Moar sorting

Use yield on aarch in spin locks
Fix regression in DNS client
2026-11-08 08:08:04 -10:00 · 2023-05-11 09:09:46 -07:00 · 2023-05-11 02:55:41 -07:00 · 2023-05-10 23:27:10 -07:00 · 2023-05-10 21:32:04 -07:00 · 2023-05-10 21:19:54 -07:00
21224 changed files with 481713 additions and 131206 deletions
@@ -8,6 +8,7 @@ AlwaysBreakBeforeMultilineStrings: false
 AllowShortFunctionsOnASingleLine: false
 KeepEmptyLinesAtTheStartOfBlocks: true
 ConstructorInitializerAllOnOneLineOrOnePerLine: true
+IncludeBlocks: Merge
 ---
 Language: Cpp
 AllowShortFunctionsOnASingleLine: false
@@ -1,3 +1,5 @@
 # -*- conf -*-
+*.gz                         binary
+/build/bootstrap/*.com       binary
 /build/bootstrap/*.com       binary
 /usr/share/zoneinfo/*        binary
@@ -4,6 +4,7 @@ on:
  push:
    branches:
      - "master"
+      - "flake"
  pull_request:
    branches:
      - "master"
@@ -20,8 +21,11 @@ jobs:
    steps:
      - uses: actions/checkout@v3

-      - name: support ape bins
-        run: sudo sh -c "echo ':APE:M::MZqFpD::/bin/sh:' >/proc/sys/fs/binfmt_misc/register"
+      - name: support ape bins 1
+        run: sudo cp build/bootstrap/ape.elf /usr/bin/ape
+
+      - name: support ape bins 2
+        run: sudo sh -c "echo ':APE:M::MZqFpD::/usr/bin/ape:' >/proc/sys/fs/binfmt_misc/register"

      - name: make matrix
        run: V=0 make -j2 MODE=${{ matrix.mode }}
@@ -1,6 +1,7 @@
 # -*- conf -*-

 /o
+/.prompt.jtlp

 # TODO: Find some way to have Python write to o/
 __pycache__
@@ -11,3 +12,5 @@ __pycache__
 /TAGS
 /bx_enh_dbg.ini
 /tool/emacs/*.elc
+/perf.data
+/perf.data.old
@@ -43,10 +43,10 @@
 				"forcealignargpointer=",
 				"noasan=",
 				"noubsan=",
-				"testonly=",
 				"donothing=",
 				"nosideeffect=",
-				"unreachable=",
+				"unreachable=",,
+				"notpossible=",
 				"thatispacked=",
 				"dontthrow=",
 				"nocallback=",
@@ -59,14 +59,21 @@
 #
 #   build/config.mk

-SHELL   = build/bootstrap/cocmd.com
-HOSTS  ?= freebsd openbsd netbsd rhel7 rhel5 win7 win10 xnu
+SHELL      = build/bootstrap/cocmd.com
+HOSTS     ?= freebsd openbsd netbsd rhel7 rhel5 xnu win10
+MAKEFLAGS += --no-builtin-rules

 .SUFFIXES:
 .DELETE_ON_ERROR:
 .FEATURES: output-sync
 .PHONY: all o bins check test depend tags

+ifneq ($(m),)
+ifeq ($(MODE),)
+MODE := $(m)
+endif
+endif
+
 all:	o
 o:	o/$(MODE)

@@ -80,6 +87,22 @@ o/$(MODE):			\
 	o/$(MODE)/examples	\
 	o/$(MODE)/third_party

+.STRICT = 1
+.PLEDGE = stdio rpath wpath cpath fattr proc
+.UNVEIL =			\
+	libc/integral		\
+	libc/disclaimer.inc	\
+	rwc:/dev/shm		\
+	rx:build/bootstrap	\
+	rx:o/third_party/gcc	\
+	/proc/stat		\
+	rw:/dev/null		\
+	w:o/stack.log		\
+	/etc/hosts		\
+	~/.runit.psk		\
+	/proc/self/status	\
+	/sys/devices/system/cpu/cpu0/cpufreq/scaling_governor
+
 PKGS =

 -include ~/.cosmo.mk
@@ -94,31 +117,36 @@ include libc/sysv/sysv.mk			# ├──SYSTEM SUPPORT
 include libc/nt/nt.mk				# │  You can do math
 include libc/intrin/intrin.mk			# │  You can use the stack
 include libc/linux/linux.mk			# │  You can manipulate arrays
-include libc/tinymath/tinymath.mk		# │  You can issue raw system calls
-include third_party/compiler_rt/compiler_rt.mk	# │
-include libc/bits/bits.mk			# │
+include third_party/compiler_rt/compiler_rt.mk	# │  You can issue raw system calls
+include libc/tinymath/tinymath.mk		# │
 include libc/str/str.mk				# │
 include third_party/xed/xed.mk			# │
+include third_party/puff/puff.mk		# │
 include third_party/zlib/zlib.mk		# │
+include third_party/double-conversion/dc.mk	# │
 include libc/elf/elf.mk				# │
 include ape/ape.mk				# │
-include libc/fmt/fmt.mk				#─┘
+include libc/fmt/fmt.mk				# │
+include libc/vga/vga.mk				#─┘
 include libc/calls/calls.mk			#─┐
+include third_party/getopt/getopt.mk		# │
 include libc/runtime/runtime.mk			# ├──SYSTEMS RUNTIME
 include libc/crt/crt.mk				# │  You can issue system calls
-include libc/rand/rand.mk			# │
-include libc/unicode/unicode.mk			# │
+include tool/hello/hello.mk			# │
+include third_party/nsync/nsync.mk		# │
 include third_party/dlmalloc/dlmalloc.mk	#─┘
 include libc/mem/mem.mk				#─┐
-include libc/zipos/zipos.mk			# ├──DYNAMIC RUNTIME
-include third_party/gdtoa/gdtoa.mk		# │  You can now use stdio
-include libc/time/time.mk			# │  You can finally call malloc()
-include libc/thread/thread.mk			# │
-include libc/alg/alg.mk				# │
+include third_party/gdtoa/gdtoa.mk		# ├──DYNAMIC RUNTIME
+include third_party/nsync/mem/mem.mk		# │  You can now use stdio
+include libc/thread/thread.mk			# │  You can finally call malloc()
+include libc/zipos/zipos.mk			# │
+include libc/time/time.mk			# │
 include libc/stdio/stdio.mk			# │
 include third_party/libcxx/libcxx.mk		# │
 include net/net.mk				# │
+include third_party/vqsort/vqsort.mk		# │
 include libc/log/log.mk				# │
+include third_party/ggml/ggml.mk		# │
 include third_party/bzip2/bzip2.mk		# │
 include dsp/core/core.mk			# │
 include libc/x/x.mk				# │
@@ -128,7 +156,8 @@ include dsp/mpeg/mpeg.mk			# │
 include dsp/dsp.mk				# │
 include third_party/zlib/gz/gz.mk		# │
 include third_party/musl/musl.mk		# │
-include third_party/getopt/getopt.mk		# │
+include third_party/intel/intel.mk		# │
+include third_party/aarch64/aarch64.mk		# │
 include libc/libc.mk				#─┘
 include libc/sock/sock.mk			#─┐
 include dsp/tty/tty.mk				# ├──ONLINE RUNTIME
@@ -139,15 +168,22 @@ include net/https/https.mk			# │
 include third_party/regex/regex.mk		#─┘
 include third_party/tidy/tidy.mk
 include third_party/third_party.mk
+include third_party/nsync/testing/testing.mk
 include libc/testlib/testlib.mk
 include tool/viz/lib/vizlib.mk
 include tool/args/args.mk
 include test/tool/args/test.mk
 include third_party/linenoise/linenoise.mk
 include third_party/maxmind/maxmind.mk
-include third_party/double-conversion/double-conversion.mk
+include net/finger/finger.mk
+include third_party/double-conversion/test/test.mk
 include third_party/lua/lua.mk
+include third_party/tr/tr.mk
+include third_party/sed/sed.mk
+include third_party/awk/awk.mk
+include third_party/hiredis/hiredis.mk
 include third_party/make/make.mk
+include third_party/ctags/ctags.mk
 include third_party/finger/finger.mk
 include third_party/argon2/argon2.mk
 include third_party/smallz4/smallz4.mk
@@ -164,6 +200,8 @@ include third_party/python/python.mk
 include tool/build/emucrt/emucrt.mk
 include tool/build/emubin/emubin.mk
 include tool/build/build.mk
+include tool/curl/curl.mk
+include third_party/qemu/qemu.mk
 include examples/examples.mk
 include examples/pyapp/pyapp.mk
 include examples/pylife/pylife.mk
@@ -178,7 +216,7 @@ include tool/hash/hash.mk
 include tool/net/net.mk
 include tool/viz/viz.mk
 include tool/tool.mk
-include test/libc/alg/test.mk
+include net/turfwar/turfwar.mk
 include test/libc/tinymath/test.mk
 include test/libc/intrin/test.mk
 include test/libc/mem/test.mk
@@ -186,22 +224,22 @@ include test/libc/nexgen32e/test.mk
 include test/libc/runtime/test.mk
 include test/libc/thread/test.mk
 include test/libc/sock/test.mk
-include test/libc/bits/test.mk
 include test/libc/str/test.mk
 include test/libc/log/test.mk
-include test/libc/unicode/test.mk
+include test/libc/str/test.mk
 include test/libc/calls/test.mk
 include test/libc/x/test.mk
 include test/libc/xed/test.mk
 include test/libc/fmt/test.mk
 include test/libc/dns/test.mk
-include test/libc/rand/test.mk
 include test/libc/time/test.mk
 include test/libc/stdio/test.mk
+include test/libc/zipos/test.mk
 include test/libc/release/test.mk
 include test/libc/test.mk
 include test/net/http/test.mk
 include test/net/https/test.mk
+include test/net/finger/test.mk
 include test/net/test.mk
 include test/tool/build/lib/test.mk
 include test/tool/build/test.mk
@@ -249,119 +287,123 @@ o/$(MODE)/srcs-old.txt: o/$(MODE)/.x $(MAKEFILES) $(call uniq,$(foreach x,$(SRCS
 o/$(MODE)/hdrs-old.txt: o/$(MODE)/.x $(MAKEFILES) $(call uniq,$(foreach x,$(HDRS) $(INCS),$(dir $(x))))
 	$(file >$@) $(foreach x,$(HDRS) $(INCS),$(file >>$@,$(x)))

-TAGS:	o/$(MODE)/srcs-old.txt $(SRCS)
+TAGS: private .UNSANDBOXED = 1
+TAGS:	o/$(MODE)/srcs-old.txt $(SRCS) #o/$(MODE)/third_party/ctags/ctags.com
 	@$(RM) $@
-	@$(COMPILE) -ATAGS -T$@ $(TAGS) $(TAGSFLAGS) -L $< -o $@
+	@o/$(MODE)/third_party/ctags/ctags.com $(TAGSFLAGS) -L $< -o $@

-HTAGS:	o/$(MODE)/hdrs-old.txt $(HDRS)
+HTAGS: private .UNSANDBOXED = 1
+HTAGS:	o/$(MODE)/hdrs-old.txt $(HDRS) #o/$(MODE)/third_party/ctags/ctags.com
 	@$(RM) $@
-	@$(COMPILE) -ATAGS -T$@ build/htags -L $< -o $@
+	@build/htags o/$(MODE)/third_party/ctags/ctags.com -L $< -o $@

+loc: private .UNSANDBOXED = 1
 loc: o/$(MODE)/tool/build/summy.com
 	find -name \*.h -or -name \*.c -or -name \*.S | \
 	$(XARGS) wc -l | grep total | awk '{print $$1}' | $<

-COSMOPOLITAN_OBJECTS =		\
-	NET_HTTP		\
-	LIBC_DNS		\
-	LIBC_SOCK		\
-	LIBC_NT_WS2_32		\
-	LIBC_NT_IPHLPAPI	\
-	LIBC_NT_MSWSOCK		\
-	LIBC_X			\
-	THIRD_PARTY_GETOPT	\
-	LIBC_LOG		\
-	LIBC_UNICODE		\
-	LIBC_TIME		\
-	LIBC_ZIPOS		\
-	THIRD_PARTY_ZLIB	\
-	THIRD_PARTY_MUSL	\
-	LIBC_STDIO		\
-	THIRD_PARTY_GDTOA	\
-	THIRD_PARTY_REGEX	\
-	LIBC_ALG		\
-	LIBC_MEM		\
-	THIRD_PARTY_DLMALLOC	\
-	LIBC_BITS		\
-	LIBC_RUNTIME		\
-	LIBC_ELF		\
-	LIBC_CALLS		\
-	LIBC_RAND		\
-	LIBC_SYSV_CALLS		\
-	LIBC_NT_PSAPI		\
-	LIBC_NT_POWRPROF	\
-	LIBC_NT_PDH		\
-	LIBC_NT_SHELL32		\
-	LIBC_NT_GDI32		\
-	LIBC_NT_COMDLG32	\
-	LIBC_NT_NETAPI32	\
-	LIBC_NT_URL		\
-	LIBC_NT_USER32		\
-	LIBC_NT_NTDLL		\
-	LIBC_NT_ADVAPI32	\
-	LIBC_FMT		\
-	THIRD_PARTY_COMPILER_RT	\
-	LIBC_THREAD		\
-	LIBC_TINYMATH		\
-	THIRD_PARTY_XED		\
-	LIBC_STR		\
-	LIBC_SYSV		\
-	LIBC_INTRIN		\
-	LIBC_NT_KERNEL32	\
+# PLEASE: MAINTAIN TOPOLOGICAL ORDER
+# FROM HIGHEST LEVEL TO LOWEST LEVEL
+COSMOPOLITAN_OBJECTS =			\
+	NET_HTTP			\
+	LIBC_DNS			\
+	LIBC_SOCK			\
+	LIBC_NT_WS2_32			\
+	LIBC_NT_IPHLPAPI		\
+	LIBC_NT_MSWSOCK			\
+	LIBC_X				\
+	THIRD_PARTY_GETOPT		\
+	LIBC_LOG			\
+	LIBC_TIME			\
+	LIBC_ZIPOS			\
+	THIRD_PARTY_MUSL		\
+	LIBC_STDIO			\
+	THIRD_PARTY_GDTOA		\
+	THIRD_PARTY_REGEX		\
+	LIBC_THREAD			\
+	THIRD_PARTY_NSYNC_MEM		\
+	LIBC_MEM			\
+	THIRD_PARTY_DLMALLOC		\
+	LIBC_RUNTIME			\
+	THIRD_PARTY_NSYNC		\
+	LIBC_ELF			\
+	LIBC_CALLS			\
+	LIBC_SYSV_CALLS			\
+	LIBC_VGA			\
+	LIBC_NT_PSAPI			\
+	LIBC_NT_POWRPROF		\
+	LIBC_NT_PDH			\
+	LIBC_NT_GDI32			\
+	LIBC_NT_COMDLG32		\
+	LIBC_NT_URL			\
+	LIBC_NT_USER32			\
+	LIBC_NT_NTDLL			\
+	LIBC_NT_ADVAPI32		\
+	LIBC_NT_SYNCHRONIZATION		\
+	LIBC_FMT			\
+	THIRD_PARTY_PUFF		\
+	THIRD_PARTY_COMPILER_RT		\
+	LIBC_TINYMATH			\
+	THIRD_PARTY_XED			\
+	LIBC_STR			\
+	LIBC_SYSV			\
+	LIBC_INTRIN			\
+	LIBC_NT_KERNEL32		\
 	LIBC_NEXGEN32E

-COSMOPOLITAN_HEADERS =		\
-	LIBC			\
-	LIBC_ALG		\
-	LIBC_BITS		\
-	LIBC_CALLS		\
-	LIBC_DNS		\
-	LIBC_ELF		\
-	LIBC_FMT		\
-	LIBC_INTRIN		\
-	LIBC_LOG		\
-	LIBC_MEM		\
-	LIBC_NEXGEN32E		\
-	LIBC_NT			\
-	LIBC_RAND		\
-	LIBC_RUNTIME		\
-	LIBC_SOCK		\
-	LIBC_STDIO		\
-	THIRD_PARTY_XED		\
-	LIBC_STR		\
-	LIBC_SYSV		\
-	LIBC_THREAD		\
-	LIBC_TIME		\
-	LIBC_TINYMATH		\
-	LIBC_UNICODE		\
-	LIBC_X			\
-	LIBC_ZIPOS		\
-	NET_HTTP		\
-	THIRD_PARTY_DLMALLOC	\
-	THIRD_PARTY_GDTOA	\
-	THIRD_PARTY_GETOPT	\
-	THIRD_PARTY_MUSL	\
-	THIRD_PARTY_ZLIB	\
-	THIRD_PARTY_ZLIB_GZ	\
+COSMOPOLITAN_HEADERS =			\
+	APE				\
+	LIBC				\
+	LIBC_CALLS			\
+	LIBC_DNS			\
+	LIBC_ELF			\
+	LIBC_FMT			\
+	LIBC_INTRIN			\
+	LIBC_LOG			\
+	LIBC_MEM			\
+	LIBC_NEXGEN32E			\
+	LIBC_NT				\
+	LIBC_RUNTIME			\
+	LIBC_SOCK			\
+	LIBC_STDIO			\
+	THIRD_PARTY_NSYNC		\
+	THIRD_PARTY_XED			\
+	LIBC_STR			\
+	LIBC_SYSV			\
+	LIBC_THREAD			\
+	LIBC_TIME			\
+	LIBC_TINYMATH			\
+	LIBC_X				\
+	LIBC_ZIPOS			\
+	LIBC_VGA			\
+	NET_HTTP			\
+	THIRD_PARTY_DLMALLOC		\
+	THIRD_PARTY_GDTOA		\
+	THIRD_PARTY_GETOPT		\
+	THIRD_PARTY_MUSL		\
 	THIRD_PARTY_REGEX

-o/$(MODE)/cosmopolitan.a:	\
+o/$(MODE)/cosmopolitan.a:		\
 		$(foreach x,$(COSMOPOLITAN_OBJECTS),$($(x)_A_OBJS))

-o/cosmopolitan.h:				\
-		o/$(MODE)/tool/build/rollup.com	\
-		libc/integral/normalize.inc	\
-		$(foreach x,$(COSMOPOLITAN_HEADERS),$($(x)_HDRS))
-	$(file >$@.args,libc/integral/normalize.inc $(foreach x,$(COSMOPOLITAN_HEADERS),$($(x)_HDRS)))
-	@$(COMPILE) -AROLLUP -T$@ o/$(MODE)/tool/build/rollup.com @$@.args >$@
+o/cosmopolitan.h:							\
+		o/$(MODE)/tool/build/rollup.com				\
+		libc/integral/normalize.inc				\
+		$(foreach x,$(COSMOPOLITAN_HEADERS),$($(x)_HDRS))	\
+		$(foreach x,$(COSMOPOLITAN_HEADERS),$($(x)_INCS))
+	$(file >$(TMPDIR)/$(subst /,_,$@),libc/integral/normalize.inc $(foreach x,$(COSMOPOLITAN_HEADERS),$($(x)_HDRS)))
+	@$(COMPILE) -AROLLUP -T$@ o/$(MODE)/tool/build/rollup.com @$(TMPDIR)/$(subst /,_,$@) >$@

+o/cosmopolitan.html: private .UNSANDBOXED = 1
 o/cosmopolitan.html:							\
 		o/$(MODE)/third_party/chibicc/chibicc.com.dbg		\
-		$(filter-out %.s,$(foreach x,$(COSMOPOLITAN_OBJECTS),$($(x)_SRCS)))
-	$(file >$@.args,$(filter-out %.s,$(foreach x,$(COSMOPOLITAN_OBJECTS),$($(x)_SRCS))))
+		$(filter-out %.s,$(foreach x,$(COSMOPOLITAN_OBJECTS),$($(x)_SRCS)))	\
+		$(SRCS)							\
+		$(HDRS)
+	$(file >$(TMPDIR)/$(subst /,_,$@),$(filter-out %.s,$(foreach x,$(COSMOPOLITAN_OBJECTS),$($(x)_SRCS))))
 	o/$(MODE)/third_party/chibicc/chibicc.com.dbg -J		\
 		-fno-common -include libc/integral/normalize.inc -o $@	\
-		@$@.args
+		@$(TMPDIR)/$(subst /,_,$@)

 $(SRCS):					\
 	libc/integral/normalize.inc		\
@@ -371,6 +413,16 @@ $(SRCS):					\
 	libc/integral/lp64arg.inc		\
 	libc/integral/lp64.inc

+.PHONY: toolchain
+toolchain:	o/cosmopolitan.h				\
+		o/$(MODE)/ape/public/ape.lds			\
+		o/$(MODE)/libc/crt/crt.o			\
+		o/$(MODE)/ape/ape.o				\
+		o/$(MODE)/ape/ape-copy-self.o			\
+		o/$(MODE)/ape/ape-no-modify-self.o		\
+		o/$(MODE)/cosmopolitan.a			\
+		o/$(MODE)/third_party/libcxx/libcxx.a
+
 # UNSPECIFIED PREREQUISITES TUTORIAL
 #
 # A build rule must exist for all files that make needs to consider in
@@ -395,9 +447,9 @@ $(SRCS):
 $(HDRS):
 $(INCS):
 .DEFAULT:
-	@$(ECHO) >&2
-	@$(ECHO) NOTE: deleting o/$(MODE)/depend because of an unspecified prerequisite: $@ >&2
-	@$(ECHO) >&2
+	@$(ECHO)
+	@$(ECHO) NOTE: deleting o/$(MODE)/depend because of an unspecified prerequisite: $@
+	@$(ECHO)
 	$(RM) o/$(MODE)/depend

 -include o/$(MODE)/depend
@@ -23,12 +23,12 @@ If you're doing your development work on Linux or BSD then you need just
 five files to get started. Here's what you do on Linux:

 ```sh
-wget https://justine.lol/cosmopolitan/cosmopolitan.zip
-unzip cosmopolitan.zip
+wget https://justine.lol/cosmopolitan/cosmopolitan-amalgamation-2.2.zip
+unzip cosmopolitan-amalgamation-2.2.zip
 printf 'main() { printf("hello world\\n"); }\n' >hello.c
 gcc -g -Os -static -nostdlib -nostdinc -fno-pie -no-pie -mno-red-zone \
-  -fno-omit-frame-pointer -pg -mnop-mcount \
-  -o hello.com.dbg hello.c -fuse-ld=bfd -Wl,-T,ape.lds \
+  -fno-omit-frame-pointer -pg -mnop-mcount -mno-tls-direct-seg-refs -gdwarf-4 \
+  -o hello.com.dbg hello.c -fuse-ld=bfd -Wl,-T,ape.lds -Wl,--gc-sections \
  -include cosmopolitan.h crt.o ape-no-modify-self.o cosmopolitan.a
 objcopy -S -O binary hello.com.dbg hello.com
 ```
@@ -40,11 +40,30 @@ You now have a portable program.
 bash -c './hello.com'  # zsh/fish workaround (we patched them in 2021)
 ```

+If `./hello.com` executed on Linux throws an error about not finding an
+interpreter, it should be fixed by running the following command (although
+note that it may not survive a system restart):
+
+```sh
+sudo sh -c "echo ':APE:M::MZqFpD::/bin/sh:' >/proc/sys/fs/binfmt_misc/register"
+```
+
+If the same command produces puzzling errors on WSL or WINE when using
+Redbean 2.x, they may be fixed by disabling binfmt_misc:
+
+```sh
+sudo sh -c 'echo -1 >/proc/sys/fs/binfmt_misc/status'
+```
+
 Since we used the `ape-no-modify-self.o` bootloader (rather than
 `ape.o`) your executable will not modify itself when it's run. What
-it'll instead do, is extract a 4kb program to `${TMPDIR:-${HOME:-.}}`
-that maps your program into memory without needing to copy it. It's
-possible to install the APE loader systemwide as follows.
+it'll instead do, is extract a 4kb program (the [APE loader](https://justine.lol/apeloader/))
+to `${TMPDIR:-${HOME:-.}}` that maps your program into memory without
+needing to copy it. The APE loader must be in an executable location
+(e.g. not stored on a `noexec` mount) for it to run. See below for
+alternatives:
+
+It's possible to install the APE loader systemwide as follows.

