mirror of
https://github.com/libretro/virtualjaguar-libretro.git
synced 2026-11-07 04:36:00 -10:00
Joseph MattielloandClaude Opus 5
8344a70399
The Skyhammer clipping sentinel has been inert. The Makefile looked for "Skyhammer_(1999).jag"; no such file exists in the corpus (the ROM is in fact absent entirely, under any spelling). The check printed one SKIP line among hundreds and the suite still exited 0, so the repo's canonical "loud broken audio" sentinel was not running at all. This is the masked-failure class PR #170 was closed over: a skip that reads as a pass is a silenced regression wearing a different hat. - scripts/find-rom.sh: resolve private ROMs by case-insensitive name pattern instead of one hardcoded literal path. Filters to loadable media (a bare "Atari Karts*" otherwise matched 7zips/*.7z) and ranks by path depth so the canonical top-level copy beats duplicates buried in sub-collections. - scripts/test-skip.sh: skip ledger. Every optional check that does not run records one line; `make test` prints a consolidated summary at the end. Non-fatal by default (CI has none of the private ROMs); VJ_REQUIRE_ROMS=1 escalates any skip to a hard failure. The ledger is truncated at the start of the run so stale rows cannot resurface as fresh skips. - Makefile: route the four audio-ROM lookups and the option-visibility disc through the above. Drop the `[ -f ]` guards on yarc.j64 and jagniccc.j64: both are committed in-tree, so absence means a broken checkout and must fail rather than read as a pass -- the rationale test_state_compat and test_frontend_pacing already state. - test_audio_presence: print the "=== Presence check: <label> ===" banner even under --quiet, matching test_audio_clipping. The suite was emitting a bare "PASS: audio is present..." naming no title. Skyhammer stays wired as an assert-clean check, identical to Iron Soldier 2. CLAUDE.md says it must keep failing; the Makefile comment and the MMULT secondary-bank fix notes say it was resolved. With no ROM available neither claim is verifiable, so the disagreement is recorded in a comment rather than settled by guess. No emulation behaviour changed. Co-Authored-By: Claude Opus 5 <[email protected]>
88 lines
3.3 KiB
Bash
Executable File
88 lines
3.3 KiB
Bash
Executable File
#!/usr/bin/env bash
|
|
# test-skip.sh -- skip ledger for `make test`.
|
|
#
|
|
# WHY THIS EXISTS
|
|
# ---------------
|
|
# Several checks in the suite only run when an optional private ROM / disc
|
|
# image is present. Historically each such site just printed "SKIP: ..." and
|
|
# fell through, so the suite still exited 0 -- a skip was indistinguishable
|
|
# from a pass. That let the Skyhammer clipping sentinel sit inert for an
|
|
# unknown number of releases: the Makefile looked for a filename that does not
|
|
# exist in the corpus, printed one SKIP line among hundreds of lines of test
|
|
# output, and reported success.
|
|
#
|
|
# This is the same masked-failure class the repo was burned by in PR #170,
|
|
# where a "fix" silenced Iron Soldier 2 and the clipping test passed because
|
|
# silence has 0% saturation. A skip that reads as a pass is that bug wearing
|
|
# a different hat.
|
|
#
|
|
# The ledger makes skips additive and visible: every skip site records one
|
|
# line, and the suite prints a consolidated summary at the end. CI has none
|
|
# of the private ROMs, so skipping stays non-fatal by default -- but it is no
|
|
# longer silent, and `VJ_REQUIRE_ROMS=1` turns it into a hard failure for
|
|
# anyone who does have the corpus and wants the full suite enforced.
|
|
#
|
|
# Usage:
|
|
# test-skip.sh reset truncate the ledger (call FIRST)
|
|
# test-skip.sh record "<check>" "<why>" print + append one skip
|
|
# test-skip.sh summary print the roll-up; exit 1 under
|
|
# VJ_REQUIRE_ROMS=1 if any skips
|
|
#
|
|
# Ledger path defaults to test/.skipped-checks; override with VJ_SKIP_LOG.
|
|
|
|
set -u
|
|
|
|
log="${VJ_SKIP_LOG:-test/.skipped-checks}"
|
|
mode="${1:-}"
|
|
|
|
case "$mode" in
|
|
reset)
|
|
# Truncate at the START of the run, never at the end. A ledger left over
|
|
# from a previous invocation would resurrect old rows as fresh skips --
|
|
# the same stale-row failure mode documented for cd_boot_matrix.sh, where
|
|
# a resumed sweep replayed an ancient Battle Morph result as a new one.
|
|
mkdir -p "$(dirname "$log")" 2>/dev/null
|
|
: > "$log"
|
|
;;
|
|
|
|
record)
|
|
check="${2:-unnamed check}"
|
|
reason="${3:-no reason given}"
|
|
# Print inline too, so the skip is visible in-context as well as in the
|
|
# roll-up. Both come from this one call, so they cannot drift apart.
|
|
printf ' SKIP: %s -- %s\n' "$check" "$reason"
|
|
mkdir -p "$(dirname "$log")" 2>/dev/null
|
|
printf '%s\t%s\n' "$check" "$reason" >> "$log"
|
|
;;
|
|
|
|
summary)
|
|
echo ""
|
|
if [ ! -s "$log" ]; then
|
|
echo "=== Skipped checks: none -- every optional check ran ==="
|
|
exit 0
|
|
fi
|
|
|
|
count=$(wc -l < "$log" | tr -d ' ')
|
|
echo "=== Skipped checks ($count) ==="
|
|
# Two columns: check name, then why. awk (not column) so the output is
|
|
# identical on macOS and Linux.
|
|
awk -F'\t' '{ printf " %-42s %s\n", $1, $2 }' "$log"
|
|
echo ""
|
|
echo " These checks did NOT run. They are not failures, but they are"
|
|
echo " not passes either -- most need optional private ROMs/discs that"
|
|
echo " CI does not have. Re-run with VJ_REQUIRE_ROMS=1 to make any"
|
|
echo " skip a hard failure once you have the corpus in place."
|
|
|
|
if [ "${VJ_REQUIRE_ROMS:-0}" = "1" ]; then
|
|
echo ""
|
|
echo "ERROR: VJ_REQUIRE_ROMS=1 and $count check(s) were skipped." >&2
|
|
exit 1
|
|
fi
|
|
;;
|
|
|
|
*)
|
|
echo "usage: test-skip.sh {reset|record <check> <reason>|summary}" >&2
|
|
exit 2
|
|
;;
|
|
esac
|