &dir);
+        assert(pos >= 0);
 
         lfs_dir_seek(&lfs, &dir, pos) => 0;
+        char path[1024];
         sprintf(path, "kitty%03d", j);
         lfs_dir_read(&lfs, &dir, &info) => 1;
         assert(strcmp(info.name, path) == 0);
         assert(info.type == LFS_TYPE_DIR);
 
         lfs_dir_rewind(&lfs, &dir) => 0;
-        sprintf(path, "kitty%03d", 0);
+        sprintf(path, "kitty%03u", 0);
         lfs_dir_read(&lfs, &dir, &info) => 1;
         assert(strcmp(info.name, ".") == 0);
         assert(info.type == LFS_TYPE_DIR);
@@ -774,22 +861,73 @@ code = '''
         lfs_dir_close(&lfs, &dir) => 0;
         lfs_unmount(&lfs) => 0;
     }
+
+    // try seeking to end of dir
+    lfs_mount(&lfs, cfg) => 0;
+    lfs_dir_t dir;
+    lfs_dir_open(&lfs, &dir, "hello") => 0;
+    struct lfs_info info;
+    lfs_dir_read(&lfs, &dir, &info) => 1;
+    assert(strcmp(info.name, ".") == 0);
+    assert(info.type == LFS_TYPE_DIR);
+    lfs_dir_read(&lfs, &dir, &info) => 1;
+    assert(strcmp(info.name, "..") == 0);
+    assert(info.type == LFS_TYPE_DIR);
+
+    for (int i = 0; i < COUNT; i++) {
+        char path[1024];
+        sprintf(path, "kitty%03d", i);
+        lfs_dir_read(&lfs, &dir, &info) => 1;
+        assert(strcmp(info.name, path) == 0);
+        assert(info.type == LFS_TYPE_DIR);
+    }
+    lfs_soff_t pos = lfs_dir_tell(&lfs, &dir);
+    assert(pos >= 0);
+
+    lfs_dir_read(&lfs, &dir, &info) => 0;
+
+    lfs_dir_seek(&lfs, &dir, pos) => 0;
+    lfs_dir_read(&lfs, &dir, &info) => 0;
+
+    lfs_dir_rewind(&lfs, &dir) => 0;
+    char path[1024];
+    sprintf(path, "kitty%03d", 0);
+    lfs_dir_read(&lfs, &dir, &info) => 1;
+    assert(strcmp(info.name, ".") == 0);
+    assert(info.type == LFS_TYPE_DIR);
+    lfs_dir_read(&lfs, &dir, &info) => 1;
+    assert(strcmp(info.name, "..") == 0);
+    assert(info.type == LFS_TYPE_DIR);
+    lfs_dir_read(&lfs, &dir, &info) => 1;
+    assert(strcmp(info.name, path) == 0);
+    assert(info.type == LFS_TYPE_DIR);
+
+    lfs_dir_seek(&lfs, &dir, pos) => 0;
+    lfs_dir_read(&lfs, &dir, &info) => 0;
+
+    lfs_dir_close(&lfs, &dir) => 0;
+    lfs_unmount(&lfs) => 0;
 '''
 
-[[case]] # root seek
-define.COUNT = [4, 128, 132]
+[cases.test_dirs_toot_seek]
+defines.COUNT = [4, 128, 132]
+if = 'COUNT < BLOCK_COUNT/2'
 code = '''
-    lfs_format(&lfs, &cfg) => 0;
-    lfs_mount(&lfs, &cfg) => 0;
+    lfs_t lfs;
+    lfs_format(&lfs, cfg) => 0;
+    lfs_mount(&lfs, cfg) => 0;
     for (int i = 0; i < COUNT; i++) {
+        char path[1024];
         sprintf(path, "hi%03d", i);
         lfs_mkdir(&lfs, path) => 0;
     }
     lfs_unmount(&lfs) => 0;
 
-    for (int j = 2; j < COUNT; j++) {
-        lfs_mount(&lfs, &cfg) => 0;
+    for (int j = 0; j < COUNT; j++) {
+        lfs_mount(&lfs, cfg) => 0;
+        lfs_dir_t dir;
         lfs_dir_open(&lfs, &dir, "/") => 0;
+        struct lfs_info info;
         lfs_dir_read(&lfs, &dir, &info) => 1;
         assert(strcmp(info.name, ".") == 0);
         assert(info.type == LFS_TYPE_DIR);
@@ -797,24 +935,25 @@ code = '''
         assert(strcmp(info.name, "..") == 0);
         assert(info.type == LFS_TYPE_DIR);
 
-        lfs_soff_t pos;
         for (int i = 0; i < j; i++) {
+            char path[1024];
             sprintf(path, "hi%03d", i);
             lfs_dir_read(&lfs, &dir, &info) => 1;
             assert(strcmp(info.name, path) == 0);
             assert(info.type == LFS_TYPE_DIR);
-            pos = lfs_dir_tell(&lfs, &dir);
-            assert(pos >= 0);
         }
+        lfs_soff_t pos = lfs_dir_tell(&lfs, &dir);
+        assert(pos >= 0);
 
         lfs_dir_seek(&lfs, &dir, pos) => 0;
+        char path[1024];
         sprintf(path, "hi%03d", j);
         lfs_dir_read(&lfs, &dir, &info) => 1;
         assert(strcmp(info.name, path) == 0);
         assert(info.type == LFS_TYPE_DIR);
 
         lfs_dir_rewind(&lfs, &dir) => 0;
-        sprintf(path, "hi%03d", 0);
+        sprintf(path, "hi%03u", 0);
         lfs_dir_read(&lfs, &dir, &info) => 1;
         assert(strcmp(info.name, ".") == 0);
         assert(info.type == LFS_TYPE_DIR);
@@ -834,5 +973,51 @@ code = '''
         lfs_dir_close(&lfs, &dir) => 0;
         lfs_unmount(&lfs) => 0;
     }
+
+    // try seeking to end of dir
+    lfs_mount(&lfs, cfg) => 0;
+    lfs_dir_t dir;
+    lfs_dir_open(&lfs, &dir, "/") => 0;
+    struct lfs_info info;
+    lfs_dir_read(&lfs, &dir, &info) => 1;
+    assert(strcmp(info.name, ".") == 0);
+    assert(info.type == LFS_TYPE_DIR);
+    lfs_dir_read(&lfs, &dir, &info) => 1;
+    assert(strcmp(info.name, "..") == 0);
+    assert(info.type == LFS_TYPE_DIR);
+
+    for (int i = 0; i < COUNT; i++) {
+        char path[1024];
+        sprintf(path, "hi%03d", i);
+        lfs_dir_read(&lfs, &dir, &info) => 1;
+        assert(strcmp(info.name, path) == 0);
+        assert(info.type == LFS_TYPE_DIR);
+    }
+    lfs_soff_t pos = lfs_dir_tell(&lfs, &dir);
+    assert(pos >= 0);
+
+    lfs_dir_read(&lfs, &dir, &info) => 0;
+
+    lfs_dir_seek(&lfs, &dir, pos) => 0;
+    lfs_dir_read(&lfs, &dir, &info) => 0;
+
+    lfs_dir_rewind(&lfs, &dir) => 0;
+    char path[1024];
+    sprintf(path, "hi%03d", 0);
+    lfs_dir_read(&lfs, &dir, &info) => 1;
+    assert(strcmp(info.name, ".") == 0);
+    assert(info.type == LFS_TYPE_DIR);
+    lfs_dir_read(&lfs, &dir, &info) => 1;
+    assert(strcmp(info.name, "..") == 0);
+    assert(info.type == LFS_TYPE_DIR);
+    lfs_dir_read(&lfs, &dir, &info) => 1;
+    assert(strcmp(info.name, path) == 0);
+    assert(info.type == LFS_TYPE_DIR);
+
+    lfs_dir_seek(&lfs, &dir, pos) => 0;
+    lfs_dir_read(&lfs, &dir, &info) => 0;
+
+    lfs_dir_close(&lfs, &dir) => 0;
+    lfs_unmount(&lfs) => 0;
 '''
 
diff --git a/tests/test_entries.toml b/tests/test_entries.toml
index 81e175f..7aa551e 100644
--- a/tests/test_entries.toml
+++ b/tests/test_entries.toml
@@ -2,19 +2,23 @@
 # Note that these tests are intended for 512 byte inline sizes. They should
 # still pass with other inline sizes but wouldn't be testing anything.
 
-define.LFS_CACHE_SIZE = 512
-if = 'LFS_CACHE_SIZE % LFS_PROG_SIZE == 0 && LFS_CACHE_SIZE == 512'
+defines.CACHE_SIZE = 512
+if = 'CACHE_SIZE % PROG_SIZE == 0 && CACHE_SIZE == 512'
 
-[[case]] # entry grow test
+[cases.test_entries_grow]
 code = '''
     uint8_t wbuffer[1024];
     uint8_t rbuffer[1024];
 
-    lfs_format(&lfs, &cfg) => 0;
-    lfs_mount(&lfs, &cfg) => 0;
+    lfs_t lfs;
+    lfs_format(&lfs, cfg) => 0;
+    lfs_mount(&lfs, cfg) => 0;
 
     // write hi0 20
+    char path[1024];
+    lfs_size_t size;
     sprintf(path, "hi0"); size = 20;
+    lfs_file_t file;
     lfs_file_open(&lfs, &file, path,
             LFS_O_WRONLY | LFS_O_CREAT | LFS_O_TRUNC) => 0;
     memset(wbuffer, 'c', size);
@@ -94,16 +98,20 @@ code = '''
     lfs_unmount(&lfs) => 0;
 '''
 
-[[case]] # entry shrink test
+[cases.test_entries_shrink]
 code = '''
     uint8_t wbuffer[1024];
     uint8_t rbuffer[1024];
 
-    lfs_format(&lfs, &cfg) => 0;
-    lfs_mount(&lfs, &cfg) => 0;
+    lfs_t lfs;
+    lfs_format(&lfs, cfg) => 0;
+    lfs_mount(&lfs, cfg) => 0;
 
     // write hi0 20
+    char path[1024];
+    lfs_size_t size;
     sprintf(path, "hi0"); size = 20;
+    lfs_file_t file;
     lfs_file_open(&lfs, &file, path,
             LFS_O_WRONLY | LFS_O_CREAT | LFS_O_TRUNC) => 0;
     memset(wbuffer, 'c', size);
@@ -183,16 +191,20 @@ code = '''
     lfs_unmount(&lfs) => 0;
 '''
 
-[[case]] # entry spill test
+[cases.test_entries_spill]
 code = '''
     uint8_t wbuffer[1024];
     uint8_t rbuffer[1024];
 
-    lfs_format(&lfs, &cfg) => 0;
-    lfs_mount(&lfs, &cfg) => 0;
+    lfs_t lfs;
+    lfs_format(&lfs, cfg) => 0;
+    lfs_mount(&lfs, cfg) => 0;
 
     // write hi0 200
+    char path[1024];
+    lfs_size_t size;
     sprintf(path, "hi0"); size = 200;
+    lfs_file_t file;
     lfs_file_open(&lfs, &file, path,
             LFS_O_WRONLY | LFS_O_CREAT | LFS_O_TRUNC) => 0;
     memset(wbuffer, 'c', size);
@@ -256,16 +268,20 @@ code = '''
     lfs_unmount(&lfs) => 0;
 '''
 
-[[case]] # entry push spill test
+[cases.test_entries_push_spill]
 code = '''
     uint8_t wbuffer[1024];
     uint8_t rbuffer[1024];
 
-    lfs_format(&lfs, &cfg) => 0;
-    lfs_mount(&lfs, &cfg) => 0;
+    lfs_t lfs;
+    lfs_format(&lfs, cfg) => 0;
+    lfs_mount(&lfs, cfg) => 0;
 
     // write hi0 200
+    char path[1024];
+    lfs_size_t size;
     sprintf(path, "hi0"); size = 200;
+    lfs_file_t file;
     lfs_file_open(&lfs, &file, path,
             LFS_O_WRONLY | LFS_O_CREAT | LFS_O_TRUNC) => 0;
     memset(wbuffer, 'c', size);
@@ -345,16 +361,20 @@ code = '''
     lfs_unmount(&lfs) => 0;
 '''
 
-[[case]] # entry push spill two test
+[cases.test_entries_push_spill_two]
 code = '''
     uint8_t wbuffer[1024];
     uint8_t rbuffer[1024];
 
-    lfs_format(&lfs, &cfg) => 0;
-    lfs_mount(&lfs, &cfg) => 0;
+    lfs_t lfs;
+    lfs_format(&lfs, cfg) => 0;
+    lfs_mount(&lfs, cfg) => 0;
 
     // write hi0 200
+    char path[1024];
+    lfs_size_t size;
     sprintf(path, "hi0"); size = 200;
+    lfs_file_t file;
     lfs_file_open(&lfs, &file, path,
             LFS_O_WRONLY | LFS_O_CREAT | LFS_O_TRUNC) => 0;
     memset(wbuffer, 'c', size);
@@ -449,16 +469,20 @@ code = '''
     lfs_unmount(&lfs) => 0;
 '''
 
-[[case]] # entry drop test
+[cases.test_entries_drop]
 code = '''
     uint8_t wbuffer[1024];
     uint8_t rbuffer[1024];
 
-    lfs_format(&lfs, &cfg) => 0;
-    lfs_mount(&lfs, &cfg) => 0;
+    lfs_t lfs;
+    lfs_format(&lfs, cfg) => 0;
+    lfs_mount(&lfs, cfg) => 0;
 
     // write hi0 200
+    char path[1024];
+    lfs_size_t size;
     sprintf(path, "hi0"); size = 200;
+    lfs_file_t file;
     lfs_file_open(&lfs, &file, path,
             LFS_O_WRONLY | LFS_O_CREAT | LFS_O_TRUNC) => 0;
     memset(wbuffer, 'c', size);
@@ -491,6 +515,7 @@ code = '''
     lfs_file_close(&lfs, &file) => 0;
 
     lfs_remove(&lfs, "hi1") => 0;
+    struct lfs_info info;
     lfs_stat(&lfs, "hi1", &info) => LFS_ERR_NOENT;
     // read hi0 200
     sprintf(path, "hi0"); size = 200;
@@ -547,15 +572,18 @@ code = '''
     lfs_unmount(&lfs) => 0;
 '''
 
-[[case]] # create too big
+[cases.test_entries_create_too_big]
 code = '''
-    lfs_format(&lfs, &cfg) => 0;
+    lfs_t lfs;
+    lfs_format(&lfs, cfg) => 0;
 
-    lfs_mount(&lfs, &cfg) => 0;
+    lfs_mount(&lfs, cfg) => 0;
+    char path[1024];
     memset(path, 'm', 200);
     path[200] = '\0';
 
-    size = 400;
+    lfs_size_t size = 400;
+    lfs_file_t file;
     lfs_file_open(&lfs, &file, path,
             LFS_O_WRONLY | LFS_O_CREAT | LFS_O_TRUNC) => 0;
     uint8_t wbuffer[1024];
@@ -572,15 +600,18 @@ code = '''
     lfs_unmount(&lfs) => 0;
 '''
 
-[[case]] # resize too big
+[cases.test_entries_resize_too_big]
 code = '''
-    lfs_format(&lfs, &cfg) => 0;
+    lfs_t lfs;
+    lfs_format(&lfs, cfg) => 0;
 
-    lfs_mount(&lfs, &cfg) => 0;
+    lfs_mount(&lfs, cfg) => 0;
+    char path[1024];
     memset(path, 'm', 200);
     path[200] = '\0';
 
-    size = 40;
+    lfs_size_t size = 40;
+    lfs_file_t file;
     lfs_file_open(&lfs, &file, path,
             LFS_O_WRONLY | LFS_O_CREAT | LFS_O_TRUNC) => 0;
     uint8_t wbuffer[1024];
diff --git a/tests/test_evil.toml b/tests/test_evil.toml
index 920d3a0..4acd5ef 100644
--- a/tests/test_evil.toml
+++ b/tests/test_evil.toml
@@ -3,16 +3,17 @@
 
 # invalid pointer tests (outside of block_count)
 
-[[case]] # invalid tail-pointer test
-define.TAIL_TYPE = ['LFS_TYPE_HARDTAIL', 'LFS_TYPE_SOFTTAIL']
-define.INVALSET = [0x3, 0x1, 0x2]
+[cases.test_evil_invalid_tail_pointer]
+defines.TAIL_TYPE = ['LFS_TYPE_HARDTAIL', 'LFS_TYPE_SOFTTAIL']
+defines.INVALSET = [0x3, 0x1, 0x2]
 in = "lfs.c"
 code = '''
     // create littlefs
-    lfs_format(&lfs, &cfg) => 0;
+    lfs_t lfs;
+    lfs_format(&lfs, cfg) => 0;
 
     // change tail-pointer to invalid pointers
-    lfs_init(&lfs, &cfg) => 0;
+    lfs_init(&lfs, cfg) => 0;
     lfs_mdir_t mdir;
     lfs_dir_fetch(&lfs, &mdir, (lfs_block_t[2]){0, 1}) => 0;
     lfs_dir_commit(&lfs, &mdir, LFS_MKATTRS(
@@ -23,25 +24,27 @@ code = '''
     lfs_deinit(&lfs) => 0;
 
     // test that mount fails gracefully
-    lfs_mount(&lfs, &cfg) => LFS_ERR_CORRUPT;
+    lfs_mount(&lfs, cfg) => LFS_ERR_CORRUPT;
 '''
 
-[[case]] # invalid dir pointer test
-define.INVALSET = [0x3, 0x1, 0x2]
+[cases.test_evil_invalid_dir_pointer]
+defines.INVALSET = [0x3, 0x1, 0x2]
 in = "lfs.c"
 code = '''
     // create littlefs
-    lfs_format(&lfs, &cfg) => 0;
+    lfs_t lfs;
+    lfs_format(&lfs, cfg) => 0;
     // make a dir
-    lfs_mount(&lfs, &cfg) => 0;
+    lfs_mount(&lfs, cfg) => 0;
     lfs_mkdir(&lfs, "dir_here") => 0;
     lfs_unmount(&lfs) => 0;
 
     // change the dir pointer to be invalid
-    lfs_init(&lfs, &cfg) => 0;
+    lfs_init(&lfs, cfg) => 0;
     lfs_mdir_t mdir;
     lfs_dir_fetch(&lfs, &mdir, (lfs_block_t[2]){0, 1}) => 0;
     // make sure id 1 == our directory
+    uint8_t buffer[1024];
     lfs_dir_get(&lfs, &mdir,
             LFS_MKTAG(0x700, 0x3ff, 0),
             LFS_MKTAG(LFS_TYPE_NAME, 1, strlen("dir_here")), buffer)
@@ -57,14 +60,17 @@ code = '''
 
     // test that accessing our bad dir fails, note there's a number
     // of ways to access the dir, some can fail, but some don't
-    lfs_mount(&lfs, &cfg) => 0;
+    lfs_mount(&lfs, cfg) => 0;
+    struct lfs_info info;
     lfs_stat(&lfs, "dir_here", &info) => 0;
     assert(strcmp(info.name, "dir_here") == 0);
     assert(info.type == LFS_TYPE_DIR);
 
+    lfs_dir_t dir;
     lfs_dir_open(&lfs, &dir, "dir_here") => LFS_ERR_CORRUPT;
     lfs_stat(&lfs, "dir_here/file_here", &info) => LFS_ERR_CORRUPT;
     lfs_dir_open(&lfs, &dir, "dir_here/dir_here") => LFS_ERR_CORRUPT;
+    lfs_file_t file;
     lfs_file_open(&lfs, &file, "dir_here/file_here",
             LFS_O_RDONLY) => LFS_ERR_CORRUPT;
     lfs_file_open(&lfs, &file, "dir_here/file_here",
@@ -72,24 +78,27 @@ code = '''
     lfs_unmount(&lfs) => 0;
 '''
 
-[[case]] # invalid file pointer test
+[cases.test_evil_invalid_file_pointer]
 in = "lfs.c"
-define.SIZE = [10, 1000, 100000] # faked file size
+defines.SIZE = [10, 1000, 100000] # faked file size
 code = '''
     // create littlefs
-    lfs_format(&lfs, &cfg) => 0;
+    lfs_t lfs;
+    lfs_format(&lfs, cfg) => 0;
     // make a file
-    lfs_mount(&lfs, &cfg) => 0;
+    lfs_mount(&lfs, cfg) => 0;
+    lfs_file_t file;
     lfs_file_open(&lfs, &file, "file_here",
             LFS_O_WRONLY | LFS_O_CREAT) => 0;
     lfs_file_close(&lfs, &file) => 0;
     lfs_unmount(&lfs) => 0;
 
     // change the file pointer to be invalid
-    lfs_init(&lfs, &cfg) => 0;
+    lfs_init(&lfs, cfg) => 0;
     lfs_mdir_t mdir;
     lfs_dir_fetch(&lfs, &mdir, (lfs_block_t[2]){0, 1}) => 0;
     // make sure id 1 == our file
+    uint8_t buffer[1024];
     lfs_dir_get(&lfs, &mdir,
             LFS_MKTAG(0x700, 0x3ff, 0),
             LFS_MKTAG(LFS_TYPE_NAME, 1, strlen("file_here")), buffer)
@@ -103,7 +112,8 @@ code = '''
 
     // test that accessing our bad file fails, note there's a number
     // of ways to access the dir, some can fail, but some don't
-    lfs_mount(&lfs, &cfg) => 0;
+    lfs_mount(&lfs, cfg) => 0;
+    struct lfs_info info;
     lfs_stat(&lfs, "file_here", &info) => 0;
     assert(strcmp(info.name, "file_here") == 0);
     assert(info.type == LFS_TYPE_REG);
@@ -114,20 +124,22 @@ code = '''
     lfs_file_close(&lfs, &file) => 0;
 
     // any allocs that traverse CTZ must unfortunately must fail
-    if (SIZE > 2*LFS_BLOCK_SIZE) {
+    if (SIZE > 2*BLOCK_SIZE) {
         lfs_mkdir(&lfs, "dir_here") => LFS_ERR_CORRUPT;
     }
     lfs_unmount(&lfs) => 0;
 '''
 
-[[case]] # invalid pointer in CTZ skip-list test
-define.SIZE = ['2*LFS_BLOCK_SIZE', '3*LFS_BLOCK_SIZE', '4*LFS_BLOCK_SIZE']
+[cases.test_evil_invalid_ctz_pointer] # invalid pointer in CTZ skip-list test
+defines.SIZE = ['2*BLOCK_SIZE', '3*BLOCK_SIZE', '4*BLOCK_SIZE']
 in = "lfs.c"
 code = '''
     // create littlefs
-    lfs_format(&lfs, &cfg) => 0;
+    lfs_t lfs;
+    lfs_format(&lfs, cfg) => 0;
     // make a file
-    lfs_mount(&lfs, &cfg) => 0;
+    lfs_mount(&lfs, cfg) => 0;
+    lfs_file_t file;
     lfs_file_open(&lfs, &file, "file_here",
             LFS_O_WRONLY | LFS_O_CREAT) => 0;
     for (int i = 0; i < SIZE; i++) {
@@ -137,10 +149,11 @@ code = '''
     lfs_file_close(&lfs, &file) => 0;
     lfs_unmount(&lfs) => 0;
     // change pointer in CTZ skip-list to be invalid
-    lfs_init(&lfs, &cfg) => 0;
+    lfs_init(&lfs, cfg) => 0;
     lfs_mdir_t mdir;
     lfs_dir_fetch(&lfs, &mdir, (lfs_block_t[2]){0, 1}) => 0;
     // make sure id 1 == our file and get our CTZ structure
+    uint8_t buffer[4*BLOCK_SIZE];
     lfs_dir_get(&lfs, &mdir,
             LFS_MKTAG(0x700, 0x3ff, 0),
             LFS_MKTAG(LFS_TYPE_NAME, 1, strlen("file_here")), buffer)
@@ -153,18 +166,19 @@ code = '''
                 => LFS_MKTAG(LFS_TYPE_CTZSTRUCT, 1, sizeof(struct lfs_ctz));
     lfs_ctz_fromle32(&ctz);
     // rewrite block to contain bad pointer
-    uint8_t bbuffer[LFS_BLOCK_SIZE];
-    cfg.read(&cfg, ctz.head, 0, bbuffer, LFS_BLOCK_SIZE) => 0;
+    uint8_t bbuffer[BLOCK_SIZE];
+    cfg->read(cfg, ctz.head, 0, bbuffer, BLOCK_SIZE) => 0;
     uint32_t bad = lfs_tole32(0xcccccccc);
     memcpy(&bbuffer[0], &bad, sizeof(bad));
     memcpy(&bbuffer[4], &bad, sizeof(bad));
-    cfg.erase(&cfg, ctz.head) => 0;
-    cfg.prog(&cfg, ctz.head, 0, bbuffer, LFS_BLOCK_SIZE) => 0;
+    cfg->erase(cfg, ctz.head) => 0;
+    cfg->prog(cfg, ctz.head, 0, bbuffer, BLOCK_SIZE) => 0;
     lfs_deinit(&lfs) => 0;
 
     // test that accessing our bad file fails, note there's a number
     // of ways to access the dir, some can fail, but some don't
-    lfs_mount(&lfs, &cfg) => 0;
+    lfs_mount(&lfs, cfg) => 0;
+    struct lfs_info info;
     lfs_stat(&lfs, "file_here", &info) => 0;
     assert(strcmp(info.name, "file_here") == 0);
     assert(info.type == LFS_TYPE_REG);
@@ -175,22 +189,23 @@ code = '''
     lfs_file_close(&lfs, &file) => 0;
 
     // any allocs that traverse CTZ must unfortunately must fail
-    if (SIZE > 2*LFS_BLOCK_SIZE) {
+    if (SIZE > 2*BLOCK_SIZE) {
         lfs_mkdir(&lfs, "dir_here") => LFS_ERR_CORRUPT;
     }
     lfs_unmount(&lfs) => 0;
 '''
 
 
-[[case]] # invalid gstate pointer
-define.INVALSET = [0x3, 0x1, 0x2]
+[cases.test_evil_invalid_gstate_pointer]
+defines.INVALSET = [0x3, 0x1, 0x2]
 in = "lfs.c"
 code = '''
     // create littlefs
-    lfs_format(&lfs, &cfg) => 0;
+    lfs_t lfs;
+    lfs_format(&lfs, cfg) => 0;
 
     // create an invalid gstate
-    lfs_init(&lfs, &cfg) => 0;
+    lfs_init(&lfs, cfg) => 0;
     lfs_mdir_t mdir;
     lfs_dir_fetch(&lfs, &mdir, (lfs_block_t[2]){0, 1}) => 0;
     lfs_fs_prepmove(&lfs, 1, (lfs_block_t [2]){
@@ -202,21 +217,22 @@ code = '''
     // test that mount fails gracefully
     // mount may not fail, but our first alloc should fail when
     // we try to fix the gstate
-    lfs_mount(&lfs, &cfg) => 0;
+    lfs_mount(&lfs, cfg) => 0;
     lfs_mkdir(&lfs, "should_fail") => LFS_ERR_CORRUPT;
     lfs_unmount(&lfs) => 0;
 '''
 
 # cycle detection/recovery tests
 
-[[case]] # metadata-pair threaded-list loop test
+[cases.test_evil_mdir_loop] # metadata-pair threaded-list loop test
 in = "lfs.c"
 code = '''
     // create littlefs
-    lfs_format(&lfs, &cfg) => 0;
+    lfs_t lfs;
+    lfs_format(&lfs, cfg) => 0;
 
     // change tail-pointer to point to ourself
-    lfs_init(&lfs, &cfg) => 0;
+    lfs_init(&lfs, cfg) => 0;
     lfs_mdir_t mdir;
     lfs_dir_fetch(&lfs, &mdir, (lfs_block_t[2]){0, 1}) => 0;
     lfs_dir_commit(&lfs, &mdir, LFS_MKATTRS(
@@ -225,20 +241,21 @@ code = '''
     lfs_deinit(&lfs) => 0;
 
     // test that mount fails gracefully
-    lfs_mount(&lfs, &cfg) => LFS_ERR_CORRUPT;
+    lfs_mount(&lfs, cfg) => LFS_ERR_CORRUPT;
 '''
 
-[[case]] # metadata-pair threaded-list 2-length loop test
+[cases.test_evil_mdir_loop2] # metadata-pair threaded-list 2-length loop test
 in = "lfs.c"
 code = '''
     // create littlefs with child dir
-    lfs_format(&lfs, &cfg) => 0;
-    lfs_mount(&lfs, &cfg) => 0;
+    lfs_t lfs;
+    lfs_format(&lfs, cfg) => 0;
+    lfs_mount(&lfs, cfg) => 0;
     lfs_mkdir(&lfs, "child") => 0;
     lfs_unmount(&lfs) => 0;
 
     // find child
-    lfs_init(&lfs, &cfg) => 0;
+    lfs_init(&lfs, cfg) => 0;
     lfs_mdir_t mdir;
     lfs_block_t pair[2];
     lfs_dir_fetch(&lfs, &mdir, (lfs_block_t[2]){0, 1}) => 0;
@@ -255,20 +272,21 @@ code = '''
     lfs_deinit(&lfs) => 0;
 
     // test that mount fails gracefully
-    lfs_mount(&lfs, &cfg) => LFS_ERR_CORRUPT;
+    lfs_mount(&lfs, cfg) => LFS_ERR_CORRUPT;
 '''
 
-[[case]] # metadata-pair threaded-list 1-length child loop test
+[cases.test_evil_mdir_loop_child] # metadata-pair threaded-list 1-length child loop test
 in = "lfs.c"
 code = '''
     // create littlefs with child dir
-    lfs_format(&lfs, &cfg) => 0;
-    lfs_mount(&lfs, &cfg) => 0;
+    lfs_t lfs;
+    lfs_format(&lfs, cfg) => 0;
+    lfs_mount(&lfs, cfg) => 0;
     lfs_mkdir(&lfs, "child") => 0;
     lfs_unmount(&lfs) => 0;
 
     // find child
-    lfs_init(&lfs, &cfg) => 0;
+    lfs_init(&lfs, cfg) => 0;
     lfs_mdir_t mdir;
     lfs_block_t pair[2];
     lfs_dir_fetch(&lfs, &mdir, (lfs_block_t[2]){0, 1}) => 0;
@@ -284,5 +302,5 @@ code = '''
     lfs_deinit(&lfs) => 0;
 