 ```sh
 # System-Wide APE Install
@@ -59,7 +78,7 @@ ape/apeuninstall.sh
 ```

 It's also possible to convert APE binaries into the system-local format
-by using the `--assimilate` flag. Plesae note that if binfmt_misc is in
+by using the `--assimilate` flag. Please note that if binfmt_misc is in
 play, you'll need to unregister it temporarily before doing this, since
 the assimilate feature is part of the shell script header.

@@ -85,7 +104,7 @@ heavyweight features like tracing to your binaries by default. In that
 case, you may want to consider using our build system:

 ```sh
-make -j8 MODE=tiny
+make m=tiny
 ```

 Which will cause programs such as `hello.com` and `life.com` to shrink
@@ -140,7 +159,7 @@ cd cosmopolitan
 This will build the entire repository and run all the tests:

 ```sh
-build/bootstrap/make.com -j16
+build/bootstrap/make.com
 o//examples/hello.com
 find o -name \*.com | xargs ls -rShal | less
 ```
@@ -159,7 +178,7 @@ having minimal deterministic builds. For example, if you wanted to build
 only hello.com then you could do that as follows:

 ```sh
-build/bootstrap/make.com -j16 o//examples/hello.com
+build/bootstrap/make.com o//examples/hello.com
 ```

 Sometimes it's desirable to build a subset of targets, without having to
@@ -168,7 +187,7 @@ directory name. For example, if you wanted to build only the targets and
 subtargets of the chibicc package including its tests, you would say:

 ```sh
-build/bootstrap/make.com -j16 o//third_party/chibicc
+build/bootstrap/make.com o//third_party/chibicc
 o//third_party/chibicc/chibicc.com --help
 ```

@@ -176,18 +195,19 @@ Cosmopolitan provides a variety of build modes. For example, if you want
 really tiny binaries (as small as 12kb in size) then you'd say:

 ```sh
-build/bootstrap/make.com -j16 MODE=tiny
+build/bootstrap/make.com m=tiny
 ```

 Here's some other build modes you can try:

 ```sh
-build/bootstrap/make.com -j16 MODE=dbg       # asan + ubsan + debug
-build/bootstrap/make.com -j16 MODE=asan      # production memory safety
-build/bootstrap/make.com -j16 MODE=opt       # -march=native optimizations
-build/bootstrap/make.com -j16 MODE=rel       # traditional release binaries
-build/bootstrap/make.com -j16 MODE=optlinux  # optimal linux-only performance
-build/bootstrap/make.com -j16 MODE=tinylinux # tiniest linux-only 4kb binaries
+build/bootstrap/make.com m=dbg       # asan + ubsan + debug
+build/bootstrap/make.com m=asan      # production memory safety
+build/bootstrap/make.com m=opt       # -march=native optimizations
+build/bootstrap/make.com m=rel       # traditional release binaries
+build/bootstrap/make.com m=optlinux  # optimal linux-only performance
+build/bootstrap/make.com m=fastbuild # build 28% faster w/o debugging
+build/bootstrap/make.com m=tinylinux # tiniest linux-only 4kb binaries
 ```

 For further details, see [//build/config.mk](build/config.mk).
@@ -196,7 +216,7 @@ For further details, see [//build/config.mk](build/config.mk).

 Here's the recommended `~/.gdbinit` config:

-```
+```gdb
 set host-charset UTF-8
 set target-charset UTF-8
 set target-wide-charset UTF-8
@@ -223,27 +243,38 @@ You normally run the `.com.dbg` file under gdb. If you need to debug the
 gdb foo.com -ex 'add-symbol-file foo.com.dbg 0x401000'
 ```

+## Discord Chatroom
+
+The Cosmopolitan development team collaborates on the Redbean Discord
+server. You're welcome to join us! <https://discord.gg/FwAVVu7eJ4>
+
 ## Support Vector

 | Platform        | Min Version | Circa |
 | :---            | ---:        | ---:  |
 | AMD             | K8 Venus    | 2005  |
 | Intel           | Core        | 2006  |
-| New Technology  | Vista       | 2006  |
-| GNU/Systemd     | 2.6.18      | 2007  |
-| XNU's Not UNIX! | 15.6        | 2018  |
-| FreeBSD         | 13          | 2020  |
+| Linux           | 2.6.18      | 2007  |
+| Windows         | 8 [1]       | 2012  |
+| Mac OS X        | 15.6        | 2018  |
 | OpenBSD         | 6.4         | 2018  |
+| FreeBSD         | 13          | 2020  |
 | NetBSD          | 9.2         | 2021  |

+[1] See our [vista branch](https://github.com/jart/cosmopolitan/tree/vista)
+    for a community supported version of Cosmopolitan that works on Windows
+    Vista and Windows 7.
+
 ## Special Thanks

 Funding for this project is crowdsourced using
 [GitHub Sponsors](https://github.com/sponsors/jart) and
 [Patreon](https://www.patreon.com/jart). Your support is what makes this
 project possible. Thank you! We'd also like to give special thanks to
-the following individuals:
+the following groups and individuals:

 - [Joe Drumgoole](https://github.com/jdrumgoole)
+- [Rob Figueiredo](https://github.com/robfig)
+- [Wasmer](https://wasmer.io/)

 For publicly sponsoring our work at the highest tier.
@@ -0,0 +1,9 @@
+#ifndef COSMOPOLITAN_APE_APE_INTERNAL_H_
+#define COSMOPOLITAN_APE_APE_INTERNAL_H_
+#include "libc/dce.h"
+
+#if SupportsWindows() || SupportsMetal() || SupportsXnu()
+#define APE_IS_SHELL_SCRIPT
+#endif
+
+#endif /* COSMOPOLITAN_APE_APE_INTERNAL_H_ */
@@ -180,8 +180,16 @@
 #include "libc/elf/def.h"
 #include "libc/elf/pf2prot.internal.h"
 #include "libc/nt/pedef.internal.h"
+#include "libc/thread/tls.h"
+#include "ape/ape.internal.h"
 #include "libc/zip.h"

+#ifdef __x86__
+#define CODE_GRANULE 1
+#else
+#define CODE_GRANULE 4
+#endif
+
 ENTRY(_start)

 PHDRS {
@@ -215,6 +223,8 @@ SECTIONS {
    KEEP(*(.elf.phdrs))
    HIDDEN(ape_phdrs_end = .);

+    KEEP(*(.emush))
+
    /* OpenBSD */
    . = ALIGN(__SIZEOF_POINTER__);
    HIDDEN(ape_note = .);
@@ -247,10 +257,11 @@ SECTIONS {
    KEEP(*(SORT_BY_NAME(.sort.text.real.*)))
    /* Code we want earlier in the binary w/o modifications */
    KEEP(*(.ape.loader))
+    . = ALIGN(CODE_GRANULE);
    HIDDEN(_ereal = .);
 /*END: realmode addressability guarantee */
 /*BEGIN: morphable code */
-    . += 1;
+    . += CODE_GRANULE;

    /* Normal Code */
    *(.start)
@@ -272,6 +283,7 @@ SECTIONS {
    KEEP(*(.textwindowsprologue))
    *(.text.windows)
    KEEP(*(.textwindowsepilogue))
+    KEEP(*(.blink))
    *(SORT_BY_ALIGNMENT(.text.modernity))
    *(SORT_BY_ALIGNMENT(.text.modernity.*))
    *(SORT_BY_ALIGNMENT(.text.hot))
@@ -279,6 +291,7 @@ SECTIONS {
    KEEP(*(.keep.text))
    *(.text .stub .text.*)
    KEEP(*(SORT_BY_NAME(.sort.text.*)))
+    *(.subrs)

    KEEP(*(.ape.pad.test));
    *(.test.unlikely)
@@ -286,13 +299,12 @@ SECTIONS {

    /* Privileged code invulnerable to magic */
    KEEP(*(.ape.pad.privileged));
-    . += . > 0 ? 1 : 0;
+    . = ALIGN(__privileged_end > __privileged_start ? PAGESIZE : 1);
 /*END: morphable code */
    HIDDEN(__privileged_start = .);
-    . += . > 0 ? 1 : 0;
    *(.privileged)
    HIDDEN(__privileged_end = .);
-    . += . > 0 ? 1 : 0;
+    . += . > 0 ? CODE_GRANULE : 0;

 /*BEGIN: Read Only Data */

@@ -347,9 +359,30 @@ SECTIONS {
    HIDDEN(_etext = .);
    PROVIDE_HIDDEN(etext = .);
 /*END: Read Only Data (only needed for initialization) */
-/*END: Read Only Data */
  } :Rom

+/* initialization image for thread-local storage, this is copied */
+/* out to actual TLS areas at runtime, so just make it read-only */
+  .tdata . : {
+    _tdata_start = .;
+    *(SORT_BY_ALIGNMENT(.tdata))
+    *(SORT_BY_ALIGNMENT(.tdata.*))
+    _tdata_end = .;
+    . = ALIGN(PAGESIZE);
+  } :Tls :Rom
+/*END: Read Only Data */
+
+/* this only tells the linker about the layout of uninitialized */
+/* TLS data, and does not advance the linker's location counter */
+  .tbss . : {
+    _tbss_start = .;
+    *(SORT_BY_ALIGNMENT(.tbss))
+    *(SORT_BY_ALIGNMENT(.tbss.*))
+    . = ALIGN(TLS_ALIGNMENT);
+    /* the %fs register is based on this location */
+    _tbss_end = .;
+  } :Tls
+
  .data ALIGN(PAGESIZE) : {
 /*BEGIN: Read/Write Data */
    KEEP(*(SORT_BY_NAME(.piro.data.sort.iat.*)))
@@ -357,7 +390,7 @@ SECTIONS {
    KEEP(*(.dataprologue))
    *(.data .data.*)
    KEEP(*(SORT_BY_NAME(.sort.data.*)))
-    . += . > 0 ? 1 : 0;
+    . += . > 0 ? CODE_GRANULE : 0;

    KEEP(*(.gotprologue))
    *(.got)
@@ -380,27 +413,10 @@ SECTIONS {
    PROVIDE_HIDDEN(edata = .);
    KEEP(*(SORT_BY_NAME(.zip.*)))
    HIDDEN(_ezip = .);
+    . = ALIGN(PAGESIZE);
  } :Ram

-  .tdata . : {
-    _tdata_start = .;
-    *(SORT_BY_ALIGNMENT(.tdata))
-    *(SORT_BY_ALIGNMENT(.tdata.*))
-    _tdata_end = .;
-    . = ALIGN(PAGESIZE);
-  } :Tls
-  . = ALIGN(PAGESIZE);
-
 /*END:   file content that's loaded by o/s */
-/*BEGIN: bss memory void */
-
-  .tbss . : {
-    _tbss_start = .;
-    *(SORT_BY_ALIGNMENT(.tbss))
-    *(SORT_BY_ALIGNMENT(.tbss.*))
-    _tbss_end = .;
-  } :Tls
-
 /*END:   file content */
 /*BEGIN: bss memory that's addressable */

@@ -411,7 +427,7 @@ SECTIONS {
    *(.piro.bss)
    KEEP(*(SORT_BY_NAME(.piro.bss.sort.*)))
    HIDDEN(__piro_end = .);
-    . += . > 0 ? 1 : 0;
+    . += . > 0 ? CODE_GRANULE : 0;
 /*END: Post-Initialization Read-Only */

    /* Statically Allocated Empty Space */
@@ -423,7 +439,7 @@ SECTIONS {

    KEEP(*(.bssepilogue))
 /*END: NT FORK COPYING */
-    . = ALIGN(FRAMESIZE); /* for brk()/sbrk() allocation */
+    . = ALIGN(PAGESIZE);
    HIDDEN(_end = .);
    PROVIDE_HIDDEN(end = .);
  } :Bss
@@ -454,6 +470,7 @@ SECTIONS {
  .debug_line        0 : { *(.debug_line .debug_line.* .debug_line_end ) }
  .debug_frame       0 : { *(.debug_frame) }
  .debug_str         0 : { *(.debug_str) }
+  .debug_line_str    0 : { *(.debug_line_str) }
  .debug_loc         0 : { *(.debug_loc) }
  .debug_macinfo     0 : { *(.debug_macinfo) }
  .debug_weaknames   0 : { *(.debug_weaknames) }
@@ -462,6 +479,7 @@ SECTIONS {
  .debug_varnames    0 : { *(.debug_varnames) }
  .debug_pubtypes    0 : { *(.debug_pubtypes) }
  .debug_ranges      0 : { *(.debug_ranges) }
+  .debug_rnglists    0 : { *(.debug_rnglists) }
  .debug_macro       0 : { *(.debug_macro) }
  .debug_addr        0 : { *(.debug_addr) }
  .gnu.attributes    0 : { KEEP(*(.gnu.attributes)) }
@@ -482,8 +500,10 @@ PFSTUB4(ape_elf_phnum, (ape_phdrs_end - ape_phdrs) / 56);
 PFSTUB4(ape_elf_shnum, 0);
 PFSTUB4(ape_elf_shstrndx, 0);

-HIDDEN(_tdata_size = _tdata_end - _tdata_start);
 HIDDEN(_tls_size = _tbss_end - _tdata_start);
+HIDDEN(_tdata_size = _tdata_end - _tdata_start);
+HIDDEN(_tbss_size = _tbss_end - _tbss_start);
+HIDDEN(_tbss_offset = _tbss_start - _tdata_start);
 HIDDEN(_tls_content = (_tdata_end - _tdata_start) + (_tbss_end - _tbss_start));

 HIDDEN(__privileged_addr = ROUNDDOWN(__privileged_start, PAGESIZE));
@@ -501,7 +521,7 @@ HIDDEN(ape_rom_rva = RVA(ape_rom_vaddr));
 HIDDEN(ape_ram_offset = ape_rom_offset + ape_rom_filesz);
 HIDDEN(ape_ram_vaddr = ADDR(.data));
 HIDDEN(ape_ram_paddr = LOADADDR(.data));
-HIDDEN(ape_ram_filesz = SIZEOF(.data) + SIZEOF(.tdata));
+HIDDEN(ape_ram_filesz = SIZEOF(.data));
 HIDDEN(ape_ram_memsz = ADDR(.bss) + SIZEOF(.bss) - ape_ram_vaddr);
 HIDDEN(ape_ram_align = PAGESIZE);
 HIDDEN(ape_ram_rva = RVA(ape_ram_vaddr));
@@ -525,16 +545,16 @@ HIDDEN(ape_note_align = __SIZEOF_POINTER__);
 HIDDEN(ape_text_offset = ape_rom_offset + LOADADDR(.text) - ape_rom_paddr);
 HIDDEN(ape_text_paddr = LOADADDR(.text));
 HIDDEN(ape_text_vaddr = ADDR(.text));
-HIDDEN(ape_text_filesz = SIZEOF(.text));
-HIDDEN(ape_text_memsz = SIZEOF(.text));
+HIDDEN(ape_text_filesz = SIZEOF(.text) + SIZEOF(.tdata));
+HIDDEN(ape_text_memsz = SIZEOF(.text) + SIZEOF(.tdata));
 HIDDEN(ape_text_align = PAGESIZE);
 HIDDEN(ape_text_rva = RVA(ape_text_vaddr));

 HIDDEN(ape_data_offset = ape_ram_offset + LOADADDR(.data) - ape_ram_paddr);
 HIDDEN(ape_data_paddr = LOADADDR(.data));
 HIDDEN(ape_data_vaddr = ADDR(.data));
-HIDDEN(ape_data_filesz = SIZEOF(.data) + SIZEOF(.tdata));
-HIDDEN(ape_data_memsz = SIZEOF(.data) + SIZEOF(.tdata));
+HIDDEN(ape_data_filesz = SIZEOF(.data));
+HIDDEN(ape_data_memsz = SIZEOF(.data));
 HIDDEN(ape_data_align = PAGESIZE);
 HIDDEN(ape_data_rva = RVA(ape_data_vaddr));

@@ -554,13 +574,63 @@ SHSTUB2(ape_loader_dd_count,
            ? ROUNDUP(ape_loader_end - ape_loader, PAGESIZE) / 64
            : 0);

+#if defined(APE_IS_SHELL_SCRIPT) && !IsTiny()
+
+HIDDEN(blink_aarch64_b0 = RVA(blink_aarch64) % 10 + 48);
+HIDDEN(blink_aarch64_b1 = RVA(blink_aarch64) < 10 ? 32 : RVA(blink_aarch64) / 10 % 10 + 48);
+HIDDEN(blink_aarch64_b2 = RVA(blink_aarch64) < 100 ? 32 : RVA(blink_aarch64) / 100 % 10 + 48);
+HIDDEN(blink_aarch64_b3 = RVA(blink_aarch64) < 1000 ? 32 : RVA(blink_aarch64) / 1000 % 10 + 48);
+HIDDEN(blink_aarch64_b4 = RVA(blink_aarch64) < 10000 ? 32 : RVA(blink_aarch64) / 10000 % 10 + 48);
+HIDDEN(blink_aarch64_b5 = RVA(blink_aarch64) < 100000 ? 32 : RVA(blink_aarch64) / 100000 % 10 + 48);
+HIDDEN(blink_aarch64_b6 = RVA(blink_aarch64) < 1000000 ? 32 : RVA(blink_aarch64) / 1000000 % 10 + 48);
+HIDDEN(blink_aarch64_b7 = RVA(blink_aarch64) < 10000000 ? 32 : RVA(blink_aarch64) / 10000000 % 10 + 48);
+HIDDEN(blink_aarch64_b8 = RVA(blink_aarch64) < 100000000 ? 32 : RVA(blink_aarch64) / 100000000 % 10 + 48);
+HIDDEN(blink_aarch64_b9 = RVA(blink_aarch64) < 1000000000 ? 32 : RVA(blink_aarch64) / 1000000000 % 10 + 48);
+
+HIDDEN(blink_aarch64_size_b0 = blink_aarch64_size % 10 + 48);
+HIDDEN(blink_aarch64_size_b1 = blink_aarch64_size < 10 ? 32 : blink_aarch64_size / 10 % 10 + 48);
+HIDDEN(blink_aarch64_size_b2 = blink_aarch64_size < 100 ? 32 : blink_aarch64_size / 100 % 10 + 48);
+HIDDEN(blink_aarch64_size_b3 = blink_aarch64_size < 1000 ? 32 : blink_aarch64_size / 1000 % 10 + 48);
+HIDDEN(blink_aarch64_size_b4 = blink_aarch64_size < 10000 ? 32 : blink_aarch64_size / 10000 % 10 + 48);
+HIDDEN(blink_aarch64_size_b5 = blink_aarch64_size < 100000 ? 32 : blink_aarch64_size / 100000 % 10 + 48);
+HIDDEN(blink_aarch64_size_b6 = blink_aarch64_size < 1000000 ? 32 : blink_aarch64_size / 1000000 % 10 + 48);
+HIDDEN(blink_aarch64_size_b7 = blink_aarch64_size < 10000000 ? 32 : blink_aarch64_size / 10000000 % 10 + 48);
+HIDDEN(blink_aarch64_size_b8 = blink_aarch64_size < 100000000 ? 32 : blink_aarch64_size / 100000000 % 10 + 48);
+HIDDEN(blink_aarch64_size_b9 = blink_aarch64_size < 1000000000 ? 32 : blink_aarch64_size / 1000000000 % 10 + 48);
+
+HIDDEN(blink_darwin_arm64_b0 = RVA(blink_darwin_arm64) % 10 + 48);
+HIDDEN(blink_darwin_arm64_b1 = RVA(blink_darwin_arm64) < 10 ? 32 : RVA(blink_darwin_arm64) / 10 % 10 + 48);
+HIDDEN(blink_darwin_arm64_b2 = RVA(blink_darwin_arm64) < 100 ? 32 : RVA(blink_darwin_arm64) / 100 % 10 + 48);
+HIDDEN(blink_darwin_arm64_b3 = RVA(blink_darwin_arm64) < 1000 ? 32 : RVA(blink_darwin_arm64) / 1000 % 10 + 48);
+HIDDEN(blink_darwin_arm64_b4 = RVA(blink_darwin_arm64) < 10000 ? 32 : RVA(blink_darwin_arm64) / 10000 % 10 + 48);
+HIDDEN(blink_darwin_arm64_b5 = RVA(blink_darwin_arm64) < 100000 ? 32 : RVA(blink_darwin_arm64) / 100000 % 10 + 48);
+HIDDEN(blink_darwin_arm64_b6 = RVA(blink_darwin_arm64) < 1000000 ? 32 : RVA(blink_darwin_arm64) / 1000000 % 10 + 48);
+HIDDEN(blink_darwin_arm64_b7 = RVA(blink_darwin_arm64) < 10000000 ? 32 : RVA(blink_darwin_arm64) / 10000000 % 10 + 48);
+HIDDEN(blink_darwin_arm64_b8 = RVA(blink_darwin_arm64) < 100000000 ? 32 : RVA(blink_darwin_arm64) / 100000000 % 10 + 48);
+HIDDEN(blink_darwin_arm64_b9 = RVA(blink_darwin_arm64) < 1000000000 ? 32 : RVA(blink_darwin_arm64) / 1000000000 % 10 + 48);
+
+HIDDEN(blink_darwin_arm64_size_b0 = blink_darwin_arm64_size % 10 + 48);
+HIDDEN(blink_darwin_arm64_size_b1 = blink_darwin_arm64_size < 10 ? 32 : blink_darwin_arm64_size / 10 % 10 + 48);
+HIDDEN(blink_darwin_arm64_size_b2 = blink_darwin_arm64_size < 100 ? 32 : blink_darwin_arm64_size / 100 % 10 + 48);
+HIDDEN(blink_darwin_arm64_size_b3 = blink_darwin_arm64_size < 1000 ? 32 : blink_darwin_arm64_size / 1000 % 10 + 48);
+HIDDEN(blink_darwin_arm64_size_b4 = blink_darwin_arm64_size < 10000 ? 32 : blink_darwin_arm64_size / 10000 % 10 + 48);
+HIDDEN(blink_darwin_arm64_size_b5 = blink_darwin_arm64_size < 100000 ? 32 : blink_darwin_arm64_size / 100000 % 10 + 48);
+HIDDEN(blink_darwin_arm64_size_b6 = blink_darwin_arm64_size < 1000000 ? 32 : blink_darwin_arm64_size / 1000000 % 10 + 48);
+HIDDEN(blink_darwin_arm64_size_b7 = blink_darwin_arm64_size < 10000000 ? 32 : blink_darwin_arm64_size / 10000000 % 10 + 48);
+HIDDEN(blink_darwin_arm64_size_b8 = blink_darwin_arm64_size < 100000000 ? 32 : blink_darwin_arm64_size / 100000000 % 10 + 48);
+HIDDEN(blink_darwin_arm64_size_b9 = blink_darwin_arm64_size < 1000000000 ? 32 : blink_darwin_arm64_size / 1000000000 % 10 + 48);
+
+#endif /* APE_IS_SHELL_SCRIPT */
+
 #if SupportsMetal()
 HIDDEN(v_ape_realsectors =
-           MIN(0x70000 - IMAGE_BASE_REAL,
-               ROUNDUP(RVA(_edata), 4096)) / 512);
-HIDDEN(v_ape_realpages = v_ape_realsectors / (4096 / 512));
-HIDDEN(v_ape_highsectors =
-           (ROUNDUP(RVA(_edata), 512) / 512) - v_ape_realsectors);
+           MIN(0x70000 - IMAGE_BASE_REAL, ROUNDUP(RVA(_ezip), 512)) / 512);
+HIDDEN(v_ape_realbytes = v_ape_realsectors * 512);
+HIDDEN(v_ape_realdwords = v_ape_realsectors * (512 / 4));
+HIDDEN(v_ape_allsectors = ROUNDUP(RVA(_ezip), 512) / 512);
+HIDDEN(v_ape_allbytes = v_ape_allsectors * 512);
+HIDDEN(v_ape_highsectors = MIN(0xffff, v_ape_allsectors - v_ape_realsectors));
+TSSDESCSTUB2(_tss, _tss, _tss_end ? _tss_end - _tss - 1 : 0);
 #endif

