Conversation
The #391 rename left release.yml building 'make glm' (an alias that now produces 'colibri') and then asserting/copying 'c/glm', which no longer exists. CI never catches this: only a tag push runs this workflow, so the v1.1.0 tag build failed on macOS at 'ls -lh glm' with the other jobs cancelled and the release skipped. Co-Authored-By: Claude Fable 5 <noreply@anthropic.com>
…o ROCm and Windows (#509) The engine read the sampling temperature from the env var TEMP, a name the platform already owns: the ROCm runtime (comgr, MIOpen, roctracer) reads $TEMP as a temp-directory path, so 'coli --temp 0.6' made HIP context init fail inside comgr's lazy first-use — SIGSEGV in the runtime's error-unwind on gfx1100, clean hipErrorOutOfMemory on gfx1030. And on native Windows builds TEMP is ALWAYS set (C:\...\Temp), so atof() returned 0 and every run without an explicit override was silently greedy. Three complementary pieces, per the discussion in #509: - coli --temp now exports COLI_TEMP; the engine reads COLI_TEMP first. - TEMP survives as a legacy alias but only when fully numeric (strtod must consume the whole string): existing TEMP=0 scripts keep working, while a Windows/ROCm directory path no longer masquerades as a temperature. - Defense-in-depth in coli_cuda_init (HIP builds only): a $TEMP that is not a real directory is dropped before the first ROCm call — covers a stray numeric TEMP set manually on the raw binary; a genuine temp-dir is preserved (noobdev-ph's scrub from the #509 thread). The parse lives in temp_from_env() so tests/test_temp_env.c can pin the contract (11 cases: precedence, legacy alias, Windows path, auto). Co-Authored-By: Claude Fable 5 <noreply@anthropic.com>
release workflow: build/copy the engine under its post-rename name (v1.1.0 tag build failed)
sampling: COLI_TEMP replaces the TEMP env var — $TEMP belongs to ROCm and Windows (#509)
…me step The README and quickstart still told Windows users to rename the downloaded engine to glm.exe: that was the pre-#508 archive layout. Since #508 the archive ships the engine plainly named colibri.exe and the launcher finds it with zero setup — the rename step now just confuses people (and the file it names no longer exists). Same sweep for the build docs: 'make glm.exe' is not a target anymore (windows.md recommended it), direct-run examples said ./glm(.exe), metal.md built 'make glm METAL=1'. All now use the real names; troubleshooting rows keep a '(glm.exe pre-rename)' note where old trees are the subject. Co-Authored-By: Claude Fable 5 <noreply@anthropic.com>
docs: no more glm.exe — the archive ships colibri.exe ready to run (obsolete rename step removed)
This file contains hidden or bidirectional Unicode text that may be interpreted or compiled differently than what appears below. To review, open the file in an editor that reveals hidden Unicode characters.
Learn more about bidirectional Unicode characters
Sign up for free
to join this conversation on GitHub.
Already have an account?
Sign in to comment
1 participant
Add this suggestion to a batch that can be applied as a single commit.This suggestion is invalid because no changes were made to the code.Suggestions cannot be applied while the pull request is closed.Suggestions cannot be applied while viewing a subset of changes.Only one suggestion per line can be applied in a batch.Add this suggestion to a batch that can be applied as a single commit.Applying suggestions on deleted lines is not supported.You must change the existing code in this line in order to create a valid suggestion.Outdated suggestions cannot be applied.This suggestion has been applied or marked resolved.Suggestions cannot be applied from pending reviews.Suggestions cannot be applied on multi-line comments.Suggestions cannot be applied while the pull request is queued to merge.Suggestion cannot be applied right now. Please check back later.
The v1.1.0 tag build died before publishing anything:
release.ymlstill built and copied the engine under its pre-#391 name (failed run). This brings main the fix plus one real bug fix that landed meanwhile:make colibriand copiesc/colibri, matching ci.yml. Only a tag push exercises this file, which is why every CI was green while the tag build failed.COLI_TEMP(fixes [Bug]: TEMP env var conflicts with ROCm on AMD GPUs causing SIGSEGV #509):$TEMPis the ROCm/Windows temp-directory, so--temp 0.6crashed HIP init on AMD, and on native Windowsatof("C:\\...\\Temp")==0silently forced greedy on every run. Legacy numericTEMPstill accepted; HIP-only env scrub as defense-in-depth; contract pinned bytests/test_temp_env.c.After merge: delete and re-push the
v1.1.0tag on the new main head (safe — no release was published).🤖 Generated with Claude Code