     // test that mount fails gracefully
-    lfs_mount(&lfs, &cfg) => LFS_ERR_CORRUPT;
+    lfs_mount(&lfs, cfg) => LFS_ERR_CORRUPT;
 '''
diff --git a/tests/test_exhaustion.toml b/tests/test_exhaustion.toml
index 569611c..273dd98 100644
--- a/tests/test_exhaustion.toml
+++ b/tests/test_exhaustion.toml
@@ -1,46 +1,50 @@
-[[case]] # test running a filesystem to exhaustion
-define.LFS_ERASE_CYCLES = 10
-define.LFS_BLOCK_COUNT = 256 # small bd so test runs faster
-define.LFS_BLOCK_CYCLES = 'LFS_ERASE_CYCLES / 2'
-define.LFS_BADBLOCK_BEHAVIOR = [
-    'LFS_TESTBD_BADBLOCK_PROGERROR',
-    'LFS_TESTBD_BADBLOCK_ERASEERROR',
-    'LFS_TESTBD_BADBLOCK_READERROR',
-    'LFS_TESTBD_BADBLOCK_PROGNOOP',
-    'LFS_TESTBD_BADBLOCK_ERASENOOP',
+# test running a filesystem to exhaustion
+[cases.test_exhaustion_normal]
+defines.ERASE_CYCLES = 10
+defines.ERASE_COUNT = 256 # small bd so test runs faster
+defines.BLOCK_CYCLES = 'ERASE_CYCLES / 2'
+defines.BADBLOCK_BEHAVIOR = [
+    'LFS_EMUBD_BADBLOCK_PROGERROR',
+    'LFS_EMUBD_BADBLOCK_ERASEERROR',
+    'LFS_EMUBD_BADBLOCK_READERROR',
+    'LFS_EMUBD_BADBLOCK_PROGNOOP',
+    'LFS_EMUBD_BADBLOCK_ERASENOOP',
 ]
-define.FILES = 10
+defines.FILES = 10
 code = '''
-    lfs_format(&lfs, &cfg) => 0;
-    lfs_mount(&lfs, &cfg) => 0;
+    lfs_t lfs;
+    lfs_format(&lfs, cfg) => 0;
+    lfs_mount(&lfs, cfg) => 0;
     lfs_mkdir(&lfs, "roadrunner") => 0;
     lfs_unmount(&lfs) => 0;
 
     uint32_t cycle = 0;
     while (true) {
-        lfs_mount(&lfs, &cfg) => 0;
+        lfs_mount(&lfs, cfg) => 0;
         for (uint32_t i = 0; i < FILES; i++) {
             // chose name, roughly random seed, and random 2^n size
+            char path[1024];
             sprintf(path, "roadrunner/test%d", i);
-            srand(cycle * i);
-            size = 1 << ((rand() % 10)+2);
+            uint32_t prng = cycle * i;
+            lfs_size_t size = 1 << ((TEST_PRNG(&prng) % 10)+2);
 
+            lfs_file_t file;
             lfs_file_open(&lfs, &file, path,
                     LFS_O_WRONLY | LFS_O_CREAT | LFS_O_TRUNC) => 0;
 
             for (lfs_size_t j = 0; j < size; j++) {
-                char c = 'a' + (rand() % 26);
+                char c = 'a' + (TEST_PRNG(&prng) % 26);
                 lfs_ssize_t res = lfs_file_write(&lfs, &file, &c, 1);
                 assert(res == 1 || res == LFS_ERR_NOSPC);
                 if (res == LFS_ERR_NOSPC) {
-                    err = lfs_file_close(&lfs, &file);
+                    int err = lfs_file_close(&lfs, &file);
                     assert(err == 0 || err == LFS_ERR_NOSPC);
                     lfs_unmount(&lfs) => 0;
                     goto exhausted;
                 }
             }
 
-            err = lfs_file_close(&lfs, &file);
+            int err = lfs_file_close(&lfs, &file);
             assert(err == 0 || err == LFS_ERR_NOSPC);
             if (err == LFS_ERR_NOSPC) {
                 lfs_unmount(&lfs) => 0;
@@ -50,13 +54,15 @@ code = '''
 
         for (uint32_t i = 0; i < FILES; i++) {
             // check for errors
+            char path[1024];
             sprintf(path, "roadrunner/test%d", i);
-            srand(cycle * i);
-            size = 1 << ((rand() % 10)+2);
+            uint32_t prng = cycle * i;
+            lfs_size_t size = 1 << ((TEST_PRNG(&prng) % 10)+2);
 
+            lfs_file_t file;
             lfs_file_open(&lfs, &file, path, LFS_O_RDONLY) => 0;
             for (lfs_size_t j = 0; j < size; j++) {
-                char c = 'a' + (rand() % 26);
+                char c = 'a' + (TEST_PRNG(&prng) % 26);
                 char r;
                 lfs_file_read(&lfs, &file, &r, 1) => 1;
                 assert(r == c);
@@ -71,10 +77,12 @@ code = '''
 
 exhausted:
     // should still be readable
-    lfs_mount(&lfs, &cfg) => 0;
+    lfs_mount(&lfs, cfg) => 0;
     for (uint32_t i = 0; i < FILES; i++) {
         // check for errors
+        char path[1024];
         sprintf(path, "roadrunner/test%d", i);
+        struct lfs_info info;
         lfs_stat(&lfs, path, &info) => 0;
     }
     lfs_unmount(&lfs) => 0;
@@ -82,47 +90,51 @@ exhausted:
     LFS_WARN("completed %d cycles", cycle);
 '''
 
-[[case]] # test running a filesystem to exhaustion
-         # which also requires expanding superblocks
-define.LFS_ERASE_CYCLES = 10
-define.LFS_BLOCK_COUNT = 256 # small bd so test runs faster
-define.LFS_BLOCK_CYCLES = 'LFS_ERASE_CYCLES / 2'
-define.LFS_BADBLOCK_BEHAVIOR = [
-    'LFS_TESTBD_BADBLOCK_PROGERROR',
-    'LFS_TESTBD_BADBLOCK_ERASEERROR',
-    'LFS_TESTBD_BADBLOCK_READERROR',
-    'LFS_TESTBD_BADBLOCK_PROGNOOP',
-    'LFS_TESTBD_BADBLOCK_ERASENOOP',
+# test running a filesystem to exhaustion
+# which also requires expanding superblocks
+[cases.test_exhaustion_superblocks]
+defines.ERASE_CYCLES = 10
+defines.ERASE_COUNT = 256 # small bd so test runs faster
+defines.BLOCK_CYCLES = 'ERASE_CYCLES / 2'
+defines.BADBLOCK_BEHAVIOR = [
+    'LFS_EMUBD_BADBLOCK_PROGERROR',
+    'LFS_EMUBD_BADBLOCK_ERASEERROR',
+    'LFS_EMUBD_BADBLOCK_READERROR',
+    'LFS_EMUBD_BADBLOCK_PROGNOOP',
+    'LFS_EMUBD_BADBLOCK_ERASENOOP',
 ]
-define.FILES = 10
+defines.FILES = 10
 code = '''
-    lfs_format(&lfs, &cfg) => 0;
+    lfs_t lfs;
+    lfs_format(&lfs, cfg) => 0;
 
     uint32_t cycle = 0;
     while (true) {
-        lfs_mount(&lfs, &cfg) => 0;
+        lfs_mount(&lfs, cfg) => 0;
         for (uint32_t i = 0; i < FILES; i++) {
             // chose name, roughly random seed, and random 2^n size
+            char path[1024];
             sprintf(path, "test%d", i);
-            srand(cycle * i);
-            size = 1 << ((rand() % 10)+2);
+            uint32_t prng = cycle * i;
+            lfs_size_t size = 1 << ((TEST_PRNG(&prng) % 10)+2);
 
+            lfs_file_t file;
             lfs_file_open(&lfs, &file, path,
                     LFS_O_WRONLY | LFS_O_CREAT | LFS_O_TRUNC) => 0;
 
             for (lfs_size_t j = 0; j < size; j++) {
-                char c = 'a' + (rand() % 26);
+                char c = 'a' + (TEST_PRNG(&prng) % 26);
                 lfs_ssize_t res = lfs_file_write(&lfs, &file, &c, 1);
                 assert(res == 1 || res == LFS_ERR_NOSPC);
                 if (res == LFS_ERR_NOSPC) {
-                    err = lfs_file_close(&lfs, &file);
+                    int err = lfs_file_close(&lfs, &file);
                     assert(err == 0 || err == LFS_ERR_NOSPC);
                     lfs_unmount(&lfs) => 0;
                     goto exhausted;
                 }
             }
 
-            err = lfs_file_close(&lfs, &file);
+            int err = lfs_file_close(&lfs, &file);
             assert(err == 0 || err == LFS_ERR_NOSPC);
             if (err == LFS_ERR_NOSPC) {
                 lfs_unmount(&lfs) => 0;
@@ -132,13 +144,15 @@ code = '''
 
         for (uint32_t i = 0; i < FILES; i++) {
             // check for errors
+            char path[1024];
             sprintf(path, "test%d", i);
-            srand(cycle * i);
-            size = 1 << ((rand() % 10)+2);
+            uint32_t prng = cycle * i;
+            lfs_size_t size = 1 << ((TEST_PRNG(&prng) % 10)+2);
 
+            lfs_file_t file;
             lfs_file_open(&lfs, &file, path, LFS_O_RDONLY) => 0;
             for (lfs_size_t j = 0; j < size; j++) {
-                char c = 'a' + (rand() % 26);
+                char c = 'a' + (TEST_PRNG(&prng) % 26);
                 char r;
                 lfs_file_read(&lfs, &file, &r, 1) => 1;
                 assert(r == c);
@@ -153,9 +167,11 @@ code = '''
 
 exhausted:
     // should still be readable
-    lfs_mount(&lfs, &cfg) => 0;
+    lfs_mount(&lfs, cfg) => 0;
     for (uint32_t i = 0; i < FILES; i++) {
         // check for errors
+        char path[1024];
+        struct lfs_info info;
         sprintf(path, "test%d", i);
         lfs_stat(&lfs, path, &info) => 0;
     }
@@ -169,51 +185,55 @@ exhausted:
 # into increasing the block devices lifetime. This is something we can actually
 # check for.
 
-[[case]] # wear-level test running a filesystem to exhaustion
-define.LFS_ERASE_CYCLES = 20
-define.LFS_BLOCK_COUNT = 256 # small bd so test runs faster
-define.LFS_BLOCK_CYCLES = 'LFS_ERASE_CYCLES / 2'
-define.FILES = 10
+# wear-level test running a filesystem to exhaustion
+[cases.test_exhuastion_wear_leveling]
+defines.ERASE_CYCLES = 20
+defines.ERASE_COUNT = 256 # small bd so test runs faster
+defines.BLOCK_CYCLES = 'ERASE_CYCLES / 2'
+defines.FILES = 10
 code = '''
     uint32_t run_cycles[2];
-    const uint32_t run_block_count[2] = {LFS_BLOCK_COUNT/2, LFS_BLOCK_COUNT};
+    const uint32_t run_block_count[2] = {BLOCK_COUNT/2, BLOCK_COUNT};
 
     for (int run = 0; run < 2; run++) {
-        for (lfs_block_t b = 0; b < LFS_BLOCK_COUNT; b++) {
-            lfs_testbd_setwear(&cfg, b,
-                    (b < run_block_count[run]) ? 0 : LFS_ERASE_CYCLES) => 0;
+        for (lfs_block_t b = 0; b < BLOCK_COUNT; b++) {
+            lfs_emubd_setwear(cfg, b,
+                    (b < run_block_count[run]) ? 0 : ERASE_CYCLES) => 0;
         }
 
-        lfs_format(&lfs, &cfg) => 0;
-        lfs_mount(&lfs, &cfg) => 0;
+        lfs_t lfs;
+        lfs_format(&lfs, cfg) => 0;
+        lfs_mount(&lfs, cfg) => 0;
         lfs_mkdir(&lfs, "roadrunner") => 0;
         lfs_unmount(&lfs) => 0;
 
         uint32_t cycle = 0;
         while (true) {
-            lfs_mount(&lfs, &cfg) => 0;
+            lfs_mount(&lfs, cfg) => 0;
             for (uint32_t i = 0; i < FILES; i++) {
                 // chose name, roughly random seed, and random 2^n size
+                char path[1024];
                 sprintf(path, "roadrunner/test%d", i);
-                srand(cycle * i);
-                size = 1 << ((rand() % 10)+2);
+                uint32_t prng = cycle * i;
+                lfs_size_t size = 1 << ((TEST_PRNG(&prng) % 10)+2);
 
+                lfs_file_t file;
                 lfs_file_open(&lfs, &file, path,
                         LFS_O_WRONLY | LFS_O_CREAT | LFS_O_TRUNC) => 0;
 
                 for (lfs_size_t j = 0; j < size; j++) {
-                    char c = 'a' + (rand() % 26);
+                    char c = 'a' + (TEST_PRNG(&prng) % 26);
                     lfs_ssize_t res = lfs_file_write(&lfs, &file, &c, 1);
                     assert(res == 1 || res == LFS_ERR_NOSPC);
                     if (res == LFS_ERR_NOSPC) {
-                        err = lfs_file_close(&lfs, &file);
+                        int err = lfs_file_close(&lfs, &file);
                         assert(err == 0 || err == LFS_ERR_NOSPC);
                         lfs_unmount(&lfs) => 0;
                         goto exhausted;
                     }
                 }
 
-                err = lfs_file_close(&lfs, &file);
+                int err = lfs_file_close(&lfs, &file);
                 assert(err == 0 || err == LFS_ERR_NOSPC);
                 if (err == LFS_ERR_NOSPC) {
                     lfs_unmount(&lfs) => 0;
@@ -223,13 +243,15 @@ code = '''
 
             for (uint32_t i = 0; i < FILES; i++) {
                 // check for errors
+                char path[1024];
                 sprintf(path, "roadrunner/test%d", i);
-                srand(cycle * i);
-                size = 1 << ((rand() % 10)+2);
+                uint32_t prng = cycle * i;
+                lfs_size_t size = 1 << ((TEST_PRNG(&prng) % 10)+2);
 
+                lfs_file_t file;
                 lfs_file_open(&lfs, &file, path, LFS_O_RDONLY) => 0;
                 for (lfs_size_t j = 0; j < size; j++) {
-                    char c = 'a' + (rand() % 26);
+                    char c = 'a' + (TEST_PRNG(&prng) % 26);
                     char r;
                     lfs_file_read(&lfs, &file, &r, 1) => 1;
                     assert(r == c);
@@ -244,9 +266,11 @@ code = '''
 
 exhausted:
         // should still be readable
-        lfs_mount(&lfs, &cfg) => 0;
+        lfs_mount(&lfs, cfg) => 0;
         for (uint32_t i = 0; i < FILES; i++) {
             // check for errors
+            char path[1024];
+            struct lfs_info info;
             sprintf(path, "roadrunner/test%d", i);
             lfs_stat(&lfs, path, &info) => 0;
         }
@@ -261,48 +285,52 @@ exhausted:
     LFS_ASSERT(run_cycles[1]*110/100 > 2*run_cycles[0]);
 '''
 
-[[case]] # wear-level test + expanding superblock
-define.LFS_ERASE_CYCLES = 20
-define.LFS_BLOCK_COUNT = 256 # small bd so test runs faster
-define.LFS_BLOCK_CYCLES = 'LFS_ERASE_CYCLES / 2'
-define.FILES = 10
+# wear-level test + expanding superblock
+[cases.test_exhaustion_wear_leveling_superblocks]
+defines.ERASE_CYCLES = 20
+defines.ERASE_COUNT = 256 # small bd so test runs faster
+defines.BLOCK_CYCLES = 'ERASE_CYCLES / 2'
+defines.FILES = 10
 code = '''
     uint32_t run_cycles[2];
-    const uint32_t run_block_count[2] = {LFS_BLOCK_COUNT/2, LFS_BLOCK_COUNT};
+    const uint32_t run_block_count[2] = {BLOCK_COUNT/2, BLOCK_COUNT};
 
     for (int run = 0; run < 2; run++) {
-        for (lfs_block_t b = 0; b < LFS_BLOCK_COUNT; b++) {
-            lfs_testbd_setwear(&cfg, b,
-                    (b < run_block_count[run]) ? 0 : LFS_ERASE_CYCLES) => 0;
+        for (lfs_block_t b = 0; b < BLOCK_COUNT; b++) {
+            lfs_emubd_setwear(cfg, b,
+                    (b < run_block_count[run]) ? 0 : ERASE_CYCLES) => 0;
         }
 
-        lfs_format(&lfs, &cfg) => 0;
+        lfs_t lfs;
+        lfs_format(&lfs, cfg) => 0;
 
         uint32_t cycle = 0;
         while (true) {
-            lfs_mount(&lfs, &cfg) => 0;
+            lfs_mount(&lfs, cfg) => 0;
             for (uint32_t i = 0; i < FILES; i++) {
                 // chose name, roughly random seed, and random 2^n size
+                char path[1024];
                 sprintf(path, "test%d", i);
-                srand(cycle * i);
-                size = 1 << ((rand() % 10)+2);
+                uint32_t prng = cycle * i;
+                lfs_size_t size = 1 << ((TEST_PRNG(&prng) % 10)+2);
 
+                lfs_file_t file;
                 lfs_file_open(&lfs, &file, path,
                         LFS_O_WRONLY | LFS_O_CREAT | LFS_O_TRUNC) => 0;
 
                 for (lfs_size_t j = 0; j < size; j++) {
-                    char c = 'a' + (rand() % 26);
+                    char c = 'a' + (TEST_PRNG(&prng) % 26);
                     lfs_ssize_t res = lfs_file_write(&lfs, &file, &c, 1);
                     assert(res == 1 || res == LFS_ERR_NOSPC);
                     if (res == LFS_ERR_NOSPC) {
-                        err = lfs_file_close(&lfs, &file);
+                        int err = lfs_file_close(&lfs, &file);
                         assert(err == 0 || err == LFS_ERR_NOSPC);
                         lfs_unmount(&lfs) => 0;
                         goto exhausted;
                     }
                 }
 
-                err = lfs_file_close(&lfs, &file);
+                int err = lfs_file_close(&lfs, &file);
                 assert(err == 0 || err == LFS_ERR_NOSPC);
                 if (err == LFS_ERR_NOSPC) {
                     lfs_unmount(&lfs) => 0;
@@ -312,13 +340,15 @@ code = '''
 
             for (uint32_t i = 0; i < FILES; i++) {
                 // check for errors
+                char path[1024];
                 sprintf(path, "test%d", i);
-                srand(cycle * i);
-                size = 1 << ((rand() % 10)+2);
+                uint32_t prng = cycle * i;
+                lfs_size_t size = 1 << ((TEST_PRNG(&prng) % 10)+2);
 
+                lfs_file_t file;
                 lfs_file_open(&lfs, &file, path, LFS_O_RDONLY) => 0;
                 for (lfs_size_t j = 0; j < size; j++) {
-                    char c = 'a' + (rand() % 26);
+                    char c = 'a' + (TEST_PRNG(&prng) % 26);
                     char r;
                     lfs_file_read(&lfs, &file, &r, 1) => 1;
                     assert(r == c);
@@ -333,9 +363,11 @@ code = '''
 
 exhausted:
         // should still be readable
-        lfs_mount(&lfs, &cfg) => 0;
+        lfs_mount(&lfs, cfg) => 0;
         for (uint32_t i = 0; i < FILES; i++) {
             // check for errors
+            char path[1024];
+            struct lfs_info info;
             sprintf(path, "test%d", i);
             lfs_stat(&lfs, path, &info) => 0;
         }
@@ -350,44 +382,48 @@ exhausted:
     LFS_ASSERT(run_cycles[1]*110/100 > 2*run_cycles[0]);
 '''
 
-[[case]] # test that we wear blocks roughly evenly
-define.LFS_ERASE_CYCLES = 0xffffffff
-define.LFS_BLOCK_COUNT = 256 # small bd so test runs faster
-define.LFS_BLOCK_CYCLES = [5, 4, 3, 2, 1]
-define.CYCLES = 100
-define.FILES = 10
-if = 'LFS_BLOCK_CYCLES < CYCLES/10'
+# test that we wear blocks roughly evenly
+[cases.test_exhaustion_wear_distribution]
+defines.ERASE_CYCLES = 0xffffffff
+defines.ERASE_COUNT = 256 # small bd so test runs faster
+defines.BLOCK_CYCLES = [5, 4, 3, 2, 1]
+defines.CYCLES = 100
+defines.FILES = 10
+if = 'BLOCK_CYCLES < CYCLES/10'
 code = '''
-    lfs_format(&lfs, &cfg) => 0;
-    lfs_mount(&lfs, &cfg) => 0;
+    lfs_t lfs;
+    lfs_format(&lfs, cfg) => 0;
+    lfs_mount(&lfs, cfg) => 0;
     lfs_mkdir(&lfs, "roadrunner") => 0;
     lfs_unmount(&lfs) => 0;
 
     uint32_t cycle = 0;
     while (cycle < CYCLES) {
-        lfs_mount(&lfs, &cfg) => 0;
+        lfs_mount(&lfs, cfg) => 0;
         for (uint32_t i = 0; i < FILES; i++) {
             // chose name, roughly random seed, and random 2^n size
+            char path[1024];
             sprintf(path, "roadrunner/test%d", i);
-            srand(cycle * i);
-            size = 1 << 4; //((rand() % 10)+2);
+            uint32_t prng = cycle * i;
+            lfs_size_t size = 1 << 4; //((TEST_PRNG(&prng) % 10)+2);
 
+            lfs_file_t file;
             lfs_file_open(&lfs, &file, path,
                     LFS_O_WRONLY | LFS_O_CREAT | LFS_O_TRUNC) => 0;
 
             for (lfs_size_t j = 0; j < size; j++) {
-                char c = 'a' + (rand() % 26);
+                char c = 'a' + (TEST_PRNG(&prng) % 26);
                 lfs_ssize_t res = lfs_file_write(&lfs, &file, &c, 1);
                 assert(res == 1 || res == LFS_ERR_NOSPC);
                 if (res == LFS_ERR_NOSPC) {
-                    err = lfs_file_close(&lfs, &file);
+                    int err = lfs_file_close(&lfs, &file);
                     assert(err == 0 || err == LFS_ERR_NOSPC);
                     lfs_unmount(&lfs) => 0;
                     goto exhausted;
                 }
             }
 
-            err = lfs_file_close(&lfs, &file);
+            int err = lfs_file_close(&lfs, &file);
             assert(err == 0 || err == LFS_ERR_NOSPC);
             if (err == LFS_ERR_NOSPC) {
                 lfs_unmount(&lfs) => 0;
@@ -397,13 +433,15 @@ code = '''
 
         for (uint32_t i = 0; i < FILES; i++) {
             // check for errors
+            char path[1024];
             sprintf(path, "roadrunner/test%d", i);
-            srand(cycle * i);
-            size = 1 << 4; //((rand() % 10)+2);
+            uint32_t prng = cycle * i;
+            lfs_size_t size = 1 << 4; //((TEST_PRNG(&prng) % 10)+2);
 
+            lfs_file_t file;
             lfs_file_open(&lfs, &file, path, LFS_O_RDONLY) => 0;
             for (lfs_size_t j = 0; j < size; j++) {
-                char c = 'a' + (rand() % 26);
+                char c = 'a' + (TEST_PRNG(&prng) % 26);
                 char r;
                 lfs_file_read(&lfs, &file, &r, 1) => 1;
                 assert(r == c);
@@ -418,9 +456,11 @@ code = '''
 
 exhausted:
     // should still be readable
-    lfs_mount(&lfs, &cfg) => 0;
+    lfs_mount(&lfs, cfg) => 0;
     for (uint32_t i = 0; i < FILES; i++) {
         // check for errors
+        char path[1024];
+        struct lfs_info info;
         sprintf(path, "roadrunner/test%d", i);
         lfs_stat(&lfs, path, &info) => 0;
     }
@@ -429,12 +469,12 @@ exhausted:
     LFS_WARN("completed %d cycles", cycle);
 
     // check the wear on our block device
-    lfs_testbd_wear_t minwear = -1;
-    lfs_testbd_wear_t totalwear = 0;
-    lfs_testbd_wear_t maxwear = 0;
+    lfs_emubd_wear_t minwear = -1;
+    lfs_emubd_wear_t totalwear = 0;
+    lfs_emubd_wear_t maxwear = 0;
     // skip 0 and 1 as superblock movement is intentionally avoided
-    for (lfs_block_t b = 2; b < LFS_BLOCK_COUNT; b++) {
-        lfs_testbd_wear_t wear = lfs_testbd_getwear(&cfg, b);
+    for (lfs_block_t b = 2; b < BLOCK_COUNT; b++) {
+        lfs_emubd_wear_t wear = lfs_emubd_wear(cfg, b);
         printf("%08x: wear %d\n", b, wear);
         assert(wear >= 0);
         if (wear < minwear) {
@@ -445,17 +485,17 @@ exhausted:
         }
         totalwear += wear;
     }
-    lfs_testbd_wear_t avgwear = totalwear / LFS_BLOCK_COUNT;
+    lfs_emubd_wear_t avgwear = totalwear / BLOCK_COUNT;
     LFS_WARN("max wear: %d cycles", maxwear);
-    LFS_WARN("avg wear: %d cycles", totalwear / LFS_BLOCK_COUNT);
+    LFS_WARN("avg wear: %d cycles", totalwear / (int)BLOCK_COUNT);
     LFS_WARN("min wear: %d cycles", minwear);
 
     // find standard deviation^2
-    lfs_testbd_wear_t dev2 = 0;
-    for (lfs_block_t b = 2; b < LFS_BLOCK_COUNT; b++) {
-        lfs_testbd_wear_t wear = lfs_testbd_getwear(&cfg, b);
+    lfs_emubd_wear_t dev2 = 0;
+    for (lfs_block_t b = 2; b < BLOCK_COUNT; b++) {
+        lfs_emubd_wear_t wear = lfs_emubd_wear(cfg, b);
         assert(wear >= 0);
-        lfs_testbd_swear_t diff = wear - avgwear;
+        lfs_emubd_swear_t diff = wear - avgwear;
         dev2 += diff*diff;
     }
     dev2 /= totalwear;
diff --git a/tests/test_files.toml b/tests/test_files.toml
index 565e665..afb0811 100644
--- a/tests/test_files.toml
+++ b/tests/test_files.toml
@@ -1,17 +1,20 @@
 
-[[case]] # simple file test
+[cases.test_files_simple]
 code = '''
-    lfs_format(&lfs, &cfg) => 0;
-    lfs_mount(&lfs, &cfg) => 0;
+    lfs_t lfs;
+    lfs_format(&lfs, cfg) => 0;
+    lfs_mount(&lfs, cfg) => 0;
+    lfs_file_t file;
     lfs_file_open(&lfs, &file, "hello",
             LFS_O_WRONLY | LFS_O_CREAT | LFS_O_EXCL) => 0;
-    size = strlen("Hello World!")+1;
+    lfs_size_t size = strlen("Hello World!")+1;
+    uint8_t buffer[1024];
     strcpy((char*)buffer, "Hello World!");
     lfs_file_write(&lfs, &file, buffer, size) => size;
     lfs_file_close(&lfs, &file) => 0;
     lfs_unmount(&lfs) => 0;
 
-    lfs_mount(&lfs, &cfg) => 0;
+    lfs_mount(&lfs, cfg) => 0;
     lfs_file_open(&lfs, &file, "hello", LFS_O_RDONLY) => 0;
     lfs_file_read(&lfs, &file, buffer, size) => size;
     assert(strcmp((char*)buffer, "Hello World!") == 0);
@@ -19,21 +22,24 @@ code = '''
     lfs_unmount(&lfs) => 0;
 '''
 
-[[case]] # larger files
-define.SIZE = [32, 8192, 262144, 0, 7, 8193]
-define.CHUNKSIZE = [31, 16, 33, 1, 1023]
+[cases.test_files_large]
+defines.SIZE = [32, 8192, 262144, 0, 7, 8193]
+defines.CHUNKSIZE = [31, 16, 33, 1, 1023]
 code = '''
-    lfs_format(&lfs, &cfg) => 0;
+    lfs_t lfs;
+    lfs_format(&lfs, cfg) => 0;
 
     // write
-    lfs_mount(&lfs, &cfg) => 0;
+    lfs_mount(&lfs, cfg) => 0;
+    lfs_file_t file;
     lfs_file_open(&lfs, &file, "avacado",
             LFS_O_WRONLY | LFS_O_CREAT | LFS_O_EXCL) => 0;
-    srand(1);
+    uint32_t prng = 1;
+    uint8_t buffer[1024];
     for (lfs_size_t i = 0; i < SIZE; i += CHUNKSIZE) {
         lfs_size_t chunk = lfs_min(CHUNKSIZE, SIZE-i);
         for (lfs_size_t b = 0; b < chunk; b++) {
-            buffer[b] = rand() & 0xff;
+            buffer[b] = TEST_PRNG(&prng) & 0xff;
         }
         lfs_file_write(&lfs, &file, buffer, chunk) => chunk;
     }
@@ -41,15 +47,15 @@ code = '''
     lfs_unmount(&lfs) => 0;
 
     // read
-    lfs_mount(&lfs, &cfg) => 0;
+    lfs_mount(&lfs, cfg) => 0;
     lfs_file_open(&lfs, &file, "avacado", LFS_O_RDONLY) => 0;
     lfs_file_size(&lfs, &file) => SIZE;
-    srand(1);
+    prng = 1;
     for (lfs_size_t i = 0; i < SIZE; i += CHUNKSIZE) {
         lfs_size_t chunk = lfs_min(CHUNKSIZE, SIZE-i);
         lfs_file_read(&lfs, &file, buffer, chunk) => chunk;
         for (lfs_size_t b = 0; b < chunk; b++) {
-            assert(buffer[b] == (rand() & 0xff));
+            assert(buffer[b] == (TEST_PRNG(&prng) & 0xff));
         }
     }
     lfs_file_read(&lfs, &file, buffer, CHUNKSIZE) => 0;
@@ -57,22 +63,25 @@ code = '''
     lfs_unmount(&lfs) => 0;
 '''
 
-[[case]] # rewriting files
-define.SIZE1 = [32, 8192, 131072, 0, 7, 8193]
-define.SIZE2 = [32, 8192, 131072, 0, 7, 8193]
-define.CHUNKSIZE = [31, 16, 1]
+[cases.test_files_rewrite]
+defines.SIZE1 = [32, 8192, 131072, 0, 7, 8193]
+defines.SIZE2 = [32, 8192, 131072, 0, 7, 8193]
+defines.CHUNKSIZE = [31, 16, 1]
 code = '''
-    lfs_format(&lfs, &cfg) => 0;
+    lfs_t lfs;
+    lfs_format(&lfs, cfg) => 0;
 
     // write
-    lfs_mount(&lfs, &cfg) => 0;
+    lfs_mount(&lfs, cfg) => 0;
+    lfs_file_t file;
+    uint8_t buffer[1024];
     lfs_file_open(&lfs, &file, "avacado",
             LFS_O_WRONLY | LFS_O_CREAT | LFS_O_EXCL) => 0;
-    srand(1);
+    uint32_t prng = 1;
     for (lfs_size_t i = 0; i < SIZE1; i += CHUNKSIZE) {
         lfs_size_t chunk = lfs_min(CHUNKSIZE, SIZE1-i);
         for (lfs_size_t b = 0; b < chunk; b++) {
-            buffer[b] = rand() & 0xff;
+            buffer[b] = TEST_PRNG(&prng) & 0xff;
         }
         lfs_file_write(&lfs, &file, buffer, chunk) => chunk;
     }
@@ -80,15 +89,15 @@ code = '''
     lfs_unmount(&lfs) => 0;
 
     // read
-    lfs_mount(&lfs, &cfg) => 0;
+    lfs_mount(&lfs, cfg) => 0;
     lfs_file_open(&lfs, &file, "avacado", LFS_O_RDONLY) => 0;
     lfs_file_size(&lfs, &file) => SIZE1;
-    srand(1);
+    prng = 1;
     for (lfs_size_t i = 0; i < SIZE1; i += CHUNKSIZE) {
         lfs_size_t chunk = lfs_min(CHUNKSIZE, SIZE1-i);
         lfs_file_read(&lfs, &file, buffer, chunk) => chunk;
         for (lfs_size_t b = 0; b < chunk; b++) {
-            assert(buffer[b] == (rand() & 0xff));
+            assert(buffer[b] == (TEST_PRNG(&prng) & 0xff));
         }
     }
     lfs_file_read(&lfs, &file, buffer, CHUNKSIZE) => 0;
@@ -96,13 +105,13 @@ code = '''
     lfs_unmount(&lfs) => 0;
 