 /* ZIP End of Central Directory header */
@@ -642,7 +712,7 @@ CHURN(ADDR(.bss));
 CHURN(_start);
 CHURN(ape_phdrs);
 #if SupportsMetal()
-CHURN(v_ape_realsectors);
+CHURN(v_ape_allsectors);
 #endif
 #if SupportsXnu()
 CHURN(ape_macho);
@@ -705,8 +775,11 @@ ASSERT(!DEFINED(_start16) || REAL(_end) < 65536,
 ASSERT(IS2POW(ape_stack_memsz),
       "ape_stack_memsz must be a two power");

-ASSERT(!(ape_stack_vaddr & (ape_stack_memsz - 1)),
-       "ape_stack_vaddr must have ape_stack_memsz alignment; try using STATIC_STACK_ADDR(0x700000000000 - ape_stack_memsz);");
+ASSERT(ape_stack_vaddr % ape_stack_memsz == 0,
+       "ape_stack_vaddr must have ape_stack_memsz alignment; try using STATIC_STACK_ADDR(0x700000040000 & -ape_stack_memsz);");
+
+ASSERT(ALIGNOF(.tdata) <= TLS_ALIGNMENT && ALIGNOF(.tbss) <= TLS_ALIGNMENT,
+       "_Thread_local _Alignof can't exceed TLS_ALIGNMENT");

 /* Let's not be like Knight Capital. */
 /* NOCROSSREFS_TO(.test .text) */
@@ -15,85 +15,151 @@

 PKGS += APE

-APE =	o/$(MODE)/ape/ape.o				\
-	o/$(MODE)/ape/ape.lds
-
-APE_NO_MODIFY_SELF =					\
-	o/$(MODE)/ape/ape.lds				\
-	o/$(MODE)/ape/ape-no-modify-self.o
-
-APE_COPY_SELF =						\
-	o/$(MODE)/ape/ape.lds				\
-	o/$(MODE)/ape/ape-copy-self.o
-
-APELINK =						\
-	$(COMPILE)					\
-	-ALINK.ape					\
-	$(LINK)						\
-	$(LINKARGS)					\
-	$(OUTPUT_OPTION)
-
-APE_LOADER_FLAGS =					\
-	-DNDEBUG					\
-	-iquote.					\
-	-Wall						\
-	-Wextra						\
-	-fpie						\
-	-Os						\
-	-ffreestanding					\
-	-mgeneral-regs-only				\
-	-mno-red-zone					\
-	-fno-ident					\
-	-fno-gnu-unique					\
-	-c						\
-	 $(OUTPUT_OPTION)				\
-	$<
-
 APE_FILES := $(wildcard ape/*.*)
 APE_HDRS = $(filter %.h,$(APE_FILES))
 APE_INCS = $(filter %.inc,$(APE_FILES))
+
+APE =	o/$(MODE)/ape/ape.o			\
+	o/$(MODE)/ape/ape.lds
+
+APELINK =					\
+	$(COMPILE)				\
+	-ALINK.ape				\
+	$(LINK)					\
+	$(LINKARGS)				\
+	$(OUTPUT_OPTION)
+
+ifeq ($(ARCH), aarch64)
+
+APE_SRCS = ape/ape.S
+APE_OBJS = o/$(MODE)/ape/ape.o
+APE_NO_MODIFY_SELF = $(APE)
+APE_COPY_SELF = $(APE)
+
+.PHONY: o/$(MODE)/ape
+o/$(MODE)/ape: $(APE)
+
+else
+
+APE_NO_MODIFY_SELF =				\
+	o/$(MODE)/ape/ape.lds			\
+	o/$(MODE)/ape/ape-no-modify-self.o
+
+APE_COPY_SELF =					\
+	o/$(MODE)/ape/ape.lds			\
+	o/$(MODE)/ape/ape-copy-self.o
+
+APE_LOADER_FLAGS =				\
+	-DNDEBUG				\
+	-iquote.				\
+	-Wall					\
+	-Wextra					\
+	-fpie					\
+	-Os					\
+	-ffreestanding				\
+	-mgeneral-regs-only			\
+	-mno-red-zone				\
+	-fno-ident				\
+	-fno-gnu-unique				\
+	-c					\
+	 $(OUTPUT_OPTION)			\
+	$<
+
 APE_SRCS_C = ape/loader.c
 APE_SRCS_S = $(filter %.S,$(APE_FILES))
 APE_SRCS = $(APE_SRCS_C) $(APE_SRCS_S)
 APE_OBJS = $(APE_SRCS_S:%.S=o/$(MODE)/%.o)
 APE_CHECKS = $(APE_HDRS:%=o/%.ok)

-o/$(MODE)/ape/ape.lds:					\
-		ape/ape.lds				\
-		ape/macros.internal.h			\
-		libc/dce.h				\
+o/$(MODE)/ape/public/ape.lds: OVERRIDE_CPPFLAGS += -UCOSMO
+o/$(MODE)/ape/public/ape.lds:			\
+		ape/public/ape.lds		\
+		ape/ape.lds			\
+		ape/ape.internal.h		\
+		ape/macros.internal.h		\
+		ape/relocations.h		\
+		libc/intrin/bits.h		\
+		libc/thread/tls.h		\
+		libc/calls/struct/timespec.h	\
+		libc/thread/thread.h		\
+		libc/dce.h			\
+		libc/elf/def.h			\
+		libc/elf/pf2prot.internal.h	\
+		libc/macros.internal.h		\
+		libc/nt/pedef.internal.h	\
+		libc/str/str.h			\
 		libc/zip.h

-o/ape/idata.inc:					\
-		ape/idata.internal.h			\
+o/ape/idata.inc:				\
+		ape/idata.internal.h		\
 		ape/relocations.h

-o/$(MODE)/ape/ape-no-modify-self.o:			\
-		ape/ape.S				\
+o/$(MODE)/ape/ape-no-modify-self.o:		\
+		ape/ape.S			\
+		ape/macros.internal.h		\
+		ape/notice.inc			\
+		ape/relocations.h		\
+		ape/ape.internal.h		\
+		libc/dce.h			\
+		libc/elf/def.h			\
+		libc/intrin/asancodes.h		\
+		libc/macho.internal.h		\
+		libc/macros.internal.h		\
+		libc/nexgen32e/uart.internal.h	\
+		libc/calls/metalfile.internal.h	\
+		libc/nexgen32e/vidya.internal.h	\
+		libc/nt/pedef.internal.h	\
+		libc/runtime/e820.internal.h	\
+		libc/runtime/mman.internal.h	\
+		libc/runtime/pc.internal.h	\
+		libc/runtime/symbolic.h		\
+		libc/sysv/consts/prot.h		\
+		ape/blink-aarch64.gz		\
+		ape/blink-darwin-arm64.gz	\
 		o/$(MODE)/ape/ape.elf
-	@$(COMPILE)					\
-		-AOBJECTIFY.S				\
-		$(OBJECTIFY.S)				\
-		$(OUTPUT_OPTION)			\
-		-DAPE_NO_MODIFY_SELF			\
+	@$(COMPILE)				\
+		-AOBJECTIFY.S			\
+		$(OBJECTIFY.S)			\
+		$(OUTPUT_OPTION)		\
+		-DAPE_NO_MODIFY_SELF		\
 		-DAPE_LOADER='"o/$(MODE)/ape/ape.elf"' $<

-o/$(MODE)/ape/ape-copy-self.o:				\
-		ape/ape.S
-	@$(COMPILE)					\
-		-AOBJECTIFY.S				\
-		$(OBJECTIFY.S)				\
-		$(OUTPUT_OPTION)			\
+o/$(MODE)/ape/ape-copy-self.o:			\
+		ape/ape.S			\
+		ape/macros.internal.h		\
+		ape/notice.inc			\
+		ape/relocations.h		\
+		ape/ape.internal.h		\
+		libc/dce.h			\
+		libc/elf/def.h			\
+		libc/intrin/asancodes.h		\
+		libc/macho.internal.h		\
+		libc/macros.internal.h		\
+		libc/nexgen32e/uart.internal.h	\
+		libc/calls/metalfile.internal.h	\
+		libc/nexgen32e/vidya.internal.h	\
+		libc/nt/pedef.internal.h	\
+		libc/runtime/e820.internal.h	\
+		libc/runtime/mman.internal.h	\
+		libc/runtime/pc.internal.h	\
+		libc/runtime/symbolic.h		\
+		libc/sysv/consts/prot.h		\
+		ape/blink-aarch64.gz		\
+		ape/blink-darwin-arm64.gz
+	@$(COMPILE)				\
+		-AOBJECTIFY.S			\
+		$(OBJECTIFY.S)			\
+		$(OUTPUT_OPTION)		\
 		-DAPE_NO_MODIFY_SELF $<

-o/$(MODE)/ape/loader.o: ape/loader.c
+o/$(MODE)/ape/loader.o: ape/loader.c ape/loader.h
 	@$(COMPILE) -AOBJECTIFY.c $(CC) -DSUPPORT_VECTOR=0b01111001 -g $(APE_LOADER_FLAGS)
 o/$(MODE)/ape/loader-gcc.asm: ape/loader.c
 	@$(COMPILE) -AOBJECTIFY.c $(CC) -DSUPPORT_VECTOR=0b01111001 -S -g0 $(APE_LOADER_FLAGS)
 o/$(MODE)/ape/loader-clang.asm: ape/loader.c
 	@$(COMPILE) -AOBJECTIFY.c $(CLANG) -DSUPPORT_VECTOR=0b01111001 -S -g0 $(APE_LOADER_FLAGS)

-o/$(MODE)/ape/loader-xnu.o: ape/loader.c
+o/$(MODE)/ape/loader-xnu.o: ape/loader.c ape/loader.h
 	@$(COMPILE) -AOBJECTIFY.c $(CC) -DSUPPORT_VECTOR=0b00001000 -g $(APE_LOADER_FLAGS)
 o/$(MODE)/ape/loader-xnu-gcc.asm: ape/loader.c
 	@$(COMPILE) -AOBJECTIFY.c $(CC) -DSUPPORT_VECTOR=0b00001000 -S -g0 $(APE_LOADER_FLAGS)
@@ -103,23 +169,51 @@ o/$(MODE)/ape/loader-xnu-clang.asm: ape/loader.c
 o/$(MODE)/ape/ape.elf: o/$(MODE)/ape/ape.elf.dbg
 o/$(MODE)/ape/ape.macho: o/$(MODE)/ape/ape.macho.dbg

-o/$(MODE)/ape/ape.elf.dbg:				\
-		o/$(MODE)/ape/loader.o			\
-		o/$(MODE)/ape/loader-elf.o		\
+o/$(MODE)/ape/ape.elf.dbg:			\
+		o/$(MODE)/ape/loader.o		\
+		o/$(MODE)/ape/loader-elf.o	\
 		ape/loader.lds
 	@$(ELFLINK) -z max-page-size=0x10

-o/$(MODE)/ape/ape.macho.dbg:				\
-		o/$(MODE)/ape/loader-xnu.o		\
-		o/$(MODE)/ape/loader-macho.o		\
+o/$(MODE)/ape/ape.macho.dbg:			\
+		o/$(MODE)/ape/loader-xnu.o	\
+		o/$(MODE)/ape/loader-macho.o	\
 		ape/loader-macho.lds
 	@$(ELFLINK) -z max-page-size=0x10

 .PHONY: o/$(MODE)/ape
-o/$(MODE)/ape:	$(APE_CHECKS)				\
-		o/$(MODE)/ape/ape.o			\
-		o/$(MODE)/ape/ape.lds			\
-		o/$(MODE)/ape/ape.elf			\
-		o/$(MODE)/ape/ape.macho			\
-		o/$(MODE)/ape/ape-copy-self.o		\
+o/$(MODE)/ape:	$(APE_CHECKS)			\
+		o/$(MODE)/ape/ape.o		\
+		o/$(MODE)/ape/ape.lds		\
+		o/$(MODE)/ape/public/ape.lds	\
+		o/$(MODE)/ape/ape.elf		\
+		o/$(MODE)/ape/ape.macho		\
+		o/$(MODE)/ape/ape-copy-self.o	\
 		o/$(MODE)/ape/ape-no-modify-self.o
+
+endif
+
+# these assembly files are safe to build on aarch64
+o/$(MODE)/ape/ape.o: ape/ape.S
+	@$(COMPILE) -AOBJECTIFY.S $(OBJECTIFY.S) $(OUTPUT_OPTION) -c $<
+
+o/$(MODE)/ape/ape.o:				\
+		ape/blink-aarch64.gz		\
+		ape/blink-darwin-arm64.gz
+
+o/$(MODE)/ape/ape.lds:				\
+		ape/ape.lds			\
+		ape/macros.internal.h		\
+		ape/relocations.h		\
+		ape/ape.internal.h		\
+		libc/intrin/bits.h		\
+		libc/thread/tls.h		\
+		libc/calls/struct/timespec.h	\
+		libc/thread/thread.h		\
+		libc/dce.h			\
+		libc/elf/def.h			\
+		libc/elf/pf2prot.internal.h	\
+		libc/macros.internal.h		\
+		libc/nt/pedef.internal.h	\
+		libc/str/str.h			\
+		libc/zip.h
@@ -1,6 +1,6 @@
 #!/bin/sh

-if [ $UID -eq 0 ]; then
+if [ "$(id -u)" -eq 0 ]; then
  SUDO=
 else
  SUDO=sudo
@@ -97,8 +97,19 @@ if [ x"$(uname -s)" = xLinux ]; then
  echo you may need to edit configs to persist across reboot >&2
  echo '$SUDO sh -c "echo '"'"':APE:M::MZqFpD::/usr/bin/ape:'"'"' >/proc/sys/fs/binfmt_misc/register"' >&2
  $SUDO sh -c "echo ':APE:M::MZqFpD::/usr/bin/ape:' >/proc/sys/fs/binfmt_misc/register" || exit
+  echo '$SUDO sh -c "echo '"'"':APE-jart:M::jartsr::/usr/bin/ape:'"'"' >/proc/sys/fs/binfmt_misc/register"' >&2
+  $SUDO sh -c "echo ':APE-jart:M::jartsr::/usr/bin/ape:' >/proc/sys/fs/binfmt_misc/register" || exit
  echo done >&2

+  if [ x"$(cat /proc/sys/fs/binfmt_misc/status)" = xdisabled ]; then
+    echo >&2
+    echo enabling binfmt_misc >&2
+    echo you may need to edit configs to persist across reboot >&2
+    echo $SUDO sh -c 'echo 1 >/proc/sys/fs/binfmt_misc/status' >&2
+    $SUDO sh -c 'echo 1 >/proc/sys/fs/binfmt_misc/status' || exit
+    echo done >&2
+  fi
+
 fi

 ################################################################################
@@ -8,17 +8,19 @@ fi

 {
  echo
-  echo "APE Uninstaller intends to run"
+  echo "APE Uninstaller intends to run (in pseudo-shell)"
  echo
-  echo "    $SUDO sh -c 'echo -1 >/proc/sys/fs/binfmt_misc/APE'"
-  echo "    $SUDO rm -f /usr/bin/ape ~/.ape o/tmp/.ape /tmp/.ape"
+  echo "    sudo echo -1 into /proc/sys/fs/binfmt_misc/APE*"
+  echo "    sudo rm -f /usr/bin/ape ~/.ape o/tmp/.ape /tmp/.ape"
  echo
  echo "You may then use ape/apeinstall.sh to reinstall it"
  echo
 } >&2

 set -ex
-if [ -f /proc/sys/fs/binfmt_misc/APE ]; then
-  $SUDO sh -c 'echo -1 >/proc/sys/fs/binfmt_misc/APE' || exit
-fi
-$SUDO rm -f /usr/bin/ape ~/.ape o/tmp/.ape /tmp/.ape || exit
+for f in /proc/sys/fs/binfmt_misc/APE*; do
+  if [ -f $f ]; then
+    $SUDO sh -c "echo -1 >$f" || exit
+  fi
+done
+$SUDO rm -f /usr/bin/ape ~/.ape o/tmp/.ape o/tmp/ape /tmp/.ape /tmp/ape || exit
@@ -32,79 +32,87 @@
 //	@see	ape/ape.lds
 //	@see	winimp
 .macro	.imp	dll:req fn:req actual:req hint
-	.dll	\dll
-	.section .piro.data.sort.iat.2.\dll\().2.\actual,"aw",@progbits
+#ifdef __x86_64__
+	.dll	"\dll"
+	.section ".piro.data.sort.iat.2.\dll\().2.\actual","aw",@progbits
 	.type	\fn,@object
 	.align	__SIZEOF_POINTER__
-\fn:	.quad	RVA((\dll\().\actual))
+\fn:	.quad	RVA(("\dll\().\actual"))
 	.size	\fn,.-\fn
 	.globl	\fn
 	.hidden	\fn
 	.previous
-	.section .idata.ro.ilt.\dll\().2.\actual,"a",@progbits
-.Lidata.ilt.\dll\().\actual:
-	.quad	RVA((\dll\().\actual))
-	.type	.Lidata.ilt.\dll\().\actual,@object
-	.size	.Lidata.ilt.\dll\().\actual,.-.Lidata.ilt.\dll\().\actual
+	.section ".idata.ro.ilt.\dll\().2.\actual","a",@progbits
+".Lidata.ilt.\dll\().\actual":
+	.quad	RVA("\dll\().\actual")
+	.type	".Lidata.ilt.\dll\().\actual",@object
+	.size	".Lidata.ilt.\dll\().\actual",.-".Lidata.ilt.\dll\().\actual"
 	.previous
-	.section .idata.ro.hnt.\dll\().2.\actual,"a",@progbits
-\dll\().\actual:
-	.ifnb	\hint			# hint i.e. guess function ordinal
+	.section ".idata.ro.hnt.\dll\().2.\actual","a",@progbits
+"\dll\().\actual":
+	.ifnb	\hint			// hint i.e. guess function ordinal
 	.short	\hint
 	.else
 	.short	0
 	.endif
 	.asciz	"\actual"
-	.align	2			# documented requirement
-	.globl	\dll\().\actual
-	.hidden	\dll\().\actual
-	.type	\dll\().\actual,@object
-	.size	\dll\().\actual,.-\dll\().\actual
+	.align	2			// documented requirement
+	.globl	"\dll\().\actual"
+	.hidden	"\dll\().\actual"
+	.type	"\dll\().\actual",@object
+	.size	"\dll\().\actual",.-"\dll\().\actual"
 	.previous
+#else
+	.section ".data.nt.\actual","aw",@progbits
+	.globl	"\fn"
+	.balign	8
+	.weak	"\actual"
+"\fn":	.quad	"\actual"
+#endif
 .endm

 //	Defines DLL import.
 //	@note	this is an implementation detail of .imp
 .macro	.dll	name:req
-  .section .idata.ro.idt.2.\name,"aG",@progbits,\name,comdat
-	.equ	.Lidata.idt.\name,.
-	.long	RVA(idata.ilt.\name)		# ImportLookupTable
-	.long	0				# TimeDateStamp
-	.long	0				# ForwarderChain
-	.long	RVA(.Lidata.str.\name)		# DllNameRva
-	.long	RVA(idata.iat.\name)		# ImportAddressTable
-	.type	.Lidata.idt.\name,@object
-	.size	.Lidata.idt.\name,.-.Lidata.idt.\name
+  .section ".idata.ro.idt.2.\name","aG",@progbits,"\name",comdat
+	.equ	".Lidata.idt.\name",.
+	.long	RVA("idata.ilt.\name")		// ImportLookupTable
+	.long	0				// TimeDateStamp
+	.long	0				// ForwarderChain
+	.long	RVA(".Lidata.str.\name")	// DllNameRva
+	.long	RVA("idata.iat.\name")		// ImportAddressTable
+	.type	".Lidata.idt.\name",@object
+	.size	".Lidata.idt.\name",.-".Lidata.idt.\name"
  .previous
-  .section .idata.ro.ilt.\name\().1,"aG",@progbits,\name,comdat
+  .section ".idata.ro.ilt.\name\().1","aG",@progbits,"\name",comdat
 	.align	__SIZEOF_POINTER__
-	.type	idata.ilt.\name,@object
-idata.ilt.\name:
+	.type	"idata.ilt.\name",@object
+"idata.ilt.\name":
 	.previous/*
 	...
 	decentralized content
 	...
-	*/.section .idata.ro.ilt.\name\().3,"aG",@progbits,\name,comdat
+	*/.section ".idata.ro.ilt.\name\().3","aG",@progbits,"\name",comdat
 	.quad	0
  .previous
-  .section .idata.ro.hnt.\name\().1,"aG",@progbits,\name,comdat
+  .section ".idata.ro.hnt.\name\().1","aG",@progbits,"\name",comdat
 	.align	__SIZEOF_POINTER__
-	.type	idata.hnt.\name,@object
-	.equ	idata.hnt.\name,.
+	.type	"idata.hnt.\name",@object
+	.equ	"idata.hnt.\name",.
  .previous
-  .section .piro.data.sort.iat.2.\name\().1,"awG",@progbits,\name,comdat
+  .section ".piro.data.sort.iat.2.\name\().1","awG",@progbits,"\name",comdat
 	.align	__SIZEOF_POINTER__
-	.type	idata.iat.\name,@object
-idata.iat.\name:
+	.type	"idata.iat.\name",@object
+"idata.iat.\name":
 	.previous/*
 	...
 	decentralized content
 	...
-	*/.section .piro.data.sort.iat.2.\name\().3,"awG",@progbits,\name,comdat
+	*/.section ".piro.data.sort.iat.2.\name\().3","awG",@progbits,"\name",comdat
 	.quad	0
  .previous
  .section .rodata.str1.1,"aSM",@progbits,1
-.Lidata.str.\name:
+".Lidata.str.\name":
 	.asciz	"\name\().dll"
  .previous
 .endm
@@ -25,7 +25,7 @@
 //	APE Loader Executable Structure
 //	Linux, FreeBSD, NetBSD, OpenBSD

-	.align	8
+	.balign	8
 ehdr:	.ascii	"\177ELF"
 	.byte	ELFCLASS64
 	.byte	ELFDATA2LSB
@@ -47,7 +47,7 @@ ehdr:	.ascii	"\177ELF"
 	.word	0			# e_shstrndx
 	.endobj	ehdr,globl

-	.align	8
+	.balign	8
 phdrs:	.long	PT_LOAD			# p_type
 	.long	PF_R|PF_X		# p_flags
 	.quad	0			# p_offset
@@ -89,18 +89,18 @@ note:	.long	2f-1f
 	.long	4f-3f
 	.long	1
 1:	.asciz	"OpenBSD"
-2:	.align	4
+2:	.balign	4
 3:	.long	0
 4:	.long	2f-1f
 	.long	4f-3f
 	.long	1
 1:	.asciz	"NetBSD"
-2:	.align	4
+2:	.balign	4
 3:	.long	901000000
 4:	.endobj	note
 	notesize = . - note

-	.align	64,0				# for ape.S dd
+	.balign	64,0				# for ape.S dd
 	.org	0x180				# for ape.S dd

 //	APE Loader XNU Header
@@ -175,7 +175,7 @@ macho:	.long	0xFEEDFACE+1
 510:	.quad	0				# rax
 	.quad	0				# rbx
 	.quad	0				# rcx
-	.quad	XNU				# rdx
+	.quad	_HOSTXNU			# rdx
 	.quad	0				# rdi
 	.quad	0				# rsi
 	.quad	0				# rbp
@@ -197,7 +197,7 @@ macho:	.long	0xFEEDFACE+1
 60:
 	.endobj	macho

