Summary: make engine on a fresh git clone --recursive fails at the libm parity gate because the reference files it requires are not committed to the repository.
FATAL: libm parity gate cannot run — missing <repo>/testkit/libm-refs/double-fn-vm26.txt
(lane=submodule) Refusing to report a build as good without it.
scripts/build-engine.sh requires testkit/libm-refs/double-fn-vm26.txt (submodule lane) or double-fn-2025lane-clang.txt (glibc-import lane), but testkit/ exists nowhere in the repo or its git history. The gate's own comment (dated 2026-08-07) says it was made fail-closed because the references used to live in gitignored logs/ — but the files were never committed, so every fresh clone hits the FATAL.
Environment: Apple Silicon (arm64), macOS 25.5 (Darwin), clean git clone --recursive of main. Everything up to the gate builds and links fine (after brew install cmake ninja gcc@15 coreutils, see below).
Request: commit testkit/libm-refs/*.txt so the parity gate can run outside the maintainer machine.
Related small gaps hit on a fresh machine, all fixable locally:
scripts/build-mesa-kk.sh brew bootstrap list is missing cmake, ninja (SPIRV-LLVM-Translator step) and gcc@15 (scripts/gcc-dbl64-swap.sh).- The path-leak audit in
build-mesa-kk.sh pipes binary grep matches into sort without LC_ALL=C ("Illegal byte sequence" in a UTF-8 locale, silently dropping hits), and it flags the install names the script itself repoints to $MESA_PREFIX when the checkout lives under /Users/.
build-engine.sh calls timeout, absent on stock macOS (gtimeout from brew coreutils works).- A fork clone has no upstream numeric release tags, so
fetch_spring_version fails with Invalid version format: <sha> main until tags are fetched from beyond-all-reason/RecoilEngine.
Summary:
make engineon a freshgit clone --recursivefails at the libm parity gate because the reference files it requires are not committed to the repository.scripts/build-engine.shrequirestestkit/libm-refs/double-fn-vm26.txt(submodule lane) ordouble-fn-2025lane-clang.txt(glibc-import lane), buttestkit/exists nowhere in the repo or its git history. The gate's own comment (dated 2026-08-07) says it was made fail-closed because the references used to live in gitignoredlogs/— but the files were never committed, so every fresh clone hits the FATAL.Environment: Apple Silicon (arm64), macOS 25.5 (Darwin), clean
git clone --recursiveofmain. Everything up to the gate builds and links fine (afterbrew install cmake ninja gcc@15 coreutils, see below).Request: commit
testkit/libm-refs/*.txtso the parity gate can run outside the maintainer machine.Related small gaps hit on a fresh machine, all fixable locally:
scripts/build-mesa-kk.shbrew bootstrap list is missingcmake,ninja(SPIRV-LLVM-Translator step) andgcc@15(scripts/gcc-dbl64-swap.sh).build-mesa-kk.shpipes binary grep matches intosortwithoutLC_ALL=C("Illegal byte sequence" in a UTF-8 locale, silently dropping hits), and it flags the install names the script itself repoints to$MESA_PREFIXwhen the checkout lives under/Users/.build-engine.shcallstimeout, absent on stock macOS (gtimeoutfrom brew coreutils works).fetch_spring_versionfails withInvalid version format: <sha> mainuntil tags are fetched frombeyond-all-reason/RecoilEngine.