     // rewrite
-    lfs_mount(&lfs, &cfg) => 0;
+    lfs_mount(&lfs, cfg) => 0;
     lfs_file_open(&lfs, &file, "avacado", LFS_O_WRONLY) => 0;
-    srand(2);
+    prng = 2;
     for (lfs_size_t i = 0; i < SIZE2; i += CHUNKSIZE) {
         lfs_size_t chunk = lfs_min(CHUNKSIZE, SIZE2-i);
         for (lfs_size_t b = 0; b < chunk; b++) {
-            buffer[b] = rand() & 0xff;
+            buffer[b] = TEST_PRNG(&prng) & 0xff;
         }
         lfs_file_write(&lfs, &file, buffer, chunk) => chunk;
     }
@@ -110,27 +119,27 @@ code = '''
     lfs_unmount(&lfs) => 0;
 
     // read
-    lfs_mount(&lfs, &cfg) => 0;
+    lfs_mount(&lfs, cfg) => 0;
     lfs_file_open(&lfs, &file, "avacado", LFS_O_RDONLY) => 0;
     lfs_file_size(&lfs, &file) => lfs_max(SIZE1, SIZE2);
-    srand(2);
+    prng = 2;
     for (lfs_size_t i = 0; i < SIZE2; i += CHUNKSIZE) {
         lfs_size_t chunk = lfs_min(CHUNKSIZE, SIZE2-i);
         lfs_file_read(&lfs, &file, buffer, chunk) => chunk;
         for (lfs_size_t b = 0; b < chunk; b++) {
-            assert(buffer[b] == (rand() & 0xff));
+            assert(buffer[b] == (TEST_PRNG(&prng) & 0xff));
         }
     }
     if (SIZE1 > SIZE2) {
-        srand(1);
+        prng = 1;
         for (lfs_size_t b = 0; b < SIZE2; b++) {
-            rand();
+            TEST_PRNG(&prng);
         }
         for (lfs_size_t i = SIZE2; i < SIZE1; i += CHUNKSIZE) {
             lfs_size_t chunk = lfs_min(CHUNKSIZE, SIZE1-i);
             lfs_file_read(&lfs, &file, buffer, chunk) => chunk;
             for (lfs_size_t b = 0; b < chunk; b++) {
-                assert(buffer[b] == (rand() & 0xff));
+                assert(buffer[b] == (TEST_PRNG(&prng) & 0xff));
             }
         }
     }
@@ -139,22 +148,25 @@ code = '''
     lfs_unmount(&lfs) => 0;
 '''
 
-[[case]] # appending files
-define.SIZE1 = [32, 8192, 131072, 0, 7, 8193]
-define.SIZE2 = [32, 8192, 131072, 0, 7, 8193]
-define.CHUNKSIZE = [31, 16, 1]
+[cases.test_files_append]
+defines.SIZE1 = [32, 8192, 131072, 0, 7, 8193]
+defines.SIZE2 = [32, 8192, 131072, 0, 7, 8193]
+defines.CHUNKSIZE = [31, 16, 1]
 code = '''
-    lfs_format(&lfs, &cfg) => 0;
+    lfs_t lfs;
+    lfs_format(&lfs, cfg) => 0;
 
     // write
-    lfs_mount(&lfs, &cfg) => 0;
+    lfs_mount(&lfs, cfg) => 0;
+    lfs_file_t file;
+    uint8_t buffer[1024];
     lfs_file_open(&lfs, &file, "avacado",
             LFS_O_WRONLY | LFS_O_CREAT | LFS_O_EXCL) => 0;
-    srand(1);
+    uint32_t prng = 1;
     for (lfs_size_t i = 0; i < SIZE1; i += CHUNKSIZE) {
         lfs_size_t chunk = lfs_min(CHUNKSIZE, SIZE1-i);
         for (lfs_size_t b = 0; b < chunk; b++) {
-            buffer[b] = rand() & 0xff;
+            buffer[b] = TEST_PRNG(&prng) & 0xff;
         }
         lfs_file_write(&lfs, &file, buffer, chunk) => chunk;
     }
@@ -162,15 +174,15 @@ code = '''
     lfs_unmount(&lfs) => 0;
 
     // read
-    lfs_mount(&lfs, &cfg) => 0;
+    lfs_mount(&lfs, cfg) => 0;
     lfs_file_open(&lfs, &file, "avacado", LFS_O_RDONLY) => 0;
     lfs_file_size(&lfs, &file) => SIZE1;
-    srand(1);
+    prng = 1;
     for (lfs_size_t i = 0; i < SIZE1; i += CHUNKSIZE) {
         lfs_size_t chunk = lfs_min(CHUNKSIZE, SIZE1-i);
         lfs_file_read(&lfs, &file, buffer, chunk) => chunk;
         for (lfs_size_t b = 0; b < chunk; b++) {
-            assert(buffer[b] == (rand() & 0xff));
+            assert(buffer[b] == (TEST_PRNG(&prng) & 0xff));
         }
     }
     lfs_file_read(&lfs, &file, buffer, CHUNKSIZE) => 0;
@@ -178,13 +190,13 @@ code = '''
     lfs_unmount(&lfs) => 0;
 
     // append
-    lfs_mount(&lfs, &cfg) => 0;
+    lfs_mount(&lfs, cfg) => 0;
     lfs_file_open(&lfs, &file, "avacado", LFS_O_WRONLY | LFS_O_APPEND) => 0;
-    srand(2);
+    prng = 2;
     for (lfs_size_t i = 0; i < SIZE2; i += CHUNKSIZE) {
         lfs_size_t chunk = lfs_min(CHUNKSIZE, SIZE2-i);
         for (lfs_size_t b = 0; b < chunk; b++) {
-            buffer[b] = rand() & 0xff;
+            buffer[b] = TEST_PRNG(&prng) & 0xff;
         }
         lfs_file_write(&lfs, &file, buffer, chunk) => chunk;
     }
@@ -192,23 +204,23 @@ code = '''
     lfs_unmount(&lfs) => 0;
 
     // read
-    lfs_mount(&lfs, &cfg) => 0;
+    lfs_mount(&lfs, cfg) => 0;
     lfs_file_open(&lfs, &file, "avacado", LFS_O_RDONLY) => 0;
     lfs_file_size(&lfs, &file) => SIZE1 + SIZE2;
-    srand(1);
+    prng = 1;
     for (lfs_size_t i = 0; i < SIZE1; i += CHUNKSIZE) {
         lfs_size_t chunk = lfs_min(CHUNKSIZE, SIZE1-i);
         lfs_file_read(&lfs, &file, buffer, chunk) => chunk;
         for (lfs_size_t b = 0; b < chunk; b++) {
-            assert(buffer[b] == (rand() & 0xff));
+            assert(buffer[b] == (TEST_PRNG(&prng) & 0xff));
         }
     }
-    srand(2);
+    prng = 2;
     for (lfs_size_t i = 0; i < SIZE2; i += CHUNKSIZE) {
         lfs_size_t chunk = lfs_min(CHUNKSIZE, SIZE2-i);
         lfs_file_read(&lfs, &file, buffer, chunk) => chunk;
         for (lfs_size_t b = 0; b < chunk; b++) {
-            assert(buffer[b] == (rand() & 0xff));
+            assert(buffer[b] == (TEST_PRNG(&prng) & 0xff));
         }
     }
     lfs_file_read(&lfs, &file, buffer, CHUNKSIZE) => 0;
@@ -216,22 +228,25 @@ code = '''
     lfs_unmount(&lfs) => 0;
 '''
 
-[[case]] # truncating files
-define.SIZE1 = [32, 8192, 131072, 0, 7, 8193]
-define.SIZE2 = [32, 8192, 131072, 0, 7, 8193]
-define.CHUNKSIZE = [31, 16, 1]
+[cases.test_files_truncate]
+defines.SIZE1 = [32, 8192, 131072, 0, 7, 8193]
+defines.SIZE2 = [32, 8192, 131072, 0, 7, 8193]
+defines.CHUNKSIZE = [31, 16, 1]
 code = '''
-    lfs_format(&lfs, &cfg) => 0;
+    lfs_t lfs;
+    lfs_format(&lfs, cfg) => 0;
 
     // write
-    lfs_mount(&lfs, &cfg) => 0;
+    lfs_mount(&lfs, cfg) => 0;
+    lfs_file_t file;
+    uint8_t buffer[1024];
     lfs_file_open(&lfs, &file, "avacado",
             LFS_O_WRONLY | LFS_O_CREAT | LFS_O_EXCL) => 0;
-    srand(1);
+    uint32_t prng = 1;
     for (lfs_size_t i = 0; i < SIZE1; i += CHUNKSIZE) {
         lfs_size_t chunk = lfs_min(CHUNKSIZE, SIZE1-i);
         for (lfs_size_t b = 0; b < chunk; b++) {
-            buffer[b] = rand() & 0xff;
+            buffer[b] = TEST_PRNG(&prng) & 0xff;
         }
         lfs_file_write(&lfs, &file, buffer, chunk) => chunk;
     }
@@ -239,15 +254,15 @@ code = '''
     lfs_unmount(&lfs) => 0;
 
     // read
-    lfs_mount(&lfs, &cfg) => 0;
+    lfs_mount(&lfs, cfg) => 0;
     lfs_file_open(&lfs, &file, "avacado", LFS_O_RDONLY) => 0;
     lfs_file_size(&lfs, &file) => SIZE1;
-    srand(1);
+    prng = 1;
     for (lfs_size_t i = 0; i < SIZE1; i += CHUNKSIZE) {
         lfs_size_t chunk = lfs_min(CHUNKSIZE, SIZE1-i);
         lfs_file_read(&lfs, &file, buffer, chunk) => chunk;
         for (lfs_size_t b = 0; b < chunk; b++) {
-            assert(buffer[b] == (rand() & 0xff));
+            assert(buffer[b] == (TEST_PRNG(&prng) & 0xff));
         }
     }
     lfs_file_read(&lfs, &file, buffer, CHUNKSIZE) => 0;
@@ -255,13 +270,13 @@ code = '''
     lfs_unmount(&lfs) => 0;
 
     // truncate
-    lfs_mount(&lfs, &cfg) => 0;
+    lfs_mount(&lfs, cfg) => 0;
     lfs_file_open(&lfs, &file, "avacado", LFS_O_WRONLY | LFS_O_TRUNC) => 0;
-    srand(2);
+    prng = 2;
     for (lfs_size_t i = 0; i < SIZE2; i += CHUNKSIZE) {
         lfs_size_t chunk = lfs_min(CHUNKSIZE, SIZE2-i);
         for (lfs_size_t b = 0; b < chunk; b++) {
-            buffer[b] = rand() & 0xff;
+            buffer[b] = TEST_PRNG(&prng) & 0xff;
         }
         lfs_file_write(&lfs, &file, buffer, chunk) => chunk;
     }
@@ -269,15 +284,15 @@ code = '''
     lfs_unmount(&lfs) => 0;
 
     // read
-    lfs_mount(&lfs, &cfg) => 0;
+    lfs_mount(&lfs, cfg) => 0;
     lfs_file_open(&lfs, &file, "avacado", LFS_O_RDONLY) => 0;
     lfs_file_size(&lfs, &file) => SIZE2;
-    srand(2);
+    prng = 2;
     for (lfs_size_t i = 0; i < SIZE2; i += CHUNKSIZE) {
         lfs_size_t chunk = lfs_min(CHUNKSIZE, SIZE2-i);
         lfs_file_read(&lfs, &file, buffer, chunk) => chunk;
         for (lfs_size_t b = 0; b < chunk; b++) {
-            assert(buffer[b] == (rand() & 0xff));
+            assert(buffer[b] == (TEST_PRNG(&prng) & 0xff));
         }
     }
     lfs_file_read(&lfs, &file, buffer, CHUNKSIZE) => 0;
@@ -285,33 +300,36 @@ code = '''
     lfs_unmount(&lfs) => 0;
 '''
 
-[[case]] # reentrant file writing
-define.SIZE = [32, 0, 7, 2049]
-define.CHUNKSIZE = [31, 16, 65]
+[cases.test_files_reentrant_write]
+defines.SIZE = [32, 0, 7, 2049]
+defines.CHUNKSIZE = [31, 16, 65]
 reentrant = true
 code = '''
-    err = lfs_mount(&lfs, &cfg);
+    lfs_t lfs;
+    int err = lfs_mount(&lfs, cfg);
     if (err) {
-        lfs_format(&lfs, &cfg) => 0;
-        lfs_mount(&lfs, &cfg) => 0;
+        lfs_format(&lfs, cfg) => 0;
+        lfs_mount(&lfs, cfg) => 0;
     }
 
+    lfs_file_t file;
+    uint8_t buffer[1024];
     err = lfs_file_open(&lfs, &file, "avacado", LFS_O_RDONLY);
     assert(err == LFS_ERR_NOENT || err == 0);
     if (err == 0) {
         // can only be 0 (new file) or full size
-        size = lfs_file_size(&lfs, &file);
+        lfs_size_t size = lfs_file_size(&lfs, &file);
         assert(size == 0 || size == SIZE);
         lfs_file_close(&lfs, &file) => 0;
     }
 
     // write
     lfs_file_open(&lfs, &file, "avacado", LFS_O_WRONLY | LFS_O_CREAT) => 0;
-    srand(1);
+    uint32_t prng = 1;
     for (lfs_size_t i = 0; i < SIZE; i += CHUNKSIZE) {
         lfs_size_t chunk = lfs_min(CHUNKSIZE, SIZE-i);
         for (lfs_size_t b = 0; b < chunk; b++) {
-            buffer[b] = rand() & 0xff;
+            buffer[b] = TEST_PRNG(&prng) & 0xff;
         }
         lfs_file_write(&lfs, &file, buffer, chunk) => chunk;
     }
@@ -320,12 +338,12 @@ code = '''
     // read
     lfs_file_open(&lfs, &file, "avacado", LFS_O_RDONLY) => 0;
     lfs_file_size(&lfs, &file) => SIZE;
-    srand(1);
+    prng = 1;
     for (lfs_size_t i = 0; i < SIZE; i += CHUNKSIZE) {
         lfs_size_t chunk = lfs_min(CHUNKSIZE, SIZE-i);
         lfs_file_read(&lfs, &file, buffer, chunk) => chunk;
         for (lfs_size_t b = 0; b < chunk; b++) {
-            assert(buffer[b] == (rand() & 0xff));
+            assert(buffer[b] == (TEST_PRNG(&prng) & 0xff));
         }
     }
     lfs_file_read(&lfs, &file, buffer, CHUNKSIZE) => 0;
@@ -333,8 +351,8 @@ code = '''
     lfs_unmount(&lfs) => 0;
 '''
 
-[[case]] # reentrant file writing with syncs
-define = [
+[cases.test_files_reentrant_write_sync]
+defines = [
     # append (O(n))
     {MODE='LFS_O_APPEND',   SIZE=[32, 0, 7, 2049],  CHUNKSIZE=[31, 16, 65]},
     # truncate (O(n^2))
@@ -344,24 +362,27 @@ define = [
 ]
 reentrant = true
 code = '''
-    err = lfs_mount(&lfs, &cfg);
+    lfs_t lfs;
+    int err = lfs_mount(&lfs, cfg);
     if (err) {
-        lfs_format(&lfs, &cfg) => 0;
-        lfs_mount(&lfs, &cfg) => 0;
+        lfs_format(&lfs, cfg) => 0;
+        lfs_mount(&lfs, cfg) => 0;
     }
 
+    lfs_file_t file;
+    uint8_t buffer[1024];
     err = lfs_file_open(&lfs, &file, "avacado", LFS_O_RDONLY);
     assert(err == LFS_ERR_NOENT || err == 0);
     if (err == 0) {
         // with syncs we could be any size, but it at least must be valid data
-        size = lfs_file_size(&lfs, &file);
+        lfs_size_t size = lfs_file_size(&lfs, &file);
         assert(size <= SIZE);
-        srand(1);
+        uint32_t prng = 1;
         for (lfs_size_t i = 0; i < size; i += CHUNKSIZE) {
             lfs_size_t chunk = lfs_min(CHUNKSIZE, size-i);
             lfs_file_read(&lfs, &file, buffer, chunk) => chunk;
             for (lfs_size_t b = 0; b < chunk; b++) {
-                assert(buffer[b] == (rand() & 0xff));
+                assert(buffer[b] == (TEST_PRNG(&prng) & 0xff));
             }
         }
         lfs_file_close(&lfs, &file) => 0;
@@ -370,17 +391,17 @@ code = '''
     // write
     lfs_file_open(&lfs, &file, "avacado",
         LFS_O_WRONLY | LFS_O_CREAT | MODE) => 0;
-    size = lfs_file_size(&lfs, &file);
+    lfs_size_t size = lfs_file_size(&lfs, &file);
     assert(size <= SIZE);
-    srand(1);
+    uint32_t prng = 1;
     lfs_size_t skip = (MODE == LFS_O_APPEND) ? size : 0;
     for (lfs_size_t b = 0; b < skip; b++) {
-        rand();
+        TEST_PRNG(&prng);
     }
     for (lfs_size_t i = skip; i < SIZE; i += CHUNKSIZE) {
         lfs_size_t chunk = lfs_min(CHUNKSIZE, SIZE-i);
         for (lfs_size_t b = 0; b < chunk; b++) {
-            buffer[b] = rand() & 0xff;
+            buffer[b] = TEST_PRNG(&prng) & 0xff;
         }
         lfs_file_write(&lfs, &file, buffer, chunk) => chunk;
         lfs_file_sync(&lfs, &file) => 0;
@@ -390,12 +411,12 @@ code = '''
     // read
     lfs_file_open(&lfs, &file, "avacado", LFS_O_RDONLY) => 0;
     lfs_file_size(&lfs, &file) => SIZE;
-    srand(1);
+    prng = 1;
     for (lfs_size_t i = 0; i < SIZE; i += CHUNKSIZE) {
         lfs_size_t chunk = lfs_min(CHUNKSIZE, SIZE-i);
         lfs_file_read(&lfs, &file, buffer, chunk) => chunk;
         for (lfs_size_t b = 0; b < chunk; b++) {
-            assert(buffer[b] == (rand() & 0xff));
+            assert(buffer[b] == (TEST_PRNG(&prng) & 0xff));
         }
     }
     lfs_file_read(&lfs, &file, buffer, CHUNKSIZE) => 0;
@@ -403,19 +424,22 @@ code = '''
     lfs_unmount(&lfs) => 0;
 '''
 
-[[case]] # many files
-define.N = 300
+[cases.test_files_many]
+defines.N = 300
 code = '''
-    lfs_format(&lfs, &cfg) => 0;
+    lfs_t lfs;
+    lfs_format(&lfs, cfg) => 0;
     // create N files of 7 bytes
-    lfs_mount(&lfs, &cfg) => 0;
+    lfs_mount(&lfs, cfg) => 0;
     for (int i = 0; i < N; i++) {
+        lfs_file_t file;
+        char path[1024];
         sprintf(path, "file_%03d", i);
         lfs_file_open(&lfs, &file, path,
                 LFS_O_WRONLY | LFS_O_CREAT | LFS_O_EXCL) => 0;
         char wbuffer[1024];
-        size = 7;
-        snprintf(wbuffer, size, "Hi %03d", i);
+        lfs_size_t size = 7;
+        sprintf(wbuffer, "Hi %03d", i);
         lfs_file_write(&lfs, &file, wbuffer, size) => size;
         lfs_file_close(&lfs, &file) => 0;
 
@@ -428,25 +452,28 @@ code = '''
     lfs_unmount(&lfs) => 0;
 '''
 
-[[case]] # many files with power cycle
-define.N = 300
+[cases.test_files_many_power_cycle]
+defines.N = 300
 code = '''
-    lfs_format(&lfs, &cfg) => 0;
+    lfs_t lfs;
+    lfs_format(&lfs, cfg) => 0;
     // create N files of 7 bytes
-    lfs_mount(&lfs, &cfg) => 0;
+    lfs_mount(&lfs, cfg) => 0;
     for (int i = 0; i < N; i++) {
+        lfs_file_t file;
+        char path[1024];
         sprintf(path, "file_%03d", i);
         lfs_file_open(&lfs, &file, path,
                 LFS_O_WRONLY | LFS_O_CREAT | LFS_O_EXCL) => 0;
         char wbuffer[1024];
-        size = 7;
-        snprintf(wbuffer, size, "Hi %03d", i);
+        lfs_size_t size = 7;
+        sprintf(wbuffer, "Hi %03d", i);
         lfs_file_write(&lfs, &file, wbuffer, size) => size;
         lfs_file_close(&lfs, &file) => 0;
         lfs_unmount(&lfs) => 0;
 
         char rbuffer[1024];
-        lfs_mount(&lfs, &cfg) => 0;
+        lfs_mount(&lfs, cfg) => 0;
         lfs_file_open(&lfs, &file, path, LFS_O_RDONLY) => 0;
         lfs_file_read(&lfs, &file, rbuffer, size) => size;
         assert(strcmp(rbuffer, wbuffer) == 0);
@@ -455,22 +482,25 @@ code = '''
     lfs_unmount(&lfs) => 0;
 '''
 
-[[case]] # many files with power loss
-define.N = 300
+[cases.test_files_many_power_loss]
+defines.N = 300
 reentrant = true
 code = '''
-    err = lfs_mount(&lfs, &cfg);
+    lfs_t lfs;
+    int err = lfs_mount(&lfs, cfg);
     if (err) {
-        lfs_format(&lfs, &cfg) => 0;
-        lfs_mount(&lfs, &cfg) => 0;
+        lfs_format(&lfs, cfg) => 0;
+        lfs_mount(&lfs, cfg) => 0;
     }
     // create N files of 7 bytes
     for (int i = 0; i < N; i++) {
+        lfs_file_t file;
+        char path[1024];
         sprintf(path, "file_%03d", i);
         err = lfs_file_open(&lfs, &file, path, LFS_O_WRONLY | LFS_O_CREAT);
         char wbuffer[1024];
-        size = 7;
-        snprintf(wbuffer, size, "Hi %03d", i);
+        lfs_size_t size = 7;
+        sprintf(wbuffer, "Hi %03d", i);
         if ((lfs_size_t)lfs_file_size(&lfs, &file) != size) {
             lfs_file_write(&lfs, &file, wbuffer, size) => size;
         }
diff --git a/tests/test_interspersed.toml b/tests/test_interspersed.toml
index 87a0578..d7143f6 100644
--- a/tests/test_interspersed.toml
+++ b/tests/test_interspersed.toml
@@ -1,13 +1,15 @@
 
-[[case]] # interspersed file test
-define.SIZE = [10, 100]
-define.FILES = [4, 10, 26] 
+[cases.test_interspersed_files]
+defines.SIZE = [10, 100]
+defines.FILES = [4, 10, 26] 
 code = '''
+    lfs_t lfs;
     lfs_file_t files[FILES];
     const char alphas[] = "abcdefghijklmnopqrstuvwxyz";
-    lfs_format(&lfs, &cfg) => 0;
-    lfs_mount(&lfs, &cfg) => 0;
+    lfs_format(&lfs, cfg) => 0;
+    lfs_mount(&lfs, cfg) => 0;
     for (int j = 0; j < FILES; j++) {
+        char path[1024];
         sprintf(path, "%c", alphas[j]);
         lfs_file_open(&lfs, &files[j], path,
                 LFS_O_WRONLY | LFS_O_CREAT | LFS_O_EXCL) => 0;
@@ -23,7 +25,9 @@ code = '''
         lfs_file_close(&lfs, &files[j]);
     }
 
+    lfs_dir_t dir;
     lfs_dir_open(&lfs, &dir, "/") => 0;
+    struct lfs_info info;
     lfs_dir_read(&lfs, &dir, &info) => 1;
     assert(strcmp(info.name, ".") == 0);
     assert(info.type == LFS_TYPE_DIR);
@@ -31,6 +35,7 @@ code = '''
     assert(strcmp(info.name, "..") == 0);
     assert(info.type == LFS_TYPE_DIR);
     for (int j = 0; j < FILES; j++) {
+        char path[1024];
         sprintf(path, "%c", alphas[j]);
         lfs_dir_read(&lfs, &dir, &info) => 1;
         assert(strcmp(info.name, path) == 0);
@@ -41,12 +46,14 @@ code = '''
     lfs_dir_close(&lfs, &dir) => 0;
 
     for (int j = 0; j < FILES; j++) {
+        char path[1024];
         sprintf(path, "%c", alphas[j]);
         lfs_file_open(&lfs, &files[j], path, LFS_O_RDONLY) => 0;
     }
 
     for (int i = 0; i < 10; i++) {
         for (int j = 0; j < FILES; j++) {
+            uint8_t buffer[1024];
             lfs_file_read(&lfs, &files[j], buffer, 1) => 1;
             assert(buffer[0] == alphas[j]);
         }
@@ -59,15 +66,18 @@ code = '''
     lfs_unmount(&lfs) => 0;
 '''
 
-[[case]] # interspersed remove file test
-define.SIZE = [10, 100]
-define.FILES = [4, 10, 26]
+[cases.test_interspersed_remove_files]
+defines.SIZE = [10, 100]
+defines.FILES = [4, 10, 26]
 code = '''
+    lfs_t lfs;
     const char alphas[] = "abcdefghijklmnopqrstuvwxyz";
-    lfs_format(&lfs, &cfg) => 0;
-    lfs_mount(&lfs, &cfg) => 0;
+    lfs_format(&lfs, cfg) => 0;
+    lfs_mount(&lfs, cfg) => 0;
     for (int j = 0; j < FILES; j++) {
+        char path[1024];
         sprintf(path, "%c", alphas[j]);
+        lfs_file_t file;
         lfs_file_open(&lfs, &file, path,
                 LFS_O_WRONLY | LFS_O_CREAT | LFS_O_EXCL) => 0;
         for (int i = 0; i < SIZE; i++) {
@@ -77,18 +87,22 @@ code = '''
     }
     lfs_unmount(&lfs) => 0;
 
-    lfs_mount(&lfs, &cfg) => 0;
+    lfs_mount(&lfs, cfg) => 0;
+    lfs_file_t file;
     lfs_file_open(&lfs, &file, "zzz", LFS_O_WRONLY | LFS_O_CREAT) => 0;
     for (int j = 0; j < FILES; j++) {
         lfs_file_write(&lfs, &file, (const void*)"~", 1) => 1;
         lfs_file_sync(&lfs, &file) => 0;
 
+        char path[1024];
         sprintf(path, "%c", alphas[j]);
         lfs_remove(&lfs, path) => 0;
     }
     lfs_file_close(&lfs, &file);
 
+    lfs_dir_t dir;
     lfs_dir_open(&lfs, &dir, "/") => 0;
+    struct lfs_info info;
     lfs_dir_read(&lfs, &dir, &info) => 1;
     assert(strcmp(info.name, ".") == 0);
     assert(info.type == LFS_TYPE_DIR);
@@ -104,6 +118,7 @@ code = '''
 
     lfs_file_open(&lfs, &file, "zzz", LFS_O_RDONLY) => 0;
     for (int i = 0; i < FILES; i++) {
+        uint8_t buffer[1024];
         lfs_file_read(&lfs, &file, buffer, 1) => 1;
         assert(buffer[0] == '~');
     }
@@ -112,11 +127,12 @@ code = '''
     lfs_unmount(&lfs) => 0;
 '''
 
-[[case]] # remove inconveniently test
-define.SIZE = [10, 100]
+[cases.test_interspersed_remove_inconveniently]
+defines.SIZE = [10, 100]
 code = '''
-    lfs_format(&lfs, &cfg) => 0;
-    lfs_mount(&lfs, &cfg) => 0;
+    lfs_t lfs;
+    lfs_format(&lfs, cfg) => 0;
+    lfs_mount(&lfs, cfg) => 0;
     lfs_file_t files[3];
     lfs_file_open(&lfs, &files[0], "e", LFS_O_WRONLY | LFS_O_CREAT) => 0;
     lfs_file_open(&lfs, &files[1], "f", LFS_O_WRONLY | LFS_O_CREAT) => 0;
@@ -140,7 +156,9 @@ code = '''
     lfs_file_close(&lfs, &files[1]);
     lfs_file_close(&lfs, &files[2]);
 
+    lfs_dir_t dir;
     lfs_dir_open(&lfs, &dir, "/") => 0;
+    struct lfs_info info;
     lfs_dir_read(&lfs, &dir, &info) => 1;
     assert(strcmp(info.name, ".") == 0);
     assert(info.type == LFS_TYPE_DIR);
@@ -161,6 +179,7 @@ code = '''
     lfs_file_open(&lfs, &files[0], "e", LFS_O_RDONLY) => 0;
     lfs_file_open(&lfs, &files[1], "g", LFS_O_RDONLY) => 0;
     for (int i = 0; i < SIZE; i++) {
+        uint8_t buffer[1024];
         lfs_file_read(&lfs, &files[0], buffer, 1) => 1;
         assert(buffer[0] == 'e');
         lfs_file_read(&lfs, &files[1], buffer, 1) => 1;
@@ -172,21 +191,23 @@ code = '''
     lfs_unmount(&lfs) => 0;
 '''
 
-[[case]] # reentrant interspersed file test
-define.SIZE = [10, 100]
-define.FILES = [4, 10, 26] 
+[cases.test_interspersed_reentrant_files]
+defines.SIZE = [10, 100]
+defines.FILES = [4, 10, 26] 
 reentrant = true
 code = '''
+    lfs_t lfs;
     lfs_file_t files[FILES];
     const char alphas[] = "abcdefghijklmnopqrstuvwxyz";
 
-    err = lfs_mount(&lfs, &cfg);
+    int err = lfs_mount(&lfs, cfg);
     if (err) {
-        lfs_format(&lfs, &cfg) => 0;
-        lfs_mount(&lfs, &cfg) => 0;
+        lfs_format(&lfs, cfg) => 0;
+        lfs_mount(&lfs, cfg) => 0;
     }
 
     for (int j = 0; j < FILES; j++) {
+        char path[1024];
         sprintf(path, "%c", alphas[j]);
         lfs_file_open(&lfs, &files[j], path,
                 LFS_O_WRONLY | LFS_O_CREAT | LFS_O_APPEND) => 0;
@@ -194,8 +215,8 @@ code = '''
 
     for (int i = 0; i < SIZE; i++) {
         for (int j = 0; j < FILES; j++) {
-            size = lfs_file_size(&lfs, &files[j]);
-            assert((int)size >= 0);
+            lfs_ssize_t size = lfs_file_size(&lfs, &files[j]);
+            assert(size >= 0);
             if ((int)size <= i) {
                 lfs_file_write(&lfs, &files[j], &alphas[j], 1) => 1;
                 lfs_file_sync(&lfs, &files[j]) => 0;
@@ -207,7 +228,9 @@ code = '''
         lfs_file_close(&lfs, &files[j]);
     }
 