-	.align	64,0				# for ape.S dd
+	.balign	64,0				# for ape.S dd
 	.org	0x400				# for ape.S dd

 //	Ape Loader Entrpoint
@@ -221,7 +221,7 @@ _start:
 	jne	0f
 	cmp	$+1,%edx
 	jne	0f
-	mov	$XNU,%dl
+	mov	$_HOSTXNU,%dl
 	xor	%ecx,%ecx

 0:	mov	%rsp,%rsi
@@ -23,7 +23,7 @@

 //	APE Loader Executable Structure for XNU

-	.align	4096
+	.balign	4096
 macho:	.long	0xFEEDFACE+1
 	.long	MAC_CPU_NEXGEN32E
 	.long	MAC_CPU_NEXGEN32E_ALL
@@ -90,7 +90,7 @@ macho:	.long	0xFEEDFACE+1
 510:	.quad	0				# rax
 	.quad	0				# rbx
 	.quad	0				# rcx
-	.quad	XNU				# rdx
+	.quad	_HOSTXNU			# rdx
 	.quad	0				# rdi
 	.quad	0				# rsi
 	.quad	0				# rbp
@@ -112,7 +112,7 @@ macho:	.long	0xFEEDFACE+1
 60:
 	.endobj	macho,globl

-	.align	64
+	.balign	64
 _start:

 //	Hack for detecting M1 Rosetta environment.
@@ -121,7 +121,7 @@ _start:
 	jne	0f
 	cmp	$+1,%edx
 	jne	0f
-	mov	$XNU,%dl
+	mov	$_HOSTXNU,%dl
 	xor	%ecx,%ecx

 0:	mov	%rsp,%rsi
@@ -18,6 +18,7 @@
 ╚─────────────────────────────────────────────────────────────────────────────*/
 #include "ape/loader.h"

+#define SET_EXE_FILE    0 /* needs root ;_; */
 #define TROUBLESHOOT    0
 #define TROUBLESHOOT_OS LINUX

@@ -99,30 +100,32 @@
 #define IsOpenbsd() (SupportsOpenbsd() && os == OPENBSD)
 #define IsNetbsd()  (SupportsNetbsd() && os == NETBSD)

-#define O_RDONLY         0
-#define PROT_READ        1
-#define PROT_WRITE       2
-#define PROT_EXEC        4
-#define MAP_SHARED       1
-#define MAP_PRIVATE      2
-#define MAP_FIXED        16
-#define MAP_ANONYMOUS    (IsLinux() ? 32 : 4096)
-#define AT_EXECFN_LINUX  31
-#define AT_EXECFN_NETBSD 2014
-#define ELFCLASS64       2
-#define ELFDATA2LSB      1
-#define EM_NEXGEN32E     62
-#define ET_EXEC          2
-#define PT_LOAD          1
-#define PT_DYNAMIC       2
-#define EI_CLASS         4
-#define EI_DATA          5
-#define PF_X             1
-#define PF_W             2
-#define PF_R             4
-#define X_OK             1
-#define XCR0_SSE         2
-#define XCR0_AVX         4
+#define O_RDONLY           0
+#define PROT_READ          1
+#define PROT_WRITE         2
+#define PROT_EXEC          4
+#define MAP_SHARED         1
+#define MAP_PRIVATE        2
+#define MAP_FIXED          16
+#define MAP_ANONYMOUS      (IsLinux() ? 32 : 4096)
+#define AT_EXECFN_LINUX    31
+#define AT_EXECFN_NETBSD   2014
+#define ELFCLASS64         2
+#define ELFDATA2LSB        1
+#define EM_NEXGEN32E       62
+#define ET_EXEC            2
+#define PT_LOAD            1
+#define PT_DYNAMIC         2
+#define EI_CLASS           4
+#define EI_DATA            5
+#define PF_X               1
+#define PF_W               2
+#define PF_R               4
+#define X_OK               1
+#define XCR0_SSE           2
+#define XCR0_AVX           4
+#define PR_SET_MM          35
+#define PR_SET_MM_EXE_FILE 13

 #define Read32(S)                                                      \
  ((unsigned)(255 & (S)[3]) << 030 | (unsigned)(255 & (S)[2]) << 020 | \
@@ -174,7 +177,10 @@ struct ElfPhdr {
  unsigned long p_align;
 };

-static void *syscall;
+extern char ehdr[];
+extern char _end[];
+static void *syscall_;
+static char relocated;
 static struct PathSearcher ps;
 extern char __syscall_loader[];

@@ -182,7 +188,7 @@ static int ToLower(int c) {
  return 'A' <= c && c <= 'Z' ? c + ('a' - 'A') : c;
 }

-static char *MemCpy(char *d, const char *s, unsigned long n) {
+char *MemCpy(char *d, const char *s, unsigned long n) {
  unsigned long i = 0;
  for (; i < n; ++i) d[i] = s[i];
  return d + n;
@@ -269,7 +275,7 @@ __attribute__((__noreturn__)) static void Exit(int rc, int os) {
  asm volatile("call\t*%2"
               : /* no outputs */
               : "a"((IsLinux() ? 60 : 1) | (IsXnu() ? 0x2000000 : 0)), "D"(rc),
-                 "rm"(syscall)
+                 "rm"(syscall_)
               : "memory");
  __builtin_unreachable();
 }
@@ -279,7 +285,7 @@ static void Close(int fd, int os) {
  asm volatile("call\t*%4"
               : "=a"(ax), "=D"(di)
               : "0"((IsLinux() ? 3 : 6) | (IsXnu() ? 0x2000000 : 0)), "1"(fd),
-                 "rm"(syscall)
+                 "rm"(syscall_)
               : "rcx", "rdx", "rsi", "r8", "r9", "r10", "r11", "memory", "cc");
 }

@@ -289,7 +295,7 @@ static int Read(int fd, void *data, int size, int os) {
  asm volatile("call\t*%8"
               : "=a"(ax), "=D"(di), "=S"(si), "=d"(dx)
               : "0"((IsLinux() ? 0 : 3) | (IsXnu() ? 0x2000000 : 0)), "1"(fd),
-                 "2"(data), "3"(size), "rm"(syscall)
+                 "2"(data), "3"(size), "rm"(syscall_)
               : "rcx", "r8", "r9", "r10", "r11", "memory");
  return ax;
 }
@@ -300,7 +306,7 @@ static void Write(int fd, const void *data, int size, int os) {
  asm volatile("call\t*%8"
               : "=a"(ax), "=D"(di), "=S"(si), "=d"(dx)
               : "0"((IsLinux() ? 1 : 4) | (IsXnu() ? 0x2000000 : 0)), "1"(fd),
-                 "2"(data), "3"(size), "rm"(syscall)
+                 "2"(data), "3"(size), "rm"(syscall_)
               : "rcx", "r8", "r9", "r10", "r11", "memory", "cc");
 }

@@ -309,7 +315,7 @@ static void Execve(const char *prog, char **argv, char **envp, int os) {
  asm volatile("call\t*%8"
               : "=a"(ax), "=D"(di), "=S"(si), "=d"(dx)
               : "0"(59 | (IsXnu() ? 0x2000000 : 0)), "1"(prog), "2"(argv),
-                 "3"(envp), "rm"(syscall)
+                 "3"(envp), "rm"(syscall_)
               : "rcx", "r8", "r9", "r10", "r11", "memory", "cc");
 }

@@ -319,7 +325,7 @@ static int Access(const char *path, int mode, int os) {
  asm volatile("call\t*%7"
               : "=a"(ax), "=D"(di), "=S"(si), "=d"(dx)
               : "0"((IsLinux() ? 21 : 33) | (IsXnu() ? 0x2000000 : 0)),
-                 "1"(path), "2"(mode), "rm"(syscall)
+                 "1"(path), "2"(mode), "rm"(syscall_)
               : "rcx", "r8", "r9", "r10", "r11", "memory", "cc");
  return ax;
 }
@@ -332,7 +338,7 @@ static int Msyscall(long p, long n, int os) {
  } else {
    asm volatile("call\t*%6"
                 : "=a"(ax), "=D"(di), "=S"(si)
-                 : "0"(37), "1"(p), "2"(n), "rm"(syscall)
+                 : "0"(37), "1"(p), "2"(n), "rm"(syscall_)
                 : "rcx", "rdx", "r8", "r9", "r10", "r11", "memory", "cc");
    return ax;
  }
@@ -344,7 +350,7 @@ static int Open(const char *path, int flags, int mode, int os) {
  asm volatile("call\t*%8"
               : "=a"(ax), "=D"(di), "=S"(si), "=d"(dx)
               : "0"((IsLinux() ? 2 : 5) | (IsXnu() ? 0x2000000 : 0)),
-                 "1"(path), "2"(flags), "3"(mode), "rm"(syscall)
+                 "1"(path), "2"(flags), "3"(mode), "rm"(syscall_)
               : "rcx", "r8", "r9", "r10", "r11", "memory", "cc");
  return ax;
 }
@@ -363,7 +369,7 @@ __attribute__((__noinline__)) static long Mmap(long addr, long size, int prot,
               "pop\t%%r9"
               : "=a"(ax), "=D"(di), "=S"(si), "=d"(dx), "+r"(flags_),
                 "+r"(fd_), "+r"(off_)
-               : "rm"(syscall),
+               : "rm"(syscall_),
                 "0"((IsLinux()     ? 9
                      : IsFreebsd() ? 477
                                    : 197) |
@@ -373,6 +379,26 @@ __attribute__((__noinline__)) static long Mmap(long addr, long size, int prot,
  return ax;
 }

+int MunmapLinux(const void *addr, unsigned long size) {
+  int ax;
+  asm volatile("syscall"
+               : "=a"(ax)
+               : "0"(11), "D"(addr), "S"(size)
+               : "rcx", "r11", "memory");
+  return ax;
+}
+
+int PrctlLinux(int op, long a, long b, long c, long d) {
+  int rc;
+  asm volatile("mov\t%5,%%r10\n\t"
+               "mov\t%6,%%r8\n\t"
+               "syscall"
+               : "=a"(rc)
+               : "0"(157), "D"(op), "S"(a), "d"(b), "g"(c), "g"(d)
+               : "rcx", "r8", "r10", "r11", "memory");
+  return rc;
+}
+
 static void Emit(int os, const char *s) {
  Write(2, s, StrLen(s), os);
 }
@@ -539,6 +565,23 @@ __attribute__((__noreturn__)) static void Spawn(int os, const char *exe, int fd,
  if (!code) {
    Pexit(os, exe, 0, "ELF needs PT_LOAD phdr w/ PF_X");
  }
+
+#if SET_EXE_FILE
+  // change /proc/pid/exe to new executable path
+  if (IsLinux() && relocated) {
+    MunmapLinux((char *)0x200000, (long)(_end - ehdr));
+    PrctlLinux(PR_SET_MM, PR_SET_MM_EXE_FILE, fd, 0, 0);
+  }
+#endif
+
+  if (relocated) {
+    int ax;
+    asm volatile("syscall"
+                 : "=a"(ax)
+                 : "0"(1), "D"(1), "S"("hello\n"), "d"(6)
+                 : "rcx", "r11", "memory");
+  }
+
  Close(fd, os);

  // authorize only the loaded program to issue system calls. if this
@@ -546,7 +589,7 @@ __attribute__((__noreturn__)) static void Spawn(int os, const char *exe, int fd,
  // since it probably means a userspace program executed this loader
  // and passed us a custom syscall function earlier.
  if (Msyscall(code, codesize, os) != -1) {
-    syscall = 0;
+    syscall_ = 0;
  }

 #if TROUBLESHOOT
@@ -557,16 +600,16 @@ __attribute__((__noreturn__)) static void Spawn(int os, const char *exe, int fd,
  // to extend the behavior of this loader in the future. we don't need
  // to clear the xmm registers since the ape loader should be compiled
  // with the -mgeneral-regs-only flag.
-  register void *r8 asm("r8") = syscall;
+  register void *r8 asm("r8") = syscall_;
  asm volatile("xor\t%%eax,%%eax\n\t"
               "xor\t%%r9d,%%r9d\n\t"
               "xor\t%%r10d,%%r10d\n\t"
               "xor\t%%r11d,%%r11d\n\t"
-               "xor\t%%ebx,%%ebx\n\t"    // netbsd dosen't clear this
-               "xor\t%%r12d,%%r12d\n\t"  // netbsd dosen't clear this
-               "xor\t%%r13d,%%r13d\n\t"  // netbsd dosen't clear this
-               "xor\t%%r14d,%%r14d\n\t"  // netbsd dosen't clear this
-               "xor\t%%r15d,%%r15d\n\t"  // netbsd dosen't clear this
+               "xor\t%%ebx,%%ebx\n\t"    // netbsd doesnt't clear this
+               "xor\t%%r12d,%%r12d\n\t"  // netbsd doesnt't clear this
+               "xor\t%%r13d,%%r13d\n\t"  // netbsd doesnt't clear this
+               "xor\t%%r14d,%%r14d\n\t"  // netbsd doesnt't clear this
+               "xor\t%%r15d,%%r15d\n\t"  // netbsd doesnt't clear this
               "mov\t%%rdx,%%rsp\n\t"
               "xor\t%%edx,%%edx\n\t"
               "push\t%%rsi\n\t"
@@ -584,7 +627,7 @@ __attribute__((__noreturn__)) void ApeLoader(long di, long *sp, char dl,
                                             struct ApeLoader *handoff) {
  int rc;
  long *auxv;
-  struct ElfEhdr *ehdr;
+  struct ElfEhdr *eh;
  int c, i, fd, os, argc;
  char *p, *exe, *prog, **argv, **envp, *page;
  static union {
@@ -616,10 +659,10 @@ __attribute__((__noreturn__)) void ApeLoader(long di, long *sp, char dl,
  }

  // get syscall function pointer
-  if (handoff && handoff->syscall) {
-    syscall = handoff->syscall;
+  if (handoff && handoff->systemcall) {
+    syscall_ = handoff->systemcall;
  } else {
-    syscall = __syscall_loader;
+    syscall_ = __syscall_loader;
  }

  if (handoff) {
@@ -630,7 +673,7 @@ __attribute__((__noreturn__)) void ApeLoader(long di, long *sp, char dl,
    fd = handoff->fd;
    exe = handoff->prog;
    page = handoff->page;
-    ehdr = (struct ElfEhdr *)handoff->page;
+    eh = (struct ElfEhdr *)handoff->page;
  } else {

    // detect openbsd
@@ -653,6 +696,20 @@ __attribute__((__noreturn__)) void ApeLoader(long di, long *sp, char dl,
      os = LINUX;
    }

+#if SET_EXE_FILE
+    if (IsLinux() && !relocated) {
+      char *b1 = (char *)0x200000;
+      char *b2 = (char *)0x300000;
+      void (*pApeLoader)(long, long *, char, struct ApeLoader *);
+      Mmap((long)b2, (long)(_end - ehdr), PROT_READ | PROT_WRITE | PROT_EXEC,
+           MAP_FIXED | MAP_PRIVATE | MAP_ANONYMOUS, -1, 0, os);
+      relocated = 1;
+      MemCpy(b2, b1, (long)(_end - ehdr));
+      pApeLoader = (void *)((char *)&ApeLoader - b1 + b2);
+      pApeLoader(di, sp, dl, handoff);
+    }
+#endif
+
    // we can load via shell, shebang, or binfmt_misc
    if (argc >= 3 && !StrCmp(argv[1], "-")) {
      // if the first argument is a hyphen then we give the user the
@@ -686,7 +743,7 @@ __attribute__((__noreturn__)) void ApeLoader(long di, long *sp, char dl,
    }

    page = u.p;
-    ehdr = &u.ehdr;
+    eh = &u.ehdr;
  }

 #if TROUBLESHOOT
@@ -726,7 +783,7 @@ __attribute__((__noreturn__)) void ApeLoader(long di, long *sp, char dl,
      page[i++] = c;
    }
    if (i >= 64 && Read32(page) == Read32("\177ELF")) {
-      Spawn(os, exe, fd, sp, page, ehdr);
+      Spawn(os, exe, fd, sp, page, eh);
    }
  }

@@ -14,7 +14,7 @@ struct ApeLoader {
  int os;
  char *prog;
  char *page;
-  void *syscall;
+  void *systemcall;
 };

 #endif /* COSMOPOLITAN_APE_LOADER_H_ */
@@ -32,7 +32,8 @@ SECTIONS {
    *(.bss)
    . = ALIGN(4096);
  }
-  memsz = . - ehdr;
+  _end = .;
+  memsz = _end - ehdr;
  /DISCARD/ : {
    *(.*)
  }
@@ -54,7 +54,7 @@
 	jmp	297f
 	.byte	0xbf			# mov $0x????xxxx,%edi
 	.long	\address
-297:	
+297:
 .endm

 //	Loads 16-bit CONSTEXPR into Qw-register w/ optional zero-extend.
@@ -180,6 +180,20 @@
 .stub	\name\()_bcs\n,long
 .endm

+//	Task State Segment Descriptor Entries.
+.macro	.tssdescstub name:req
+ .ifndef \name
+  .weak	\name
+  .set	\name,0
+ .endif
+ .ifndef \name\()_end
+  .weak	\name\()_end
+  .set	\name\()_end,0
+ .endif
+ .stub	\name\()_desc_ent0,quad
+ .stub	\name\()_desc_ent1,quad
+.endm
+
 /* clang-format on */
 #elif defined(__LINKER__)

@@ -221,44 +235,58 @@
 /**
 * Binary coded decimal support.
 *
- * <p>This allows linker scripts to generate dd commands. Leading spaces
- * need to be inserted so Mac doesn't consider them octal; therefore,
- * parameters must be quoted; and eight digits should be good enough.
+ * <p>This allows linker scripts to generate dd commands, e.g. ape.lds.
+ * There are a few ways to pad each number to the necessary 8 bytes.
+ * Spaces cannot be prepended because busybox refuses to parse them.
+ * Zeros cannot be prepended because Mac will take numbers as octal.
+ * That leaves appending spaces. The user's shell ought to treat any
+ * unquoted run of spaces as if there was only one, so this is safe.
 */
-#define SHSTUB2(SYM, X)           \
-  HIDDEN(SYM##_bcs0 = BCD10K(X)); \
-  HIDDEN(SYM##_bcs1 = BCD(X))
-#define BCD(X)                                                                 \
-  ((X) == 0                                                                    \
-       ? 0x20202030                                                            \
-       : (X) < 10 ? 0x30202020 + (((X) % 10) << 24)                            \
-                  : (X) < 100 ? 0x30302020 + (((X) % 10) << 24) +              \
-                                    (((X) / 10 % 10) << 16)                    \
-                              : (X) < 1000 ? 0x30303020 + (((X) % 10) << 24) + \
-                                                 (((X) / 10 % 10) << 16) +     \
-                                                 (((X) / 100 % 10) << 8)       \
-                                           : 0x30303030 + (((X) % 10) << 24) + \
-                                                 (((X) / 10 % 10) << 16) +     \
-                                                 (((X) / 100 % 10) << 8) +     \
-                                                 (((X) / 1000 % 10) << 0))
-#define BCD10K(X)                                                          \
-  ((X) < 10000                                                             \
-       ? 0x20202020                                                        \
-       : (X) < 100000                                                      \
-             ? 0x30202020 + (((X) / 10000 % 10) << 24)                     \
-             : (X) < 1000000                                               \
-                   ? 0x30302020 + (((X) / 10000 % 10) << 24) +             \
-                         (((X) / 100000 % 10) << 16)                       \
-                   : (X) < 10000000                                        \
-                         ? 0x30303020 + (((X) / 10000 % 10) << 24) +       \
-                               (((X) / 100000 % 10) << 16) +               \
-                               (((X) / 1000000 % 10) << 8)                 \
-                         : (X) < 100000000                                 \
-                               ? 0x30303030 + (((X) / 10000 % 10) << 24) + \
-                                     (((X) / 100000 % 10) << 16) +         \
-                                     (((X) / 1000000 % 10) << 8) +         \
-                                     (((X) / 10000000 % 10) << 0)          \
-                               : 0xffffffffffffffff)
+#define SHSTUB2(SYM, X)             \
+  HIDDEN(SYM##_bcs0 = BCD_LEFT(X)); \
+  HIDDEN(SYM##_bcs1 = BCD_RIGHT(X))
+#define BCD_SMEAR(X) ((X) + (X) * 10000)
+#define BCD_LEFT(X)                                      \
+  (((X)) < 10000     ? BCD_RIGHT(BCD_SMEAR(X)) | 0x10    \
+   : (X) < 100000    ? BCD_RIGHT(BCD_SMEAR((X) / 10))    \
+   : (X) < 1000000   ? BCD_RIGHT(BCD_SMEAR((X) / 100))   \
+   : (X) < 10000000  ? BCD_RIGHT(BCD_SMEAR((X) / 1000))  \
+   : (X) < 100000000 ? BCD_RIGHT(BCD_SMEAR((X) / 10000)) \
+                     : 0xffffffffffffffff)
+#define BCD_RIGHT(X) \
+  (((X)) < 10000     ? 0x20202020                  \
+   : (X) < 100000    ? 0x20202030 +                \
+                       (X) % 10                    \
+   : (X) < 1000000   ? 0x20203030 +                \
+                       ((X) / 10) % 10 +           \
+                       (X) % 10 * 0x100            \
+   : (X) < 10000000  ? 0x20303030 +                \
+                       ((X) / 100) % 10 +          \
+                       ((X) / 10) % 10 * 0x100 +   \
+                       (X) % 10 * 0x10000          \
+   : (X) < 100000000 ? 0x30303030 +                \
+                       ((X) / 1000) % 10 +         \
+                       ((X) / 100) % 10 * 0x100 +  \
+                       ((X) / 10) % 10 * 0x10000 + \
+                       (X) % 10 * 0x1000000        \
+                     : 0xffffffffffffffff)
+
+/**
+ * Laying out the GDT entries for a TSS for bare metal operation.
+ */
+#define TSSDESCSTUB2(SYM, BASE, LIM)                 \
+  HIDDEN(SYM##_desc_ent0 = TSSDESC_ENT0(BASE, LIM)); \
+  HIDDEN(SYM##_desc_ent1 = TSSDESC_ENT1(BASE));      \
+  ASSERT((LIM) >= 0 && (LIM) <= 0xffff, "bare metal TSS is suspiciously fat")
+#define TSSDESC_ENT0(BASE, LIM)                \
+  (((LIM)        <<  0 & 0x000000000000ffff) | \
+   ((BASE)       << 16 & 0x000000ffffff0000) | \
+    0x89         << 40                       | \
+   ((LIM)  >> 16 << 48 & 0x000f000000000000) | \
+    0x2          << 52                       | \
+   ((BASE) >> 24 << 56 & 0xff00000000000000))
+#define TSSDESC_ENT1(BASE)                     \
+   ((BASE) >> 32 <<  0 & 0x00000000ffffffff)

 #endif /* __ASSEMBLER__ */
 #endif /* APE_MACROS_H_ */
@@ -0,0 +1 @@
+#include "ape/ape.lds"
@@ -36,8 +36,8 @@
 */
 #define REAL(x) ((x) - (IMAGE_BASE_VIRTUAL - IMAGE_BASE_REAL))

