Repository navigation
[BUG]: CI with Coverage fails on some cuda_core/tests/test_build_hooks #3344
Description
Is this a duplicate?
- I confirmed there appear to be no duplicate issues for this bug and that I agree to the Code of Conduct
Type of Bug
Runtime Error
Component
cuda.core
Describe the bug
There are two tests that are failing on both Linux and Windows.
Coverage (Linux): cuda_core/tests/test_build_hooks.py#L440
TestGeneratedSourceDirIsKeyed.test_dir_is_anchored_not_relative_to_cwd AssertionError: assert False + where False = is_absolute() + where is_absolute = PosixPath('.').is_absolute
Coverage (Windows): cuda_core\tests\test_build_hooks.py#L440
TestGeneratedSourceDirIsKeyed.test_dir_is_anchored_not_relative_to_cwd AssertionError: assert False + where False = is_absolute() + where is_absolute = WindowsPath('.').is_absolute
Coverage (Linux): cuda_core/tests/test_build_hooks.py#L429
TestGeneratedSourceDirIsKeyed.test_majors_use_different_dirs AssertionError: assert PosixPath('.') != PosixPath('.')
Coverage (Windows): cuda_core\tests\test_build_hooks.py#L429
TestGeneratedSourceDirIsKeyed.test_majors_use_different_dirs AssertionError: assert WindowsPath('.') != WindowsPath('.')
How to Reproduce
Run CI: Coverage in cuda-python CDCI.
Expected behavior
Those asserts look weird, so not that.
Operating System
No response
nvidia-smi output
No response
Activity
jpascucci-nv commentedon Nov 3, 2026
Additional info from Codex Sol 6.1:
PR #2472, “build: make Cython/extension artifacts CUDA-major aware,” merged August 27, 2026, introduced these two tests and their coverage failures. Merge commit: 9cab1db.
I verified both exact assertion failures in the first nightly run afterward:
The cause is a test setup oversight: coverage sets CUDA_PYTHON_COVERAGE=1, which deliberately makes build_dir=".". The new tests assume ordinary build behavior, so both CUDA majors return the same relative directory—breaking both assertions.
These failures can coexist with green coverage jobs because the test steps use continue-on-error: true. I also confirmed them in a later successful run.
jpascucci-nv commentedon Nov 4, 2026
This should be fixed by #3354