+    lfs_dir_t dir;
     lfs_dir_open(&lfs, &dir, "/") => 0;
+    struct lfs_info info;
     lfs_dir_read(&lfs, &dir, &info) => 1;
     assert(strcmp(info.name, ".") == 0);
     assert(info.type == LFS_TYPE_DIR);
@@ -215,6 +238,7 @@ code = '''
     assert(strcmp(info.name, "..") == 0);
     assert(info.type == LFS_TYPE_DIR);
     for (int j = 0; j < FILES; j++) {
+        char path[1024];
         sprintf(path, "%c", alphas[j]);
         lfs_dir_read(&lfs, &dir, &info) => 1;
         assert(strcmp(info.name, path) == 0);
@@ -225,12 +249,14 @@ code = '''
     lfs_dir_close(&lfs, &dir) => 0;
 
     for (int j = 0; j < FILES; j++) {
+        char path[1024];
         sprintf(path, "%c", alphas[j]);
         lfs_file_open(&lfs, &files[j], path, LFS_O_RDONLY) => 0;
     }
 
     for (int i = 0; i < 10; i++) {
         for (int j = 0; j < FILES; j++) {
+            uint8_t buffer[1024];
             lfs_file_read(&lfs, &files[j], buffer, 1) => 1;
             assert(buffer[0] == alphas[j]);
         }
diff --git a/tests/test_move.toml b/tests/test_move.toml
index bb3b713..0537f48 100644
--- a/tests/test_move.toml
+++ b/tests/test_move.toml
@@ -1,11 +1,13 @@
-[[case]] # move file
+[cases.test_move_file]
 code = '''
-    lfs_format(&lfs, &cfg) => 0;
-    lfs_mount(&lfs, &cfg) => 0;
+    lfs_t lfs;
+    lfs_format(&lfs, cfg) => 0;
+    lfs_mount(&lfs, cfg) => 0;
     lfs_mkdir(&lfs, "a") => 0;
     lfs_mkdir(&lfs, "b") => 0;
     lfs_mkdir(&lfs, "c") => 0;
     lfs_mkdir(&lfs, "d") => 0;
+    lfs_file_t file;
     lfs_file_open(&lfs, &file, "a/hello", LFS_O_CREAT | LFS_O_WRONLY) => 0;
     lfs_file_write(&lfs, &file, "hola\n", 5) => 5;
     lfs_file_write(&lfs, &file, "bonjour\n", 8) => 8;
@@ -13,11 +15,13 @@ code = '''
     lfs_file_close(&lfs, &file) => 0;
     lfs_unmount(&lfs) => 0;
 
-    lfs_mount(&lfs, &cfg) => 0;
+    lfs_mount(&lfs, cfg) => 0;
     lfs_rename(&lfs, "a/hello", "c/hello") => 0;
     lfs_unmount(&lfs) => 0;
 
-    lfs_mount(&lfs, &cfg) => 0;
+    lfs_mount(&lfs, cfg) => 0;
+    lfs_dir_t dir;
+    struct lfs_info info;
     lfs_dir_open(&lfs, &dir, "a") => 0;
     lfs_dir_read(&lfs, &dir, &info) => 1;
     assert(strcmp(info.name, ".") == 0);
@@ -44,6 +48,7 @@ code = '''
     lfs_file_open(&lfs, &file, "a/hello", LFS_O_RDONLY) => LFS_ERR_NOENT;
     lfs_file_open(&lfs, &file, "b/hello", LFS_O_RDONLY) => LFS_ERR_NOENT;
     lfs_file_open(&lfs, &file, "c/hello", LFS_O_RDONLY) => 0;
+    uint8_t buffer[1024];
     lfs_file_read(&lfs, &file, buffer, 5) => 5;
     memcmp(buffer, "hola\n", 5) => 0;
     lfs_file_read(&lfs, &file, buffer, 8) => 8;
@@ -55,31 +60,35 @@ code = '''
     lfs_unmount(&lfs) => 0;
 '''
 
-[[case]] # noop move, yes this is legal
+[cases.test_move_nop] # yes this is legal
 code = '''
-    lfs_format(&lfs, &cfg) => 0;
-    lfs_mount(&lfs, &cfg) => 0;
+    lfs_t lfs;
+    lfs_format(&lfs, cfg) => 0;
+    lfs_mount(&lfs, cfg) => 0;
     lfs_mkdir(&lfs, "hi") => 0;
     lfs_rename(&lfs, "hi", "hi") => 0;
     lfs_mkdir(&lfs, "hi/hi") => 0;
     lfs_rename(&lfs, "hi/hi", "hi/hi") => 0;
     lfs_mkdir(&lfs, "hi/hi/hi") => 0;
     lfs_rename(&lfs, "hi/hi/hi", "hi/hi/hi") => 0;
+    struct lfs_info info;
     lfs_stat(&lfs, "hi/hi/hi", &info) => 0;
     assert(strcmp(info.name, "hi") == 0);
     assert(info.type == LFS_TYPE_DIR);
     lfs_unmount(&lfs) => 0;
 '''
 
-[[case]] # move file corrupt source
+[cases.test_move_file_corrupt_source]
 in = "lfs.c"
 code = '''
-    lfs_format(&lfs, &cfg) => 0;
-    lfs_mount(&lfs, &cfg) => 0;
+    lfs_t lfs;
+    lfs_format(&lfs, cfg) => 0;
+    lfs_mount(&lfs, cfg) => 0;
     lfs_mkdir(&lfs, "a") => 0;
     lfs_mkdir(&lfs, "b") => 0;
     lfs_mkdir(&lfs, "c") => 0;
     lfs_mkdir(&lfs, "d") => 0;
+    lfs_file_t file;
     lfs_file_open(&lfs, &file, "a/hello", LFS_O_CREAT | LFS_O_WRONLY) => 0;
     lfs_file_write(&lfs, &file, "hola\n", 5) => 5;
     lfs_file_write(&lfs, &file, "bonjour\n", 8) => 8;
@@ -87,28 +96,30 @@ code = '''
     lfs_file_close(&lfs, &file) => 0;
     lfs_unmount(&lfs) => 0;
 
-    lfs_mount(&lfs, &cfg) => 0;
+    lfs_mount(&lfs, cfg) => 0;
     lfs_rename(&lfs, "a/hello", "c/hello") => 0;
     lfs_unmount(&lfs) => 0;
 
     // corrupt the source
-    lfs_mount(&lfs, &cfg) => 0;
+    lfs_mount(&lfs, cfg) => 0;
+    lfs_dir_t dir;
+    struct lfs_info info;
     lfs_dir_open(&lfs, &dir, "a") => 0;
     lfs_block_t block = dir.m.pair[0];
     lfs_dir_close(&lfs, &dir) => 0;
     lfs_unmount(&lfs) => 0;
-    uint8_t bbuffer[LFS_BLOCK_SIZE];
-    cfg.read(&cfg, block, 0, bbuffer, LFS_BLOCK_SIZE) => 0;
-    int off = LFS_BLOCK_SIZE-1;
-    while (off >= 0 && bbuffer[off] == LFS_ERASE_VALUE) {
+    uint8_t buffer[BLOCK_SIZE];
+    cfg->read(cfg, block, 0, buffer, BLOCK_SIZE) => 0;
+    int off = BLOCK_SIZE-1;
+    while (off >= 0 && buffer[off] == ERASE_VALUE) {
         off -= 1;
     }
-    memset(&bbuffer[off-3], LFS_BLOCK_SIZE, 3);
-    cfg.erase(&cfg, block) => 0;
-    cfg.prog(&cfg, block, 0, bbuffer, LFS_BLOCK_SIZE) => 0;
-    cfg.sync(&cfg) => 0;
+    memset(&buffer[off-3], BLOCK_SIZE, 3);
+    cfg->erase(cfg, block) => 0;
+    cfg->prog(cfg, block, 0, buffer, BLOCK_SIZE) => 0;
+    cfg->sync(cfg) => 0;
 
-    lfs_mount(&lfs, &cfg) => 0;
+    lfs_mount(&lfs, cfg) => 0;
     lfs_dir_open(&lfs, &dir, "a") => 0;
     lfs_dir_read(&lfs, &dir, &info) => 1;
     assert(strcmp(info.name, ".") == 0);
@@ -146,16 +157,19 @@ code = '''
     lfs_unmount(&lfs) => 0;
 '''
 
-[[case]] # move file corrupt source and dest
+# move file corrupt source and dest
+[cases.test_move_file_corrupt_source_dest]
 in = "lfs.c"
-if = 'LFS_PROG_SIZE <= 0x3fe' # only works with one crc per commit
+if = 'PROG_SIZE <= 0x3fe' # only works with one crc per commit
 code = '''
-    lfs_format(&lfs, &cfg) => 0;
-    lfs_mount(&lfs, &cfg) => 0;
+    lfs_t lfs;
+    lfs_format(&lfs, cfg) => 0;
+    lfs_mount(&lfs, cfg) => 0;
     lfs_mkdir(&lfs, "a") => 0;
     lfs_mkdir(&lfs, "b") => 0;
     lfs_mkdir(&lfs, "c") => 0;
     lfs_mkdir(&lfs, "d") => 0;
+    lfs_file_t file;
     lfs_file_open(&lfs, &file, "a/hello", LFS_O_CREAT | LFS_O_WRONLY) => 0;
     lfs_file_write(&lfs, &file, "hola\n", 5) => 5;
     lfs_file_write(&lfs, &file, "bonjour\n", 8) => 8;
@@ -163,44 +177,46 @@ code = '''
     lfs_file_close(&lfs, &file) => 0;
     lfs_unmount(&lfs) => 0;
 
-    lfs_mount(&lfs, &cfg) => 0;
+    lfs_mount(&lfs, cfg) => 0;
     lfs_rename(&lfs, "a/hello", "c/hello") => 0;
     lfs_unmount(&lfs) => 0;
 
     // corrupt the source
-    lfs_mount(&lfs, &cfg) => 0;
+    lfs_mount(&lfs, cfg) => 0;
+    lfs_dir_t dir;
+    struct lfs_info info;
     lfs_dir_open(&lfs, &dir, "a") => 0;
     lfs_block_t block = dir.m.pair[0];
     lfs_dir_close(&lfs, &dir) => 0;
     lfs_unmount(&lfs) => 0;
-    uint8_t bbuffer[LFS_BLOCK_SIZE];
-    cfg.read(&cfg, block, 0, bbuffer, LFS_BLOCK_SIZE) => 0;
-    int off = LFS_BLOCK_SIZE-1;
-    while (off >= 0 && bbuffer[off] == LFS_ERASE_VALUE) {
+    uint8_t buffer[BLOCK_SIZE];
+    cfg->read(cfg, block, 0, buffer, BLOCK_SIZE) => 0;
+    int off = BLOCK_SIZE-1;
+    while (off >= 0 && buffer[off] == ERASE_VALUE) {
         off -= 1;
     }
-    memset(&bbuffer[off-3], LFS_BLOCK_SIZE, 3);
-    cfg.erase(&cfg, block) => 0;
-    cfg.prog(&cfg, block, 0, bbuffer, LFS_BLOCK_SIZE) => 0;
-    cfg.sync(&cfg) => 0;
+    memset(&buffer[off-3], BLOCK_SIZE, 3);
+    cfg->erase(cfg, block) => 0;
+    cfg->prog(cfg, block, 0, buffer, BLOCK_SIZE) => 0;
+    cfg->sync(cfg) => 0;
 
     // corrupt the destination
-    lfs_mount(&lfs, &cfg) => 0;
+    lfs_mount(&lfs, cfg) => 0;
     lfs_dir_open(&lfs, &dir, "c") => 0;
     block = dir.m.pair[0];
     lfs_dir_close(&lfs, &dir) => 0;
     lfs_unmount(&lfs) => 0;
-    cfg.read(&cfg, block, 0, bbuffer, LFS_BLOCK_SIZE) => 0;
-    off = LFS_BLOCK_SIZE-1;
-    while (off >= 0 && bbuffer[off] == LFS_ERASE_VALUE) {
+    cfg->read(cfg, block, 0, buffer, BLOCK_SIZE) => 0;
+    off = BLOCK_SIZE-1;
+    while (off >= 0 && buffer[off] == ERASE_VALUE) {
         off -= 1;
     }
-    memset(&bbuffer[off-3], LFS_BLOCK_SIZE, 3);
-    cfg.erase(&cfg, block) => 0;
-    cfg.prog(&cfg, block, 0, bbuffer, LFS_BLOCK_SIZE) => 0;
-    cfg.sync(&cfg) => 0;
+    memset(&buffer[off-3], BLOCK_SIZE, 3);
+    cfg->erase(cfg, block) => 0;
+    cfg->prog(cfg, block, 0, buffer, BLOCK_SIZE) => 0;
+    cfg->sync(cfg) => 0;
 
-    lfs_mount(&lfs, &cfg) => 0;
+    lfs_mount(&lfs, cfg) => 0;
     lfs_dir_open(&lfs, &dir, "a") => 0;
     lfs_dir_read(&lfs, &dir, &info) => 1;
     assert(strcmp(info.name, ".") == 0);
@@ -238,16 +254,18 @@ code = '''
     lfs_unmount(&lfs) => 0;
 '''
 
-[[case]] # move file after corrupt
+[cases.test_move_file_after_corrupt]
 in = "lfs.c"
-if = 'LFS_PROG_SIZE <= 0x3fe' # only works with one crc per commit
+if = 'PROG_SIZE <= 0x3fe' # only works with one crc per commit
 code = '''
-    lfs_format(&lfs, &cfg) => 0;
-    lfs_mount(&lfs, &cfg) => 0;
+    lfs_t lfs;
+    lfs_format(&lfs, cfg) => 0;
+    lfs_mount(&lfs, cfg) => 0;
     lfs_mkdir(&lfs, "a") => 0;
     lfs_mkdir(&lfs, "b") => 0;
     lfs_mkdir(&lfs, "c") => 0;
     lfs_mkdir(&lfs, "d") => 0;
+    lfs_file_t file;
     lfs_file_open(&lfs, &file, "a/hello", LFS_O_CREAT | LFS_O_WRONLY) => 0;
     lfs_file_write(&lfs, &file, "hola\n", 5) => 5;
     lfs_file_write(&lfs, &file, "bonjour\n", 8) => 8;
@@ -255,49 +273,51 @@ code = '''
     lfs_file_close(&lfs, &file) => 0;
     lfs_unmount(&lfs) => 0;
 
-    lfs_mount(&lfs, &cfg) => 0;
+    lfs_mount(&lfs, cfg) => 0;
     lfs_rename(&lfs, "a/hello", "c/hello") => 0;
     lfs_unmount(&lfs) => 0;
 
     // corrupt the source
-    lfs_mount(&lfs, &cfg) => 0;
+    lfs_mount(&lfs, cfg) => 0;
+    lfs_dir_t dir;
+    struct lfs_info info;
     lfs_dir_open(&lfs, &dir, "a") => 0;
     lfs_block_t block = dir.m.pair[0];
     lfs_dir_close(&lfs, &dir) => 0;
     lfs_unmount(&lfs) => 0;
-    uint8_t bbuffer[LFS_BLOCK_SIZE];
-    cfg.read(&cfg, block, 0, bbuffer, LFS_BLOCK_SIZE) => 0;
-    int off = LFS_BLOCK_SIZE-1;
-    while (off >= 0 && bbuffer[off] == LFS_ERASE_VALUE) {
+    uint8_t buffer[BLOCK_SIZE];
+    cfg->read(cfg, block, 0, buffer, BLOCK_SIZE) => 0;
+    int off = BLOCK_SIZE-1;
+    while (off >= 0 && buffer[off] == ERASE_VALUE) {
         off -= 1;
     }
-    memset(&bbuffer[off-3], LFS_BLOCK_SIZE, 3);
-    cfg.erase(&cfg, block) => 0;
-    cfg.prog(&cfg, block, 0, bbuffer, LFS_BLOCK_SIZE) => 0;
-    cfg.sync(&cfg) => 0;
+    memset(&buffer[off-3], BLOCK_SIZE, 3);
+    cfg->erase(cfg, block) => 0;
+    cfg->prog(cfg, block, 0, buffer, BLOCK_SIZE) => 0;
+    cfg->sync(cfg) => 0;
 
     // corrupt the destination
-    lfs_mount(&lfs, &cfg) => 0;
+    lfs_mount(&lfs, cfg) => 0;
     lfs_dir_open(&lfs, &dir, "c") => 0;
     block = dir.m.pair[0];
     lfs_dir_close(&lfs, &dir) => 0;
     lfs_unmount(&lfs) => 0;
-    cfg.read(&cfg, block, 0, bbuffer, LFS_BLOCK_SIZE) => 0;
-    off = LFS_BLOCK_SIZE-1;
-    while (off >= 0 && bbuffer[off] == LFS_ERASE_VALUE) {
+    cfg->read(cfg, block, 0, buffer, BLOCK_SIZE) => 0;
+    off = BLOCK_SIZE-1;
+    while (off >= 0 && buffer[off] == ERASE_VALUE) {
         off -= 1;
     }
-    memset(&bbuffer[off-3], LFS_BLOCK_SIZE, 3);
-    cfg.erase(&cfg, block) => 0;
-    cfg.prog(&cfg, block, 0, bbuffer, LFS_BLOCK_SIZE) => 0;
-    cfg.sync(&cfg) => 0;
+    memset(&buffer[off-3], BLOCK_SIZE, 3);
+    cfg->erase(cfg, block) => 0;
+    cfg->prog(cfg, block, 0, buffer, BLOCK_SIZE) => 0;
+    cfg->sync(cfg) => 0;
 
     // continue move
-    lfs_mount(&lfs, &cfg) => 0;
+    lfs_mount(&lfs, cfg) => 0;
     lfs_rename(&lfs, "a/hello", "c/hello") => 0;
     lfs_unmount(&lfs) => 0;
 
-    lfs_mount(&lfs, &cfg) => 0;
+    lfs_mount(&lfs, cfg) => 0;
     lfs_dir_open(&lfs, &dir, "a") => 0;
     lfs_dir_read(&lfs, &dir, &info) => 1;
     assert(strcmp(info.name, ".") == 0);
@@ -335,13 +355,14 @@ code = '''
     lfs_unmount(&lfs) => 0;
 '''
 
-[[case]] # simple reentrant move file
+[cases.test_move_reentrant_file]
 reentrant = true
 code = '''
-    err = lfs_mount(&lfs, &cfg);
+    lfs_t lfs;
+    int err = lfs_mount(&lfs, cfg);
     if (err) {
-        lfs_format(&lfs, &cfg) => 0;
-        lfs_mount(&lfs, &cfg) => 0;
+        lfs_format(&lfs, cfg) => 0;
+        lfs_mount(&lfs, cfg) => 0;
     }
     err = lfs_mkdir(&lfs, "a");
     assert(!err || err == LFS_ERR_EXIST);
@@ -354,9 +375,10 @@ code = '''
     lfs_unmount(&lfs) => 0;
 
     while (true) {
-        lfs_mount(&lfs, &cfg) => 0;
+        lfs_mount(&lfs, cfg) => 0;
         // there should never exist _2_ hello files
         int count = 0;
+        struct lfs_info info;
         if (lfs_stat(&lfs, "a/hello", &info) == 0) {
             assert(strcmp(info.name, "hello") == 0);
             assert(info.type == LFS_TYPE_REG);
@@ -384,7 +406,7 @@ code = '''
         assert(count <= 1);
         lfs_unmount(&lfs) => 0;
 
-        lfs_mount(&lfs, &cfg) => 0;
+        lfs_mount(&lfs, cfg) => 0;
         if (lfs_stat(&lfs, "a/hello", &info) == 0 && info.size > 0) {
             lfs_rename(&lfs, "a/hello", "b/hello") => 0;
         } else if (lfs_stat(&lfs, "b/hello", &info) == 0) {
@@ -397,6 +419,7 @@ code = '''
             break;
         } else {
             // create file
+            lfs_file_t file;
             lfs_file_open(&lfs, &file, "a/hello",
                     LFS_O_WRONLY | LFS_O_CREAT) => 0;
             lfs_file_write(&lfs, &file, "hola\n", 5) => 5;
@@ -407,7 +430,9 @@ code = '''
         lfs_unmount(&lfs) => 0;
     }
 
-    lfs_mount(&lfs, &cfg) => 0;
+    lfs_mount(&lfs, cfg) => 0;
+    lfs_dir_t dir;
+    struct lfs_info info;
     lfs_dir_open(&lfs, &dir, "a") => 0;
     lfs_dir_read(&lfs, &dir, &info) => 1;
     assert(strcmp(info.name, ".") == 0);
@@ -431,10 +456,12 @@ code = '''
     lfs_dir_read(&lfs, &dir, &info) => 0;
     lfs_dir_close(&lfs, &dir) => 0;
 
+    lfs_file_t file;
     lfs_file_open(&lfs, &file, "a/hello", LFS_O_RDONLY) => LFS_ERR_NOENT;
     lfs_file_open(&lfs, &file, "b/hello", LFS_O_RDONLY) => LFS_ERR_NOENT;
     lfs_file_open(&lfs, &file, "c/hello", LFS_O_RDONLY) => LFS_ERR_NOENT;
     lfs_file_open(&lfs, &file, "d/hello", LFS_O_RDONLY) => 0;
+    uint8_t buffer[1024];
     lfs_file_read(&lfs, &file, buffer, 5) => 5;
     memcmp(buffer, "hola\n", 5) => 0;
     lfs_file_read(&lfs, &file, buffer, 8) => 8;
@@ -445,10 +472,11 @@ code = '''
     lfs_unmount(&lfs) => 0;
 '''
 
-[[case]] # move dir
+[cases.test_move_dir]
 code = '''
-    lfs_format(&lfs, &cfg) => 0;
-    lfs_mount(&lfs, &cfg) => 0;
+    lfs_t lfs;
+    lfs_format(&lfs, cfg) => 0;
+    lfs_mount(&lfs, cfg) => 0;
     lfs_mkdir(&lfs, "a") => 0;
     lfs_mkdir(&lfs, "b") => 0;
     lfs_mkdir(&lfs, "c") => 0;
@@ -459,11 +487,13 @@ code = '''
     lfs_mkdir(&lfs, "a/hi/ohayo") => 0;
     lfs_unmount(&lfs) => 0;
 
-    lfs_mount(&lfs, &cfg) => 0;
+    lfs_mount(&lfs, cfg) => 0;
     lfs_rename(&lfs, "a/hi", "c/hi") => 0;
     lfs_unmount(&lfs) => 0;
 
-    lfs_mount(&lfs, &cfg) => 0;
+    lfs_mount(&lfs, cfg) => 0;
+    lfs_dir_t dir;
+    struct lfs_info info;
     lfs_dir_open(&lfs, &dir, "a") => 0;
     lfs_dir_read(&lfs, &dir, &info) => 1;
     assert(strcmp(info.name, ".") == 0);
@@ -510,11 +540,12 @@ code = '''
     lfs_unmount(&lfs) => 0;
 '''
 
-[[case]] # move dir corrupt source
+[cases.test_move_dir_corrupt_source]
 in = "lfs.c"
 code = '''
-    lfs_format(&lfs, &cfg) => 0;
-    lfs_mount(&lfs, &cfg) => 0;
+    lfs_t lfs;
+    lfs_format(&lfs, cfg) => 0;
+    lfs_mount(&lfs, cfg) => 0;
     lfs_mkdir(&lfs, "a") => 0;
     lfs_mkdir(&lfs, "b") => 0;
     lfs_mkdir(&lfs, "c") => 0;
@@ -525,28 +556,30 @@ code = '''
     lfs_mkdir(&lfs, "a/hi/ohayo") => 0;
     lfs_unmount(&lfs) => 0;
 
-    lfs_mount(&lfs, &cfg) => 0;
+    lfs_mount(&lfs, cfg) => 0;
     lfs_rename(&lfs, "a/hi", "c/hi") => 0;
     lfs_unmount(&lfs) => 0;
 
     // corrupt the source
-    lfs_mount(&lfs, &cfg) => 0;
+    lfs_mount(&lfs, cfg) => 0;
+    lfs_dir_t dir;
+    struct lfs_info info;
     lfs_dir_open(&lfs, &dir, "a") => 0;
     lfs_block_t block = dir.m.pair[0];
     lfs_dir_close(&lfs, &dir) => 0;
     lfs_unmount(&lfs) => 0;
-    uint8_t bbuffer[LFS_BLOCK_SIZE];
-    cfg.read(&cfg, block, 0, bbuffer, LFS_BLOCK_SIZE) => 0;
-    int off = LFS_BLOCK_SIZE-1;
-    while (off >= 0 && bbuffer[off] == LFS_ERASE_VALUE) {
+    uint8_t buffer[BLOCK_SIZE];
+    cfg->read(cfg, block, 0, buffer, BLOCK_SIZE) => 0;
+    int off = BLOCK_SIZE-1;
+    while (off >= 0 && buffer[off] == ERASE_VALUE) {
         off -= 1;
     }
-    memset(&bbuffer[off-3], LFS_BLOCK_SIZE, 3);
-    cfg.erase(&cfg, block) => 0;
-    cfg.prog(&cfg, block, 0, bbuffer, LFS_BLOCK_SIZE) => 0;
-    cfg.sync(&cfg) => 0;
+    memset(&buffer[off-3], BLOCK_SIZE, 3);
+    cfg->erase(cfg, block) => 0;
+    cfg->prog(cfg, block, 0, buffer, BLOCK_SIZE) => 0;
+    cfg->sync(cfg) => 0;
 
-    lfs_mount(&lfs, &cfg) => 0;
+    lfs_mount(&lfs, cfg) => 0;
     lfs_dir_open(&lfs, &dir, "a") => 0;
     lfs_dir_read(&lfs, &dir, &info) => 1;
     assert(strcmp(info.name, ".") == 0);
@@ -593,12 +626,13 @@ code = '''
     lfs_unmount(&lfs) => 0;
 '''
 
-[[case]] # move dir corrupt source and dest
+[cases.test_move_dir_corrupt_source_dest]
 in = "lfs.c"
-if = 'LFS_PROG_SIZE <= 0x3fe' # only works with one crc per commit
+if = 'PROG_SIZE <= 0x3fe' # only works with one crc per commit
 code = '''
-    lfs_format(&lfs, &cfg) => 0;
-    lfs_mount(&lfs, &cfg) => 0;
+    lfs_t lfs;
+    lfs_format(&lfs, cfg) => 0;
+    lfs_mount(&lfs, cfg) => 0;
     lfs_mkdir(&lfs, "a") => 0;
     lfs_mkdir(&lfs, "b") => 0;
     lfs_mkdir(&lfs, "c") => 0;
@@ -609,44 +643,46 @@ code = '''
     lfs_mkdir(&lfs, "a/hi/ohayo") => 0;
     lfs_unmount(&lfs) => 0;
 
-    lfs_mount(&lfs, &cfg) => 0;
+    lfs_mount(&lfs, cfg) => 0;
     lfs_rename(&lfs, "a/hi", "c/hi") => 0;
     lfs_unmount(&lfs) => 0;
 
     // corrupt the source
-    lfs_mount(&lfs, &cfg) => 0;
+    lfs_mount(&lfs, cfg) => 0;
+    lfs_dir_t dir;
+    struct lfs_info info;
     lfs_dir_open(&lfs, &dir, "a") => 0;
     lfs_block_t block = dir.m.pair[0];
     lfs_dir_close(&lfs, &dir) => 0;
     lfs_unmount(&lfs) => 0;
-    uint8_t bbuffer[LFS_BLOCK_SIZE];
-    cfg.read(&cfg, block, 0, bbuffer, LFS_BLOCK_SIZE) => 0;
-    int off = LFS_BLOCK_SIZE-1;
-    while (off >= 0 && bbuffer[off] == LFS_ERASE_VALUE) {
+    uint8_t buffer[BLOCK_SIZE];
+    cfg->read(cfg, block, 0, buffer, BLOCK_SIZE) => 0;
+    int off = BLOCK_SIZE-1;
+    while (off >= 0 && buffer[off] == ERASE_VALUE) {
         off -= 1;
     }
-    memset(&bbuffer[off-3], LFS_BLOCK_SIZE, 3);
-    cfg.erase(&cfg, block) => 0;
-    cfg.prog(&cfg, block, 0, bbuffer, LFS_BLOCK_SIZE) => 0;
-    cfg.sync(&cfg) => 0;
+    memset(&buffer[off-3], BLOCK_SIZE, 3);
+    cfg->erase(cfg, block) => 0;
+    cfg->prog(cfg, block, 0, buffer, BLOCK_SIZE) => 0;
+    cfg->sync(cfg) => 0;
 
     // corrupt the destination
-    lfs_mount(&lfs, &cfg) => 0;
+    lfs_mount(&lfs, cfg) => 0;
     lfs_dir_open(&lfs, &dir, "c") => 0;
     block = dir.m.pair[0];
     lfs_dir_close(&lfs, &dir) => 0;
     lfs_unmount(&lfs) => 0;
-    cfg.read(&cfg, block, 0, bbuffer, LFS_BLOCK_SIZE) => 0;
-    off = LFS_BLOCK_SIZE-1;
-    while (off >= 0 && bbuffer[off] == LFS_ERASE_VALUE) {
+    cfg->read(cfg, block, 0, buffer, BLOCK_SIZE) => 0;
+    off = BLOCK_SIZE-1;
+    while (off >= 0 && buffer[off] == ERASE_VALUE) {
         off -= 1;
     }
-    memset(&bbuffer[off-3], LFS_BLOCK_SIZE, 3);
-    cfg.erase(&cfg, block) => 0;
-    cfg.prog(&cfg, block, 0, bbuffer, LFS_BLOCK_SIZE) => 0;
-    cfg.sync(&cfg) => 0;
+    memset(&buffer[off-3], BLOCK_SIZE, 3);
+    cfg->erase(cfg, block) => 0;
+    cfg->prog(cfg, block, 0, buffer, BLOCK_SIZE) => 0;
+    cfg->sync(cfg) => 0;
 
-    lfs_mount(&lfs, &cfg) => 0;
+    lfs_mount(&lfs, cfg) => 0;
     lfs_dir_open(&lfs, &dir, "a") => 0;
     lfs_dir_read(&lfs, &dir, &info) => 1;
     assert(strcmp(info.name, ".") == 0);
@@ -693,12 +729,13 @@ code = '''
     lfs_unmount(&lfs) => 0;
 '''
 
-[[case]] # move dir after corrupt
+[cases.test_move_dir_after_corrupt]
 in = "lfs.c"
-if = 'LFS_PROG_SIZE <= 0x3fe' # only works with one crc per commit
+if = 'PROG_SIZE <= 0x3fe' # only works with one crc per commit
 code = '''
-    lfs_format(&lfs, &cfg) => 0;
-    lfs_mount(&lfs, &cfg) => 0;
+    lfs_t lfs;
+    lfs_format(&lfs, cfg) => 0;
+    lfs_mount(&lfs, cfg) => 0;
     lfs_mkdir(&lfs, "a") => 0;
     lfs_mkdir(&lfs, "b") => 0;
     lfs_mkdir(&lfs, "c") => 0;
@@ -709,49 +746,51 @@ code = '''
     lfs_mkdir(&lfs, "a/hi/ohayo") => 0;
     lfs_unmount(&lfs) => 0;
 
-    lfs_mount(&lfs, &cfg) => 0;
+    lfs_mount(&lfs, cfg) => 0;
     lfs_rename(&lfs, "a/hi", "c/hi") => 0;
     lfs_unmount(&lfs) => 0;
 
     // corrupt the source
-    lfs_mount(&lfs, &cfg) => 0;
+    lfs_mount(&lfs, cfg) => 0;
+    lfs_dir_t dir;
+    struct lfs_info info;
     lfs_dir_open(&lfs, &dir, "a") => 0;
     lfs_block_t block = dir.m.pair[0];
     lfs_dir_close(&lfs, &dir) => 0;
     lfs_unmount(&lfs) => 0;
-    uint8_t bbuffer[LFS_BLOCK_SIZE];
-    cfg.read(&cfg, block, 0, bbuffer, LFS_BLOCK_SIZE) => 0;
-    int off = LFS_BLOCK_SIZE-1;
-    while (off >= 0 && bbuffer[off] == LFS_ERASE_VALUE) {
+    uint8_t buffer[BLOCK_SIZE];
+    cfg->read(cfg, block, 0, buffer, BLOCK_SIZE) => 0;
+    int off = BLOCK_SIZE-1;
+    while (off >= 0 && buffer[off] == ERASE_VALUE) {
         off -= 1;
     }
-    memset(&bbuffer[off-3], LFS_BLOCK_SIZE, 3);
-    cfg.erase(&cfg, block) => 0;
-    cfg.prog(&cfg, block, 0, bbuffer, LFS_BLOCK_SIZE) => 0;
-    cfg.sync(&cfg) => 0;
+    memset(&buffer[off-3], BLOCK_SIZE, 3);
+    cfg->erase(cfg, block) => 0;
+    cfg->prog(cfg, block, 0, buffer, BLOCK_SIZE) => 0;
+    cfg->sync(cfg) => 0;
 