-#if IMAGE_BASE_VIRTUAL % 0x200000 != 0
-#error "IMAGE_BASE_VIRTUAL must be 2mb aligned"
+#if IMAGE_BASE_VIRTUAL % 0x1000 != 0
+#error "IMAGE_BASE_VIRTUAL must be 4kb aligned"
 #endif
 #if IMAGE_BASE_PHYSICAL % 0x1000 != 0
 #error "IMAGE_BASE_PHYSICAL must be 4kb aligned"
@@ -0,0 +1,34 @@
+#ifndef COSMOPOLITAN_APE_SECTIONS_INTERNAL_H_
+#define COSMOPOLITAN_APE_SECTIONS_INTERNAL_H_
+#if !(__ASSEMBLER__ + __LINKER__ + 0)
+COSMOPOLITAN_C_START_
+
+extern unsigned char _base[];
+extern unsigned char _ehead[];
+extern unsigned char _etext[];
+extern unsigned char _edata[];
+extern unsigned char _ezip[];
+extern unsigned char _end[];
+extern unsigned char _ereal[];
+extern unsigned char _tdata_start[];
+extern unsigned char _tdata_end[];
+extern unsigned char _tbss_start[];
+extern unsigned char _tbss_end[];
+extern unsigned char __privileged_start[];
+extern unsigned char __privileged_addr[];
+extern unsigned char __privileged_size[];
+extern unsigned char __privileged_end[];
+extern unsigned char __test_start[];
+extern unsigned char __ro[];
+extern unsigned char *__relo_start[];
+extern unsigned char *__relo_end[];
+extern uint8_t __zip_start[];
+extern uint8_t __zip_end[];
+extern uint8_t __data_start[];
+extern uint8_t __data_end[];
+extern uint8_t __bss_start[];
+extern uint8_t __bss_end[];
+
+COSMOPOLITAN_C_END_
+#endif /* !(__ASSEMBLER__ + __LINKER__ + 0) */
+#endif /* COSMOPOLITAN_APE_SECTIONS_INTERNAL_H_ */
@@ -4,20 +4,40 @@
 # Default Mode
 #
 #   - `make`
+#   - Optimized
 #   - Backtraces
+#   - Debuggable
 #   - Syscall tracing
 #   - Function tracing
 #   - Reasonably small
-#   - Reasonably optimized
-#   - Reasonably debuggable
+#
 ifeq ($(MODE),)
-CONFIG_CCFLAGS +=		\
-	$(BACKTRACES)		\
-	$(FTRACE)		\
-	-DSYSDEBUG		\
-	-Og
-TARGET_ARCH ?=			\
-	-msse3
+CONFIG_CCFLAGS += $(BACKTRACES) $(FTRACE) -O2
+CONFIG_CPPFLAGS += -DSYSDEBUG
+TARGET_ARCH ?= -msse3
+endif
+
+ifeq ($(MODE), aarch64)
+CONFIG_CCFLAGS += -O2 $(BACKTRACES) #$(FTRACE)
+CONFIG_CPPFLAGS += -DSYSDEBUG
+endif
+
+# Fast Build Mode
+#
+#   - `make MODE=fastbuild`
+#   - No debugging
+#   - Syscall tracing
+#   - Function tracing
+#   - Some optimizations
+#   - Limited Backtraces
+#   - Compiles 28% faster
+#
+ifeq ($(MODE), fastbuild)
+CONFIG_CCFLAGS += $(BACKTRACES) $(FTRACE) -O
+CONFIG_CPPFLAGS += -DSYSDEBUG -DDWARFLESS
+CONFIG_OFLAGS += -g0
+CONFIG_LDFLAGS += -S
+TARGET_ARCH ?= -msse3
 endif

 # Optimized Mode
@@ -31,19 +51,11 @@ endif
 #   - No memory corruption detection
 #   - assert() / CHECK_xx() may leak code into binary for debuggability
 #   - GCC 8+ hoists check fails into .text.cold, thus minimizing impact
+#
 ifeq ($(MODE), opt)
-CONFIG_CPPFLAGS +=		\
-	-DNDEBUG		\
-	-msse2avx		\
-	-Wa,-msse2avx
-CONFIG_CCFLAGS +=		\
-	$(BACKTRACES)		\
-	$(FTRACE)		\
-	-DSYSDEBUG		\
-	-O3			\
-	-fmerge-all-constants
-TARGET_ARCH ?=			\
-	-march=native
+CONFIG_CPPFLAGS += -DNDEBUG -DSYSDEBUG -msse2avx -Wa,-msse2avx
+CONFIG_CCFLAGS += $(BACKTRACES) $(FTRACE) -O3 -fmerge-all-constants
+TARGET_ARCH ?= -march=native
 endif

 # Optimized Linux Mode
@@ -54,18 +66,12 @@ endif
 #   - Turns off function tracing
 #   - Turns off support for older cpu models
 #   - Turns off support for other operating systems
+#
 ifeq ($(MODE), optlinux)
-CONFIG_CPPFLAGS +=		\
-	-DNDEBUG		\
-	-msse2avx		\
-	-Wa,-msse2avx		\
-	-DSUPPORT_VECTOR=1
-CONFIG_CCFLAGS +=		\
-	-O3 -fmerge-all-constants
-DEFAULT_COPTS +=		\
-	-mred-zone
-TARGET_ARCH ?=			\
-	-march=native
+CONFIG_CPPFLAGS += -DNDEBUG -msse2avx -Wa,-msse2avx -DSUPPORT_VECTOR=1
+CONFIG_CCFLAGS += -O3 -fmerge-all-constants
+DEFAULT_COPTS += -mred-zone
+TARGET_ARCH ?= -march=native
 endif

 # Release Mode
@@ -81,16 +87,12 @@ endif
 #   - DCHECK_xx() statements removed
 #   - No memory corruption detection
 #   - CHECK_xx() won't leak strings into binary
+#
 ifeq ($(MODE), rel)
-CONFIG_CPPFLAGS +=		\
-	-DNDEBUG
-CONFIG_CCFLAGS +=		\
-	$(BACKTRACES)		\
-	-O2
-TARGET_ARCH ?=			\
-	-msse3
-PYFLAGS +=			\
-	-O1
+CONFIG_CPPFLAGS += -DNDEBUG
+CONFIG_CCFLAGS += $(BACKTRACES) -O2
+TARGET_ARCH ?= -msse3
+PYFLAGS += -O1
 endif

 # Asan Mode
@@ -103,14 +105,11 @@ endif
 #   - Backtraces
 #   - Debuggability
 #   - Larger binaries
+#
 ifeq ($(MODE), asan)
-CONFIG_CCFLAGS +=		\
-	$(BACKTRACES)		\
-	-O2
-CONFIG_COPTS +=			\
-	-fsanitize=address
-TARGET_ARCH ?=			\
-	-msse3
+CONFIG_CCFLAGS += $(BACKTRACES) -O2 -DSYSDEBUG
+CONFIG_COPTS += -fsanitize=address
+TARGET_ARCH ?= -msse3
 endif

 # Debug Mode
@@ -122,22 +121,29 @@ endif
 #   - Stack canaries
 #   - No optimization (TODO)
 #   - Enormous binaries
+#
 ifeq ($(MODE), dbg)
-CONFIG_CPPFLAGS +=		\
-	-DMODE_DBG
-CONFIG_CCFLAGS +=		\
-	$(BACKTRACES)		\
-	$(FTRACE)		\
-	-DSYSDEBUG		\
-	-O2			\
-	-fno-inline
-CONFIG_COPTS +=			\
-	-fsanitize=address	\
-	-fsanitize=undefined
-TARGET_ARCH ?=			\
-	-msse3
-OVERRIDE_CCFLAGS +=		\
-	-fno-pie
+CONFIG_CPPFLAGS += -DMODE_DBG
+CONFIG_CCFLAGS += $(BACKTRACES) $(FTRACE) -DSYSDEBUG -O -fno-inline
+CONFIG_COPTS += -fsanitize=address -fsanitize=undefined
+TARGET_ARCH ?= -msse3
+OVERRIDE_CCFLAGS += -fno-pie
+endif
+
+# System Five Mode
+#
+#   - `make MODE=sysv`
+#   - Optimized
+#   - Backtraces
+#   - Debuggable
+#   - Syscall tracing
+#   - Function tracing
+#   - No Windows bloat!
+#
+ifeq ($(MODE), sysv)
+CONFIG_CCFLAGS += $(BACKTRACES) $(FTRACE) -O2
+CONFIG_CPPFLAGS += -DSYSDEBUG -DSUPPORT_VECTOR=121
+TARGET_ARCH ?= -msse3
 endif

 # Tiny Mode
@@ -151,6 +157,7 @@ endif
 #   - No backtraces
 #   - No algorithmics
 #   - YOLO
+#
 ifeq ($(MODE), tiny)
 CONFIG_CPPFLAGS +=			\
 	-DTINY				\
@@ -165,7 +172,12 @@ CONFIG_CCFLAGS +=			\
 	-fschedule-insns2		\
 	-fomit-frame-pointer		\
 	-momit-leaf-frame-pointer	\
-	-foptimize-sibling-calls
+	-foptimize-sibling-calls	\
+	-DDWARFLESS
+CONFIG_OFLAGS +=			\
+	-g0
+CONFIG_LDFLAGS +=			\
+	-S
 TARGET_ARCH ?=				\
 	-msse3
 PYFLAGS +=				\
@@ -173,6 +185,31 @@ PYFLAGS +=				\
 	-B
 endif

+ifeq ($(MODE), aarch64-tiny)
+CONFIG_CPPFLAGS +=			\
+	-DTINY				\
+	-DNDEBUG			\
+	-DTRUSTWORTHY
+CONFIG_CCFLAGS +=			\
+	-Os				\
+	-fno-align-functions		\
+	-fno-align-jumps		\
+	-fno-align-labels		\
+	-fno-align-loops		\
+	-fschedule-insns2		\
+	-fomit-frame-pointer		\
+	-momit-leaf-frame-pointer	\
+	-foptimize-sibling-calls	\
+	-DDWARFLESS
+CONFIG_OFLAGS +=			\
+	-g0
+CONFIG_LDFLAGS +=			\
+	-S
+PYFLAGS +=				\
+	-O2				\
+	-B
+endif
+
 # Linux-Only Tiny Mode
 #
 #   - `make MODE=tinylinux`
@@ -185,14 +222,20 @@ endif
 #   - No portability
 #   - No algorithmics
 #   - YOLO
+#
 ifeq ($(MODE), tinylinux)
 CONFIG_CPPFLAGS +=			\
 	-DTINY				\
 	-DNDEBUG			\
 	-DTRUSTWORTHY			\
-	-DSUPPORT_VECTOR=1
+	-DSUPPORT_VECTOR=1		\
+	-DDWARFLESS
 DEFAULT_COPTS +=			\
 	-mred-zone
+CONFIG_OFLAGS +=			\
+	-g0
+CONFIG_LDFLAGS +=			\
+	-S
 CONFIG_CCFLAGS +=			\
 	-Os				\
 	-fno-align-functions		\
@@ -216,14 +259,20 @@ endif
 #   - No backtraces
 #   - No algorithmics
 #   - YOLO
+#
 ifeq ($(MODE), tinylinuxbsd)
 CONFIG_CPPFLAGS +=		\
 	-DTINY			\
 	-DNDEBUG		\
 	-DTRUSTWORTHY		\
-	-DSUPPORT_VECTOR=113
+	-DSUPPORT_VECTOR=113	\
+	-DDWARFLESS
 DEFAULT_COPTS +=		\
 	-mred-zone
+CONFIG_OFLAGS +=		\
+	-g0
+CONFIG_LDFLAGS +=		\
+	-S
 CONFIG_CCFLAGS +=		\
 	-Os			\
 	-fno-align-functions	\
@@ -246,12 +295,14 @@ endif
 #   - No backtraces
 #   - No algorithmics
 #   - YOLO
+#
 ifeq ($(MODE), tinysysv)
 CONFIG_CPPFLAGS +=		\
 	-DTINY			\
 	-DNDEBUG		\
 	-DTRUSTWORTHY		\
-	-DSUPPORT_VECTOR=121
+	-DSUPPORT_VECTOR=121	\
+	-DDWARFLESS
 DEFAULT_COPTS +=		\
 	-mred-zone
 CONFIG_CCFLAGS +=		\
@@ -260,6 +311,10 @@ CONFIG_CCFLAGS +=		\
 	-fno-align-jumps	\
 	-fno-align-labels	\
 	-fno-align-loops
+CONFIG_OFLAGS +=		\
+	-g0
+CONFIG_LDFLAGS +=		\
+	-S
 TARGET_ARCH ?=			\
 	-msse3
 endif
@@ -276,22 +331,50 @@ endif
 #   - No backtraces
 #   - No algorithmics
 #   - YOLO
+#
 ifeq ($(MODE), tinynowin)
 CONFIG_CPPFLAGS +=		\
 	-DTINY			\
 	-DNDEBUG		\
 	-DTRUSTWORTHY		\
-	-DSUPPORT_VECTOR=251
+	-DSUPPORT_VECTOR=251	\
+	-DDWARFLESS
 CONFIG_CCFLAGS +=		\
 	-Os			\
 	-fno-align-functions	\
 	-fno-align-jumps	\
 	-fno-align-labels	\
 	-fno-align-loops
+CONFIG_OFLAGS +=		\
+	-g0
+CONFIG_LDFLAGS +=		\
+	-S
 TARGET_ARCH ?=			\
 	-msse3
 endif

+# GCC11 Mode
+# https://justine.lol/compilers/x86_64-linux-musl__x86_64-linux-musl__g++-11.2.0.tar.xz
+ifeq ($(MODE), gcc11)
+.UNVEIL += rx:/opt/gcc11
+CONFIG_CCFLAGS += $(BACKTRACES) $(FTRACE) -DSYSDEBUG -O2
+AS = /opt/gcc11/bin/x86_64-linux-musl-as
+CC = /opt/gcc11/bin/x86_64-linux-musl-gcc
+CXX = /opt/gcc11/bin/x86_64-linux-musl-g++
+CXXFILT = /opt/gcc11/bin/x86_64-linux-musl-c++filt
+LD = /opt/gcc11/bin/x86_64-linux-musl-ld.bfd
+NM = /opt/gcc11/bin/x86_64-linux-musl-nm
+GCC = /opt/gcc11/bin/x86_64-linux-musl-gcc
+STRIP = /opt/gcc11/bin/x86_64-linux-musl-strip
+OBJCOPY = /opt/gcc11/bin/x86_64-linux-musl-objcopy
+OBJDUMP = /opt/gcc11/bin/x86_64-linux-musl-objdump
+ADDR2LINE = /opt/gcc11/bin/x86_64-linux-musl-addr2line
+CONFIG_CCFLAGS += $(BACKTRACES) $(FTRACE) -O2 -Wno-stringop-overread
+CONFIG_CFLAGS += -Wno-old-style-definition
+CONFIG_CPPFLAGS += -DNDEBUG -DSYSDEBUG
+TARGET_ARCH ?= -msse3
+endif
+
 # LLVM Mode
 ifeq ($(MODE), llvm)
 TARGET_ARCH ?= -msse3
@@ -45,7 +45,6 @@
 LC_ALL = C
 SOURCE_DATE_EPOCH = 0

-TAGS ?= /usr/bin/ctags  # emacs source builds or something breaks it
 ARFLAGS = rcsD
 ZFLAGS ?=
 XARGS ?= xargs -P4 -rs8000
@@ -57,11 +56,13 @@ TMPDIR = o/tmp
 AR = build/bootstrap/ar.com
 CP = build/bootstrap/cp.com
 RM = build/bootstrap/rm.com -f
+GZIP = build/bootstrap/gzip.com
 ECHO = build/bootstrap/echo.com
 TOUCH = build/bootstrap/touch.com
 PKG = build/bootstrap/package.com
 MKDEPS = build/bootstrap/mkdeps.com
 ZIPOBJ = build/bootstrap/zipobj.com
+FIXUPOBJ = build/bootstrap/fixupobj.com
 MKDIR = build/bootstrap/mkdir.com -p
 COMPILE = build/bootstrap/compile.com -V9 -P4096 $(QUOTA)

@@ -72,32 +73,34 @@ IMAGE_BASE_VIRTUAL ?= 0x400000
 IGNORE := $(shell $(ECHO) -2 ♥cosmo)
 IGNORE := $(shell $(MKDIR) o/tmp)

-ifneq ("$(wildcard o/third_party/gcc/bin/x86_64-pc-linux-gnu-as.exe)","")
-AS = o/third_party/gcc/bin/x86_64-pc-linux-gnu-as.exe
-CC = o/third_party/gcc/bin/x86_64-pc-linux-gnu-gcc.exe
-CXX = o/third_party/gcc/bin/x86_64-pc-linux-gnu-g++.exe
-CXXFILT = o/third_party/gcc/bin/x86_64-pc-linux-gnu-c++filt.exe
-LD = o/third_party/gcc/bin/x86_64-pc-linux-gnu-ld.bfd.exe
-NM = o/third_party/gcc/bin/x86_64-pc-linux-gnu-nm.exe
-GCC = o/third_party/gcc/bin/x86_64-pc-linux-gnu-gcc.exe
-STRIP = o/third_party/gcc/bin/x86_64-pc-linux-gnu-strip.exe
-OBJCOPY = o/third_party/gcc/bin/x86_64-pc-linux-gnu-objcopy.exe
-OBJDUMP = o/third_party/gcc/bin/x86_64-pc-linux-gnu-objdump.exe
-ADDR2LINE = $(shell build/bootstrap/pwd.com)/o/third_party/gcc/bin/x86_64-pc-linux-gnu-addr2line.exe
+ifneq ($(findstring aarch64,$(MODE)),)
+ARCH = aarch64
+VM = o/third_party/qemu/qemu-aarch64
+else
+ARCH = x86_64
+endif
+
+ifneq ("$(wildcard o/third_party/gcc/bin/x86_64-pc-linux-gnu-*)","")
+PREFIX = o/third_party/gcc/bin/x86_64-pc-linux-gnu-
 else
 IGNORE := $(shell build/bootstrap/unbundle.com)
-AS = o/third_party/gcc/bin/x86_64-linux-musl-as
-CC = o/third_party/gcc/bin/x86_64-linux-musl-gcc
-CXX = o/third_party/gcc/bin/x86_64-linux-musl-g++
-CXXFILT = o/third_party/gcc/bin/x86_64-linux-musl-c++filt
-LD = o/third_party/gcc/bin/x86_64-linux-musl-ld.bfd
-NM = o/third_party/gcc/bin/x86_64-linux-musl-nm
-GCC = o/third_party/gcc/bin/x86_64-linux-musl-gcc
-STRIP = o/third_party/gcc/bin/x86_64-linux-musl-strip
-OBJCOPY = o/third_party/gcc/bin/x86_64-linux-musl-objcopy
-OBJDUMP = o/third_party/gcc/bin/x86_64-linux-musl-objdump
-ADDR2LINE = $(shell build/bootstrap/pwd.com)/o/third_party/gcc/bin/x86_64-linux-musl-addr2line
+PREFIX = o/third_party/gcc/bin/x86_64-linux-musl-
 endif
+ifeq ($(ARCH), aarch64)
+PREFIX = o/third_party/gcc/bin/aarch64-linux-musl-
+endif
+
+AS = $(PREFIX)as
+CC = $(PREFIX)gcc
+CXX = $(PREFIX)g++
+CXXFILT = $(PREFIX)c++filt
+LD = $(PREFIX)ld.bfd
+NM = $(PREFIX)nm
+GCC = $(PREFIX)gcc
+STRIP = $(PREFIX)strip
+OBJCOPY = $(PREFIX)objcopy
+OBJDUMP = $(PREFIX)objdump
+ADDR2LINE = $(PWD)/$(PREFIX)addr2line

 export ADDR2LINE
 export LC_ALL
@@ -106,20 +109,33 @@ export MODE
 export SOURCE_DATE_EPOCH
 export TMPDIR

+ifeq ($(LANDLOCKMAKE_VERSION),)
+TMPSAFE = $(TMPDIR)/$(subst /,_,$@).tmp
+else
+TMPSAFE = $(TMPDIR)/
+endif
+
+ifneq ($(ARCH), aarch64)
+MNO_FENTRY = -mno-fentry
+endif
+
 FTRACE =								\
 	-pg

 BACKTRACES =								\
-	-fno-schedule-insns2						\
 	-fno-omit-frame-pointer						\
 	-fno-optimize-sibling-calls					\
 	-mno-omit-leaf-frame-pointer

+ifneq ($(ARCH), aarch64)
+BACKTRACES += -fno-schedule-insns2
+endif
+
 SANITIZER =								\
 	-fsanitize=address

 NO_MAGIC =								\
-	-mno-fentry							\
+	$(MNO_FENTRY)							\
 	-fno-stack-protector						\
 	-fwrapv								\
 	-fno-sanitize=all
@@ -141,20 +157,24 @@ DEFAULT_CCFLAGS =							\

 DEFAULT_OFLAGS =							\
 	-g								\
+	-gdwarf-4							\
 	-gdescribe-dies

 DEFAULT_COPTS =								\
-	-mno-red-zone							\
 	-fno-math-errno							\
-	-fno-trapping-math						\
 	-fno-fp-int-builtin-inexact					\
 	-fno-ident							\
 	-fno-common							\
 	-fno-gnu-unique							\
 	-fstrict-aliasing						\
 	-fstrict-overflow						\
-	-fno-semantic-interposition					\
+	-fno-semantic-interposition
+
+ifneq ($(ARCH), aarch64)
+DEFAULT_COPTS +=							\
+	-mno-red-zone							\
 	-mno-tls-direct-seg-refs
+endif

 MATHEMATICAL =								\
 	-O3								\
@@ -165,7 +185,7 @@ DEFAULT_CPPFLAGS =							\
 	-DMODE='"$(MODE)"'						\
 	-DIMAGE_BASE_VIRTUAL=$(IMAGE_BASE_VIRTUAL)			\
 	-nostdinc							\
-	-iquote.
+	-iquote .

