You signed in with another tab or window. Reload to refresh your session.You signed out in another tab or window. Reload to refresh your session.You switched accounts on another tab or window. Reload to refresh your session.Dismiss alert
{{ message }}
Repository navigation
[CI failure] CI: pixi run test (source build) scheduled runs #2970
This issue tracks an unhealthy unattended CI workflow.
Subsequent failures are recorded in comments; a successful run in the
same trigger scope closes the issue automatically.
My AI suggests that this change broke the nightly test.
The failure is a stale CI assertion, not a failed CUDA Core build.
The likely regression is commit 2f44c6b / PR #2903, “add CUDA_PYTHON_TOOLCHAIN override.” It changed:
cuda_core/build/.build-cuda-major into an ABI-specific .build-config. file.
Stamp contents from 13 to values such as cu13-gnu-debug.
Generated-source directories from build/cython/cu13 to build/cython/cu13-gnu-debug.
However, the nightly workflow remained unchanged and still:
cat cuda_core/build/.build-cuda-major
test -d cuda_core/build/cython/cu13
The failing job shows all three builds/imports succeeding. It correctly recorded these transitions:
None → cu13-gnu-debug
cu13-gnu-debug → cu12-gnu-debug
cu12-gnu-debug → cu13-gnu-debug
Only afterward did the obsolete cat command fail. Even if that check were updated, the following cu12/cu13 directory checks would fail for the same naming mismatch.
The regression boundary supports this conclusion: the September 29 scheduled run at 36e4d40 passed; the September 30 run at e8d9075, which includes 2f44c6b, was the first scheduled failure. The subsequent Cython-cache and uv commits do not explain this particular missing filename.
Recommended fix: update ci-pixi-source-test.yml (line 157) to discover the ABI-scoped .build-config.* stamp, expect cu13-gnu-debug, and check build/cython/cu12-gnu-debug plus cu13-gnu-debug.
Activity
github-actions commentedon Sep 30, 2026
Another unhealthy run
CI: pixi run test (source build) run #844, attempt 2 concluded with failure.
schedulemaine8d90752026-09-30T16:25:20Zjpascucci-nv commentedon Sep 30, 2026
Hi, @juenglin
My AI suggests that this change broke the nightly test.
The failure is a stale CI assertion, not a failed CUDA Core build.
The likely regression is commit 2f44c6b / PR #2903, “add CUDA_PYTHON_TOOLCHAIN override.” It changed:
However, the nightly workflow remained unchanged and still:
cat cuda_core/build/.build-cuda-major
test -d cuda_core/build/cython/cu13
The failing job shows all three builds/imports succeeding. It correctly recorded these transitions:
None → cu13-gnu-debug
cu13-gnu-debug → cu12-gnu-debug
cu12-gnu-debug → cu13-gnu-debug
Only afterward did the obsolete cat command fail. Even if that check were updated, the following cu12/cu13 directory checks would fail for the same naming mismatch.
The regression boundary supports this conclusion: the September 29 scheduled run at 36e4d40 passed; the September 30 run at e8d9075, which includes 2f44c6b, was the first scheduled failure. The subsequent Cython-cache and uv commits do not explain this particular missing filename.
Recommended fix: update ci-pixi-source-test.yml (line 157) to discover the ABI-scoped .build-config.* stamp, expect cu13-gnu-debug, and check build/cython/cu12-gnu-debug plus cu13-gnu-debug.
github-actions commentedon Oct 1, 2026
Another unhealthy run
CI: pixi run test (source build) run #861, attempt 1 concluded with failure.
schedulemain52e162b2026-10-01T03:16:14Z