     // corrupt the destination
-    lfs_mount(&lfs, &cfg) => 0;
+    lfs_mount(&lfs, cfg) => 0;
     lfs_dir_open(&lfs, &dir, "c") => 0;
     block = dir.m.pair[0];
     lfs_dir_close(&lfs, &dir) => 0;
     lfs_unmount(&lfs) => 0;
-    cfg.read(&cfg, block, 0, bbuffer, LFS_BLOCK_SIZE) => 0;
-    off = LFS_BLOCK_SIZE-1;
-    while (off >= 0 && bbuffer[off] == LFS_ERASE_VALUE) {
+    cfg->read(cfg, block, 0, buffer, BLOCK_SIZE) => 0;
+    off = BLOCK_SIZE-1;
+    while (off >= 0 && buffer[off] == ERASE_VALUE) {
         off -= 1;
     }
-    memset(&bbuffer[off-3], LFS_BLOCK_SIZE, 3);
-    cfg.erase(&cfg, block) => 0;
-    cfg.prog(&cfg, block, 0, bbuffer, LFS_BLOCK_SIZE) => 0;
-    cfg.sync(&cfg) => 0;
+    memset(&buffer[off-3], BLOCK_SIZE, 3);
+    cfg->erase(cfg, block) => 0;
+    cfg->prog(cfg, block, 0, buffer, BLOCK_SIZE) => 0;
+    cfg->sync(cfg) => 0;
 
     // continue move
-    lfs_mount(&lfs, &cfg) => 0;
+    lfs_mount(&lfs, cfg) => 0;
     lfs_rename(&lfs, "a/hi", "c/hi") => 0;
     lfs_unmount(&lfs) => 0;
 
-    lfs_mount(&lfs, &cfg) => 0;
+    lfs_mount(&lfs, cfg) => 0;
     lfs_dir_open(&lfs, &dir, "a") => 0;
     lfs_dir_read(&lfs, &dir, &info) => 1;
     assert(strcmp(info.name, ".") == 0);
@@ -798,13 +837,14 @@ code = '''
     lfs_unmount(&lfs) => 0;
 '''
 
-[[case]] # simple reentrant move dir
+[cases.test_reentrant_dir]
 reentrant = true
 code = '''
-    err = lfs_mount(&lfs, &cfg);
+    lfs_t lfs;
+    int err = lfs_mount(&lfs, cfg);
     if (err) {
-        lfs_format(&lfs, &cfg) => 0;
-        lfs_mount(&lfs, &cfg) => 0;
+        lfs_format(&lfs, cfg) => 0;
+        lfs_mount(&lfs, cfg) => 0;
     }
     err = lfs_mkdir(&lfs, "a");
     assert(!err || err == LFS_ERR_EXIST);
@@ -817,9 +857,10 @@ code = '''
     lfs_unmount(&lfs) => 0;
 
     while (true) {
-        lfs_mount(&lfs, &cfg) => 0;
+        lfs_mount(&lfs, cfg) => 0;
         // there should never exist _2_ hi directories
         int count = 0;
+        struct lfs_info info;
         if (lfs_stat(&lfs, "a/hi", &info) == 0) {
             assert(strcmp(info.name, "hi") == 0);
             assert(info.type == LFS_TYPE_DIR);
@@ -843,7 +884,7 @@ code = '''
         assert(count <= 1);
         lfs_unmount(&lfs) => 0;
 
-        lfs_mount(&lfs, &cfg) => 0;
+        lfs_mount(&lfs, cfg) => 0;
         if (lfs_stat(&lfs, "a/hi", &info) == 0) {
             lfs_rename(&lfs, "a/hi", "b/hi") => 0;
         } else if (lfs_stat(&lfs, "b/hi", &info) == 0) {
@@ -868,7 +909,9 @@ code = '''
         lfs_unmount(&lfs) => 0;
     }
 
-    lfs_mount(&lfs, &cfg) => 0;
+    lfs_mount(&lfs, cfg) => 0;
+    lfs_dir_t dir;
+    struct lfs_info info;
     lfs_dir_open(&lfs, &dir, "a") => 0;
     lfs_dir_read(&lfs, &dir, &info) => 1;
     assert(strcmp(info.name, ".") == 0);
@@ -915,14 +958,16 @@ code = '''
     lfs_unmount(&lfs) => 0;
 '''
 
-[[case]] # move state stealing
+[cases.test_move_state_stealing]
 code = '''
-    lfs_format(&lfs, &cfg) => 0;
-    lfs_mount(&lfs, &cfg) => 0;
+    lfs_t lfs;
+    lfs_format(&lfs, cfg) => 0;
+    lfs_mount(&lfs, cfg) => 0;
     lfs_mkdir(&lfs, "a") => 0;
     lfs_mkdir(&lfs, "b") => 0;
     lfs_mkdir(&lfs, "c") => 0;
     lfs_mkdir(&lfs, "d") => 0;
+    lfs_file_t file;
     lfs_file_open(&lfs, &file, "a/hello", LFS_O_CREAT | LFS_O_WRONLY) => 0;
     lfs_file_write(&lfs, &file, "hola\n", 5) => 5;
     lfs_file_write(&lfs, &file, "bonjour\n", 8) => 8;
@@ -930,21 +975,22 @@ code = '''
     lfs_file_close(&lfs, &file) => 0;
     lfs_unmount(&lfs) => 0;
 
-    lfs_mount(&lfs, &cfg) => 0;
+    lfs_mount(&lfs, cfg) => 0;
     lfs_rename(&lfs, "a/hello", "b/hello") => 0;
     lfs_unmount(&lfs) => 0;
-    lfs_mount(&lfs, &cfg) => 0;
+    lfs_mount(&lfs, cfg) => 0;
     lfs_rename(&lfs, "b/hello", "c/hello") => 0;
     lfs_unmount(&lfs) => 0;
-    lfs_mount(&lfs, &cfg) => 0;
+    lfs_mount(&lfs, cfg) => 0;
     lfs_rename(&lfs, "c/hello", "d/hello") => 0;
     lfs_unmount(&lfs) => 0;
 
-    lfs_mount(&lfs, &cfg) => 0;
+    lfs_mount(&lfs, cfg) => 0;
     lfs_file_open(&lfs, &file, "a/hello", LFS_O_RDONLY) => LFS_ERR_NOENT;
     lfs_file_open(&lfs, &file, "b/hello", LFS_O_RDONLY) => LFS_ERR_NOENT;
     lfs_file_open(&lfs, &file, "c/hello", LFS_O_RDONLY) => LFS_ERR_NOENT;
     lfs_file_open(&lfs, &file, "d/hello", LFS_O_RDONLY) => 0;
+    uint8_t buffer[1024];
     lfs_file_read(&lfs, &file, buffer, 5) => 5;
     memcmp(buffer, "hola\n", 5) => 0;
     lfs_file_read(&lfs, &file, buffer, 8) => 8;
@@ -954,12 +1000,13 @@ code = '''
     lfs_file_close(&lfs, &file) => 0;
     lfs_unmount(&lfs) => 0;
 
-    lfs_mount(&lfs, &cfg) => 0;
+    lfs_mount(&lfs, cfg) => 0;
     lfs_remove(&lfs, "b") => 0;
     lfs_remove(&lfs, "c") => 0;
     lfs_unmount(&lfs) => 0;
 
-    lfs_mount(&lfs, &cfg) => 0;
+    lfs_mount(&lfs, cfg) => 0;
+    struct lfs_info info;
     lfs_stat(&lfs, "a", &info) => 0;
     lfs_stat(&lfs, "b", &info) => LFS_ERR_NOENT;
     lfs_stat(&lfs, "c", &info) => LFS_ERR_NOENT;
@@ -979,12 +1026,16 @@ code = '''
 '''
 
 # Other specific corner cases
-[[case]] # create + delete in same commit with neighbors
+
+# create + delete in same commit with neighbors
+[cases.test_move_create_delete_same]
 code = '''
-    lfs_format(&lfs, &cfg) => 0;
-    lfs_mount(&lfs, &cfg) => 0;
+    lfs_t lfs;
+    lfs_format(&lfs, cfg) => 0;
+    lfs_mount(&lfs, cfg) => 0;
 
     // littlefs keeps files sorted, so we know the order these will be in
+    lfs_file_t file;
     lfs_file_open(&lfs, &file, "/1.move_me",
             LFS_O_WRONLY | LFS_O_CREAT) => 0;
     lfs_file_close(&lfs, &file) => 0;
@@ -1024,6 +1075,8 @@ code = '''
     lfs_file_close(&lfs, &files[2]) => 0;
 
     // check that nothing was corrupted
+    lfs_dir_t dir;
+    struct lfs_info info;
     lfs_dir_open(&lfs, &dir, "/") => 0;
     lfs_dir_read(&lfs, &dir, &info) => 1;
     assert(strcmp(info.name, ".") == 0);
@@ -1051,6 +1104,7 @@ code = '''
     lfs_dir_close(&lfs, &dir) => 0;
 
     lfs_file_open(&lfs, &file, "/0.before", LFS_O_RDONLY) => 0;
+    uint8_t buffer[1024];
     lfs_file_read(&lfs, &file, buffer, 7) => 7;
     assert(strcmp((char*)buffer, "test.4") == 0);
     lfs_file_close(&lfs, &file) => 0;
@@ -1124,13 +1178,15 @@ code = '''
     lfs_unmount(&lfs) => 0;
 '''
 
-# Other specific corner cases
-[[case]] # create + delete + delete in same commit with neighbors
+# create + delete + delete in same commit with neighbors
+[cases.test_move_create_delete_delete_same]
 code = '''
-    lfs_format(&lfs, &cfg) => 0;
-    lfs_mount(&lfs, &cfg) => 0;
+    lfs_t lfs;
+    lfs_format(&lfs, cfg) => 0;
+    lfs_mount(&lfs, cfg) => 0;
 
     // littlefs keeps files sorted, so we know the order these will be in
+    lfs_file_t file;
     lfs_file_open(&lfs, &file, "/1.move_me",
             LFS_O_WRONLY | LFS_O_CREAT) => 0;
     lfs_file_close(&lfs, &file) => 0;
@@ -1175,6 +1231,8 @@ code = '''
     lfs_file_close(&lfs, &files[2]) => 0;
 
     // check that nothing was corrupted
+    lfs_dir_t dir;
+    struct lfs_info info;
     lfs_dir_open(&lfs, &dir, "/") => 0;
     lfs_dir_read(&lfs, &dir, &info) => 1;
     assert(strcmp(info.name, ".") == 0);
@@ -1202,6 +1260,7 @@ code = '''
     lfs_dir_close(&lfs, &dir) => 0;
 
     lfs_file_open(&lfs, &file, "/0.before", LFS_O_RDONLY) => 0;
+    uint8_t buffer[1024];
     lfs_file_read(&lfs, &file, buffer, 7) => 7;
     assert(strcmp((char*)buffer, "test.4") == 0);
     lfs_file_close(&lfs, &file) => 0;
@@ -1281,14 +1340,17 @@ code = '''
     lfs_unmount(&lfs) => 0;
 '''
 
-[[case]] # create + delete in different dirs with neighbors
+# create + delete in different dirs with neighbors
+[cases.test_move_create_delete_different]
 code = '''
-    lfs_format(&lfs, &cfg) => 0;
-    lfs_mount(&lfs, &cfg) => 0;
+    lfs_t lfs;
+    lfs_format(&lfs, cfg) => 0;
+    lfs_mount(&lfs, cfg) => 0;
 
     // littlefs keeps files sorted, so we know the order these will be in
     lfs_mkdir(&lfs, "/dir.1") => 0;
     lfs_mkdir(&lfs, "/dir.2") => 0;
+    lfs_file_t file;
     lfs_file_open(&lfs, &file, "/dir.1/1.move_me",
             LFS_O_WRONLY | LFS_O_CREAT) => 0;
     lfs_file_close(&lfs, &file) => 0;
@@ -1340,6 +1402,8 @@ code = '''
     lfs_file_close(&lfs, &files[3]) => 0;
 
     // check that nothing was corrupted
+    lfs_dir_t dir;
+    struct lfs_info info;
     lfs_dir_open(&lfs, &dir, "/") => 0;
     lfs_dir_read(&lfs, &dir, &info) => 1;
     assert(strcmp(info.name, ".") == 0);
@@ -1397,6 +1461,7 @@ code = '''
     lfs_dir_close(&lfs, &dir) => 0;
 
     lfs_file_open(&lfs, &file, "/dir.1/0.before", LFS_O_RDONLY) => 0;
+    uint8_t buffer[1024];
     lfs_file_read(&lfs, &file, buffer, 7) => 7;
     assert(strcmp((char*)buffer, "test.5") == 0);
     lfs_file_close(&lfs, &file) => 0;
@@ -1518,17 +1583,20 @@ code = '''
     lfs_unmount(&lfs) => 0;
 '''
 
-[[case]] # move fix in relocation
+# move fix in relocation
+[cases.test_move_fix_relocation]
 in = "lfs.c"
-define.RELOCATIONS = 'range(0x3+1)'
-define.LFS_ERASE_CYCLES = 0xffffffff
+defines.RELOCATIONS = 'range(4)'
+defines.ERASE_CYCLES = 0xffffffff
 code = '''
-    lfs_format(&lfs, &cfg) => 0;
-    lfs_mount(&lfs, &cfg) => 0;
+    lfs_t lfs;
+    lfs_format(&lfs, cfg) => 0;
+    lfs_mount(&lfs, cfg) => 0;
 
     lfs_mkdir(&lfs, "/parent") => 0;
     lfs_mkdir(&lfs, "/parent/child") => 0;
 
+    lfs_file_t file;
     lfs_file_open(&lfs, &file, "/parent/1.move_me",
             LFS_O_WRONLY | LFS_O_CREAT) => 0;
     lfs_file_write(&lfs, &file, "move me",
@@ -1568,15 +1636,17 @@ code = '''
 
     // force specific directories to relocate
     if (RELOCATIONS & 0x1) {
+        lfs_dir_t dir;
         lfs_dir_open(&lfs, &dir, "/parent");
-        lfs_testbd_setwear(&cfg, dir.m.pair[0], 0xffffffff) => 0;
-        lfs_testbd_setwear(&cfg, dir.m.pair[1], 0xffffffff) => 0;
+        lfs_emubd_setwear(cfg, dir.m.pair[0], 0xffffffff) => 0;
+        lfs_emubd_setwear(cfg, dir.m.pair[1], 0xffffffff) => 0;
         lfs_dir_close(&lfs, &dir) => 0;
     }
     if (RELOCATIONS & 0x2) {
+        lfs_dir_t dir;
         lfs_dir_open(&lfs, &dir, "/parent/child");
-        lfs_testbd_setwear(&cfg, dir.m.pair[0], 0xffffffff) => 0;
-        lfs_testbd_setwear(&cfg, dir.m.pair[1], 0xffffffff) => 0;
+        lfs_emubd_setwear(cfg, dir.m.pair[0], 0xffffffff) => 0;
+        lfs_emubd_setwear(cfg, dir.m.pair[1], 0xffffffff) => 0;
         lfs_dir_close(&lfs, &dir) => 0;
     }
 
@@ -1593,6 +1663,8 @@ code = '''
     lfs_file_close(&lfs, &files[3]) => 0;
 
     // check that nothing was corrupted
+    lfs_dir_t dir;
+    struct lfs_info info;
     lfs_dir_open(&lfs, &dir, "/parent") => 0;
     lfs_dir_read(&lfs, &dir, &info) => 1;
     assert(strcmp(info.name, ".") == 0);
@@ -1637,6 +1709,7 @@ code = '''
     lfs_dir_close(&lfs, &dir) => 0;
 
     lfs_file_open(&lfs, &file, "/parent/0.before", LFS_O_RDONLY) => 0;
+    uint8_t buffer[1024];
     lfs_file_read(&lfs, &file, buffer, 7) => 7;
     assert(strcmp((char*)buffer, "test.5") == 0);
     lfs_file_close(&lfs, &file) => 0;
@@ -1655,18 +1728,21 @@ code = '''
     lfs_unmount(&lfs) => 0;
 '''
 
-[[case]] # move fix in relocation with predecessor
+# move fix in relocation with predecessor
+[cases.test_move_fix_relocation_predecessor]
 in = "lfs.c"
-define.RELOCATIONS = 'range(0x7+1)'
-define.LFS_ERASE_CYCLES = 0xffffffff
+defines.RELOCATIONS = 'range(8)'
+defines.ERASE_CYCLES = 0xffffffff
 code = '''
-    lfs_format(&lfs, &cfg) => 0;
-    lfs_mount(&lfs, &cfg) => 0;
+    lfs_t lfs;
+    lfs_format(&lfs, cfg) => 0;
+    lfs_mount(&lfs, cfg) => 0;
 
     lfs_mkdir(&lfs, "/parent") => 0;
     lfs_mkdir(&lfs, "/parent/child") => 0;
     lfs_mkdir(&lfs, "/parent/sibling") => 0;
 
+    lfs_file_t file;
     lfs_file_open(&lfs, &file, "/parent/sibling/1.move_me",
             LFS_O_WRONLY | LFS_O_CREAT) => 0;
     lfs_file_write(&lfs, &file, "move me",
@@ -1706,21 +1782,24 @@ code = '''
 
     // force specific directories to relocate
     if (RELOCATIONS & 0x1) {
+        lfs_dir_t dir;
         lfs_dir_open(&lfs, &dir, "/parent");
-        lfs_testbd_setwear(&cfg, dir.m.pair[0], 0xffffffff) => 0;
-        lfs_testbd_setwear(&cfg, dir.m.pair[1], 0xffffffff) => 0;
+        lfs_emubd_setwear(cfg, dir.m.pair[0], 0xffffffff) => 0;
+        lfs_emubd_setwear(cfg, dir.m.pair[1], 0xffffffff) => 0;
         lfs_dir_close(&lfs, &dir) => 0;
     }
     if (RELOCATIONS & 0x2) {
+        lfs_dir_t dir;
         lfs_dir_open(&lfs, &dir, "/parent/sibling");
-        lfs_testbd_setwear(&cfg, dir.m.pair[0], 0xffffffff) => 0;
-        lfs_testbd_setwear(&cfg, dir.m.pair[1], 0xffffffff) => 0;
+        lfs_emubd_setwear(cfg, dir.m.pair[0], 0xffffffff) => 0;
+        lfs_emubd_setwear(cfg, dir.m.pair[1], 0xffffffff) => 0;
         lfs_dir_close(&lfs, &dir) => 0;
     }
     if (RELOCATIONS & 0x4) {
+        lfs_dir_t dir;
         lfs_dir_open(&lfs, &dir, "/parent/child");
-        lfs_testbd_setwear(&cfg, dir.m.pair[0], 0xffffffff) => 0;
-        lfs_testbd_setwear(&cfg, dir.m.pair[1], 0xffffffff) => 0;
+        lfs_emubd_setwear(cfg, dir.m.pair[0], 0xffffffff) => 0;
+        lfs_emubd_setwear(cfg, dir.m.pair[1], 0xffffffff) => 0;
         lfs_dir_close(&lfs, &dir) => 0;
     }
 
@@ -1739,6 +1818,8 @@ code = '''
     lfs_file_close(&lfs, &files[3]) => 0;
 
     // check that nothing was corrupted
+    lfs_dir_t dir;
+    struct lfs_info info;
     lfs_dir_open(&lfs, &dir, "/parent") => 0;
     lfs_dir_read(&lfs, &dir, &info) => 1;
     assert(strcmp(info.name, ".") == 0);
@@ -1796,6 +1877,7 @@ code = '''
     lfs_dir_close(&lfs, &dir) => 0;
 
     lfs_file_open(&lfs, &file, "/parent/sibling/0.before", LFS_O_RDONLY) => 0;
+    uint8_t buffer[1024];
     lfs_file_read(&lfs, &file, buffer, 7) => 7;
     assert(strcmp((char*)buffer, "test.5") == 0);
     lfs_file_close(&lfs, &file) => 0;
diff --git a/tests/test_orphans.toml b/tests/test_orphans.toml
index 241e273..2c8405a 100644
--- a/tests/test_orphans.toml
+++ b/tests/test_orphans.toml
@@ -1,9 +1,10 @@
-[[case]] # orphan test
+[cases.test_orphans_normal]
 in = "lfs.c"
-if = 'LFS_PROG_SIZE <= 0x3fe' # only works with one crc per commit
+if = 'PROG_SIZE <= 0x3fe' # only works with one crc per commit
 code = '''
-    lfs_format(&lfs, &cfg) => 0;
-    lfs_mount(&lfs, &cfg) => 0;
+    lfs_t lfs;
+    lfs_format(&lfs, cfg) => 0;
+    lfs_mount(&lfs, cfg) => 0;
     lfs_mkdir(&lfs, "parent") => 0;
     lfs_mkdir(&lfs, "parent/orphan") => 0;
     lfs_mkdir(&lfs, "parent/child") => 0;
@@ -13,29 +14,31 @@ code = '''
     // corrupt the child's most recent commit, this should be the update
     // to the linked-list entry, which should orphan the orphan. Note this
     // makes a lot of assumptions about the remove operation.
-    lfs_mount(&lfs, &cfg) => 0;
+    lfs_mount(&lfs, cfg) => 0;
+    lfs_dir_t dir;
     lfs_dir_open(&lfs, &dir, "parent/child") => 0;
     lfs_block_t block = dir.m.pair[0];
     lfs_dir_close(&lfs, &dir) => 0;
     lfs_unmount(&lfs) => 0;
-    uint8_t bbuffer[LFS_BLOCK_SIZE];
-    cfg.read(&cfg, block, 0, bbuffer, LFS_BLOCK_SIZE) => 0;
-    int off = LFS_BLOCK_SIZE-1;
-    while (off >= 0 && bbuffer[off] == LFS_ERASE_VALUE) {
+    uint8_t buffer[BLOCK_SIZE];
+    cfg->read(cfg, block, 0, buffer, BLOCK_SIZE) => 0;
+    int off = BLOCK_SIZE-1;
+    while (off >= 0 && buffer[off] == ERASE_VALUE) {
         off -= 1;
     }
-    memset(&bbuffer[off-3], LFS_BLOCK_SIZE, 3);
-    cfg.erase(&cfg, block) => 0;
-    cfg.prog(&cfg, block, 0, bbuffer, LFS_BLOCK_SIZE) => 0;
-    cfg.sync(&cfg) => 0;
+    memset(&buffer[off-3], BLOCK_SIZE, 3);
+    cfg->erase(cfg, block) => 0;
+    cfg->prog(cfg, block, 0, buffer, BLOCK_SIZE) => 0;
+    cfg->sync(cfg) => 0;
 
-    lfs_mount(&lfs, &cfg) => 0;
+    lfs_mount(&lfs, cfg) => 0;
+    struct lfs_info info;
     lfs_stat(&lfs, "parent/orphan", &info) => LFS_ERR_NOENT;
     lfs_stat(&lfs, "parent/child", &info) => 0;
     lfs_fs_size(&lfs) => 8;
     lfs_unmount(&lfs) => 0;
 
-    lfs_mount(&lfs, &cfg) => 0;
+    lfs_mount(&lfs, cfg) => 0;
     lfs_stat(&lfs, "parent/orphan", &info) => LFS_ERR_NOENT;
     lfs_stat(&lfs, "parent/child", &info) => 0;
     lfs_fs_size(&lfs) => 8;
@@ -48,7 +51,7 @@ code = '''
     lfs_fs_size(&lfs) => 8;
     lfs_unmount(&lfs) => 0;
 
-    lfs_mount(&lfs, &cfg) => 0;
+    lfs_mount(&lfs, cfg) => 0;
     lfs_stat(&lfs, "parent/orphan", &info) => LFS_ERR_NOENT;
     lfs_stat(&lfs, "parent/child", &info) => 0;
     lfs_stat(&lfs, "parent/otherchild", &info) => 0;
@@ -56,43 +59,192 @@ code = '''
     lfs_unmount(&lfs) => 0;
 '''
 
-[[case]] # reentrant testing for orphans, basically just spam mkdir/remove
+# test that we only run deorphan once per power-cycle
+[cases.test_orphans_no_orphans]
+in = 'lfs.c'
+code = '''
+    lfs_t lfs;
+    lfs_format(&lfs, cfg) => 0;
+
+    lfs_mount(&lfs, cfg) => 0;
+    // mark the filesystem as having orphans
+    lfs_fs_preporphans(&lfs, +1) => 0;
+    lfs_mdir_t mdir;
+    lfs_dir_fetch(&lfs, &mdir, (lfs_block_t[2]){0, 1}) => 0;
+    lfs_dir_commit(&lfs, &mdir, NULL, 0) => 0;
+
+    // we should have orphans at this state
+    assert(lfs_gstate_hasorphans(&lfs.gstate));
+    lfs_unmount(&lfs) => 0;
+
+    // mount
+    lfs_mount(&lfs, cfg) => 0;
+    // we should detect orphans
+    assert(lfs_gstate_hasorphans(&lfs.gstate));
+    // force consistency
+    lfs_fs_forceconsistency(&lfs) => 0;
+    // we should no longer have orphans
+    assert(!lfs_gstate_hasorphans(&lfs.gstate));
+
+    lfs_unmount(&lfs) => 0;
+'''
+
+[cases.test_orphans_one_orphan]
+in = 'lfs.c'
+code = '''
+    lfs_t lfs;
+    lfs_format(&lfs, cfg) => 0;
+
+    lfs_mount(&lfs, cfg) => 0;
+    // create an orphan
+    lfs_mdir_t orphan;
+    lfs_alloc_ack(&lfs);
+    lfs_dir_alloc(&lfs, &orphan) => 0;
+    lfs_dir_commit(&lfs, &orphan, NULL, 0) => 0;
+
+    // append our orphan and mark the filesystem as having orphans
+    lfs_fs_preporphans(&lfs, +1) => 0;
+    lfs_mdir_t mdir;
+    lfs_dir_fetch(&lfs, &mdir, (lfs_block_t[2]){0, 1}) => 0;
+    lfs_pair_tole32(orphan.pair);
+    lfs_dir_commit(&lfs, &mdir, LFS_MKATTRS(
+            {LFS_MKTAG(LFS_TYPE_SOFTTAIL, 0x3ff, 8), orphan.pair})) => 0;
+
+    // we should have orphans at this state
+    assert(lfs_gstate_hasorphans(&lfs.gstate));
+    lfs_unmount(&lfs) => 0;
+
+    // mount
+    lfs_mount(&lfs, cfg) => 0;
+    // we should detect orphans
+    assert(lfs_gstate_hasorphans(&lfs.gstate));
+    // force consistency
+    lfs_fs_forceconsistency(&lfs) => 0;
+    // we should no longer have orphans
+    assert(!lfs_gstate_hasorphans(&lfs.gstate));
+
+    lfs_unmount(&lfs) => 0;
+'''
+
+# test that we can persist gstate with lfs_fs_mkconsistent
+[cases.test_orphans_mkconsistent_no_orphans]
+in = 'lfs.c'
+code = '''
+    lfs_t lfs;
+    lfs_format(&lfs, cfg) => 0;
+
+    lfs_mount(&lfs, cfg) => 0;
+    // mark the filesystem as having orphans
+    lfs_fs_preporphans(&lfs, +1) => 0;
+    lfs_mdir_t mdir;
+    lfs_dir_fetch(&lfs, &mdir, (lfs_block_t[2]){0, 1}) => 0;
+    lfs_dir_commit(&lfs, &mdir, NULL, 0) => 0;
+
+    // we should have orphans at this state
+    assert(lfs_gstate_hasorphans(&lfs.gstate));
+    lfs_unmount(&lfs) => 0;
+
+    // mount
+    lfs_mount(&lfs, cfg) => 0;
+    // we should detect orphans
+    assert(lfs_gstate_hasorphans(&lfs.gstate));
+    // force consistency
+    lfs_fs_mkconsistent(&lfs) => 0;
+    // we should no longer have orphans
+    assert(!lfs_gstate_hasorphans(&lfs.gstate));
+
+    // remount
+    lfs_unmount(&lfs) => 0;
+    lfs_mount(&lfs, cfg) => 0;
+    // we should still have no orphans
+    assert(!lfs_gstate_hasorphans(&lfs.gstate));
+    lfs_unmount(&lfs) => 0;
+'''
+
+[cases.test_orphans_mkconsistent_one_orphan]
+in = 'lfs.c'
+code = '''
+    lfs_t lfs;
+    lfs_format(&lfs, cfg) => 0;
+
+    lfs_mount(&lfs, cfg) => 0;
+    // create an orphan
+    lfs_mdir_t orphan;
+    lfs_alloc_ack(&lfs);
+    lfs_dir_alloc(&lfs, &orphan) => 0;
+    lfs_dir_commit(&lfs, &orphan, NULL, 0) => 0;
+
+    // append our orphan and mark the filesystem as having orphans
+    lfs_fs_preporphans(&lfs, +1) => 0;
+    lfs_mdir_t mdir;
+    lfs_dir_fetch(&lfs, &mdir, (lfs_block_t[2]){0, 1}) => 0;
+    lfs_pair_tole32(orphan.pair);
+    lfs_dir_commit(&lfs, &mdir, LFS_MKATTRS(
+            {LFS_MKTAG(LFS_TYPE_SOFTTAIL, 0x3ff, 8), orphan.pair})) => 0;
+
+    // we should have orphans at this state
+    assert(lfs_gstate_hasorphans(&lfs.gstate));
+    lfs_unmount(&lfs) => 0;
+
+    // mount
+    lfs_mount(&lfs, cfg) => 0;
+    // we should detect orphans
+    assert(lfs_gstate_hasorphans(&lfs.gstate));
+    // force consistency
+    lfs_fs_mkconsistent(&lfs) => 0;
+    // we should no longer have orphans
+    assert(!lfs_gstate_hasorphans(&lfs.gstate));
+
+    // remount
+    lfs_unmount(&lfs) => 0;
+    lfs_mount(&lfs, cfg) => 0;
+    // we should still have no orphans
+    assert(!lfs_gstate_hasorphans(&lfs.gstate));
+    lfs_unmount(&lfs) => 0;
+'''
+
+# reentrant testing for orphans, basically just spam mkdir/remove
+[cases.test_orphans_reentrant]
 reentrant = true
 # TODO fix this case, caused by non-DAG trees
-if = '!(DEPTH == 3 && LFS_CACHE_SIZE != 64)'
-define = [
+if = '!(DEPTH == 3 && CACHE_SIZE != 64)'
+defines = [
     {FILES=6,  DEPTH=1, CYCLES=20},
     {FILES=26, DEPTH=1, CYCLES=20},
     {FILES=3,  DEPTH=3, CYCLES=20},
 ]
 code = '''
-    err = lfs_mount(&lfs, &cfg);
+    lfs_t lfs;
+    int err = lfs_mount(&lfs, cfg);
     if (err) {
-        lfs_format(&lfs, &cfg) => 0;
-        lfs_mount(&lfs, &cfg) => 0;
+        lfs_format(&lfs, cfg) => 0;
+        lfs_mount(&lfs, cfg) => 0;
     }
 
-    srand(1);
+    uint32_t prng = 1;
     const char alpha[] = "abcdefghijklmnopqrstuvwxyz";
-    for (int i = 0; i < CYCLES; i++) {
+    for (unsigned i = 0; i < CYCLES; i++) {
         // create random path
         char full_path[256];
-        for (int d = 0; d < DEPTH; d++) {
-            sprintf(&full_path[2*d], "/%c", alpha[rand() % FILES]);
+        for (unsigned d = 0; d < DEPTH; d++) {
+            sprintf(&full_path[2*d], "/%c", alpha[TEST_PRNG(&prng) % FILES]);
         }
 