 DEFAULT_CFLAGS =							\
 	-std=gnu2x
@@ -176,7 +196,8 @@ DEFAULT_CXXFLAGS =							\
 	-fuse-cxa-atexit						\
 	-fno-threadsafe-statics						\
 	-Wno-int-in-bool-context					\
-	-Wno-narrowing
+	-Wno-narrowing							\
+	-Wno-literal-suffix

 DEFAULT_ASFLAGS =							\
 	-W								\
@@ -187,7 +208,6 @@ DEFAULT_ASFLAGS =							\
 DEFAULT_LDFLAGS =							\
 	-static								\
 	-nostdlib							\
-	-melf_x86_64							\
 	--gc-sections							\
 	--build-id=none							\
 	--no-dynamic-linker						\
@@ -231,7 +251,7 @@ cpp.flags =								\
 	$(CONFIG_CPPFLAGS)						\
 	$(CPPFLAGS)							\
 	$(OVERRIDE_CPPFLAGS)						\
-	-includelibc/integral/normalize.inc
+	-include libc/integral/normalize.inc

 copt.flags =								\
 	$(TARGET_ARCH)							\
@@ -324,7 +344,7 @@ OBJECTIFY.greg.c =							\
 	-fno-optimize-sibling-calls					\
 	-fno-sanitize=all						\
 	-ffreestanding							\
-	-mno-fentry							\
+	$(MNO_FENTRY)							\
 	-fwrapv								\
 	-c

@@ -364,7 +384,7 @@ OBJECTIFY.ncabi.c =							\
 	$(OBJECTIFY.c.flags)						\
 	-mno-sse							\
 	-mfpmath=387							\
-	-mno-fentry							\
+	$(MNO_FENTRY)							\
 	-fno-stack-protector						\
 	-fno-instrument-functions					\
 	-fno-optimize-sibling-calls					\
@@ -383,7 +403,7 @@ OBJECTIFY.ncabi.c =							\
 OBJECTIFY.initabi.c =							\
 	$(GCC)								\
 	$(OBJECTIFY.c.flags)						\
-	-mno-fentry							\
+	$(MNO_FENTRY)							\
 	-fno-stack-protector						\
 	-fno-instrument-functions					\
 	-fno-optimize-sibling-calls					\
@@ -12,5 +12,9 @@

 tail = $(wordlist 2,$(words $1),$1)
 reverse = $(if $1,$(call reverse,$(call tail,$1)) $(firstword $1))
-uniq = $(if $1,$(firstword $1) $(call uniq,$(filter-out $(firstword $1),$1)))
 uniqr = $(if $1,$(call uniqr,$(filter-out $(firstword $1),$1)) $(firstword $1))
+
+# polyfill uniq native (landlock make 1.4)
+ifneq ($(call uniq,c b c a),c b a)
+uniq = $(if $1,$(firstword $1) $(call uniq,$(filter-out $(firstword $1),$1)))
+endif
@@ -43,6 +43,9 @@
 #     '(progn
 #        (add-hook 'c-mode-common-hook 'jart-c-mode-common-hook)))

+TAGS="$1"
+shift
+
 # ctags doesn't understand atomics, e.g.
 #   extern char **environ;
 set -- --regex-c='/_Atomic(\([^)]*\))/\1/b' "$@"
@@ -63,7 +66,7 @@ set -- --regex-c='/^extern [^(]*(\*const \([^)]*\))(/\1/b' "$@"
 #   struct WorstSoftwareEver;
 set -- --regex-c='/^struct.*;$/uehocruehcroue/b' "$@"

-exec ${TAGS:-ctags}                                               \
+exec $TAGS                                                        \
     -e                                                           \
     --langmap=c:.c.h                                             \
     --exclude=libc/nt/struct/imagefileheader.internal.h          \
@@ -0,0 +1,7 @@
+#!/bin/sh
+for last; do true; done
+if printf '%s\n' "$last" | grep aarch64 >/dev/null 2>&1; then
+  exec o/third_party/gcc/bin/aarch64-linux-musl-objdump "$@"
+else
+  exec o/third_party/gcc/bin/x86_64-linux-musl-objdump "$@"
+fi
@@ -23,8 +23,9 @@
 #   - tool/build/runitd.c

 .PRECIOUS: o/$(MODE)/%.com.ok
+o/$(MODE)/%.com.ok: .PLEDGE = stdio rpath wpath cpath proc fattr inet
 o/$(MODE)/%.com.ok:				\
 		o/$(MODE)/tool/build/runit.com	\
 		o/$(MODE)/tool/build/runitd.com	\
 		o/$(MODE)/%.com
-	@$(COMPILE) -ATEST -tT$@ $^ $(HOSTS)
+	@$(COMPILE) -wATEST -tT$@ $^ $(HOSTS)
@@ -16,91 +16,223 @@

 MAKEFLAGS += --no-builtin-rules

-o/%.o: %.s                         ; @$(COMPILE) -AOBJECTIFY.s $(OBJECTIFY.s) $(OUTPUT_OPTION) $<
-o/%.o: o/%.s                       ; @$(COMPILE) -AOBJECTIFY.s $(OBJECTIFY.s) $(OUTPUT_OPTION) $<
-o/%.s: %.S                         ; @$(COMPILE) -APREPROCESS $(PREPROCESS) $(OUTPUT_OPTION) $<
-o/%.s: o/%.S                       ; @$(COMPILE) -APREPROCESS $(PREPROCESS) $(OUTPUT_OPTION) $<
-o/%.i: %.S                         ; @$(COMPILE) -APREPROCESS $(PREPROCESS) $(OUTPUT_OPTION) $<
-o/%.o: %.S                         ; @$(COMPILE) -AOBJECTIFY.S $(OBJECTIFY.S) $(OUTPUT_OPTION) $<
-o/%.o: o/%.S                       ; @$(COMPILE) -AOBJECTIFY.S $(OBJECTIFY.S) $(OUTPUT_OPTION) $<
-o/%.s: %.i                         ; @$(COMPILE) -ACOMPILE.i $(COMPILE.i) $(OUTPUT_OPTION) $<
-o/%.s: o/%.i                       ; @$(COMPILE) -ACOMPILE.i $(COMPILE.i) $(OUTPUT_OPTION) $<
-o/%.o: %.cc                        ; @$(COMPILE) -AOBJECTIFY.cxx $(OBJECTIFY.cxx) $(OUTPUT_OPTION) $<
-o/%.o: o/%.cc                      ; @$(COMPILE) -AOBJECTIFY.cxx $(OBJECTIFY.cxx) $(OUTPUT_OPTION) $<
 o/%.lds: %.lds                     ; @$(COMPILE) -APREPROCESS $(PREPROCESS.lds) $(OUTPUT_OPTION) $<
 o/%.inc: %.h                       ; @$(COMPILE) -APREPROCESS $(PREPROCESS) $(OUTPUT_OPTION) -D__ASSEMBLER__ -P $<
-o/%.h.ok: %.h                      ; @$(COMPILE) -ACHECK.h $(COMPILE.c) -xc -g0 -o $@ $<
-o/%.okk: %                         ; @$(COMPILE) -ACHECK.h $(COMPILE.cxx) -xc++ -g0 -o $@ $<
 o/%.greg.o: %.greg.c               ; @$(COMPILE) -AOBJECTIFY.greg $(OBJECTIFY.greg.c) $(OUTPUT_OPTION) $<
-o/%.zip.o: o/%                     ; @$(COMPILE) -AZIPOBJ $(ZIPOBJ) $(ZIPOBJ_FLAGS) $(OUTPUT_OPTION) $<
-
 o/$(MODE)/%: o/$(MODE)/%.dbg       ; @$(COMPILE) -AOBJCOPY -T$@ $(OBJCOPY) -S -O binary $< $@
-o/$(MODE)/%.o: %.s                 ; @$(COMPILE) -AOBJECTIFY.s $(OBJECTIFY.s) $(OUTPUT_OPTION) $<
-o/$(MODE)/%.o: o/$(MODE)/%.s       ; @$(COMPILE) -AOBJECTIFY.s $(OBJECTIFY.s) $(OUTPUT_OPTION) $<
-o/$(MODE)/%.s: %.S                 ; @$(COMPILE) -APREPROCESS $(PREPROCESS) $(OUTPUT_OPTION) $<
-o/$(MODE)/%.s: o/$(MODE)/%.S       ; @$(COMPILE) -APREPROCESS $(PREPROCESS) $(OUTPUT_OPTION) $<
-o/$(MODE)/%.o: %.c                 ; @$(COMPILE) -AOBJECTIFY.c $(OBJECTIFY.c) $(OUTPUT_OPTION) $<
 o/$(MODE)/%.o: %.f                 ; @$(COMPILE) -AOBJECTIFY.f $(OBJECTIFY.f) $(OUTPUT_OPTION) $<
 o/$(MODE)/%.o: %.F                 ; @$(COMPILE) -AOBJECTIFY.F $(OBJECTIFY.F) $(OUTPUT_OPTION) $<
-o/$(MODE)/%.o: o/$(MODE)/%.c       ; @$(COMPILE) -AOBJECTIFY.c $(OBJECTIFY.c) $(OUTPUT_OPTION) $<
-o/$(MODE)/%.ss: %.c                ; @$(COMPILE) -ACOMPILE.c $(COMPILE.c) $(OUTPUT_OPTION) $<
-o/$(MODE)/%.ss: o/$(MODE)/%.c      ; @$(COMPILE) -AOBJECTIFY.s $(COMPILE.c) $(OUTPUT_OPTION) $<
-o/$(MODE)/%.i: %.S                 ; @$(COMPILE) -APREPROCESS $(PREPROCESS) $(OUTPUT_OPTION) $<
-o/$(MODE)/%.i: %.c                 ; @$(COMPILE) -APREPROCESS $(PREPROCESS) $(OUTPUT_OPTION) $<
-o/$(MODE)/%.i: %.cc                ; @$(COMPILE) -APREPROCESS $(PREPROCESS) $(OUTPUT_OPTION) $<
-o/$(MODE)/%.i: o/$(MODE)/%.c       ; @$(COMPILE) -APREPROCESS $(PREPROCESS) $(OUTPUT_OPTION) $<
 o/$(MODE)/%.h: %.c                 ; @$(COMPILE) -AAMALGAMATE $(PREPROCESS) $(OUTPUT_OPTION) -fdirectives-only -P $<
 o/$(MODE)/%.h: o/$(MODE)/%.c       ; @$(COMPILE) -AAMALGAMATE $(PREPROCESS) $(OUTPUT_OPTION) -fdirectives-only -P $<
-o/$(MODE)/%.o: %.S                 ; @$(COMPILE) -AOBJECTIFY.S $(OBJECTIFY.S) $(OUTPUT_OPTION) $<
-o/$(MODE)/%.o: o/$(MODE)/%.S       ; @$(COMPILE) -AOBJECTIFY.S $(OBJECTIFY.S) $(OUTPUT_OPTION) $<
-o/$(MODE)/%.s: %.i                 ; @$(COMPILE) -ACOMPILE.i $(COMPILE.i) $(OUTPUT_OPTION) $<
-o/$(MODE)/%.s: o/$(MODE)/%.i       ; @$(COMPILE) -ACOMPILE.i $(COMPILE.i) $(OUTPUT_OPTION) $<
 o/$(MODE)/%.o: %.cc                ; @$(COMPILE) -AOBJECTIFY.cxx $(OBJECTIFY.cxx) $(OUTPUT_OPTION) $<
-o/$(MODE)/%.o: o/$(MODE)/%.cc      ; @$(COMPILE) -AOBJECTIFY.cxx $(OBJECTIFY.cxx) $(OUTPUT_OPTION) $<
 o/$(MODE)/%.lds: %.lds             ; @$(COMPILE) -APREPROCESS $(PREPROCESS.lds) $(OUTPUT_OPTION) $<
-o/$(MODE)/%.h.ok: %.h              ; @$(COMPILE) -ACHECK.h $(COMPILE.c) -xc -g0 -o $@ $<
-o/$(MODE)/%.hh.ok: %.hh            ; @$(COMPILE) -ACHECK.h $(COMPILE.cxx) -xc++ -g0 -o $@ $<
-o/$(MODE)/%.okk: %                 ; @$(COMPILE) -ACHECK.h $(COMPILE.cxx) -xc++ -g0 -o $@ $<
-o/$(MODE)/%.cxx.o: %.c             ; @$(COMPILE) -AOBJECTIFY.cxx $(OBJECTIFY.cxx) -xc++ $(OUTPUT_OPTION) $<
 o/$(MODE)/%.o: %.greg.c            ; @$(COMPILE) -AOBJECTIFY.greg $(OBJECTIFY.greg.c) $(OUTPUT_OPTION) $<
 o/$(MODE)/%.greg.o: %.greg.c       ; @$(COMPILE) -AOBJECTIFY.greg $(OBJECTIFY.greg.c) $(OUTPUT_OPTION) $<
-o/$(MODE)/%.ansi.o: %.ansi.c       ; @$(COMPILE) -AOBJECTIFY.ansi $(OBJECTIFY.ansi.c) $(OUTPUT_OPTION) $<
-o/$(MODE)/%.ansi.o: %.c            ; @$(COMPILE) -AOBJECTIFY.ansi $(OBJECTIFY.ansi.c) $(OUTPUT_OPTION) $<
-o/$(MODE)/%.c99.o: %.c99.c         ; @$(COMPILE) -AOBJECTIFY.c99 $(OBJECTIFY.c99.c) $(OUTPUT_OPTION) $<
-o/$(MODE)/%.c11.o: %.c11.c         ; @$(COMPILE) -AOBJECTIFY.c11 $(OBJECTIFY.c11.c) $(OUTPUT_OPTION) $<
-o/$(MODE)/%.c2x.o: %.c2x.c         ; @$(COMPILE) -AOBJECTIFY.c2x $(OBJECTIFY.c2x.c) $(OUTPUT_OPTION) $<
 o/$(MODE)/%.initabi.o: %.initabi.c ; @$(COMPILE) -AOBJECTIFY.init $(OBJECTIFY.initabi.c) $(OUTPUT_OPTION) $<
 o/$(MODE)/%.ncabi.o: %.ncabi.c     ; @$(COMPILE) -AOBJECTIFY.nc $(OBJECTIFY.ncabi.c) $(OUTPUT_OPTION) $<
 o/$(MODE)/%.real.o: %.c            ; @$(COMPILE) -AOBJECTIFY.real $(OBJECTIFY.real.c) $(OUTPUT_OPTION) $<

-o/$(MODE)/%.runs: o/$(MODE)/%      ; @$(COMPILE) -ACHECK -tT$@ $< $(TESTARGS)
-o/$(MODE)/%.zip.o: %               ; @$(COMPILE) -AZIPOBJ $(ZIPOBJ) $(ZIPOBJ_FLAGS) $(OUTPUT_OPTION) $<
-o/$(MODE)/%-gcc.asm: %.c           ; @$(COMPILE) -AOBJECTIFY.c $(OBJECTIFY.c) -S -g0 $(OUTPUT_OPTION) $<
-o/$(MODE)/%-gcc.asm: %.cc          ; @$(COMPILE) -AOBJECTIFY.c $(OBJECTIFY.cxx) -S -g0 $(OUTPUT_OPTION) $<
-o/$(MODE)/%-clang.asm: %.c         ; @$(COMPILE) -AOBJECTIFY.c $(OBJECTIFY.c) -S -g0 $(OUTPUT_OPTION) $<
-o/$(MODE)/%-clang.asm: %.cc        ; @$(COMPILE) -AOBJECTIFY.c $(OBJECTIFY.cxx) -S -g0 $(OUTPUT_OPTION) $<
-o/$(MODE)/%-clang.asm: CC = $(CLANG)
+ifneq ($(ARCH), aarch64)
+o/%.o: %.s                         ; @$(COMPILE) -AOBJECTIFY.s $(OBJECTIFY.s) $(OUTPUT_OPTION) $<
+o/%.o: o/%.s                       ; @$(COMPILE) -AOBJECTIFY.s $(OBJECTIFY.s) $(OUTPUT_OPTION) $<
+o/%.s: %.S                         ; @$(COMPILE) -APREPROCESS $(PREPROCESS) $(OUTPUT_OPTION) $<
+o/%.s: o/%.S                       ; @$(COMPILE) -APREPROCESS $(PREPROCESS) $(OUTPUT_OPTION) $<
+o/%.o: %.S                         ; @$(COMPILE) -AOBJECTIFY.S $(OBJECTIFY.S) $(OUTPUT_OPTION) $<
+o/%.o: o/%.S                       ; @$(COMPILE) -AOBJECTIFY.S $(OBJECTIFY.S) $(OUTPUT_OPTION) $<
+o/$(MODE)/%.o: %.s                 ; @$(COMPILE) -AOBJECTIFY.s $(OBJECTIFY.s) $(OUTPUT_OPTION) $<
+o/$(MODE)/%.o: o/$(MODE)/%.s       ; @$(COMPILE) -AOBJECTIFY.s $(OBJECTIFY.s) $(OUTPUT_OPTION) $<
+o/$(MODE)/%.o: %.S                 ; @$(COMPILE) -AOBJECTIFY.S $(OBJECTIFY.S) $(OUTPUT_OPTION) $<
+o/$(MODE)/%.o: o/$(MODE)/%.S       ; @$(COMPILE) -AOBJECTIFY.S $(OBJECTIFY.S) $(OUTPUT_OPTION) $<
+else
+o/%.o: %.s libc/empty.s                   ; @$(COMPILE) -AOBJECTIFY.s $(OBJECTIFY.s) $(OUTPUT_OPTION) libc/empty.s
+o/%.o: o/%.s libc/empty.s                 ; @$(COMPILE) -AOBJECTIFY.s $(OBJECTIFY.s) $(OUTPUT_OPTION) libc/empty.s
+o/%.s: %.S libc/empty.s                   ; @$(COMPILE) -APREPROCESS $(PREPROCESS) $(OUTPUT_OPTION) libc/empty.s
+o/%.s: o/%.S libc/empty.s                 ; @$(COMPILE) -APREPROCESS $(PREPROCESS) $(OUTPUT_OPTION) libc/empty.s
+o/%.o: %.S libc/empty.s                   ; @$(COMPILE) -AOBJECTIFY.S $(OBJECTIFY.s) $(OUTPUT_OPTION) libc/empty.s
+o/%.o: o/%.S libc/empty.s                 ; @$(COMPILE) -AOBJECTIFY.S $(OBJECTIFY.s) $(OUTPUT_OPTION) libc/empty.s
+o/$(MODE)/%.o: %.s libc/empty.s           ; @$(COMPILE) -AOBJECTIFY.s $(OBJECTIFY.s) $(OUTPUT_OPTION) libc/empty.s
+o/$(MODE)/%.o: o/$(MODE)/%.s libc/empty.s ; @$(COMPILE) -AOBJECTIFY.s $(OBJECTIFY.s) $(OUTPUT_OPTION) libc/empty.s
+o/$(MODE)/%.o: %.S libc/empty.s           ; @$(COMPILE) -AOBJECTIFY.S $(OBJECTIFY.s) $(OUTPUT_OPTION) libc/empty.s
+o/$(MODE)/%.o: o/$(MODE)/%.S libc/empty.s ; @$(COMPILE) -AOBJECTIFY.S $(OBJECTIFY.s) $(OUTPUT_OPTION) libc/empty.s
+endif
+
+o/%.o: %.cc
+	@$(COMPILE) -AOBJECTIFY.cxx $(OBJECTIFY.cxx) $(OUTPUT_OPTION) $<
+	@$(COMPILE) -AFIXUPOBJ -wT$@ $(FIXUPOBJ) $@
+
+o/%.o: o/%.cc
+	@$(COMPILE) -AOBJECTIFY.cxx $(OBJECTIFY.cxx) $(OUTPUT_OPTION) $<
+	@$(COMPILE) -AFIXUPOBJ -wT$@ $(FIXUPOBJ) $@
+
+o/$(MODE)/%.o: %.c
+	@$(COMPILE) -AOBJECTIFY.c $(OBJECTIFY.c) $(OUTPUT_OPTION) $<
+	@$(COMPILE) -AFIXUPOBJ -wT$@ $(FIXUPOBJ) $@
+
+o/$(MODE)/%.o: o/$(MODE)/%.c
+	@$(COMPILE) -AOBJECTIFY.c $(OBJECTIFY.c) $(OUTPUT_OPTION) $<
+	@$(COMPILE) -AFIXUPOBJ -wT$@ $(FIXUPOBJ) $@
+
+o/$(MODE)/%.o: o/$(MODE)/%.cc
+	@$(COMPILE) -AOBJECTIFY.cxx $(OBJECTIFY.cxx) $(OUTPUT_OPTION) $<
+	@$(COMPILE) -AFIXUPOBJ -wT$@ $(FIXUPOBJ) $@

 o/%.a:
-	$(file >$@.args,$^)
-	@$(COMPILE) -AARCHIVE -T$@ $(AR) $(ARFLAGS) $@ @$@.args
+	$(file >$(TMPDIR)/$(subst /,_,$@),$^)
+	@$(COMPILE) -AARCHIVE -wT$@ $(AR) $(ARFLAGS) $@ @$(TMPDIR)/$(subst /,_,$@)

 o/%.pkg:
-	$(file >$@.args,$(filter %.o,$^))
-	@$(COMPILE) -APACKAGE -T$@ $(PKG) $(OUTPUT_OPTION) $(addprefix -d,$(filter %.pkg,$^)) @$@.args
+	$(file >$(TMPSAFE).args,$(filter %.o,$^))
+	@$(COMPILE) -APACKAGE -wT$@ $(PKG) $(OUTPUT_OPTION) $(addprefix -d,$(filter %.pkg,$^)) @$(TMPSAFE).args

 o/$(MODE)/%.pkg:
-	$(file >$@.args,$(filter %.o,$^))
-	@$(COMPILE) -APACKAGE -T$@ $(PKG) $(OUTPUT_OPTION) $(addprefix -d,$(filter %.pkg,$^)) @$@.args
+	$(file >$(TMPSAFE).args,$(filter %.o,$^))
+	@$(COMPILE) -APACKAGE -wT$@ $(PKG) $(OUTPUT_OPTION) $(addprefix -d,$(filter %.pkg,$^)) @$(TMPSAFE).args

 o/$(MODE)/%.o: %.py o/$(MODE)/third_party/python/pyobj.com
-	@$(COMPILE) -APYOBJ o/$(MODE)/third_party/python/pyobj.com $(PYFLAGS) -o $@ $<
+	@$(COMPILE) -wAPYOBJ o/$(MODE)/third_party/python/pyobj.com $(PYFLAGS) -o $@ $<

 o/$(MODE)/%.pyc: %.py o/$(MODE)/third_party/python/pycomp.com
-	@$(COMPILE) -APYCOMP o/$(MODE)/third_party/python/pycomp.com $(PYCFLAGS) -o $@ $<
+	@$(COMPILE) -wAPYCOMP o/$(MODE)/third_party/python/pycomp.com $(PYCFLAGS) -o $@ $<

 o/$(MODE)/%.lua: %.lua o/$(MODE)/third_party/lua/luac.com
-	@$(COMPILE) -ALUAC o/$(MODE)/third_party/lua/luac.com -s -o $@ $<
+	@$(COMPILE) -wALUAC o/$(MODE)/third_party/lua/luac.com -s -o $@ $<