         // if it does not exist, we create it, else we destroy
+        struct lfs_info info;
         int res = lfs_stat(&lfs, full_path, &info);
         if (res == LFS_ERR_NOENT) {
             // create each directory in turn, ignore if dir already exists
-            for (int d = 0; d < DEPTH; d++) {
+            for (unsigned d = 0; d < DEPTH; d++) {
+                char path[1024];
                 strcpy(path, full_path);
                 path[2*d+2] = '\0';
                 err = lfs_mkdir(&lfs, path);
                 assert(!err || err == LFS_ERR_EXIST);
             }
 
-            for (int d = 0; d < DEPTH; d++) {
+            for (unsigned d = 0; d < DEPTH; d++) {
+                char path[1024];
                 strcpy(path, full_path);
                 path[2*d+2] = '\0';
                 lfs_stat(&lfs, path, &info) => 0;
@@ -106,6 +258,7 @@ code = '''
 
             // try to delete path in reverse order, ignore if dir is not empty
             for (int d = DEPTH-1; d >= 0; d--) {
+                char path[1024];
                 strcpy(path, full_path);
                 path[2*d+2] = '\0';
                 err = lfs_remove(&lfs, path);
diff --git a/tests/test_paths.toml b/tests/test_paths.toml
index a7474c0..97a519e 100644
--- a/tests/test_paths.toml
+++ b/tests/test_paths.toml
@@ -1,13 +1,16 @@
 
-[[case]] # simple path test
+# simple path test
+[cases.test_paths_normal]
 code = '''
-    lfs_format(&lfs, &cfg) => 0;
-    lfs_mount(&lfs, &cfg) => 0;
+    lfs_t lfs;
+    lfs_format(&lfs, cfg) => 0;
+    lfs_mount(&lfs, cfg) => 0;
     lfs_mkdir(&lfs, "tea") => 0;
     lfs_mkdir(&lfs, "tea/hottea") => 0;
     lfs_mkdir(&lfs, "tea/warmtea") => 0;
     lfs_mkdir(&lfs, "tea/coldtea") => 0;
 
+    struct lfs_info info;
     lfs_stat(&lfs, "tea/hottea", &info) => 0;
     assert(strcmp(info.name, "hottea") == 0);
     lfs_stat(&lfs, "/tea/hottea", &info) => 0;
@@ -21,15 +24,18 @@ code = '''
     lfs_unmount(&lfs) => 0;
 '''
 
-[[case]] # redundant slashes
+# redundant slashes
+[cases.test_paths_redundant_slashes]
 code = '''
-    lfs_format(&lfs, &cfg) => 0;
-    lfs_mount(&lfs, &cfg) => 0;
+    lfs_t lfs;
+    lfs_format(&lfs, cfg) => 0;
+    lfs_mount(&lfs, cfg) => 0;
     lfs_mkdir(&lfs, "tea") => 0;
     lfs_mkdir(&lfs, "tea/hottea") => 0;
     lfs_mkdir(&lfs, "tea/warmtea") => 0;
     lfs_mkdir(&lfs, "tea/coldtea") => 0;
 
+    struct lfs_info info;
     lfs_stat(&lfs, "/tea/hottea", &info) => 0;
     assert(strcmp(info.name, "hottea") == 0);
     lfs_stat(&lfs, "//tea//hottea", &info) => 0;
@@ -45,15 +51,18 @@ code = '''
     lfs_unmount(&lfs) => 0;
 '''
 
-[[case]] # dot path test
+# dot path test
+[cases.test_paths_dot]
 code = '''
-    lfs_format(&lfs, &cfg) => 0;
-    lfs_mount(&lfs, &cfg) => 0;
+    lfs_t lfs;
+    lfs_format(&lfs, cfg) => 0;
+    lfs_mount(&lfs, cfg) => 0;
     lfs_mkdir(&lfs, "tea") => 0;
     lfs_mkdir(&lfs, "tea/hottea") => 0;
     lfs_mkdir(&lfs, "tea/warmtea") => 0;
     lfs_mkdir(&lfs, "tea/coldtea") => 0;
 
+    struct lfs_info info;
     lfs_stat(&lfs, "./tea/hottea", &info) => 0;
     assert(strcmp(info.name, "hottea") == 0);
     lfs_stat(&lfs, "/./tea/hottea", &info) => 0;
@@ -71,10 +80,12 @@ code = '''
     lfs_unmount(&lfs) => 0;
 '''
 
-[[case]] # dot dot path test
+# dot dot path test
+[cases.test_paths_dot_dot]
 code = '''
-    lfs_format(&lfs, &cfg) => 0;
-    lfs_mount(&lfs, &cfg) => 0;
+    lfs_t lfs;
+    lfs_format(&lfs, cfg) => 0;
+    lfs_mount(&lfs, cfg) => 0;
     lfs_mkdir(&lfs, "tea") => 0;
     lfs_mkdir(&lfs, "tea/hottea") => 0;
     lfs_mkdir(&lfs, "tea/warmtea") => 0;
@@ -84,6 +95,7 @@ code = '''
     lfs_mkdir(&lfs, "coffee/warmcoffee") => 0;
     lfs_mkdir(&lfs, "coffee/coldcoffee") => 0;
 
+    struct lfs_info info;
     lfs_stat(&lfs, "coffee/../tea/hottea", &info) => 0;
     assert(strcmp(info.name, "hottea") == 0);
     lfs_stat(&lfs, "tea/coldtea/../hottea", &info) => 0;
@@ -101,15 +113,18 @@ code = '''
     lfs_unmount(&lfs) => 0;
 '''
 
-[[case]] # trailing dot path test
+# trailing dot path test
+[cases.test_paths_trailing_dot]
 code = '''
-    lfs_format(&lfs, &cfg) => 0;
-    lfs_mount(&lfs, &cfg) => 0;
+    lfs_t lfs;
+    lfs_format(&lfs, cfg) => 0;
+    lfs_mount(&lfs, cfg) => 0;
     lfs_mkdir(&lfs, "tea") => 0;
     lfs_mkdir(&lfs, "tea/hottea") => 0;
     lfs_mkdir(&lfs, "tea/warmtea") => 0;
     lfs_mkdir(&lfs, "tea/coldtea") => 0;
 
+    struct lfs_info info;
     lfs_stat(&lfs, "tea/hottea/", &info) => 0;
     assert(strcmp(info.name, "hottea") == 0);
     lfs_stat(&lfs, "tea/hottea/.", &info) => 0;
@@ -123,11 +138,14 @@ code = '''
     lfs_unmount(&lfs) => 0;
 '''
 
-[[case]] # leading dot path test
+# leading dot path test
+[cases.test_paths_leading_dot]
 code = '''
-    lfs_format(&lfs, &cfg) => 0;
-    lfs_mount(&lfs, &cfg) => 0;
+    lfs_t lfs;
+    lfs_format(&lfs, cfg) => 0;
+    lfs_mount(&lfs, cfg) => 0;
     lfs_mkdir(&lfs, ".milk") => 0;
+    struct lfs_info info;
     lfs_stat(&lfs, ".milk", &info) => 0;
     strcmp(info.name, ".milk") => 0;
     lfs_stat(&lfs, "tea/.././.milk", &info) => 0;
@@ -135,10 +153,12 @@ code = '''
     lfs_unmount(&lfs) => 0;
 '''
 
-[[case]] # root dot dot path test
+# root dot dot path test
+[cases.test_paths_root_dot_dot]
 code = '''
-    lfs_format(&lfs, &cfg) => 0;
-    lfs_mount(&lfs, &cfg) => 0;
+    lfs_t lfs;
+    lfs_format(&lfs, cfg) => 0;
+    lfs_mount(&lfs, cfg) => 0;
     lfs_mkdir(&lfs, "tea") => 0;
     lfs_mkdir(&lfs, "tea/hottea") => 0;
     lfs_mkdir(&lfs, "tea/warmtea") => 0;
@@ -148,6 +168,7 @@ code = '''
     lfs_mkdir(&lfs, "coffee/warmcoffee") => 0;
     lfs_mkdir(&lfs, "coffee/coldcoffee") => 0;
 
+    struct lfs_info info;
     lfs_stat(&lfs, "coffee/../../../../../../tea/hottea", &info) => 0;
     strcmp(info.name, "hottea") => 0;
 
@@ -159,10 +180,13 @@ code = '''
     lfs_unmount(&lfs) => 0;
 '''
 
-[[case]] # invalid path tests
+# invalid path tests
+[cases.test_paths_invalid]
 code = '''
-    lfs_format(&lfs, &cfg);
-    lfs_mount(&lfs, &cfg) => 0;
+    lfs_t lfs;
+    lfs_format(&lfs, cfg);
+    lfs_mount(&lfs, cfg) => 0;
+    struct lfs_info info;
     lfs_stat(&lfs, "dirt", &info) => LFS_ERR_NOENT;
     lfs_stat(&lfs, "dirt/ground", &info) => LFS_ERR_NOENT;
     lfs_stat(&lfs, "dirt/ground/earth", &info) => LFS_ERR_NOENT;
@@ -172,6 +196,7 @@ code = '''
     lfs_remove(&lfs, "dirt/ground/earth") => LFS_ERR_NOENT;
 
     lfs_mkdir(&lfs, "dirt/ground") => LFS_ERR_NOENT;
+    lfs_file_t file;
     lfs_file_open(&lfs, &file, "dirt/ground", LFS_O_WRONLY | LFS_O_CREAT)
             => LFS_ERR_NOENT;
     lfs_mkdir(&lfs, "dirt/ground/earth") => LFS_ERR_NOENT;
@@ -180,15 +205,19 @@ code = '''
     lfs_unmount(&lfs) => 0;
 '''
 
-[[case]] # root operations
+# root operations
+[cases.test_paths_root]
 code = '''
-    lfs_format(&lfs, &cfg) => 0;
-    lfs_mount(&lfs, &cfg) => 0;
+    lfs_t lfs;
+    lfs_format(&lfs, cfg) => 0;
+    lfs_mount(&lfs, cfg) => 0;
+    struct lfs_info info;
     lfs_stat(&lfs, "/", &info) => 0;
     assert(strcmp(info.name, "/") == 0);
     assert(info.type == LFS_TYPE_DIR);
 
     lfs_mkdir(&lfs, "/") => LFS_ERR_EXIST;
+    lfs_file_t file;
     lfs_file_open(&lfs, &file, "/", LFS_O_WRONLY | LFS_O_CREAT)
             => LFS_ERR_ISDIR;
 
@@ -196,10 +225,13 @@ code = '''
     lfs_unmount(&lfs) => 0;
 '''
 
-[[case]] # root representations
+# root representations
+[cases.test_paths_root_reprs]
 code = '''
-    lfs_format(&lfs, &cfg) => 0;
-    lfs_mount(&lfs, &cfg) => 0;
+    lfs_t lfs;
+    lfs_format(&lfs, cfg) => 0;
+    lfs_mount(&lfs, cfg) => 0;
+    struct lfs_info info;
     lfs_stat(&lfs, "/", &info) => 0;
     assert(strcmp(info.name, "/") == 0);
     assert(info.type == LFS_TYPE_DIR);
@@ -221,10 +253,13 @@ code = '''
     lfs_unmount(&lfs) => 0;
 '''
 
-[[case]] # superblock conflict test
+# superblock conflict test
+[cases.test_paths_superblock_conflict]
 code = '''
-    lfs_format(&lfs, &cfg) => 0;
-    lfs_mount(&lfs, &cfg) => 0;
+    lfs_t lfs;
+    lfs_format(&lfs, cfg) => 0;
+    lfs_mount(&lfs, cfg) => 0;
+    struct lfs_info info;
     lfs_stat(&lfs, "littlefs", &info) => LFS_ERR_NOENT;
     lfs_remove(&lfs, "littlefs") => LFS_ERR_NOENT;
 
@@ -237,18 +272,22 @@ code = '''
     lfs_unmount(&lfs) => 0;
 '''
 
-[[case]] # max path test
+# max path test
+[cases.test_paths_max]
 code = '''
-    lfs_format(&lfs, &cfg) => 0;
-    lfs_mount(&lfs, &cfg) => 0;
+    lfs_t lfs;
+    lfs_format(&lfs, cfg) => 0;
+    lfs_mount(&lfs, cfg) => 0;
     lfs_mkdir(&lfs, "coffee") => 0;
     lfs_mkdir(&lfs, "coffee/hotcoffee") => 0;
     lfs_mkdir(&lfs, "coffee/warmcoffee") => 0;
     lfs_mkdir(&lfs, "coffee/coldcoffee") => 0;
 
+    char path[1024];
     memset(path, 'w', LFS_NAME_MAX+1);
     path[LFS_NAME_MAX+1] = '\0';
     lfs_mkdir(&lfs, path) => LFS_ERR_NAMETOOLONG;
+    lfs_file_t file;
     lfs_file_open(&lfs, &file, path, LFS_O_WRONLY | LFS_O_CREAT)
             => LFS_ERR_NAMETOOLONG;
 
@@ -261,19 +300,23 @@ code = '''
     lfs_unmount(&lfs) => 0;
 '''
 
-[[case]] # really big path test
+# really big path test
+[cases.test_paths_really_big]
 code = '''
-    lfs_format(&lfs, &cfg) => 0;
-    lfs_mount(&lfs, &cfg) => 0;
+    lfs_t lfs;
+    lfs_format(&lfs, cfg) => 0;
+    lfs_mount(&lfs, cfg) => 0;
     lfs_mkdir(&lfs, "coffee") => 0;
     lfs_mkdir(&lfs, "coffee/hotcoffee") => 0;
     lfs_mkdir(&lfs, "coffee/warmcoffee") => 0;
     lfs_mkdir(&lfs, "coffee/coldcoffee") => 0;
 
+    char path[1024];
     memset(path, 'w', LFS_NAME_MAX);
     path[LFS_NAME_MAX] = '\0';
     lfs_mkdir(&lfs, path) => 0;
     lfs_remove(&lfs, path) => 0;
+    lfs_file_t file;
     lfs_file_open(&lfs, &file, path,
             LFS_O_WRONLY | LFS_O_CREAT) => 0;
     lfs_file_close(&lfs, &file) => 0;
diff --git a/tests/test_powerloss.toml b/tests/test_powerloss.toml
new file mode 100644
index 0000000..92c323b
--- /dev/null
+++ b/tests/test_powerloss.toml
@@ -0,0 +1,185 @@
+# There are already a number of tests that test general operations under
+# power-loss (see the reentrant attribute). These tests are for explicitly
+# testing specific corner cases.
+
+# only a revision count
+[cases.test_powerloss_only_rev]
+code = '''
+    lfs_t lfs;
+    lfs_format(&lfs, cfg) => 0;
+
+    lfs_mount(&lfs, cfg) => 0;
+    lfs_mkdir(&lfs, "notebook") => 0;
+    lfs_file_t file;
+    lfs_file_open(&lfs, &file, "notebook/paper",
+            LFS_O_WRONLY | LFS_O_CREAT | LFS_O_APPEND) => 0;
+    char buffer[256];
+    strcpy(buffer, "hello");
+    lfs_size_t size = strlen("hello");
+    for (int i = 0; i < 5; i++) {
+        lfs_file_write(&lfs, &file, buffer, size) => size;
+        lfs_file_sync(&lfs, &file) => 0;
+    }
+    lfs_file_close(&lfs, &file) => 0;
+
+    char rbuffer[256];
+    lfs_file_open(&lfs, &file, "notebook/paper", LFS_O_RDONLY) => 0;
+    for (int i = 0; i < 5; i++) {
+        lfs_file_read(&lfs, &file, rbuffer, size) => size;
+        assert(memcmp(rbuffer, buffer, size) == 0);
+    }
+    lfs_file_close(&lfs, &file) => 0;
+    lfs_unmount(&lfs) => 0;
+
+    // get pair/rev count
+    lfs_mount(&lfs, cfg) => 0;
+    lfs_dir_t dir;
+    lfs_dir_open(&lfs, &dir, "notebook") => 0;
+    lfs_block_t pair[2] = {dir.m.pair[0], dir.m.pair[1]};
+    uint32_t rev = dir.m.rev;
+    lfs_dir_close(&lfs, &dir) => 0;
+    lfs_unmount(&lfs) => 0;
+
+    // write just the revision count
+    uint8_t bbuffer[BLOCK_SIZE];
+    cfg->read(cfg, pair[1], 0, bbuffer, BLOCK_SIZE) => 0;
+
+    memcpy(bbuffer, &(uint32_t){lfs_tole32(rev+1)}, sizeof(uint32_t));
+
+    cfg->erase(cfg, pair[1]) => 0;
+    cfg->prog(cfg, pair[1], 0, bbuffer, BLOCK_SIZE) => 0;
+
+    lfs_mount(&lfs, cfg) => 0;
+
+    // can read?
+    lfs_file_open(&lfs, &file, "notebook/paper", LFS_O_RDONLY) => 0;
+    for (int i = 0; i < 5; i++) {
+        lfs_file_read(&lfs, &file, rbuffer, size) => size;
+        assert(memcmp(rbuffer, buffer, size) == 0);
+    }
+    lfs_file_close(&lfs, &file) => 0;
+
+    // can write?
+    lfs_file_open(&lfs, &file, "notebook/paper",
+            LFS_O_WRONLY | LFS_O_APPEND) => 0;
+    strcpy(buffer, "goodbye");
+    size = strlen("goodbye");
+    for (int i = 0; i < 5; i++) {
+        lfs_file_write(&lfs, &file, buffer, size) => size;
+        lfs_file_sync(&lfs, &file) => 0;
+    }
+    lfs_file_close(&lfs, &file) => 0;
+
+    lfs_file_open(&lfs, &file, "notebook/paper", LFS_O_RDONLY) => 0;
+    strcpy(buffer, "hello");
+    size = strlen("hello");
+    for (int i = 0; i < 5; i++) {
+        lfs_file_read(&lfs, &file, rbuffer, size) => size;
+        assert(memcmp(rbuffer, buffer, size) == 0);
+    }
+    strcpy(buffer, "goodbye");
+    size = strlen("goodbye");
+    for (int i = 0; i < 5; i++) {
+        lfs_file_read(&lfs, &file, rbuffer, size) => size;
+        assert(memcmp(rbuffer, buffer, size) == 0);
+    }
+    lfs_file_close(&lfs, &file) => 0;
+
+    lfs_unmount(&lfs) => 0;
+'''
+
+# partial prog, may not be byte in order!
+[cases.test_powerloss_partial_prog]
+if = '''
+    PROG_SIZE < BLOCK_SIZE
+        && (DISK_VERSION == 0 || DISK_VERSION >= 0x00020001)
+'''
+defines.BYTE_OFF = ["0", "PROG_SIZE-1", "PROG_SIZE/2"]
+defines.BYTE_VALUE = [0x33, 0xcc]
+in = "lfs.c"
+code = '''
+    lfs_t lfs;
+    lfs_format(&lfs, cfg) => 0;
+
+    lfs_mount(&lfs, cfg) => 0;
+    lfs_mkdir(&lfs, "notebook") => 0;
+    lfs_file_t file;
+    lfs_file_open(&lfs, &file, "notebook/paper",
+            LFS_O_WRONLY | LFS_O_CREAT | LFS_O_APPEND) => 0;
+    char buffer[256];
+    strcpy(buffer, "hello");
+    lfs_size_t size = strlen("hello");
+    for (int i = 0; i < 5; i++) {
+        lfs_file_write(&lfs, &file, buffer, size) => size;
+        lfs_file_sync(&lfs, &file) => 0;
+    }
+    lfs_file_close(&lfs, &file) => 0;
+
+    char rbuffer[256];
+    lfs_file_open(&lfs, &file, "notebook/paper", LFS_O_RDONLY) => 0;
+    for (int i = 0; i < 5; i++) {
+        lfs_file_read(&lfs, &file, rbuffer, size) => size;
+        assert(memcmp(rbuffer, buffer, size) == 0);
+    }
+    lfs_file_close(&lfs, &file) => 0;
+    lfs_unmount(&lfs) => 0;
+
+    // imitate a partial prog, value should not matter, if littlefs
+    // doesn't notice the partial prog testbd will assert
+
+    // get offset to next prog
+    lfs_mount(&lfs, cfg) => 0;
+    lfs_dir_t dir;
+    lfs_dir_open(&lfs, &dir, "notebook") => 0;
+    lfs_block_t block = dir.m.pair[0];
+    lfs_off_t off = dir.m.off;
+    lfs_dir_close(&lfs, &dir) => 0;
+    lfs_unmount(&lfs) => 0;
+
+    // tweak byte
+    uint8_t bbuffer[BLOCK_SIZE];
+    cfg->read(cfg, block, 0, bbuffer, BLOCK_SIZE) => 0;
+
+    bbuffer[off + BYTE_OFF] = BYTE_VALUE;
+
+    cfg->erase(cfg, block) => 0;
+    cfg->prog(cfg, block, 0, bbuffer, BLOCK_SIZE) => 0;
+
+    lfs_mount(&lfs, cfg) => 0;
+
+    // can read?
+    lfs_file_open(&lfs, &file, "notebook/paper", LFS_O_RDONLY) => 0;
+    for (int i = 0; i < 5; i++) {
+        lfs_file_read(&lfs, &file, rbuffer, size) => size;
+        assert(memcmp(rbuffer, buffer, size) == 0);
+    }
+    lfs_file_close(&lfs, &file) => 0;
+
+    // can write?
+    lfs_file_open(&lfs, &file, "notebook/paper",
+            LFS_O_WRONLY | LFS_O_APPEND) => 0;
+    strcpy(buffer, "goodbye");
+    size = strlen("goodbye");
+    for (int i = 0; i < 5; i++) {
+        lfs_file_write(&lfs, &file, buffer, size) => size;
+        lfs_file_sync(&lfs, &file) => 0;
+    }
+    lfs_file_close(&lfs, &file) => 0;
+
+    lfs_file_open(&lfs, &file, "notebook/paper", LFS_O_RDONLY) => 0;
+    strcpy(buffer, "hello");
+    size = strlen("hello");
+    for (int i = 0; i < 5; i++) {
+        lfs_file_read(&lfs, &file, rbuffer, size) => size;
+        assert(memcmp(rbuffer, buffer, size) == 0);
+    }
+    strcpy(buffer, "goodbye");
+    size = strlen("goodbye");
+    for (int i = 0; i < 5; i++) {
+        lfs_file_read(&lfs, &file, rbuffer, size) => size;
+        assert(memcmp(rbuffer, buffer, size) == 0);
+    }
+    lfs_file_close(&lfs, &file) => 0;
+
+    lfs_unmount(&lfs) => 0;
+'''
diff --git a/tests/test_relocations.toml b/tests/test_relocations.toml
index 71b1047..d20cb8c 100644
--- a/tests/test_relocations.toml
+++ b/tests/test_relocations.toml
@@ -1,15 +1,18 @@
 # specific corner cases worth explicitly testing for
-[[case]] # dangling split dir test
-define.ITERATIONS = 20
-define.COUNT = 10
-define.LFS_BLOCK_CYCLES = [8, 1]
+[cases.test_relocations_dangling_split_dir]
+defines.ITERATIONS = 20
+defines.COUNT = 10
+defines.BLOCK_CYCLES = [8, 1]
 code = '''
-    lfs_format(&lfs, &cfg) => 0;
+    lfs_t lfs;
+    lfs_format(&lfs, cfg) => 0;
     // fill up filesystem so only ~16 blocks are left
-    lfs_mount(&lfs, &cfg) => 0;
+    lfs_mount(&lfs, cfg) => 0;
+    lfs_file_t file;
     lfs_file_open(&lfs, &file, "padding", LFS_O_CREAT | LFS_O_WRONLY) => 0;
+    uint8_t buffer[512];
     memset(buffer, 0, 512);
-    while (LFS_BLOCK_COUNT - lfs_fs_size(&lfs) > 16) {
+    while (BLOCK_COUNT - lfs_fs_size(&lfs) > 16) {
         lfs_file_write(&lfs, &file, buffer, 512) => 512;
     }
     lfs_file_close(&lfs, &file) => 0;
@@ -17,18 +20,22 @@ code = '''
     lfs_mkdir(&lfs, "child") => 0;
     lfs_unmount(&lfs) => 0;
 
-    lfs_mount(&lfs, &cfg) => 0;
-    for (int j = 0; j < ITERATIONS; j++) {
-        for (int i = 0; i < COUNT; i++) {
+    lfs_mount(&lfs, cfg) => 0;
+    for (unsigned j = 0; j < ITERATIONS; j++) {
+        for (unsigned i = 0; i < COUNT; i++) {
+            char path[1024];
             sprintf(path, "child/test%03d_loooooooooooooooooong_name", i);
             lfs_file_open(&lfs, &file, path, LFS_O_CREAT | LFS_O_WRONLY) => 0;
             lfs_file_close(&lfs, &file) => 0;
         }
 
+        lfs_dir_t dir;
+        struct lfs_info info;
         lfs_dir_open(&lfs, &dir, "child") => 0;
         lfs_dir_read(&lfs, &dir, &info) => 1;
         lfs_dir_read(&lfs, &dir, &info) => 1;
-        for (int i = 0; i < COUNT; i++) {
+        for (unsigned i = 0; i < COUNT; i++) {
+            char path[1024];
             sprintf(path, "test%03d_loooooooooooooooooong_name", i);
             lfs_dir_read(&lfs, &dir, &info) => 1;
             strcmp(info.name, path) => 0;
@@ -36,46 +43,54 @@ code = '''
         lfs_dir_read(&lfs, &dir, &info) => 0;
         lfs_dir_close(&lfs, &dir) => 0;
 
-        if (j == ITERATIONS-1) {
+        if (j == (unsigned)ITERATIONS-1) {
             break;
         }
 
-        for (int i = 0; i < COUNT; i++) {
+        for (unsigned i = 0; i < COUNT; i++) {
+            char path[1024];
             sprintf(path, "child/test%03d_loooooooooooooooooong_name", i);
             lfs_remove(&lfs, path) => 0;
         }
     }
     lfs_unmount(&lfs) => 0;
 
-    lfs_mount(&lfs, &cfg) => 0;
+    lfs_mount(&lfs, cfg) => 0;
+    lfs_dir_t dir;
+    struct lfs_info info;
     lfs_dir_open(&lfs, &dir, "child") => 0;
     lfs_dir_read(&lfs, &dir, &info) => 1;
     lfs_dir_read(&lfs, &dir, &info) => 1;
-    for (int i = 0; i < COUNT; i++) {
+    for (unsigned i = 0; i < COUNT; i++) {
+        char path[1024];
         sprintf(path, "test%03d_loooooooooooooooooong_name", i);
         lfs_dir_read(&lfs, &dir, &info) => 1;
         strcmp(info.name, path) => 0;
     }
     lfs_dir_read(&lfs, &dir, &info) => 0;
     lfs_dir_close(&lfs, &dir) => 0;
-    for (int i = 0; i < COUNT; i++) {
+    for (unsigned i = 0; i < COUNT; i++) {
+        char path[1024];
         sprintf(path, "child/test%03d_loooooooooooooooooong_name", i);
         lfs_remove(&lfs, path) => 0;
     }
     lfs_unmount(&lfs) => 0;
 '''
 
-[[case]] # outdated head test
-define.ITERATIONS = 20
-define.COUNT = 10
-define.LFS_BLOCK_CYCLES = [8, 1]
+[cases.test_relocations_outdated_head]
+defines.ITERATIONS = 20
+defines.COUNT = 10
+defines.BLOCK_CYCLES = [8, 1]
 code = '''
-    lfs_format(&lfs, &cfg) => 0;
+    lfs_t lfs;
+    lfs_format(&lfs, cfg) => 0;
     // fill up filesystem so only ~16 blocks are left
-    lfs_mount(&lfs, &cfg) => 0;
+    lfs_mount(&lfs, cfg) => 0;
+    lfs_file_t file;
     lfs_file_open(&lfs, &file, "padding", LFS_O_CREAT | LFS_O_WRONLY) => 0;
+    uint8_t buffer[512];
     memset(buffer, 0, 512);
-    while (LFS_BLOCK_COUNT - lfs_fs_size(&lfs) > 16) {
+    while (BLOCK_COUNT - lfs_fs_size(&lfs) > 16) {
         lfs_file_write(&lfs, &file, buffer, 512) => 512;
     }
     lfs_file_close(&lfs, &file) => 0;
@@ -83,18 +98,22 @@ code = '''
     lfs_mkdir(&lfs, "child") => 0;
     lfs_unmount(&lfs) => 0;
 
-    lfs_mount(&lfs, &cfg) => 0;
-    for (int j = 0; j < ITERATIONS; j++) {
-        for (int i = 0; i < COUNT; i++) {
+    lfs_mount(&lfs, cfg) => 0;
+    for (unsigned j = 0; j < ITERATIONS; j++) {
+        for (unsigned i = 0; i < COUNT; i++) {
+            char path[1024];
             sprintf(path, "child/test%03d_loooooooooooooooooong_name", i);
             lfs_file_open(&lfs, &file, path, LFS_O_CREAT | LFS_O_WRONLY) => 0;
             lfs_file_close(&lfs, &file) => 0;
         }
 
+        lfs_dir_t dir;
+        struct lfs_info info;
         lfs_dir_open(&lfs, &dir, "child") => 0;
         lfs_dir_read(&lfs, &dir, &info) => 1;
         lfs_dir_read(&lfs, &dir, &info) => 1;
-        for (int i = 0; i < COUNT; i++) {
+        for (unsigned i = 0; i < COUNT; i++) {
+            char path[1024];
             sprintf(path, "test%03d_loooooooooooooooooong_name", i);
             lfs_dir_read(&lfs, &dir, &info) => 1;
             strcmp(info.name, path) => 0;
@@ -110,7 +129,8 @@ code = '''
         lfs_dir_rewind(&lfs, &dir) => 0;
         lfs_dir_read(&lfs, &dir, &info) => 1;
         lfs_dir_read(&lfs, &dir, &info) => 1;
-        for (int i = 0; i < COUNT; i++) {
+        for (unsigned i = 0; i < COUNT; i++) {
+            char path[1024];
             sprintf(path, "test%03d_loooooooooooooooooong_name", i);
             lfs_dir_read(&lfs, &dir, &info) => 1;
             strcmp(info.name, path) => 0;
@@ -126,7 +146,8 @@ code = '''
         lfs_dir_rewind(&lfs, &dir) => 0;
         lfs_dir_read(&lfs, &dir, &info) => 1;
         lfs_dir_read(&lfs, &dir, &info) => 1;
-        for (int i = 0; i < COUNT; i++) {
+        for (unsigned i = 0; i < COUNT; i++) {
+            char path[1024];
             sprintf(path, "test%03d_loooooooooooooooooong_name", i);
             lfs_dir_read(&lfs, &dir, &info) => 1;
             strcmp(info.name, path) => 0;
@@ -135,7 +156,8 @@ code = '''
         lfs_dir_read(&lfs, &dir, &info) => 0;
         lfs_dir_close(&lfs, &dir) => 0;
 
-        for (int i = 0; i < COUNT; i++) {
+        for (unsigned i = 0; i < COUNT; i++) {
+            char path[1024];
             sprintf(path, "child/test%03d_loooooooooooooooooong_name", i);
             lfs_remove(&lfs, path) => 0;
         }
@@ -143,45 +165,51 @@ code = '''
     lfs_unmount(&lfs) => 0;
 '''
 
-[[case]] # reentrant testing for relocations, this is the same as the
-         # orphan testing, except here we also set block_cycles so that
-         # almost every tree operation needs a relocation
+# reentrant testing for relocations, this is the same as the
+# orphan testing, except here we also set block_cycles so that
+# almost every tree operation needs a relocation
+[cases.test_relocations_reentrant]
 reentrant = true
 # TODO fix this case, caused by non-DAG trees
-if = '!(DEPTH == 3 && LFS_CACHE_SIZE != 64)'
-define = [
-    {FILES=6,  DEPTH=1, CYCLES=20, LFS_BLOCK_CYCLES=1},
-    {FILES=26, DEPTH=1, CYCLES=20, LFS_BLOCK_CYCLES=1},
-    {FILES=3,  DEPTH=3, CYCLES=20, LFS_BLOCK_CYCLES=1},
+# NOTE the second condition is required
+if = '!(DEPTH == 3 && CACHE_SIZE != 64) && 2*FILES < BLOCK_COUNT'
+defines = [
+    {FILES=6,  DEPTH=1, CYCLES=20, BLOCK_CYCLES=1},
+    {FILES=26, DEPTH=1, CYCLES=20, BLOCK_CYCLES=1},
+    {FILES=3,  DEPTH=3, CYCLES=20, BLOCK_CYCLES=1},
 ]
 code = '''
-    err = lfs_mount(&lfs, &cfg);
+    lfs_t lfs;
+    int err = lfs_mount(&lfs, cfg);
     if (err) {
-        lfs_format(&lfs, &cfg) => 0;
-        lfs_mount(&lfs, &cfg) => 0;
+        lfs_format(&lfs, cfg) => 0;
+        lfs_mount(&lfs, cfg) => 0;
     }
 
-    srand(1);
+    uint32_t prng = 1;
     const char alpha[] = "abcdefghijklmnopqrstuvwxyz";
-    for (int i = 0; i < CYCLES; i++) {
+    for (unsigned i = 0; i < CYCLES; i++) {
         // create random path
         char full_path[256];
-        for (int d = 0; d < DEPTH; d++) {
-            sprintf(&full_path[2*d], "/%c", alpha[rand() % FILES]);
+        for (unsigned d = 0; d < DEPTH; d++) {
+            sprintf(&full_path[2*d], "/%c", alpha[TEST_PRNG(&prng) % FILES]);
         }
 
         // if it does not exist, we create it, else we destroy
+        struct lfs_info info;
         int res = lfs_stat(&lfs, full_path, &info);
         if (res == LFS_ERR_NOENT) {
             // create each directory in turn, ignore if dir already exists
-            for (int d = 0; d < DEPTH; d++) {
+            for (unsigned d = 0; d < DEPTH; d++) {
+                char path[1024];
                 strcpy(path, full_path);
                 path[2*d+2] = '\0';
                 err = lfs_mkdir(&lfs, path);
                 assert(!err || err == LFS_ERR_EXIST);
             }
 
-            for (int d = 0; d < DEPTH; d++) {
+            for (unsigned d = 0; d < DEPTH; d++) {
+                char path[1024];
                 strcpy(path, full_path);
                 path[2*d+2] = '\0';
                 lfs_stat(&lfs, path, &info) => 0;
@@ -194,7 +222,8 @@ code = '''
             assert(info.type == LFS_TYPE_DIR);
 
             // try to delete path in reverse order, ignore if dir is not empty
-            for (int d = DEPTH-1; d >= 0; d--) {
+            for (unsigned d = DEPTH-1; d+1 > 0; d--) {
+                char path[1024];
                 strcpy(path, full_path);
                 path[2*d+2] = '\0';
                 err = lfs_remove(&lfs, path);
@@ -207,44 +236,50 @@ code = '''
     lfs_unmount(&lfs) => 0;
 '''
 
-[[case]] # reentrant testing for relocations, but now with random renames!
+# reentrant testing for relocations, but now with random renames!
+[cases.test_relocations_reentrant_renames]
 reentrant = true
 # TODO fix this case, caused by non-DAG trees
-if = '!(DEPTH == 3 && LFS_CACHE_SIZE != 64)'
-define = [
-    {FILES=6,  DEPTH=1, CYCLES=20, LFS_BLOCK_CYCLES=1},
-    {FILES=26, DEPTH=1, CYCLES=20, LFS_BLOCK_CYCLES=1},
-    {FILES=3,  DEPTH=3, CYCLES=20, LFS_BLOCK_CYCLES=1},
+# NOTE the second condition is required
+if = '!(DEPTH == 3 && CACHE_SIZE != 64) && 2*FILES < BLOCK_COUNT'
+defines = [
+    {FILES=6,  DEPTH=1, CYCLES=20, BLOCK_CYCLES=1},
+    {FILES=26, DEPTH=1, CYCLES=20, BLOCK_CYCLES=1},
+    {FILES=3,  DEPTH=3, CYCLES=20, BLOCK_CYCLES=1},
 ]
 code = '''
-    err = lfs_mount(&lfs, &cfg);
+    lfs_t lfs;
+    int err = lfs_mount(&lfs, cfg);
     if (err) {
-        lfs_format(&lfs, &cfg) => 0;
-        lfs_mount(&lfs, &cfg) => 0;
+        lfs_format(&lfs, cfg) => 0;
+        lfs_mount(&lfs, cfg) => 0;
     }
 
-    srand(1);
+    uint32_t prng = 1;
     const char alpha[] = "abcdefghijklmnopqrstuvwxyz";
-    for (int i = 0; i < CYCLES; i++) {
+    for (unsigned i = 0; i < CYCLES; i++) {
         // create random path
         char full_path[256];
-        for (int d = 0; d < DEPTH; d++) {
-            sprintf(&full_path[2*d], "/%c", alpha[rand() % FILES]);
+        for (unsigned d = 0; d < DEPTH; d++) {
+            sprintf(&full_path[2*d], "/%c", alpha[TEST_PRNG(&prng) % FILES]);
         }
 
         // if it does not exist, we create it, else we destroy
+        struct lfs_info info;
         int res = lfs_stat(&lfs, full_path, &info);
         assert(!res || res == LFS_ERR_NOENT);
         if (res == LFS_ERR_NOENT) {
             // create each directory in turn, ignore if dir already exists
-            for (int d = 0; d < DEPTH; d++) {
+            for (unsigned d = 0; d < DEPTH; d++) {
+                char path[1024];
                 strcpy(path, full_path);
                 path[2*d+2] = '\0';
                 err = lfs_mkdir(&lfs, path);
                 assert(!err || err == LFS_ERR_EXIST);
             }
 
-            for (int d = 0; d < DEPTH; d++) {
+            for (unsigned d = 0; d < DEPTH; d++) {
+                char path[1024];
                 strcpy(path, full_path);
                 path[2*d+2] = '\0';
                 lfs_stat(&lfs, path, &info) => 0;
@@ -257,8 +292,8 @@ code = '''
 
             // create new random path
             char new_path[256];
-            for (int d = 0; d < DEPTH; d++) {
-                sprintf(&new_path[2*d], "/%c", alpha[rand() % FILES]);
+            for (unsigned d = 0; d < DEPTH; d++) {
+                sprintf(&new_path[2*d], "/%c", alpha[TEST_PRNG(&prng) % FILES]);
             }
 
             // if new path does not exist, rename, otherwise destroy
@@ -266,7 +301,8 @@ code = '''
             assert(!res || res == LFS_ERR_NOENT);
             if (res == LFS_ERR_NOENT) {
                 // stop once some dir is renamed
-                for (int d = 0; d < DEPTH; d++) {
+                for (unsigned d = 0; d < DEPTH; d++) {
+                    char path[1024];
                     strcpy(&path[2*d], &full_path[2*d]);
                     path[2*d+2] = '\0';
                     strcpy(&path[128+2*d], &new_path[2*d]);
@@ -278,7 +314,8 @@ code = '''
                     }
                 }
 
-                for (int d = 0; d < DEPTH; d++) {
+                for (unsigned d = 0; d < DEPTH; d++) {
+                    char path[1024];
                     strcpy(path, new_path);
                     path[2*d+2] = '\0';
                     lfs_stat(&lfs, path, &info) => 0;
@@ -290,7 +327,8 @@ code = '''
             } else {
                 // try to delete path in reverse order,
                 // ignore if dir is not empty
-                for (int d = DEPTH-1; d >= 0; d--) {
+                for (unsigned d = DEPTH-1; d+1 > 0; d--) {
+                    char path[1024];
                     strcpy(path, full_path);
                     path[2*d+2] = '\0';
                     err = lfs_remove(&lfs, path);
diff --git a/tests/test_seek.toml b/tests/test_seek.toml
index 79d7728..b976057 100644
--- a/tests/test_seek.toml
+++ b/tests/test_seek.toml
@@ -1,6 +1,7 @@
 
-[[case]] # simple file seek
-define = [
+# simple file seek
+[cases.test_seek_read]
+defines = [
     {COUNT=132, SKIP=4},
     {COUNT=132, SKIP=128},
     {COUNT=200, SKIP=10},
@@ -9,11 +10,14 @@ define = [
     {COUNT=4,   SKIP=2},
 ]
 code = '''
-    lfs_format(&lfs, &cfg) => 0;
-    lfs_mount(&lfs, &cfg) => 0;
+    lfs_t lfs;
+    lfs_format(&lfs, cfg) => 0;
+    lfs_mount(&lfs, cfg) => 0;
+    lfs_file_t file;
     lfs_file_open(&lfs, &file, "kitty",
             LFS_O_WRONLY | LFS_O_CREAT | LFS_O_APPEND) => 0;
-    size = strlen("kittycatcat");
+    size_t size = strlen("kittycatcat");
+    uint8_t buffer[1024];
     memcpy(buffer, "kittycatcat", size);
     for (int j = 0; j < COUNT; j++) {
         lfs_file_write(&lfs, &file, buffer, size);
@@ -21,7 +25,7 @@ code = '''
     lfs_file_close(&lfs, &file) => 0;
     lfs_unmount(&lfs) => 0;
 
-    lfs_mount(&lfs, &cfg) => 0;
+    lfs_mount(&lfs, cfg) => 0;
     lfs_file_open(&lfs, &file, "kitty", LFS_O_RDONLY) => 0;
 
     lfs_soff_t pos = -1;
@@ -68,8 +72,9 @@ code = '''
     lfs_unmount(&lfs) => 0;
 '''
 
-[[case]] # simple file seek and write
-define = [
+# simple file seek and write
+[cases.test_seek_write]
+defines = [
     {COUNT=132, SKIP=4},
     {COUNT=132, SKIP=128},
     {COUNT=200, SKIP=10},
@@ -78,11 +83,14 @@ define = [
     {COUNT=4,   SKIP=2},
 ]
 code = '''
-    lfs_format(&lfs, &cfg) => 0;
-    lfs_mount(&lfs, &cfg) => 0;
+    lfs_t lfs;
+    lfs_format(&lfs, cfg) => 0;
+    lfs_mount(&lfs, cfg) => 0;
+    lfs_file_t file;
     lfs_file_open(&lfs, &file, "kitty",
             LFS_O_WRONLY | LFS_O_CREAT | LFS_O_APPEND) => 0;
-    size = strlen("kittycatcat");
+    size_t size = strlen("kittycatcat");
+    uint8_t buffer[1024];
     memcpy(buffer, "kittycatcat", size);
     for (int j = 0; j < COUNT; j++) {
         lfs_file_write(&lfs, &file, buffer, size);
@@ -90,7 +98,7 @@ code = '''
     lfs_file_close(&lfs, &file) => 0;
     lfs_unmount(&lfs) => 0;
 
-    lfs_mount(&lfs, &cfg) => 0;
+    lfs_mount(&lfs, cfg) => 0;
     lfs_file_open(&lfs, &file, "kitty", LFS_O_RDWR) => 0;
 
     lfs_soff_t pos = -1;
@@ -129,15 +137,18 @@ code = '''
     lfs_unmount(&lfs) => 0;
 '''
 
-[[case]] # boundary seek and writes
-define.COUNT = 132
-define.OFFSETS = '"{512, 1020, 513, 1021, 511, 1019, 1441}"'
+# boundary seek and writes
+[cases.test_seek_boundary_write]
+defines.COUNT = 132
 code = '''
-    lfs_format(&lfs, &cfg) => 0;
-    lfs_mount(&lfs, &cfg) => 0;
+    lfs_t lfs;
+    lfs_format(&lfs, cfg) => 0;
+    lfs_mount(&lfs, cfg) => 0;
+    lfs_file_t file;
     lfs_file_open(&lfs, &file, "kitty",
             LFS_O_WRONLY | LFS_O_CREAT | LFS_O_APPEND) => 0;
-    size = strlen("kittycatcat");
+    size_t size = strlen("kittycatcat");
+    uint8_t buffer[1024];
     memcpy(buffer, "kittycatcat", size);
     for (int j = 0; j < COUNT; j++) {
         lfs_file_write(&lfs, &file, buffer, size);
@@ -145,11 +156,11 @@ code = '''
     lfs_file_close(&lfs, &file) => 0;
     lfs_unmount(&lfs) => 0;
 
-    lfs_mount(&lfs, &cfg) => 0;
+    lfs_mount(&lfs, cfg) => 0;
     lfs_file_open(&lfs, &file, "kitty", LFS_O_RDWR) => 0;
 
     size = strlen("hedgehoghog");
-    const lfs_soff_t offsets[] = OFFSETS;
+    const lfs_soff_t offsets[] = {512, 1020, 513, 1021, 511, 1019, 1441};
 
     for (unsigned i = 0; i < sizeof(offsets) / sizeof(offsets[0]); i++) {
         lfs_soff_t off = offsets[i];
@@ -183,8 +194,9 @@ code = '''
     lfs_unmount(&lfs) => 0;
 '''
 
-[[case]] # out of bounds seek
-define = [
+# out of bounds seek
+[cases.test_seek_out_of_bounds]
+defines = [
     {COUNT=132, SKIP=4},
     {COUNT=132, SKIP=128},
     {COUNT=200, SKIP=10},
@@ -193,18 +205,21 @@ define = [
     {COUNT=4,   SKIP=3},
 ]
 code = '''
-    lfs_format(&lfs, &cfg) => 0;
-    lfs_mount(&lfs, &cfg) => 0;
+    lfs_t lfs;
+    lfs_format(&lfs, cfg) => 0;
+    lfs_mount(&lfs, cfg) => 0;
+    lfs_file_t file;
     lfs_file_open(&lfs, &file, "kitty",
             LFS_O_WRONLY | LFS_O_CREAT | LFS_O_APPEND) => 0;
-    size = strlen("kittycatcat");
+    size_t size = strlen("kittycatcat");
+    uint8_t buffer[1024];
     memcpy(buffer, "kittycatcat", size);
     for (int j = 0; j < COUNT; j++) {
         lfs_file_write(&lfs, &file, buffer, size);
     }
     lfs_file_close(&lfs, &file) => 0;
     lfs_unmount(&lfs) => 0;
-    lfs_mount(&lfs, &cfg) => 0;
+    lfs_mount(&lfs, cfg) => 0;
     lfs_file_open(&lfs, &file, "kitty", LFS_O_RDWR) => 0;
 
     size = strlen("kittycatcat");
@@ -238,16 +253,20 @@ code = '''
     lfs_unmount(&lfs) => 0;
 '''
 
-[[case]] # inline write and seek
-define.SIZE = [2, 4, 128, 132]
+# inline write and seek
+[cases.test_seek_inline_write]
+defines.SIZE = [2, 4, 128, 132]
 code = '''
-    lfs_format(&lfs, &cfg) => 0;
-    lfs_mount(&lfs, &cfg) => 0;
+    lfs_t lfs;
+    lfs_format(&lfs, cfg) => 0;
+    lfs_mount(&lfs, cfg) => 0;
+    lfs_file_t file;
     lfs_file_open(&lfs, &file, "tinykitty",
             LFS_O_RDWR | LFS_O_CREAT) => 0;
     int j = 0;
     int k = 0;
 
+    uint8_t buffer[1024];
     memcpy(buffer, "abcdefghijklmnopqrstuvwxyz", 26);
     for (unsigned i = 0; i < SIZE; i++) {
         lfs_file_write(&lfs, &file, &buffer[j++ % 26], 1) => 1;
@@ -305,16 +324,20 @@ code = '''
     lfs_unmount(&lfs) => 0;
 '''
 
-[[case]] # file seek and write with power-loss
+# file seek and write with power-loss
+[cases.test_seek_reentrant_write]
 # must be power-of-2 for quadratic probing to be exhaustive
-define.COUNT = [4, 64, 128]
+defines.COUNT = [4, 64, 128]
 reentrant = true
 code = '''
-    err = lfs_mount(&lfs, &cfg);
+    lfs_t lfs;
+    int err = lfs_mount(&lfs, cfg);
     if (err) {
-        lfs_format(&lfs, &cfg) => 0;
-        lfs_mount(&lfs, &cfg) => 0;
+        lfs_format(&lfs, cfg) => 0;
+        lfs_mount(&lfs, cfg) => 0;
     }
+    lfs_file_t file;
+    uint8_t buffer[1024];
     err = lfs_file_open(&lfs, &file, "kitty", LFS_O_RDONLY);
     assert(!err || err == LFS_ERR_NOENT);
     if (!err) {
@@ -334,14 +357,14 @@ code = '''
     if (lfs_file_size(&lfs, &file) == 0) {
         for (int j = 0; j < COUNT; j++) {
             strcpy((char*)buffer, "kittycatcat");
-            size = strlen((char*)buffer);
+            size_t size = strlen((char*)buffer);
             lfs_file_write(&lfs, &file, buffer, size) => size;
         }
     }
     lfs_file_close(&lfs, &file) => 0;
 
     strcpy((char*)buffer, "doggodogdog");
-    size = strlen((char*)buffer);
+    size_t size = strlen((char*)buffer);
 
     lfs_file_open(&lfs, &file, "kitty", LFS_O_RDWR) => 0;
     lfs_file_size(&lfs, &file) => COUNT*size;
diff --git a/tests/test_superblocks.toml b/tests/test_superblocks.toml
index 407c845..6d4a3e4 100644
--- a/tests/test_superblocks.toml
+++ b/tests/test_superblocks.toml
@@ -1,41 +1,116 @@
-[[case]] # simple formatting test
+# simple formatting test
+[cases.test_superblocks_format]
 code = '''
-    lfs_format(&lfs, &cfg) => 0;
+    lfs_t lfs;
+    lfs_format(&lfs, cfg) => 0;
 '''
 
-[[case]] # mount/unmount
+# mount/unmount
+[cases.test_superblocks_mount]
 code = '''
-    lfs_format(&lfs, &cfg) => 0;
-    lfs_mount(&lfs, &cfg) => 0;
+    lfs_t lfs;
+    lfs_format(&lfs, cfg) => 0;
+    lfs_mount(&lfs, cfg) => 0;
     lfs_unmount(&lfs) => 0;
 '''
 
-[[case]] # reentrant format
+# mount/unmount from interpretting a previous superblock block_count
+[cases.test_superblocks_mount_unknown_block_count]
+code = '''
+    lfs_t lfs;
+    lfs_format(&lfs, cfg) => 0;
+
+    memset(&lfs, 0, sizeof(lfs));
+    struct lfs_config tweaked_cfg = *cfg;
+    tweaked_cfg.block_count = 0;
+    lfs_mount(&lfs, &tweaked_cfg) => 0;
+    assert(lfs.block_count == cfg->block_count);
+    lfs_unmount(&lfs) => 0;
+'''
+
+
+# reentrant format
+[cases.test_superblocks_reentrant_format]
 reentrant = true
 code = '''
-    err = lfs_mount(&lfs, &cfg);
+    lfs_t lfs;
+    int err = lfs_mount(&lfs, cfg);
     if (err) {
-        lfs_format(&lfs, &cfg) => 0;
-        lfs_mount(&lfs, &cfg) => 0;
+        lfs_format(&lfs, cfg) => 0;
+        lfs_mount(&lfs, cfg) => 0;
     }
     lfs_unmount(&lfs) => 0;
 '''
 
-[[case]] # invalid mount
+# invalid mount
+[cases.test_superblocks_invalid_mount]
 code = '''
-    lfs_mount(&lfs, &cfg) => LFS_ERR_CORRUPT;
+    lfs_t lfs;
+    lfs_mount(&lfs, cfg) => LFS_ERR_CORRUPT;
 '''
 
-[[case]] # expanding superblock
-define.LFS_BLOCK_CYCLES = [32, 33, 1]
-define.N = [10, 100, 1000]
+# test we can read superblock info through lfs_fs_stat
+[cases.test_superblocks_stat]
+if = 'DISK_VERSION == 0'
 code = '''
-    lfs_format(&lfs, &cfg) => 0;
-    lfs_mount(&lfs, &cfg) => 0;
+    lfs_t lfs;
+    lfs_format(&lfs, cfg) => 0;
+
+    // test we can mount and read fsinfo
+    lfs_mount(&lfs, cfg) => 0;
+
+    struct lfs_fsinfo fsinfo;
+    lfs_fs_stat(&lfs, &fsinfo) => 0;
+    assert(fsinfo.disk_version == LFS_DISK_VERSION);
+    assert(fsinfo.name_max == LFS_NAME_MAX);
+    assert(fsinfo.file_max == LFS_FILE_MAX);
+    assert(fsinfo.attr_max == LFS_ATTR_MAX);
+
+    lfs_unmount(&lfs) => 0;
+'''
+
+[cases.test_superblocks_stat_tweaked]
+if = 'DISK_VERSION == 0'
+defines.TWEAKED_NAME_MAX = 63
+defines.TWEAKED_FILE_MAX = '(1 << 16)-1'
+defines.TWEAKED_ATTR_MAX = 512
+code = '''
+    // create filesystem with tweaked params
+    struct lfs_config tweaked_cfg = *cfg;
+    tweaked_cfg.name_max = TWEAKED_NAME_MAX;
+    tweaked_cfg.file_max = TWEAKED_FILE_MAX;
+    tweaked_cfg.attr_max = TWEAKED_ATTR_MAX;
+
+    lfs_t lfs;
+    lfs_format(&lfs, &tweaked_cfg) => 0;
+
+    // test we can mount and read these params with the original config
+    lfs_mount(&lfs, cfg) => 0;
+
+    struct lfs_fsinfo fsinfo;
+    lfs_fs_stat(&lfs, &fsinfo) => 0;
+    assert(fsinfo.disk_version == LFS_DISK_VERSION);
+    assert(fsinfo.name_max == TWEAKED_NAME_MAX);
+    assert(fsinfo.file_max == TWEAKED_FILE_MAX);
+    assert(fsinfo.attr_max == TWEAKED_ATTR_MAX);
+
+    lfs_unmount(&lfs) => 0;
+'''
+
+# expanding superblock
+[cases.test_superblocks_expand]
+defines.BLOCK_CYCLES = [32, 33, 1]
+defines.N = [10, 100, 1000]
+code = '''
+    lfs_t lfs;
+    lfs_format(&lfs, cfg) => 0;
+    lfs_mount(&lfs, cfg) => 0;
     for (int i = 0; i < N; i++) {
+        lfs_file_t file;
         lfs_file_open(&lfs, &file, "dummy",
                 LFS_O_WRONLY | LFS_O_CREAT | LFS_O_EXCL) => 0;
         lfs_file_close(&lfs, &file) => 0;
+        struct lfs_info info;
         lfs_stat(&lfs, "dummy", &info) => 0;
         assert(strcmp(info.name, "dummy") == 0);
         assert(info.type == LFS_TYPE_REG);
@@ -44,25 +119,30 @@ code = '''
     lfs_unmount(&lfs) => 0;
 
     // one last check after power-cycle
-    lfs_mount(&lfs, &cfg) => 0;
+    lfs_mount(&lfs, cfg) => 0;
+    lfs_file_t file;
     lfs_file_open(&lfs, &file, "dummy",
             LFS_O_WRONLY | LFS_O_CREAT | LFS_O_EXCL) => 0;
     lfs_file_close(&lfs, &file) => 0;
+    struct lfs_info info;
     lfs_stat(&lfs, "dummy", &info) => 0;
     assert(strcmp(info.name, "dummy") == 0);
     assert(info.type == LFS_TYPE_REG);
     lfs_unmount(&lfs) => 0;
 '''
 
-[[case]] # expanding superblock with power cycle
-define.LFS_BLOCK_CYCLES = [32, 33, 1]
-define.N = [10, 100, 1000]
+# expanding superblock with power cycle
+[cases.test_superblocks_expand_power_cycle]
+defines.BLOCK_CYCLES = [32, 33, 1]
+defines.N = [10, 100, 1000]
 code = '''
-    lfs_format(&lfs, &cfg) => 0;
+    lfs_t lfs;
+    lfs_format(&lfs, cfg) => 0;
     for (int i = 0; i < N; i++) {
-        lfs_mount(&lfs, &cfg) => 0;
+        lfs_mount(&lfs, cfg) => 0;
         // remove lingering dummy?
-        err = lfs_stat(&lfs, "dummy", &info);
+        struct lfs_info info;
+        int err = lfs_stat(&lfs, "dummy", &info);
         assert(err == 0 || (err == LFS_ERR_NOENT && i == 0));
         if (!err) {
             assert(strcmp(info.name, "dummy") == 0);
@@ -70,6 +150,7 @@ code = '''
             lfs_remove(&lfs, "dummy") => 0;
         }
 
+        lfs_file_t file;
         lfs_file_open(&lfs, &file, "dummy",
                 LFS_O_WRONLY | LFS_O_CREAT | LFS_O_EXCL) => 0;
         lfs_file_close(&lfs, &file) => 0;
@@ -80,26 +161,30 @@ code = '''
     }
 
     // one last check after power-cycle
-    lfs_mount(&lfs, &cfg) => 0;
+    lfs_mount(&lfs, cfg) => 0;
+    struct lfs_info info;
     lfs_stat(&lfs, "dummy", &info) => 0;
     assert(strcmp(info.name, "dummy") == 0);
     assert(info.type == LFS_TYPE_REG);
     lfs_unmount(&lfs) => 0;
 '''
 
-[[case]] # reentrant expanding superblock
-define.LFS_BLOCK_CYCLES = [2, 1]
-define.N = 24
+# reentrant expanding superblock
+[cases.test_superblocks_reentrant_expand]
+defines.BLOCK_CYCLES = [2, 1]
+defines.N = 24
 reentrant = true
 code = '''
-    err = lfs_mount(&lfs, &cfg);
+    lfs_t lfs;
+    int err = lfs_mount(&lfs, cfg);
     if (err) {
-        lfs_format(&lfs, &cfg) => 0;
-        lfs_mount(&lfs, &cfg) => 0;
+        lfs_format(&lfs, cfg) => 0;
+        lfs_mount(&lfs, cfg) => 0;
     }
 
     for (int i = 0; i < N; i++) {
         // remove lingering dummy?
+        struct lfs_info info;
         err = lfs_stat(&lfs, "dummy", &info);
         assert(err == 0 || (err == LFS_ERR_NOENT && i == 0));
         if (!err) {
@@ -108,6 +193,7 @@ code = '''
             lfs_remove(&lfs, "dummy") => 0;
         }
 
+        lfs_file_t file;
         lfs_file_open(&lfs, &file, "dummy",
                 LFS_O_WRONLY | LFS_O_CREAT | LFS_O_EXCL) => 0;
         lfs_file_close(&lfs, &file) => 0;
@@ -119,9 +205,262 @@ code = '''
     lfs_unmount(&lfs) => 0;
 