 o/$(MODE)/%.lua.runs: %.lua o/$(MODE)/tool/net/redbean.com
-	@$(COMPILE) -ALUA -tT$@ o/$(MODE)/tool/net/redbean.com $(LUAFLAGS) -i $<
+	@$(COMPILE) -wALUA -tT$@ o/$(MODE)/tool/net/redbean.com $(LUAFLAGS) -i $<
+
+o/$(MODE)/%: o/$(MODE)/%.com o/$(MODE)/tool/build/cp.com o/$(MODE)/tool/build/assimilate.com
+	@$(COMPILE) -wACP -T$@ o/$(MODE)/tool/build/cp.com $< $@
+	@$(COMPILE) -wAASSIMILATE -T$@ o/$(MODE)/tool/build/assimilate.com $@
+
+################################################################################
+# LOCAL UNIT TESTS
+#
+# We always run unit tests as part of the normal `make` invocation. you
+# may override the $(TESTARGS) variable to do things such as enable the
+# benchmarking feature. For example:
+#
+#     TESTARGS = -b
+#
+# May be specified in your ~/.cosmo.mk file. You can also use this to
+# enable things like function tracing. For example:
+#
+#     TESTARGS = ----ftrace
+#     .PLEDGE += prot_exec
+#
+# You could then run a command like:
+#
+#     make -j8 o//test/libc/calls/openbsd_test.com.runs
+#
+# You need PROT_EXEC permission since ftrace morphs the binary. It's
+# also worth mentioning that the pledge.com command can simulate what
+# Landlock Make does:
+#
+#     o//tool/build/pledge.com \
+#       -v. -p 'stdio rpath wpath cpath tty prot_exec' \
+#       o//test/libc/calls/openbsd_test.com \
+#       ----ftrace
+#
+# This is useful in the event a test binary should run by itself, but
+# fails to run beneath Landlock Make. It's also useful sometimes to
+# override the verbosity when running tests:
+#
+#     make V=5 TESTARGS=-b o//test/libc/calls/openbsd_test.com.runs
+#
+# This way, if for some reason a test should fail but calls exit(0),
+# then the stdout/stderr output, which would normally be suppressed,
+# will actually be displayed.
+
+o/$(MODE)/%.runs: o/$(MODE)/% $(VM)
+	@$(COMPILE) -ACHECK -wtT$@ $(VM) $< $(TESTARGS)
+
+################################################################################
+# ELF ZIP FILES
+#
+# zipobj.com lets us do fast incremental linking of compressed data.
+# it's nice because if we link a hundred binaries that use the time zone
+# database, then that database only needs to be DEFLATE'd once.
+
+o/%.zip.o: o/%
+	@$(COMPILE) -wAZIPOBJ $(ZIPOBJ) $(ZIPOBJ_FLAGS) $(OUTPUT_OPTION) $<
+
+o/$(MODE)/%.zip.o: %
+	@$(COMPILE) -wAZIPOBJ $(ZIPOBJ) $(ZIPOBJ_FLAGS) $(OUTPUT_OPTION) $<
+
+o/$(MODE)/%.zip.o: %
+	@$(COMPILE) -wAZIPOBJ $(ZIPOBJ) $(ZIPOBJ_FLAGS) $(OUTPUT_OPTION) $<
+
+# an issue with sandboxing arises when creating directory entries in the
+# zip file. we need the trailing slash (e.g. o//foo/.zip.o) but Landlock
+# Make avoids sandboxing directory names that have a trailing slash (so
+# they can be used to watch for deleted files, without creating overly
+# broad unveiling). such rules need to be written more explicitly.
+o/$(MODE)%/.zip.o: %
+	@$(COMPILE) -wAZIPOBJ $(ZIPOBJ) $(ZIPOBJ_FLAGS) $(OUTPUT_OPTION) $<
+
+################################################################################
+# STRICT HEADER CHECKING
+#
+# these rules are unsandboxed since they're already a sandboxing test,
+# and it would be too costly in terms of make latency to have every
+# header file depend on $(HDRS) and $(INCS).
+
+o/%.h.ok: private .UNSANDBOXED = 1
+o/%.h.ok: %.h
+	@$(COMPILE) -ACHECK.h $(COMPILE.c) -xc -g0 -o $@ $<
+
+o/$(MODE)/%.h.ok: private .UNSANDBOXED = 1
+o/$(MODE)/%.h.ok: %.h
+	@$(COMPILE) -ACHECK.h $(COMPILE.c) -xc -g0 -o $@ $<
+
+o/$(MODE)/%.hh.ok: private .UNSANDBOXED = 1
+o/$(MODE)/%.hh.ok: %.hh
+	@$(COMPILE) -ACHECK.h $(COMPILE.cxx) -xc++ -g0 -o $@ $<
+
+o/%.okk: .UNSANDBOXED = 1
+o/%.okk: %
+	@$(COMPILE) -ACHECK.h $(COMPILE.cxx) -xc++ -g0 -o $@ $<
+
+o/$(MODE)/%.okk: private .UNSANDBOXED = 1
+o/$(MODE)/%.okk: %
+	@$(COMPILE) -ACHECK.h $(COMPILE.cxx) -xc++ -g0 -o $@ $<
+
+################################################################################
+# EXECUTABLE HELPERS
+
+MAKE_OBJCOPY =					\
+	$(COMPILE) -AOBJCOPY -T$@		\
+	$(OBJCOPY) -S -O binary $< $@
+
+MAKE_SYMTAB_CREATE =				\
+	$(COMPILE) -wASYMTAB			\
+	o/$(MODE)/tool/build/symtab.com		\
+	-o $(TMPSAFE)/.symtab			\
+	$<
+
+MAKE_SYMTAB_ZIP =				\
+	$(COMPILE) -AZIP -T$@			\
+	o/$(MODE)/third_party/zip/zip.com	\
+	-b$(TMPDIR)				\
+	-9qj					\
+	$@					\
+	$(TMPSAFE)/.symtab
+
+################################################################################
+# EMACS ASSEMBLY GENERATION
+
+o/$(MODE)/%-gcc.asm: private .UNSANDBOXED = 1
+o/$(MODE)/%-gcc.asm: %.c
+	@$(COMPILE) -AOBJECTIFY.c $(OBJECTIFY.c) -S -g0 $(OUTPUT_OPTION) $<
+
+o/$(MODE)/%-gcc.asm: private .UNSANDBOXED = 1
+o/$(MODE)/%-gcc.asm: %.cc
+	@$(COMPILE) -AOBJECTIFY.c $(OBJECTIFY.cxx) -S -g0 $(OUTPUT_OPTION) $<
+
+o/$(MODE)/%-clang.asm: private .UNSANDBOXED = 1
+o/$(MODE)/%-clang.asm: %.c
+	@$(COMPILE) -AOBJECTIFY.c $(OBJECTIFY.c) -S -g0 $(OUTPUT_OPTION) $<
+
+o/$(MODE)/%-clang.asm: CC = $(CLANG)
+o/$(MODE)/%-clang.asm: private .UNSANDBOXED = 1
+o/$(MODE)/%-clang.asm: %.cc
+	@$(COMPILE) -AOBJECTIFY.c $(OBJECTIFY.cxx) -S -g0 $(OUTPUT_OPTION) $<
@@ -0,0 +1,10 @@
+#!/bin/sh
+for last; do true; done
+if printf '%s\n' "$last" | grep aarch64 >/dev/null 2>&1; then
+  if [ ! -f o/third_party/qemu/qemu-aarch64 ]; then
+    make -j8 o/third_party/qemu/qemu-aarch64
+  fi
+  exec o/third_party/qemu/qemu-aarch64 "$@"
+else
+  exec "$@"
+fi
@@ -31,8 +31,7 @@ if [ ! -f /proc/sys/fs/binfmt_misc/status ]; then
  exit 0
 fi

-STATUS="$(build/bootstrap/apetest.com)"
-if [ x"$STATUS" != xsuccess ]; then
+if ! build/bootstrap/echo.com -n; then
  cat <<'EOF' >&2

 ERROR
@@ -40,11 +40,11 @@ $(DSP_BMP_A).pkg:				\
 		$(foreach x,$(DSP_BMP_A_DIRECTDEPS),$($(x)_A).pkg)

 o/$(MODE)/dsp/bmp/float2short.o			\
-o/$(MODE)/dsp/bmp/scalevolume.o:		\
+o/$(MODE)/dsp/bmp/scalevolume.o: private	\
 		OVERRIDE_CFLAGS +=		\
 			$(MATHEMATICAL)

-o/tiny/dsp/bmp/scalevolume.o:			\
+o/tiny/dsp/bmp/scalevolume.o: private		\
 		OVERRIDE_CFLAGS +=		\
 			-Os

@@ -47,15 +47,15 @@ o/$(MODE)/dsp/core/c1331.o			\
 o/$(MODE)/dsp/core/magikarp.o			\
 o/$(MODE)/dsp/core/c93654369.o			\
 o/$(MODE)/dsp/core/float2short.o		\
-o/$(MODE)/dsp/core/scalevolume.o:		\
+o/$(MODE)/dsp/core/scalevolume.o: private	\
 		OVERRIDE_CFLAGS +=		\
 			$(MATHEMATICAL)

-o/tiny/dsp/core/scalevolume.o:			\
+o/tiny/dsp/core/scalevolume.o: private		\
 		OVERRIDE_CFLAGS +=		\
 			-Os

-o/$(MODE)/dsp/core/det3.o:			\
+o/$(MODE)/dsp/core/det3.o: private		\
 		OVERRIDE_CFLAGS +=		\
 			-ffast-math

@@ -17,7 +17,7 @@
 │ PERFORMANCE OF THIS SOFTWARE.                                                │
 ╚─────────────────────────────────────────────────────────────────────────────*/
 #include "libc/macros.internal.h"
-.align	16
+.balign	16

 //	Mixes audio.
 //
@@ -17,8 +17,8 @@
 │ PERFORMANCE OF THIS SOFTWARE.                                                │
 ╚─────────────────────────────────────────────────────────────────────────────*/
 #include "dsp/core/core.h"
-#include "libc/bits/bits.h"
-#include "libc/bits/safemacros.internal.h"
+#include "libc/intrin/bits.h"
+#include "libc/intrin/safemacros.internal.h"
 #include "libc/limits.h"

 /**
@@ -17,7 +17,7 @@
 │ PERFORMANCE OF THIS SOFTWARE.                                                │
 ╚─────────────────────────────────────────────────────────────────────────────*/
 #include "dsp/core/core.h"
-#include "libc/bits/likely.h"
+#include "libc/intrin/likely.h"

 /**
 * Expands μ-Law coded to audio sample.
@@ -37,10 +37,10 @@ This library provides several interfaces to load, demux and decode MPEG video
 and audio data. A high-level API combines the demuxer, video & audio decoders
 in an easy to use wrapper.

-Lower-level APIs for accessing the demuxer, video decoder and audio decoder, 
+Lower-level APIs for accessing the demuxer, video decoder and audio decoder,
 as well as providing different data sources are also available.

-Interfaces are written in an object orientet style, meaning you create object 
+Interfaces are written in an object orientet style, meaning you create object
 instances via various different constructor functions (plm_*create()),
 do some work on them and later dispose them via plm_*destroy().

@@ -51,11 +51,11 @@ plm_video_*  -- the MPEG1 Video ("mpeg1") decoder
 plm_audio_*  -- the MPEG1 Audio Layer II ("mp2") decoder


-This library uses malloc(), realloc() and free() to manage memory. Typically 
+This library uses malloc(), realloc() and free() to manage memory. Typically
 all allocation happens up-front when creating the interface. However, the
 default buffer size may be too small for certain inputs. In these cases plmpeg
 will realloc() the buffer with a larger size whenever needed. You can configure
-the default buffer size by defining PLM_BUFFER_DEFAULT_SIZE *before* 
+the default buffer size by defining PLM_BUFFER_DEFAULT_SIZE *before*
 including this library.

 With the high-level interface you have two options to decode video & audio:
@@ -85,7 +85,7 @@ mat4 rec601 = mat4(
 gl_FragColor = vec4(y, cb, cr, 1.0) * rec601;

 Audio data is decoded into a struct with either one single float array with the
-samples for the left and right channel interleaved, or if the 
+samples for the left and right channel interleaved, or if the
 PLM_AUDIO_SEPARATE_CHANNELS is defined *before* including this library, into
 two separate float arrays - one for each channel.

@@ -67,7 +67,7 @@ plm_buffer_t *plm_buffer_create_with_file(FILE *fh, int close_when_done) {

 plm_buffer_t *plm_buffer_create_with_memory(unsigned char *bytes, size_t length, int free_when_done) {
 	plm_buffer_t *b;
-	b = memalign(alignof(plm_buffer_t), sizeof(plm_buffer_t));
+	b = memalign(_Alignof(plm_buffer_t), sizeof(plm_buffer_t));
 	memset(b, 0, sizeof(plm_buffer_t));
 	b->capacity = length;
 	b->length = length;
@@ -79,7 +79,7 @@ plm_buffer_t *plm_buffer_create_with_memory(unsigned char *bytes, size_t length,

 plm_buffer_t * plm_buffer_create_with_capacity(size_t capacity) {
 	plm_buffer_t *b;
-	b = memalign(alignof(plm_buffer_t), sizeof(plm_buffer_t));
+	b = memalign(_Alignof(plm_buffer_t), sizeof(plm_buffer_t));
 	memset(b, 0, sizeof(plm_buffer_t));
 	b->capacity = capacity;
 	b->free_when_done = true;
@@ -115,8 +115,8 @@ plm_packet_t *plm_demux_decode(plm_demux_t *self) {
 	do {
 		code = plm_buffer_next_start_code(self->buffer);
 		if (
-			code == PLM_DEMUX_PACKET_VIDEO_1 || 
-			code == PLM_DEMUX_PACKET_PRIVATE || 
+			code == PLM_DEMUX_PACKET_VIDEO_1 ||
+			code == PLM_DEMUX_PACKET_PRIVATE ||
 			(code >= PLM_DEMUX_PACKET_AUDIO_1 && code <= PLM_DEMUX_PACKET_AUDIO_4)
 		) {
 			return plm_demux_decode_packet(self, code);
@@ -273,11 +273,11 @@ typedef plm_audio_t plm_audio_t;
 int plm_audio_decode_header(plm_audio_t *self);
 void plm_audio_decode_frame(plm_audio_t *self);
 const plm_quantizer_spec_t *plm_audio_read_allocation(plm_audio_t *self, int sb, int tab3);
-void plm_audio_read_samples(plm_audio_t *self, int ch, int sb, int part); 
+void plm_audio_read_samples(plm_audio_t *self, int ch, int sb, int part);
 void plm_audio_matrix_transform(int s[32][3], int ss, float *d, int dp);

 plm_audio_t *plm_audio_create_with_buffer(plm_buffer_t *buffer, int destroy_when_done) {
-	plm_audio_t *self = (plm_audio_t *)memalign(alignof(plm_audio_t), sizeof(plm_audio_t));
+	plm_audio_t *self = (plm_audio_t *)memalign(_Alignof(plm_audio_t), sizeof(plm_audio_t));
 	memset(self, 0, sizeof(plm_audio_t));

 	self->samples.count = PLM_AUDIO_SAMPLES_PER_FRAME;
@@ -345,7 +345,7 @@ plm_samples_t *plm_audio_decode(plm_audio_t *self) {
 	self->samples.time = self->time;

 	self->samples_decoded += PLM_AUDIO_SAMPLES_PER_FRAME;
-	self->time = (double)self->samples_decoded / 
+	self->time = (double)self->samples_decoded /
 		(double)PLM_AUDIO_SAMPLE_RATE[self->samplerate_index];

 	return &self->samples;
@@ -402,7 +402,7 @@ int plm_audio_decode_header(plm_audio_t *self) {
 		plm_buffer_skip(self->buffer, 16);
 	}

-	// Compute frame size, check if we have enough data to decode the whole 
+	// Compute frame size, check if we have enough data to decode the whole
 	// frame.
 	int bitrate = PLM_AUDIO_BIT_RATE[self->bitrate_index];
 	int samplerate = PLM_AUDIO_SAMPLE_RATE[self->samplerate_index];
@@ -561,7 +561,7 @@ void plm_audio_decode_frame(plm_audio_t *self) {
 						}
 					#else
 						for (int j = 0; j < 32; j++) {
-							self->samples.interleaved[((out_pos + j) << 1) + ch] = 
+							self->samples.interleaved[((out_pos + j) << 1) + ch] =
 								self->U[j] / 2147418112.0f;
 						}
 					#endif
@@ -37,8 +37,7 @@ DSP_MPEG_A_DIRECTDEPS =				\
 	LIBC_STUBS				\
 	LIBC_SYSV				\
 	LIBC_TIME				\
-	LIBC_TINYMATH				\
-	LIBC_UNICODE
+	LIBC_TINYMATH

 DSP_MPEG_A_DEPS :=				\
 	$(call uniq,$(foreach x,$(DSP_MPEG_A_DIRECTDEPS),$($(x))))
@@ -51,13 +50,10 @@ $(DSP_MPEG_A).pkg:				\
 		$(DSP_MPEG_A_OBJS)		\
 		$(foreach x,$(DSP_MPEG_A_DIRECTDEPS),$($(x)_A).pkg)

-o/$(MODE)/dsp/mpeg/clamp4int256-k8.o:		\
+o/$(MODE)/dsp/mpeg/clamp4int256-k8.o: private	\
 		OVERRIDE_CFLAGS +=		\
 			-Os

-#o/$(MODE)/dsp/mpeg/macroblock.o:		\
-		CC = $(CLANG)
-
 DSP_MPEG_LIBS = $(foreach x,$(DSP_MPEG_ARTIFACTS),$($(x)))
 DSP_MPEG_SRCS = $(foreach x,$(DSP_MPEG_ARTIFACTS),$($(x)_SRCS))
 DSP_MPEG_HDRS = $(foreach x,$(DSP_MPEG_ARTIFACTS),$($(x)_HDRS))
@@ -32,7 +32,6 @@
 #include "dsp/mpeg/idct.h"
 #include "dsp/mpeg/mpeg.h"
 #include "dsp/mpeg/video.h"
-#include "libc/bits/initializer.internal.h"
 #include "libc/fmt/conv.h"
 #include "libc/log/log.h"
 #include "libc/macros.internal.h"
@@ -94,9 +94,9 @@ plm_t *plm_create_with_buffer(plm_buffer_t *buffer, int destroy_when_done) {

 	self->demux = plm_demux_create(buffer, destroy_when_done);

-	// In theory we should check plm_demux_get_num_video_streams() and 
+	// In theory we should check plm_demux_get_num_video_streams() and
 	// plm_demux_get_num_audio_streams() here, but older files typically
-	// do not specify these correcly. So we just assume we have a video and
+	// do not specify these correctly. So we just assume we have a video and
 	// audio stream and create the decoders.

 	self->video_packet_type = PLM_DEMUX_PACKET_VIDEO_1;
@@ -159,7 +159,7 @@ double plm_get_framerate(plm_t *self) {

 int plm_get_num_audio_streams(plm_t *self) {
 	// Some files do not specify the number of audio streams in the system header.
-	// If the reported number of streams is 0, we check if we have a samplerate, 
+	// If the reported number of streams is 0, we check if we have a samplerate,
 	// indicating at least one audio stream.
 	int num_streams = plm_demux_get_num_audio_streams(self->demux);
 	return num_streams == 0 && plm_get_samplerate(self) ? 1 : num_streams;
@@ -40,7 +40,7 @@ asm(".include \"libc/disclaimer.inc\"");
 * @see YCbCr2RGB() in tool/viz/lib/ycbcr2rgb.c
 */
 void plm_frame_to_rgb(plm_frame_t *frame, uint8_t *rgb) {
-  // Chroma values are the same for each block of 4 pixels, so we proccess
+  // Chroma values are the same for each block of 4 pixels, so we process
  // 2 lines at a time, 2 neighboring pixels each.
  int w = frame->y.width, w2 = w >> 1;
  int y_index1 = 0, y_index2 = w, y_next_2_lines = w + (w - frame->width);
@@ -21,14 +21,14 @@
 #include "dsp/core/ituround.h"
 #include "dsp/core/q.h"
 #include "dsp/core/twixt8.h"
+#include "libc/intrin/bsr.h"
 #include "libc/limits.h"
 #include "libc/log/check.h"
 #include "libc/log/log.h"
 #include "libc/macros.internal.h"
 #include "libc/math.h"
+#include "libc/mem/gc.internal.h"
 #include "libc/mem/mem.h"
-#include "libc/nexgen32e/bsr.h"
-#include "libc/runtime/gc.internal.h"
 #include "libc/str/str.h"
 #include "libc/testlib/testlib.h"
 #include "libc/x/x.h"
@@ -203,8 +203,8 @@ void *Gyarados(long dyw, long dxw, int dst[dyw][dxw], long syw, long sxw,
      CHECK_LE(sxn, sxw);
      CHECK_LE(dyn, dyw);
      CHECK_LE(dxn, dxw);
-      CHECK_LT(bsrl(syn) + bsrl(sxn), 32);
-      CHECK_LT(bsrl(dyn) + bsrl(dxn), 32);
+      CHECK_LT(_bsrl(syn) + _bsrl(sxn), 32);
+      CHECK_LT(_bsrl(dyn) + _bsrl(dxn), 32);
      CHECK_LE(dyw, 0x7fff);
      CHECK_LE(dxw, 0x7fff);
      CHECK_LE(syw, 0x7fff);
@@ -51,7 +51,7 @@ $(DSP_SCALE_A).pkg:				\
 o/$(MODE)/dsp/scale/cdecimate2xuint8x8.o	\
 o/$(MODE)/dsp/scale/gyarados.o			\
 o/$(MODE)/dsp/scale/magikarp.o			\
-o/$(MODE)/dsp/scale/scale.o:			\
+o/$(MODE)/dsp/scale/scale.o: private		\
 		OVERRIDE_CFLAGS +=		\
 			$(MATHEMATICAL)

@@ -30,7 +30,7 @@
 * @see ttyconfig(), ttyrestore()
 */
 int ttyconfig(int ttyfd, ttyconf_f fn, int64_t arg,
-              const struct termios *opt_out_oldconf) {
+              struct termios *opt_out_oldconf) {
  struct termios conf[2];
  if (tcgetattr(ttyfd, &conf[0]) != -1 &&
      fn(memcpy(&conf[1], &conf[0], sizeof(conf[0])), arg) != -1 &&
@@ -17,7 +17,7 @@
 │ PERFORMANCE OF THIS SOFTWARE.                                                │
 ╚─────────────────────────────────────────────────────────────────────────────*/
 #include "dsp/tty/tty.h"
-#include "libc/bits/pushpop.h"
+#include "libc/intrin/pushpop.h"
 #include "libc/dce.h"
 #include "libc/log/internal.h"
 #include "libc/log/log.h"
@@ -17,14 +17,17 @@
 │ PERFORMANCE OF THIS SOFTWARE.                                                │
 ╚─────────────────────────────────────────────────────────────────────────────*/
 #include "dsp/tty/tty.h"
-#include "libc/bits/safemacros.internal.h"
-#include "libc/bits/weaken.h"
 #include "libc/calls/calls.h"
 #include "libc/calls/termios.h"
+#include "libc/dce.h"
+#include "libc/errno.h"
 #include "libc/fmt/fmt.h"
+#include "libc/intrin/safemacros.internal.h"
+#include "libc/intrin/weaken.h"
 #include "libc/mem/mem.h"
 #include "libc/runtime/runtime.h"
 #include "libc/sock/sock.h"
+#include "libc/sock/struct/pollfd.h"
 #include "libc/str/str.h"
 #include "libc/sysv/consts/poll.h"
 #include "libc/sysv/errfuns.h"
@@ -80,13 +83,15 @@ int ttyident(struct TtyIdent *ti, int ttyinfd, int ttyoutfd) {
      if (ttyident_probe(ti, ttyinfd, ttyoutfd, "\e[>c") != -1) {
        rc = 0;
        memset(&outer, 0, sizeof(outer));
-        if (ti->id == 83 /* GNU Screen */ && (ti->next || weaken(malloc)) &&
+        if (ti->id == 83 /* GNU Screen */ && (ti->next || _weaken(malloc)) &&
            ttyident_probe(&outer, ttyinfd, ttyoutfd, "\eP\e[>c\e\\") != -1 &&
-            (ti->next = (ti->next ? ti->next
-                                  : weaken(malloc)(sizeof(struct TtyIdent))))) {
+            (ti->next =
+                 (ti->next ? ti->next
+                           : _weaken(malloc)(sizeof(struct TtyIdent))))) {
          memcpy(ti->next, &outer, sizeof(outer));
        } else {
-          free_s(&ti->next);
+          free(ti->next);
+          ti->next = 0;
        }
      }
      ttyrestore(ttyinfd, &old);
@@ -26,7 +26,7 @@ static void ttyidentfree(struct TtyIdent *ti) {
  if (ti) {
    assert(ti != ti->next);
    ttyidentfree(ti->next);
-    free_s(&ti);
+    free(ti);
  }
 }

@@ -1,21 +1,22 @@
 #ifndef COSMOPOLITAN_DSP_TTY_INTERNAL_H_
 #define COSMOPOLITAN_DSP_TTY_INTERNAL_H_
+#include "dsp/tty/quant.h"
 #include "dsp/tty/ttyrgb.h"
-#include "libc/bits/xmmintrin.internal.h"
+#include "third_party/intel/xmmintrin.internal.h"
 #if !(__ASSEMBLER__ + __LINKER__ + 0)
 COSMOPOLITAN_C_START_

-struct TtyRgb rgb2tty24f_(__m128);
-struct TtyRgb rgb2ttyf2i_(__m128);
+struct TtyRgb rgb2tty24f_(ttyrgb_m128);
+struct TtyRgb rgb2ttyf2i_(ttyrgb_m128);
 struct TtyRgb rgb2ttyi2f_(int, int, int);
 struct TtyRgb rgb2ansi_(int, int, int);
 struct TtyRgb rgb2ansihash_(int, int, int);
 struct TtyRgb rgb2xterm24_(int, int, int);
-struct TtyRgb rgb2xterm256gray_(__m128);
+struct TtyRgb rgb2xterm256gray_(ttyrgb_m128);
 struct TtyRgb tty2rgb_(struct TtyRgb);
 struct TtyRgb tty2rgb24_(struct TtyRgb);
-__m128 tty2rgbf_(struct TtyRgb);
-__m128 tty2rgbf24_(struct TtyRgb);
+ttyrgb_m128 tty2rgbf_(struct TtyRgb);
+ttyrgb_m128 tty2rgbf24_(struct TtyRgb);

 char *setbg16_(char *, struct TtyRgb);
 char *setfg16_(char *, struct TtyRgb);
@@ -17,8 +17,7 @@
 │ PERFORMANCE OF THIS SOFTWARE.                                                │
 ╚─────────────────────────────────────────────────────────────────────────────*/
 #include "dsp/tty/itoa8.h"
-#include "libc/bits/bits.h"
-#include "libc/bits/initializer.internal.h"
+#include "libc/intrin/bits.h"
 #include "libc/str/str.h"

 struct Itoa8 kItoa8;
@@ -2,8 +2,7 @@
 #define DSP_TTY_QUANT_H_
 #include "dsp/tty/ttyrgb.h"
 #include "libc/assert.h"
-#include "libc/bits/bits.h"
-#include "libc/bits/xmmintrin.internal.h"
+#include "libc/intrin/bits.h"
 #include "libc/limits.h"
 #include "libc/str/str.h"
 #if !(__ASSEMBLER__ + __LINKER__ + 0)
@@ -14,12 +13,14 @@ COSMOPOLITAN_C_START_
 #define BL 2
 #define BR 3

-typedef __m128 (*tty2rgbf_f)(struct TtyRgb);
+typedef float ttyrgb_m128 __attribute__((__vector_size__(16), __may_alias__));
+
+typedef ttyrgb_m128 (*tty2rgbf_f)(struct TtyRgb);
 typedef char *(*setbg_f)(char *, struct TtyRgb);
 typedef char *(*setbgfg_f)(char *, struct TtyRgb, struct TtyRgb);
 typedef char *(*setfg_f)(char *, struct TtyRgb);
 typedef struct TtyRgb (*rgb2tty_f)(int, int, int);
-typedef struct TtyRgb (*rgb2ttyf_f)(__m128);
+typedef struct TtyRgb (*rgb2ttyf_f)(ttyrgb_m128);
 typedef struct TtyRgb (*tty2rgb_f)(struct TtyRgb);
 typedef struct TtyRgb ttypalette_t[2][8];

@@ -19,7 +19,6 @@
 #include "dsp/core/core.h"
 #include "dsp/tty/quant.h"
 #include "libc/assert.h"
-#include "libc/bits/initializer.internal.h"
 #include "libc/limits.h"
 #include "libc/log/log.h"
 #include "libc/macros.internal.h"
@@ -40,9 +39,9 @@ struct TtyRgb tty2rgb_(struct TtyRgb rgbxt) {
  return g_ansi2rgb_[rgbxt.xt];
 }

-__m128 tty2rgbf_(struct TtyRgb rgbxt) {
+ttyrgb_m128 tty2rgbf_(struct TtyRgb rgbxt) {
  rgbxt = g_ansi2rgb_[rgbxt.xt];
-  return (__m128){(int)rgbxt.r, (int)rgbxt.g, (int)rgbxt.b} / 255;
+  return (ttyrgb_m128){(int)rgbxt.r, (int)rgbxt.g, (int)rgbxt.b} / 255;
 }

 static int rgb2xterm256_(int r, int g, int b) {
@@ -17,12 +17,11 @@
 │ PERFORMANCE OF THIS SOFTWARE.                                                │
 ╚─────────────────────────────────────────────────────────────────────────────*/
 #include "dsp/tty/quant.h"
-#include "libc/bits/xmmintrin.internal.h"

-struct TtyRgb rgb2ttyf2i_(__m128 rgb) {
-  __v4si i4;
+typedef int ttyrgb_i4 __attribute__((__vector_size__(16)));
+
+struct TtyRgb rgb2ttyf2i_(ttyrgb_m128 rgb) {
  rgb *= 255;
-  /* i4 = __builtin_ia32_cvtps2dq(rgb); */
-  asm("cvttps2dq\t%0,%1" : "+%x"(rgb), "=x"(i4));
-  return rgb2tty(i4[0], i4[1], i4[2]);
+  ttyrgb_i4 rgbi = {rgb[0], rgb[1], rgb[2], rgb[3]};
+  return rgb2tty(rgbi[0], rgbi[1], rgbi[2]);
 }
@@ -21,5 +21,5 @@
 #include "libc/macros.internal.h"

 struct TtyRgb rgb2ttyi2f_(int r, int g, int b) {
-  return rgb2ttyf((__m128){r, g, b} / 255);
+  return rgb2ttyf((ttyrgb_m128){r, g, b} / 255);
 }
@@ -17,25 +17,24 @@
 │ PERFORMANCE OF THIS SOFTWARE.                                                │
 ╚─────────────────────────────────────────────────────────────────────────────*/
 #include "dsp/tty/quant.h"
+#include "libc/macros.internal.h"
 #include "libc/math.h"
+#include "third_party/intel/xmmintrin.internal.h"

-/*
-struct TtyRgb rgb2tty24f_(__m128 rgb) {
-  const __v4si kMax = {255, 255, 255, 255};
-  const __v4si kMin = {0, 0, 0, 0};
-  struct TtyRgb res;
-  __v4si rgb255;
-  rgb255 = _mm_min_ps(_mm_max_ps(_mm_cvtps_epi32(rgb * 255), kMin), kMax);
-  res = (struct TtyRgb){rgb255[0], rgb255[1], rgb255[2], rgb255[3]};
-  return res;
-}
-*/
-
-struct TtyRgb rgb2tty24f_(__m128 rgb) {
-  const __m128 kMax = {1, 1, 1, 1};
-  const __m128 kMin = {0, 0, 0, 0};
+struct TtyRgb rgb2tty24f_(ttyrgb_m128 rgb) {
+#ifdef __x86_64__
+  const ttyrgb_m128 kMax = {1, 1, 1, 1};
+  const ttyrgb_m128 kMin = {0, 0, 0, 0};
  struct TtyRgb res;
  rgb = _mm_min_ps(_mm_max_ps(rgb, kMin), kMax) * 255;
  res = (struct TtyRgb){rgb[0], rgb[1], rgb[2], rgb[3]};
  return res;
+#else
+  return (struct TtyRgb){
+      MAX(0, MIN(1, rgb[0])) * 255,
+      MAX(0, MIN(1, rgb[1])) * 255,
+      MAX(0, MIN(1, rgb[2])) * 255,
+      MAX(0, MIN(1, rgb[3])) * 255,
+  };
+#endif
 }
@@ -17,8 +17,8 @@
 │ PERFORMANCE OF THIS SOFTWARE.                                                │
 ╚─────────────────────────────────────────────────────────────────────────────*/
 #include "dsp/tty/tty.h"
-#include "libc/alg/arraylist2.internal.h"
-#include "libc/runtime/gc.internal.h"
+#include "libc/mem/arraylist2.internal.h"
+#include "libc/mem/gc.internal.h"
 #include "libc/x/x.h"

 /* TODO(jart): DELETE */
@@ -30,13 +30,17 @@
 * @param ti comes from ttyident() and null means no-op
 */
 int ttysendtitle(int ttyfd, const char *title, const struct TtyIdent *ti) {
+  int res;
  if (ti) {
+    char *p;
    if (ti->id == kTtyIdScreen) {
-      return ttysend(ttyfd, gc(xstrcat("\eP\e]0;", title, "\a\e\\")));
+      res = ttysend(ttyfd, (p = xstrcat("\eP\e]0;", title, "\a\e\\")));
    } else {
-      return ttysend(ttyfd, gc(xstrcat("\e]0;", title, "\a")));
+      res = ttysend(ttyfd, (p = xstrcat("\e]0;", title, "\a")));
    }
+    free(p);
  } else {
-    return 0;
+    res = 0;
  }
+  return res;
 }
@@ -44,7 +44,7 @@ int ttysendtitle(int, const char *, const struct TtyIdent *);
 int ttyident(struct TtyIdent *, int, int);
 void ttyidentclear(struct TtyIdent *);
 char *ttydescribe(char *, size_t, const struct TtyIdent *);
-int ttyconfig(int, ttyconf_f, int64_t, const struct termios *);
+int ttyconfig(int, ttyconf_f, int64_t, struct termios *);
 int ttyrestore(int, const struct termios *);
 int ttysetrawdeadline(struct termios *, int64_t);
 int ttysetrawmode(struct termios *, int64_t);
@@ -25,7 +25,6 @@ DSP_TTY_A_CHECKS =				\

 DSP_TTY_A_DIRECTDEPS =				\
 	DSP_CORE				\
-	LIBC_ALG				\
 	LIBC_CALLS				\
 	LIBC_FMT				\
 	LIBC_INTRIN				\
@@ -41,8 +40,7 @@ DSP_TTY_A_DIRECTDEPS =				\
 	LIBC_SYSV				\
 	LIBC_TINYMATH				\
 	LIBC_TIME				\
-	LIBC_X					\
-	LIBC_UNICODE
+	LIBC_X

 DSP_TTY_A_DEPS :=				\
 	$(call uniq,$(foreach x,$(DSP_TTY_A_DIRECTDEPS),$($(x))))
@@ -55,8 +53,8 @@ $(DSP_TTY_A).pkg:				\
 		$(DSP_TTY_A_OBJS)		\
 		$(foreach x,$(DSP_TTY_A_DIRECTDEPS),$($(x)_A).pkg)

-o/$(MODE)/dsp/tty/ttyraster.o:			\
-		OVERRIDE_CFLAGS +=		\
+o/$(MODE)/dsp/tty/ttyraster.o: private		\
+		OVERRIDE_CFLAGS +=	\
 			$(MATHEMATICAL)

 DSP_TTY_LIBS = $(foreach x,$(DSP_TTY_ARTIFACTS),$($(x)))
@@ -18,6 +18,6 @@
 ╚─────────────────────────────────────────────────────────────────────────────*/
 #include "dsp/tty/quant.h"

-__m128 tty2rgbf24_(struct TtyRgb rgbxt) {
-  return (__m128){(int)rgbxt.r, (int)rgbxt.g, (int)rgbxt.b} / 255;
+ttyrgb_m128 tty2rgbf24_(struct TtyRgb rgbxt) {
+  return (ttyrgb_m128){(int)rgbxt.r, (int)rgbxt.g, (int)rgbxt.b} / 255;
 }
@@ -18,7 +18,7 @@
 │ PERFORMANCE OF THIS SOFTWARE.                                                │
 ╚─────────────────────────────────────────────────────────────────────────────*/
 #include "dsp/tty/tty.h"
-#include "libc/alg/alg.h"
+#include "libc/mem/alg.h"
 #include "libc/nexgen32e/nexgen32e.h"
 #include "libc/str/str.h"

@@ -45,6 +45,8 @@ void ttyhisto(uint32_t histogram[hasatleast 256],
      histogram[xtcolors[i * 8]]++;
    }
  }
-  imapxlatab(dominant);
+  for (i = 0; i < 256; ++i) {
+    dominant[i] = i;
+  }
  qsort_r(dominant, 256, 1, (void *)histcmp, histogram);
 }
@@ -18,7 +18,7 @@
 ╚─────────────────────────────────────────────────────────────────────────────*/
 #include "dsp/tty/itoa8.h"
 #include "dsp/tty/tty.h"
-#include "libc/bits/safemacros.internal.h"
+#include "libc/intrin/safemacros.internal.h"
 #include "libc/limits.h"
 #include "libc/log/check.h"

@@ -18,7 +18,6 @@
 ╚─────────────────────────────────────────────────────────────────────────────*/
 #include "dsp/tty/internal.h"
 #include "dsp/tty/quant.h"
-#include "libc/bits/initializer.internal.h"
 #include "libc/dce.h"
 #include "libc/runtime/runtime.h"
 #include "libc/stdio/stdio.h"
@@ -22,8 +22,8 @@
 #include "dsp/tty/ttyrgb.h"
 #include "dsp/tty/windex.h"
 #include "libc/assert.h"
-#include "libc/bits/bits.h"
-#include "libc/bits/safemacros.internal.h"
+#include "libc/intrin/bits.h"
+#include "libc/intrin/safemacros.internal.h"
 #include "libc/limits.h"
 #include "libc/log/check.h"
 #include "libc/log/log.h"
@@ -22,11 +22,10 @@
 #include "libc/calls/struct/sigaction.h"
 #include "libc/calls/struct/siginfo.h"
 #include "libc/calls/termios.h"
-#include "libc/calls/typedef/sigaction_f.h"
 #include "libc/calls/ucontext.h"
 #include "libc/log/log.h"
 #include "libc/macros.internal.h"
-#include "libc/runtime/gc.internal.h"
+#include "libc/mem/gc.internal.h"
 #include "libc/runtime/runtime.h"
 #include "libc/str/str.h"
 #include "libc/sysv/consts/fileno.h"
@@ -34,6 +33,7 @@
 #include "libc/sysv/consts/sig.h"
 #include "libc/sysv/errfuns.h"
 #include "libc/x/x.h"
+#include "libc/x/xsigaction.h"

 /* TODO(jart): DELETE */

@@ -98,8 +98,8 @@ static relegated void ttyraw_onsig(int sig, struct siginfo *info,
  }
  for (i = 0; i < ARRAYLEN(g_ttyraw.sigs); ++i) {
    if (g_ttyraw.sigs[i] == sig) {
-      if (g_ttyraw.next[i] != SIG_IGN) {
-        if (g_ttyraw.next[i] != SIG_DFL) {
+      if (g_ttyraw.next[i] != (void *)SIG_IGN) {
+        if (g_ttyraw.next[i] != (void *)SIG_DFL) {
          if (g_ttyraw.next[i]) {
            g_ttyraw.next[i](sig, info, ctx);
          }
@@ -118,7 +118,7 @@ static textstartup void ttyraw_initsig(int sig, unsigned flags, unsigned mask) {
  struct sigaction old;
  g_ttyraw.next[i] = xsigaction(sig, ttyraw_onsig, flags, mask, &old) != -1
                         ? old.sa_sigaction
-                         : SIG_DFL;
+                         : (void *)SIG_DFL;
  g_ttyraw.sigs[i++] = sig;
 }

@@ -21,6 +21,7 @@
 #include "libc/calls/termios.h"
 #include "libc/errno.h"
 #include "libc/sock/sock.h"
+#include "libc/sock/struct/pollfd.h"
 #include "libc/sysv/consts/poll.h"
 #include "libc/sysv/consts/termios.h"

@@ -10,7 +10,7 @@
 #include "libc/log/log.h"

 /**
- * Automatically launchs GDB Debugger TUI during crash.
+ * Automatically launches GDB Debugger TUI during crash.
 *
 * Run the following inside a terminal:
 *
@@ -7,7 +7,7 @@
 │   • http://creativecommons.org/publicdomain/zero/1.0/            │
 ╚─────────────────────────────────────────────────────────────────*/
 #endif
-#include "libc/bits/bits.h"
+#include "libc/intrin/bits.h"
 #include "libc/dce.h"
 #include "libc/log/log.h"
 #include "libc/runtime/runtime.h"
@@ -7,12 +7,12 @@
 │   • http://creativecommons.org/publicdomain/zero/1.0/            │
 ╚─────────────────────────────────────────────────────────────────*/
 #endif
-#include "libc/bits/bits.h"
 #include "libc/dce.h"
+#include "libc/intrin/bits.h"
+#include "libc/intrin/kprintf.h"
 #include "libc/log/log.h"
 #include "libc/mem/mem.h"
 #include "libc/runtime/runtime.h"
-#include "libc/stdio/stdio.h"
 #include "libc/str/str.h"

 /**
@@ -59,9 +59,15 @@

 int main(int argc, char *argv[]) {
  if (!IsAsan()) {
-    printf("this example is intended for MODE=asan or MODE=dbg\n");
+    kprintf("this example is intended for MODE=asan or MODE=dbg\n");
    exit(1);
  }
+
+  kprintf("----------------\n");
+  kprintf(" THIS IS A TEST \n");
+  kprintf("SIMULATING CRASH\n");
+  kprintf("----------------\n");
+
  char *buffer;
  ShowCrashReports(); /* not needed but yoinks appropriate symbols */
  buffer = malloc(13);
@@ -7,7 +7,7 @@
 │   • http://creativecommons.org/publicdomain/zero/1.0/            │
 ╚─────────────────────────────────────────────────────────────────*/
 #endif
-#include "libc/bits/bits.h"
+#include "libc/intrin/bits.h"
 #include "libc/dce.h"
 #include "libc/log/log.h"
 #include "libc/mem/mem.h"
@@ -0,0 +1,41 @@
+#if 0
+/*─────────────────────────────────────────────────────────────────╗
+│ To the extent possible under law, Justine Tunney has waived      │
+│ all copyright and related or neighboring rights to this file,    │
+│ as it is written in the following disclaimers:                   │
+│   • http://unlicense.org/                                        │
+│   • http://creativecommons.org/publicdomain/zero/1.0/            │
+╚─────────────────────────────────────────────────────────────────*/
+#endif
+#include "libc/calls/struct/timespec.h"
+#include "libc/stdio/stdio.h"
+#include "libc/time/time.h"
+
+/**
+ * @fileoverview clock() function demo
+ */
+
+int main(int argc, char *argv[]) {
+  unsigned long i;
+  volatile unsigned long x;
+  struct timespec now, start, next, interval;
+  printf("hammering the cpu...\n");
+  next = start = timespec_mono();
+  interval = timespec_frommillis(500);
+  next = timespec_add(next, interval);
+  for (;;) {
+    for (i = 0;; ++i) {
+      x *= 7;
+      if (!(i % 256)) {
+        now = timespec_mono();
+        if (timespec_cmp(now, next) >= 0) {
+          break;
+        }
+      }
+    }
+    next = timespec_add(next, interval);
+    printf("consumed %10g seconds monotonic time and %10g seconds cpu time\n",
+           timespec_tonanos(timespec_sub(now, start)) / 1000000000.,
+           (double)clock() / CLOCKS_PER_SEC);
+  }
+}
@@ -0,0 +1,58 @@
+#if 0
+/*─────────────────────────────────────────────────────────────────╗
+│ To the extent possible under law, Justine Tunney has waived      │
+│ all copyright and related or neighboring rights to this file,    │
+│ as it is written in the following disclaimers:                   │
+│   • http://unlicense.org/                                        │
+│   • http://creativecommons.org/publicdomain/zero/1.0/            │
+╚─────────────────────────────────────────────────────────────────*/
+#endif
+#include "libc/calls/struct/timespec.h"
+#include "libc/errno.h"
+#include "libc/intrin/describeflags.internal.h"
+#include "libc/intrin/kprintf.h"
+#include "libc/str/str.h"
+#include "libc/sysv/consts/clock.h"
+
+int n;
+int shown[64];
+
+void show(int clock) {
+  int i;
+  struct timespec ts;
+  if (clock == 127) return;
+  for (i = 0; i < n; ++i) {
+    if (shown[i] == clock) {
+      return;
+    }
+  }
+  shown[n++] = clock;
+  if (clock_getres(clock, &ts) != -1) {
+    kprintf("%s %'ld ns\n", DescribeClockName(clock), timespec_tonanos(ts));
+  } else {
+    kprintf("%s %s\n", DescribeClockName(clock), _strerrno(errno));
+  }
+}
+
+int main(int argc, char *argv[]) {
+  show(CLOCK_REALTIME);
+  show(CLOCK_REALTIME_FAST);
+  show(CLOCK_REALTIME_PRECISE);
+  show(CLOCK_REALTIME_COARSE);
+  show(CLOCK_MONOTONIC);
+  show(CLOCK_MONOTONIC_RAW);
+  show(CLOCK_MONOTONIC_FAST);
+  show(CLOCK_MONOTONIC_PRECISE);
+  show(CLOCK_MONOTONIC_COARSE);
+  show(CLOCK_PROCESS_CPUTIME_ID);
+  show(CLOCK_THREAD_CPUTIME_ID);
+  show(CLOCK_PROF);
+  show(CLOCK_BOOTTIME);
+  show(CLOCK_REALTIME_ALARM);
+  show(CLOCK_BOOTTIME_ALARM);
+  show(CLOCK_TAI);
+  show(CLOCK_UPTIME);
+  show(CLOCK_UPTIME_PRECISE);
+  show(CLOCK_UPTIME_FAST);
+  show(CLOCK_SECOND);
+}
@@ -8,11 +8,12 @@
 ╚─────────────────────────────────────────────────────────────────*/
 #endif
 #include "libc/assert.h"
+#include "libc/calls/calls.h"
 #include "libc/errno.h"
 #include "libc/fmt/conv.h"
 #include "libc/log/check.h"
 #include "libc/mem/mem.h"
-#include "libc/runtime/gc.internal.h"
+#include "libc/mem/gc.internal.h"
 #include "libc/runtime/runtime.h"
 #include "libc/stdio/stdio.h"
 #include "libc/str/str.h"
@@ -9,18 +9,21 @@
 #endif
 #include "libc/calls/calls.h"
 #include "libc/calls/struct/dirent.h"
+#include "libc/calls/struct/rusage.h"
 #include "libc/calls/struct/sigaction.h"
 #include "libc/calls/struct/sigset.h"
 #include "libc/calls/struct/timespec.h"
+#include "libc/errno.h"
 #include "libc/fmt/fmt.h"
 #include "libc/fmt/itoa.h"
-#include "libc/intrin/kprintf.h"
+#include "libc/log/appendresourcereport.internal.h"
 #include "libc/log/internal.h"
 #include "libc/log/log.h"
 #include "libc/macros.internal.h"
+#include "libc/mem/mem.h"
 #include "libc/runtime/internal.h"
 #include "libc/runtime/runtime.h"
-#include "libc/stdio/append.internal.h"
+#include "libc/stdio/append.h"
 #include "libc/stdio/stdio.h"
 #include "libc/str/str.h"
 #include "libc/sysv/consts/clock.h"
@@ -151,7 +154,7 @@ int main(int argc, char *argv[]) {
  while ((line = linenoiseWithHistory(prompt, "cmd"))) {
    n = 0;
    start = line;
-    if (startswith(start, "time ")) {
+    if (_startswith(start, "time ")) {
      timeit = true;
      start += 5;
    } else {
@@ -16,7 +16,7 @@
 │ TORTIOUS ACTION, ARISING OUT OF OR IN CONNECTION WITH THE USE OR             │
 │ PERFORMANCE OF THIS SOFTWARE.                                                │
 ╚─────────────────────────────────────────────────────────────────────────────*/
-#include "libc/bits/safemacros.internal.h"
+#include "libc/intrin/safemacros.internal.h"
 #include "libc/stdio/stdio.h"
 #include "libc/str/str.h"

@@ -7,6 +7,7 @@
 │   • http://creativecommons.org/publicdomain/zero/1.0/            │
 ╚─────────────────────────────────────────────────────────────────*/
 #endif
+#include "libc/intrin/kprintf.h"
 #include "libc/log/log.h"
 #include "libc/runtime/symbols.internal.h"

@@ -24,6 +25,12 @@
 */

 noubsan int main(int argc, char *argv[]) {
+
+  kprintf("----------------\n");
+  kprintf(" THIS IS A TEST \n");
+  kprintf("SIMULATING CRASH\n");
+  kprintf("----------------\n");
+
  volatile int64_t x;
  ShowCrashReports();
  return 1 / (x = 0);
@@ -8,6 +8,7 @@
 ╚─────────────────────────────────────────────────────────────────*/
 #endif
 #include "libc/calls/calls.h"
+#include "libc/calls/struct/sigset.h"
 #include "libc/log/log.h"
 #include "libc/stdio/stdio.h"

@@ -23,7 +24,7 @@
 *
 * On supported platforms, this will cause GDB to automatically attach.
 * The nice thing about this, is you can start stepping through your
- * code at the precice instruction where the interrupt happened. See
+ * code at the precise instruction where the interrupt happened. See
 * `libc/log/attachdebugger.c` to see how it works.
 *
 * If you wish to suppress the auto-GDB behavior, then:
@@ -8,7 +8,8 @@
 ╚─────────────────────────────────────────────────────────────────*/
 #endif
 #include "libc/log/log.h"
-#include "libc/runtime/gc.internal.h"
+#include "libc/mem/gc.internal.h"
+#include "libc/runtime/runtime.h"
 #include "libc/stdio/stdio.h"
 #include "libc/x/x.h"
 #include "net/http/escape.h"