     // one last check after power-cycle
-    lfs_mount(&lfs, &cfg) => 0;
+    lfs_mount(&lfs, cfg) => 0;
+    struct lfs_info info;
     lfs_stat(&lfs, "dummy", &info) => 0;
     assert(strcmp(info.name, "dummy") == 0);
     assert(info.type == LFS_TYPE_REG);
     lfs_unmount(&lfs) => 0;
 '''
+
+# mount with unknown block_count
+[cases.test_superblocks_unknown_blocks]
+code = '''
+    lfs_t lfs;
+    lfs_format(&lfs, cfg) => 0;
+
+    // known block_size/block_count
+    cfg->block_size = BLOCK_SIZE;
+    cfg->block_count = BLOCK_COUNT;
+    lfs_mount(&lfs, cfg) => 0;
+    struct lfs_fsinfo fsinfo;
+    lfs_fs_stat(&lfs, &fsinfo) => 0;
+    assert(fsinfo.block_size == BLOCK_SIZE);
+    assert(fsinfo.block_count == BLOCK_COUNT);
+    lfs_unmount(&lfs) => 0;
+
+    // unknown block_count
+    cfg->block_size = BLOCK_SIZE;
+    cfg->block_count = 0;
+    lfs_mount(&lfs, cfg) => 0;
+    lfs_fs_stat(&lfs, &fsinfo) => 0;
+    assert(fsinfo.block_size == BLOCK_SIZE);
+    assert(fsinfo.block_count == BLOCK_COUNT);
+    lfs_unmount(&lfs) => 0;
+
+    // do some work
+    lfs_mount(&lfs, cfg) => 0;
+    lfs_fs_stat(&lfs, &fsinfo) => 0;
+    assert(fsinfo.block_size == BLOCK_SIZE);
+    assert(fsinfo.block_count == BLOCK_COUNT);
+    lfs_file_t file;
+    lfs_file_open(&lfs, &file, "test",
+            LFS_O_CREAT | LFS_O_EXCL | LFS_O_WRONLY) => 0;
+    lfs_file_write(&lfs, &file, "hello!", 6) => 6;
+    lfs_file_close(&lfs, &file) => 0;
+    lfs_unmount(&lfs) => 0;
+
+    lfs_mount(&lfs, cfg) => 0;
+    lfs_fs_stat(&lfs, &fsinfo) => 0;
+    assert(fsinfo.block_size == BLOCK_SIZE);
+    assert(fsinfo.block_count == BLOCK_COUNT);
+    lfs_file_open(&lfs, &file, "test", LFS_O_RDONLY) => 0;
+    uint8_t buffer[256];
+    lfs_file_read(&lfs, &file, buffer, sizeof(buffer)) => 6;
+    lfs_file_close(&lfs, &file) => 0;
+    assert(memcmp(buffer, "hello!", 6) == 0);
+    lfs_unmount(&lfs) => 0;
+'''
+
+# mount with blocks fewer than the erase_count
+[cases.test_superblocks_fewer_blocks]
+defines.BLOCK_COUNT = ['ERASE_COUNT/2', 'ERASE_COUNT/4', '2']
+code = '''
+    lfs_t lfs;
+    lfs_format(&lfs, cfg) => 0;
+
+    // known block_size/block_count
+    cfg->block_size = BLOCK_SIZE;
+    cfg->block_count = BLOCK_COUNT;
+    lfs_mount(&lfs, cfg) => 0;
+    struct lfs_fsinfo fsinfo;
+    lfs_fs_stat(&lfs, &fsinfo) => 0;
+    assert(fsinfo.block_size == BLOCK_SIZE);
+    assert(fsinfo.block_count == BLOCK_COUNT);
+    lfs_unmount(&lfs) => 0;
+
+    // incorrect block_count
+    cfg->block_size = BLOCK_SIZE;
+    cfg->block_count = ERASE_COUNT;
+    lfs_mount(&lfs, cfg) => LFS_ERR_INVAL;
+
+    // unknown block_count
+    cfg->block_size = BLOCK_SIZE;
+    cfg->block_count = 0;
+    lfs_mount(&lfs, cfg) => 0;
+    lfs_fs_stat(&lfs, &fsinfo) => 0;
+    assert(fsinfo.block_size == BLOCK_SIZE);
+    assert(fsinfo.block_count == BLOCK_COUNT);
+    lfs_unmount(&lfs) => 0;
+
+    // do some work
+    lfs_mount(&lfs, cfg) => 0;
+    lfs_fs_stat(&lfs, &fsinfo) => 0;
+    assert(fsinfo.block_size == BLOCK_SIZE);
+    assert(fsinfo.block_count == BLOCK_COUNT);
+    lfs_file_t file;
+    lfs_file_open(&lfs, &file, "test",
+            LFS_O_CREAT | LFS_O_EXCL | LFS_O_WRONLY) => 0;
+    lfs_file_write(&lfs, &file, "hello!", 6) => 6;
+    lfs_file_close(&lfs, &file) => 0;
+    lfs_unmount(&lfs) => 0;
+
+    lfs_mount(&lfs, cfg) => 0;
+    lfs_fs_stat(&lfs, &fsinfo) => 0;
+    assert(fsinfo.block_size == BLOCK_SIZE);
+    assert(fsinfo.block_count == BLOCK_COUNT);
+    lfs_file_open(&lfs, &file, "test", LFS_O_RDONLY) => 0;
+    uint8_t buffer[256];
+    lfs_file_read(&lfs, &file, buffer, sizeof(buffer)) => 6;
+    lfs_file_close(&lfs, &file) => 0;
+    assert(memcmp(buffer, "hello!", 6) == 0);
+    lfs_unmount(&lfs) => 0;
+'''
+
+# mount with more blocks than the erase_count
+[cases.test_superblocks_more_blocks]
+defines.FORMAT_BLOCK_COUNT = '2*ERASE_COUNT'
+in = 'lfs.c'
+code = '''
+    lfs_t lfs;
+    lfs_init(&lfs, cfg) => 0;
+    lfs.block_count = BLOCK_COUNT;
+
+    lfs_mdir_t root = {
+        .pair = {0, 0}, // make sure this goes into block 0
+        .rev = 0,
+        .off = sizeof(uint32_t),
+        .etag = 0xffffffff,
+        .count = 0,
+        .tail = {LFS_BLOCK_NULL, LFS_BLOCK_NULL},
+        .erased = false,
+        .split = false,
+    };
+
+    lfs_superblock_t superblock = {
+        .version     = LFS_DISK_VERSION,
+        .block_size  = BLOCK_SIZE,
+        .block_count = FORMAT_BLOCK_COUNT,
+        .name_max    = LFS_NAME_MAX,
+        .file_max    = LFS_FILE_MAX,
+        .attr_max    = LFS_ATTR_MAX,
+    };
+
+    lfs_superblock_tole32(&superblock);
+    lfs_dir_commit(&lfs, &root, LFS_MKATTRS(
+            {LFS_MKTAG(LFS_TYPE_CREATE, 0, 0), NULL},
+            {LFS_MKTAG(LFS_TYPE_SUPERBLOCK, 0, 8), "littlefs"},
+            {LFS_MKTAG(LFS_TYPE_INLINESTRUCT, 0, sizeof(superblock)),
+                &superblock})) => 0;
+    lfs_deinit(&lfs) => 0;
+
+    // known block_size/block_count
+    cfg->block_size = BLOCK_SIZE;
+    cfg->block_count = BLOCK_COUNT;
+    lfs_mount(&lfs, cfg) => LFS_ERR_INVAL;
+'''
+
+# mount and grow the filesystem
+[cases.test_superblocks_grow]
+defines.BLOCK_COUNT = ['ERASE_COUNT/2', 'ERASE_COUNT/4', '2']
+defines.BLOCK_COUNT_2 = 'ERASE_COUNT'
+defines.KNOWN_BLOCK_COUNT = [true, false]
+code = '''
+    lfs_t lfs;
+    lfs_format(&lfs, cfg) => 0;
+
+    if (KNOWN_BLOCK_COUNT) {
+        cfg->block_count = BLOCK_COUNT;
+    } else {
+        cfg->block_count = 0;
+    }
+
+    // mount with block_size < erase_size
+    lfs_mount(&lfs, cfg) => 0;
+    struct lfs_fsinfo fsinfo;
+    lfs_fs_stat(&lfs, &fsinfo) => 0;
+    assert(fsinfo.block_size == BLOCK_SIZE);
+    assert(fsinfo.block_count == BLOCK_COUNT);
+    lfs_unmount(&lfs) => 0;
+
+    // same size is a noop
+    lfs_mount(&lfs, cfg) => 0;
+    lfs_fs_grow(&lfs, BLOCK_COUNT) => 0;
+    lfs_fs_stat(&lfs, &fsinfo) => 0;
+    assert(fsinfo.block_size == BLOCK_SIZE);
+    assert(fsinfo.block_count == BLOCK_COUNT);
+    lfs_unmount(&lfs) => 0;
+
+    lfs_mount(&lfs, cfg) => 0;
+    lfs_fs_stat(&lfs, &fsinfo) => 0;
+    assert(fsinfo.block_size == BLOCK_SIZE);
+    assert(fsinfo.block_count == BLOCK_COUNT);
+    lfs_unmount(&lfs) => 0;
+
+    // grow to new size
+    lfs_mount(&lfs, cfg) => 0;
+    lfs_fs_grow(&lfs, BLOCK_COUNT_2) => 0;
+    lfs_fs_stat(&lfs, &fsinfo) => 0;
+    assert(fsinfo.block_size == BLOCK_SIZE);
+    assert(fsinfo.block_count == BLOCK_COUNT_2);
+    lfs_unmount(&lfs) => 0;
+
+    if (KNOWN_BLOCK_COUNT) {
+        cfg->block_count = BLOCK_COUNT_2;
+    } else {
+        cfg->block_count = 0;
+    }
+
+    lfs_mount(&lfs, cfg) => 0;
+    lfs_fs_stat(&lfs, &fsinfo) => 0;
+    assert(fsinfo.block_size == BLOCK_SIZE);
+    assert(fsinfo.block_count == BLOCK_COUNT_2);
+    lfs_unmount(&lfs) => 0;
+
+    // mounting with the previous size should fail
+    cfg->block_count = BLOCK_COUNT;
+    lfs_mount(&lfs, cfg) => LFS_ERR_INVAL;
+
+    if (KNOWN_BLOCK_COUNT) {
+        cfg->block_count = BLOCK_COUNT_2;
+    } else {
+        cfg->block_count = 0;
+    }
+
+    // same size is a noop
+    lfs_mount(&lfs, cfg) => 0;
+    lfs_fs_grow(&lfs, BLOCK_COUNT_2) => 0;
+    lfs_fs_stat(&lfs, &fsinfo) => 0;
+    assert(fsinfo.block_size == BLOCK_SIZE);
+    assert(fsinfo.block_count == BLOCK_COUNT_2);
+    lfs_unmount(&lfs) => 0;
+
+    lfs_mount(&lfs, cfg) => 0;
+    lfs_fs_stat(&lfs, &fsinfo) => 0;
+    assert(fsinfo.block_size == BLOCK_SIZE);
+    assert(fsinfo.block_count == BLOCK_COUNT_2);
+    lfs_unmount(&lfs) => 0;
+
+    // do some work
+    lfs_mount(&lfs, cfg) => 0;
+    lfs_fs_stat(&lfs, &fsinfo) => 0;
+    assert(fsinfo.block_size == BLOCK_SIZE);
+    assert(fsinfo.block_count == BLOCK_COUNT_2);
+    lfs_file_t file;
+    lfs_file_open(&lfs, &file, "test",
+            LFS_O_CREAT | LFS_O_EXCL | LFS_O_WRONLY) => 0;
+    lfs_file_write(&lfs, &file, "hello!", 6) => 6;
+    lfs_file_close(&lfs, &file) => 0;
+    lfs_unmount(&lfs) => 0;
+
+    lfs_mount(&lfs, cfg) => 0;
+    lfs_fs_stat(&lfs, &fsinfo) => 0;
+    assert(fsinfo.block_size == BLOCK_SIZE);
+    assert(fsinfo.block_count == BLOCK_COUNT_2);
+    lfs_file_open(&lfs, &file, "test", LFS_O_RDONLY) => 0;
+    uint8_t buffer[256];
+    lfs_file_read(&lfs, &file, buffer, sizeof(buffer)) => 6;
+    lfs_file_close(&lfs, &file) => 0;
+    assert(memcmp(buffer, "hello!", 6) == 0);
+    lfs_unmount(&lfs) => 0;
+'''
diff --git a/tests/test_truncate.toml b/tests/test_truncate.toml
index 850d7aa..2f10e95 100644
--- a/tests/test_truncate.toml
+++ b/tests/test_truncate.toml
@@ -1,23 +1,29 @@
-[[case]] # simple truncate
-define.MEDIUMSIZE = [32, 2048]
-define.LARGESIZE = 8192
+# simple truncate
+[cases.test_truncate_simple]
+defines.MEDIUMSIZE = [31, 32, 33, 511, 512, 513, 2047, 2048, 2049]
+defines.LARGESIZE = [32, 33, 512, 513, 2048, 2049, 8192, 8193]
+if = 'MEDIUMSIZE < LARGESIZE'
 code = '''
-    lfs_format(&lfs, &cfg) => 0;
-    lfs_mount(&lfs, &cfg) => 0;
+    lfs_t lfs;
+    lfs_format(&lfs, cfg) => 0;
+    lfs_mount(&lfs, cfg) => 0;
+    lfs_file_t file;
     lfs_file_open(&lfs, &file, "baldynoop",
             LFS_O_WRONLY | LFS_O_CREAT) => 0;
 
+    uint8_t buffer[1024];
     strcpy((char*)buffer, "hair");
-    size = strlen((char*)buffer);
+    size_t size = strlen((char*)buffer);
     for (lfs_off_t j = 0; j < LARGESIZE; j += size) {
-        lfs_file_write(&lfs, &file, buffer, size) => size;
+        lfs_file_write(&lfs, &file, buffer, lfs_min(size, LARGESIZE-j))
+                => lfs_min(size, LARGESIZE-j);
     }
     lfs_file_size(&lfs, &file) => LARGESIZE;
 
     lfs_file_close(&lfs, &file) => 0;
     lfs_unmount(&lfs) => 0;
     
-    lfs_mount(&lfs, &cfg) => 0;
+    lfs_mount(&lfs, cfg) => 0;
     lfs_file_open(&lfs, &file, "baldynoop", LFS_O_RDWR) => 0;
     lfs_file_size(&lfs, &file) => LARGESIZE;
 
@@ -27,14 +33,15 @@ code = '''
     lfs_file_close(&lfs, &file) => 0;
     lfs_unmount(&lfs) => 0;
 
-    lfs_mount(&lfs, &cfg) => 0;
+    lfs_mount(&lfs, cfg) => 0;
     lfs_file_open(&lfs, &file, "baldynoop", LFS_O_RDONLY) => 0;
     lfs_file_size(&lfs, &file) => MEDIUMSIZE;
 
     size = strlen("hair");
     for (lfs_off_t j = 0; j < MEDIUMSIZE; j += size) {
-        lfs_file_read(&lfs, &file, buffer, size) => size;
-        memcmp(buffer, "hair", size) => 0;
+        lfs_file_read(&lfs, &file, buffer, lfs_min(size, MEDIUMSIZE-j))
+                => lfs_min(size, MEDIUMSIZE-j);
+        memcmp(buffer, "hair", lfs_min(size, MEDIUMSIZE-j)) => 0;
     }
     lfs_file_read(&lfs, &file, buffer, size) => 0;
 
@@ -42,26 +49,32 @@ code = '''
     lfs_unmount(&lfs) => 0;
 '''
 
-[[case]] # truncate and read
-define.MEDIUMSIZE = [32, 2048]
-define.LARGESIZE = 8192
+# truncate and read
+[cases.test_truncate_read]
+defines.MEDIUMSIZE = [31, 32, 33, 511, 512, 513, 2047, 2048, 2049]
+defines.LARGESIZE = [32, 33, 512, 513, 2048, 2049, 8192, 8193]
+if = 'MEDIUMSIZE < LARGESIZE'
 code = '''
-    lfs_format(&lfs, &cfg) => 0;
-    lfs_mount(&lfs, &cfg) => 0;
+    lfs_t lfs;
+    lfs_format(&lfs, cfg) => 0;
+    lfs_mount(&lfs, cfg) => 0;
+    lfs_file_t file;
     lfs_file_open(&lfs, &file, "baldyread",
             LFS_O_WRONLY | LFS_O_CREAT) => 0;
 
+    uint8_t buffer[1024];
     strcpy((char*)buffer, "hair");
-    size = strlen((char*)buffer);
+    size_t size = strlen((char*)buffer);
     for (lfs_off_t j = 0; j < LARGESIZE; j += size) {
-        lfs_file_write(&lfs, &file, buffer, size) => size;
+        lfs_file_write(&lfs, &file, buffer, lfs_min(size, LARGESIZE-j))
+                => lfs_min(size, LARGESIZE-j);
     }
     lfs_file_size(&lfs, &file) => LARGESIZE;
 
     lfs_file_close(&lfs, &file) => 0;
     lfs_unmount(&lfs) => 0;
 
-    lfs_mount(&lfs, &cfg) => 0;
+    lfs_mount(&lfs, cfg) => 0;
     lfs_file_open(&lfs, &file, "baldyread", LFS_O_RDWR) => 0;
     lfs_file_size(&lfs, &file) => LARGESIZE;
 
@@ -70,22 +83,24 @@ code = '''
 
     size = strlen("hair");
     for (lfs_off_t j = 0; j < MEDIUMSIZE; j += size) {
-        lfs_file_read(&lfs, &file, buffer, size) => size;
-        memcmp(buffer, "hair", size) => 0;
+        lfs_file_read(&lfs, &file, buffer, lfs_min(size, MEDIUMSIZE-j))
+                => lfs_min(size, MEDIUMSIZE-j);
+        memcmp(buffer, "hair", lfs_min(size, MEDIUMSIZE-j)) => 0;
     }
     lfs_file_read(&lfs, &file, buffer, size) => 0;
 
     lfs_file_close(&lfs, &file) => 0;
     lfs_unmount(&lfs) => 0;
 
-    lfs_mount(&lfs, &cfg) => 0;
+    lfs_mount(&lfs, cfg) => 0;
     lfs_file_open(&lfs, &file, "baldyread", LFS_O_RDONLY) => 0;
     lfs_file_size(&lfs, &file) => MEDIUMSIZE;
 
     size = strlen("hair");
     for (lfs_off_t j = 0; j < MEDIUMSIZE; j += size) {
-        lfs_file_read(&lfs, &file, buffer, size) => size;
-        memcmp(buffer, "hair", size) => 0;
+        lfs_file_read(&lfs, &file, buffer, lfs_min(size, MEDIUMSIZE-j))
+                => lfs_min(size, MEDIUMSIZE-j);
+        memcmp(buffer, "hair", lfs_min(size, MEDIUMSIZE-j)) => 0;
     }
     lfs_file_read(&lfs, &file, buffer, size) => 0;
 
@@ -93,14 +108,18 @@ code = '''
     lfs_unmount(&lfs) => 0;
 '''
 
-[[case]] # write, truncate, and read
+# write, truncate, and read
+[cases.test_truncate_write_read]
 code = '''
-    lfs_format(&lfs, &cfg) => 0;
-    lfs_mount(&lfs, &cfg) => 0;
+    lfs_t lfs;
+    lfs_format(&lfs, cfg) => 0;
+    lfs_mount(&lfs, cfg) => 0;
+    lfs_file_t file;
     lfs_file_open(&lfs, &file, "sequence",
             LFS_O_RDWR | LFS_O_CREAT | LFS_O_TRUNC) => 0;
 
-    size = lfs_min(lfs.cfg->cache_size, sizeof(buffer)/2);
+    uint8_t buffer[1024];
+    size_t size = lfs_min(lfs.cfg->cache_size, sizeof(buffer)/2);
     lfs_size_t qsize = size / 4;
     uint8_t *wb = buffer;
     uint8_t *rb = buffer + size;
@@ -136,7 +155,7 @@ code = '''
     lfs_file_truncate(&lfs, &file, trunc) => 0;
     lfs_file_tell(&lfs, &file) => qsize;
     lfs_file_size(&lfs, &file) => trunc;
-    
+
     /* Read should produce second quarter */
     lfs_file_read(&lfs, &file, rb, size) => trunc - qsize;
     memcmp(rb, wb + qsize, trunc - qsize) => 0;
@@ -145,50 +164,60 @@ code = '''
     lfs_unmount(&lfs) => 0;
 '''
 
-[[case]] # truncate and write
-define.MEDIUMSIZE = [32, 2048]
-define.LARGESIZE = 8192
+# truncate and write
+[cases.test_truncate_write]
+defines.MEDIUMSIZE = [31, 32, 33, 511, 512, 513, 2047, 2048, 2049]
+defines.LARGESIZE = [32, 33, 512, 513, 2048, 2049, 8192, 8193]
+if = 'MEDIUMSIZE < LARGESIZE'
 code = '''
-    lfs_format(&lfs, &cfg) => 0;
-    lfs_mount(&lfs, &cfg) => 0;
+    lfs_t lfs;
+    lfs_format(&lfs, cfg) => 0;
+    lfs_mount(&lfs, cfg) => 0;
+    lfs_file_t file;
     lfs_file_open(&lfs, &file, "baldywrite",
             LFS_O_WRONLY | LFS_O_CREAT) => 0;
 
+    uint8_t buffer[1024];
     strcpy((char*)buffer, "hair");
-    size = strlen((char*)buffer);
+    size_t size = strlen((char*)buffer);
     for (lfs_off_t j = 0; j < LARGESIZE; j += size) {
-        lfs_file_write(&lfs, &file, buffer, size) => size;
+        lfs_file_write(&lfs, &file, buffer, lfs_min(size, LARGESIZE-j))
+                => lfs_min(size, LARGESIZE-j);
     }
     lfs_file_size(&lfs, &file) => LARGESIZE;
 
     lfs_file_close(&lfs, &file) => 0;
     lfs_unmount(&lfs) => 0;
 
-    lfs_mount(&lfs, &cfg) => 0;
+    lfs_mount(&lfs, cfg) => 0;
     lfs_file_open(&lfs, &file, "baldywrite", LFS_O_RDWR) => 0;
     lfs_file_size(&lfs, &file) => LARGESIZE;
 
+    /* truncate */
     lfs_file_truncate(&lfs, &file, MEDIUMSIZE) => 0;
     lfs_file_size(&lfs, &file) => MEDIUMSIZE;
 
+    /* and write */
     strcpy((char*)buffer, "bald");
     size = strlen((char*)buffer);
     for (lfs_off_t j = 0; j < MEDIUMSIZE; j += size) {
-        lfs_file_write(&lfs, &file, buffer, size) => size;
+        lfs_file_write(&lfs, &file, buffer, lfs_min(size, MEDIUMSIZE-j))
+                => lfs_min(size, MEDIUMSIZE-j);
     }
     lfs_file_size(&lfs, &file) => MEDIUMSIZE;
 
     lfs_file_close(&lfs, &file) => 0;
     lfs_unmount(&lfs) => 0;
 
-    lfs_mount(&lfs, &cfg) => 0;
+    lfs_mount(&lfs, cfg) => 0;
     lfs_file_open(&lfs, &file, "baldywrite", LFS_O_RDONLY) => 0;
     lfs_file_size(&lfs, &file) => MEDIUMSIZE;
 
     size = strlen("bald");
     for (lfs_off_t j = 0; j < MEDIUMSIZE; j += size) {
-        lfs_file_read(&lfs, &file, buffer, size) => size;
-        memcmp(buffer, "bald", size) => 0;
+        lfs_file_read(&lfs, &file, buffer, lfs_min(size, MEDIUMSIZE-j))
+                => lfs_min(size, MEDIUMSIZE-j);
+        memcmp(buffer, "bald", lfs_min(size, MEDIUMSIZE-j)) => 0;
     }
     lfs_file_read(&lfs, &file, buffer, size) => 0;
 
@@ -196,30 +225,35 @@ code = '''
     lfs_unmount(&lfs) => 0;
 '''
 
-[[case]] # truncate write under powerloss
-define.SMALLSIZE = [4, 512]
-define.MEDIUMSIZE = [32, 1024]
-define.LARGESIZE = 2048
+# truncate write under powerloss
+[cases.test_truncate_reentrant_write]
+defines.SMALLSIZE = [4, 512]
+defines.MEDIUMSIZE = [0, 3, 4, 5, 31, 32, 33, 511, 512, 513, 1023, 1024, 1025]
+defines.LARGESIZE = 2048
 reentrant = true
 code = '''
-    err = lfs_mount(&lfs, &cfg);
+    lfs_t lfs;
+    int err = lfs_mount(&lfs, cfg);
     if (err) {
-        lfs_format(&lfs, &cfg) => 0;
-        lfs_mount(&lfs, &cfg) => 0;
+        lfs_format(&lfs, cfg) => 0;
+        lfs_mount(&lfs, cfg) => 0;
     }
+    lfs_file_t file;
     err = lfs_file_open(&lfs, &file, "baldy", LFS_O_RDONLY);
     assert(!err || err == LFS_ERR_NOENT);
     if (!err) {
-        size = lfs_file_size(&lfs, &file);
+        size_t size = lfs_file_size(&lfs, &file);
         assert(size == 0 ||
-                size == LARGESIZE ||
-                size == MEDIUMSIZE ||
-                size == SMALLSIZE);
+                size == (size_t)LARGESIZE ||
+                size == (size_t)MEDIUMSIZE ||
+                size == (size_t)SMALLSIZE);
         for (lfs_off_t j = 0; j < size; j += 4) {
-            lfs_file_read(&lfs, &file, buffer, 4) => 4;
-            assert(memcmp(buffer, "hair", 4) == 0 ||
-                   memcmp(buffer, "bald", 4) == 0 ||
-                   memcmp(buffer, "comb", 4) == 0);
+            uint8_t buffer[1024];
+            lfs_file_read(&lfs, &file, buffer, lfs_min(4, size-j))
+                    => lfs_min(4, size-j);
+            assert(memcmp(buffer, "hair", lfs_min(4, size-j)) == 0 ||
+                   memcmp(buffer, "bald", lfs_min(4, size-j)) == 0 ||
+                   memcmp(buffer, "comb", lfs_min(4, size-j)) == 0);
         }
         lfs_file_close(&lfs, &file) => 0;
     }
@@ -227,22 +261,27 @@ code = '''
     lfs_file_open(&lfs, &file, "baldy",
             LFS_O_WRONLY | LFS_O_CREAT | LFS_O_TRUNC) => 0;
     lfs_file_size(&lfs, &file) => 0;
+    uint8_t buffer[1024];
     strcpy((char*)buffer, "hair");
-    size = strlen((char*)buffer);
+    size_t size = strlen((char*)buffer);
     for (lfs_off_t j = 0; j < LARGESIZE; j += size) {
-        lfs_file_write(&lfs, &file, buffer, size) => size;
+        lfs_file_write(&lfs, &file, buffer, lfs_min(size, LARGESIZE-j))
+                => lfs_min(size, LARGESIZE-j);
     }
     lfs_file_size(&lfs, &file) => LARGESIZE;
     lfs_file_close(&lfs, &file) => 0;
 
     lfs_file_open(&lfs, &file, "baldy", LFS_O_RDWR) => 0;
     lfs_file_size(&lfs, &file) => LARGESIZE;
+    /* truncate */
     lfs_file_truncate(&lfs, &file, MEDIUMSIZE) => 0;
     lfs_file_size(&lfs, &file) => MEDIUMSIZE;
+    /* and write */
     strcpy((char*)buffer, "bald");
     size = strlen((char*)buffer);
     for (lfs_off_t j = 0; j < MEDIUMSIZE; j += size) {
-        lfs_file_write(&lfs, &file, buffer, size) => size;
+        lfs_file_write(&lfs, &file, buffer, lfs_min(size, MEDIUMSIZE-j))
+                => lfs_min(size, MEDIUMSIZE-j);
     }
     lfs_file_size(&lfs, &file) => MEDIUMSIZE;
     lfs_file_close(&lfs, &file) => 0;
@@ -254,7 +293,8 @@ code = '''
     strcpy((char*)buffer, "comb");
     size = strlen((char*)buffer);
     for (lfs_off_t j = 0; j < SMALLSIZE; j += size) {
-        lfs_file_write(&lfs, &file, buffer, size) => size;
+        lfs_file_write(&lfs, &file, buffer, lfs_min(size, SMALLSIZE-j))
+                => lfs_min(size, SMALLSIZE-j);
     }
     lfs_file_size(&lfs, &file) => SMALLSIZE;
     lfs_file_close(&lfs, &file) => 0;
@@ -262,12 +302,14 @@ code = '''
     lfs_unmount(&lfs) => 0;
 '''
 
-[[case]] # more aggressive general truncation tests
-define.CONFIG = 'range(6)'
-define.SMALLSIZE = 32
-define.MEDIUMSIZE = 2048
-define.LARGESIZE = 8192
+# more aggressive general truncation tests
+[cases.test_truncate_aggressive]
+defines.CONFIG = 'range(6)'
+defines.SMALLSIZE = 32
+defines.MEDIUMSIZE = 2048
+defines.LARGESIZE = 8192
 code = '''
+    lfs_t lfs;
     #define COUNT 5
     const struct {
         lfs_off_t startsizes[COUNT];
@@ -312,16 +354,19 @@ code = '''
     const lfs_off_t *hotsizes   = configs[CONFIG].hotsizes;
     const lfs_off_t *coldsizes  = configs[CONFIG].coldsizes;
 
-    lfs_format(&lfs, &cfg) => 0;
-    lfs_mount(&lfs, &cfg) => 0;
+    lfs_format(&lfs, cfg) => 0;
+    lfs_mount(&lfs, cfg) => 0;
 
     for (unsigned i = 0; i < COUNT; i++) {
+        char path[1024];
         sprintf(path, "hairyhead%d", i);
+        lfs_file_t file;
         lfs_file_open(&lfs, &file, path,
                 LFS_O_WRONLY | LFS_O_CREAT | LFS_O_TRUNC) => 0;
 
+        uint8_t buffer[1024];
         strcpy((char*)buffer, "hair");
-        size = strlen((char*)buffer);
+        size_t size = strlen((char*)buffer);
         for (lfs_off_t j = 0; j < startsizes[i]; j += size) {
             lfs_file_write(&lfs, &file, buffer, size) => size;
         }
@@ -340,21 +385,25 @@ code = '''
 
     lfs_unmount(&lfs) => 0;
 
-    lfs_mount(&lfs, &cfg) => 0;
+    lfs_mount(&lfs, cfg) => 0;
 
     for (unsigned i = 0; i < COUNT; i++) {
+        char path[1024];
         sprintf(path, "hairyhead%d", i);
+        lfs_file_t file;
         lfs_file_open(&lfs, &file, path, LFS_O_RDWR) => 0;
         lfs_file_size(&lfs, &file) => hotsizes[i];
 
-        size = strlen("hair");
+        size_t size = strlen("hair");
         lfs_off_t j = 0;
         for (; j < startsizes[i] && j < hotsizes[i]; j += size) {
+            uint8_t buffer[1024];
             lfs_file_read(&lfs, &file, buffer, size) => size;
             memcmp(buffer, "hair", size) => 0;
         }
 
         for (; j < hotsizes[i]; j += size) {
+            uint8_t buffer[1024];
             lfs_file_read(&lfs, &file, buffer, size) => size;
             memcmp(buffer, "\0\0\0\0", size) => 0;
         }
@@ -367,22 +416,26 @@ code = '''
 
     lfs_unmount(&lfs) => 0;
 
-    lfs_mount(&lfs, &cfg) => 0;
+    lfs_mount(&lfs, cfg) => 0;
 
     for (unsigned i = 0; i < COUNT; i++) {
+        char path[1024];
         sprintf(path, "hairyhead%d", i);
+        lfs_file_t file;
         lfs_file_open(&lfs, &file, path, LFS_O_RDONLY) => 0;
         lfs_file_size(&lfs, &file) => coldsizes[i];
 
-        size = strlen("hair");
+        size_t size = strlen("hair");
         lfs_off_t j = 0;
         for (; j < startsizes[i] && j < hotsizes[i] && j < coldsizes[i];
                 j += size) {
+            uint8_t buffer[1024];
             lfs_file_read(&lfs, &file, buffer, size) => size;
             memcmp(buffer, "hair", size) => 0;
         }
 
         for (; j < coldsizes[i]; j += size) {
+            uint8_t buffer[1024];
             lfs_file_read(&lfs, &file, buffer, size) => size;
             memcmp(buffer, "\0\0\0\0", size) => 0;
         }
@@ -393,21 +446,26 @@ code = '''
     lfs_unmount(&lfs) => 0;
 '''
 
-[[case]] # noop truncate
-define.MEDIUMSIZE = [32, 2048]
+# noop truncate
+[cases.test_truncate_nop]
+defines.MEDIUMSIZE = [32, 33, 512, 513, 2048, 2049, 8192, 8193]
 code = '''
-    lfs_format(&lfs, &cfg) => 0;
-    lfs_mount(&lfs, &cfg) => 0;
+    lfs_t lfs;
+    lfs_format(&lfs, cfg) => 0;
+    lfs_mount(&lfs, cfg) => 0;
+    lfs_file_t file;
     lfs_file_open(&lfs, &file, "baldynoop",
             LFS_O_RDWR | LFS_O_CREAT) => 0;
 
+    uint8_t buffer[1024];
     strcpy((char*)buffer, "hair");
-    size = strlen((char*)buffer);
+    size_t size = strlen((char*)buffer);
     for (lfs_off_t j = 0; j < MEDIUMSIZE; j += size) {
-        lfs_file_write(&lfs, &file, buffer, size) => size;
+        lfs_file_write(&lfs, &file, buffer, lfs_min(size, MEDIUMSIZE-j))
+                => lfs_min(size, MEDIUMSIZE-j);
 
         // this truncate should do nothing
-        lfs_file_truncate(&lfs, &file, j+size) => 0;
+        lfs_file_truncate(&lfs, &file, j+lfs_min(size, MEDIUMSIZE-j)) => 0;
     }
     lfs_file_size(&lfs, &file) => MEDIUMSIZE;
 
@@ -417,8 +475,9 @@ code = '''
     lfs_file_size(&lfs, &file) => MEDIUMSIZE;
 
     for (lfs_off_t j = 0; j < MEDIUMSIZE; j += size) {
-        lfs_file_read(&lfs, &file, buffer, size) => size;
-        memcmp(buffer, "hair", size) => 0;
+        lfs_file_read(&lfs, &file, buffer, lfs_min(size, MEDIUMSIZE-j))
+                => lfs_min(size, MEDIUMSIZE-j);
+        memcmp(buffer, "hair", lfs_min(size, MEDIUMSIZE-j)) => 0;
     }
     lfs_file_read(&lfs, &file, buffer, size) => 0;
 
@@ -426,12 +485,13 @@ code = '''
     lfs_unmount(&lfs) => 0;
 
     // still there after reboot?
-    lfs_mount(&lfs, &cfg) => 0;
+    lfs_mount(&lfs, cfg) => 0;
     lfs_file_open(&lfs, &file, "baldynoop", LFS_O_RDWR) => 0;
     lfs_file_size(&lfs, &file) => MEDIUMSIZE;
     for (lfs_off_t j = 0; j < MEDIUMSIZE; j += size) {
-        lfs_file_read(&lfs, &file, buffer, size) => size;
-        memcmp(buffer, "hair", size) => 0;
+        lfs_file_read(&lfs, &file, buffer, lfs_min(size, MEDIUMSIZE-j))
+                => lfs_min(size, MEDIUMSIZE-j);
+        memcmp(buffer, "hair", lfs_min(size, MEDIUMSIZE-j)) => 0;
     }
     lfs_file_read(&lfs, &file, buffer, size) => 0;
     lfs_file_close(&lfs, &file) => 0;