Add cupy-cuda12x, nvidia-cuda-nvrtc-cu12, nvidia-cublas-cu12 to RAPID…#60
Add cupy-cuda12x, nvidia-cuda-nvrtc-cu12, nvidia-cublas-cu12 to RAPID…#60
Conversation
…S runenv. Update windows whitelists
There was a problem hiding this comment.
Code Review
This pull request empties several whitelist JSON files for Python 3.12.10 on Windows across the h5ad, rapids, sccoda, and general environments. This action contradicts the pull request title's objective of adding packages and is flagged as a potential error that could lead to regressions in the import checker for numerous packages.
I am having trouble creating individual review comments. Click here to see my feedback.
checker/whitelists/python-3.12.10/windows-x64.json (2-45)
This general whitelist for Python 3.12.10 on Windows is being completely emptied. This change affects a large number of packages (including torchvision, torchaudio, tensorflow, etc.). Given the PR title's focus on adding CUDA packages to the RAPIDS environment, clearing this general whitelist seems like a potential error or an over-broad cleanup that could lead to regressions in the import checker.
> Fix "Failed to create virtual environment" on all Windows CI builds virtualenv was unpinned, and 21.2.4 (released between April 10-22) breaks the virtualenv->venv copy trick used for embedded Python. All Windows checker jobs fail: venv creation silently exits non-zero. Pinning to 21.2.1 (last known working version).
> Restore whitelist entries that were accidentally cleared The previous commit emptied all Windows whitelist files, but these contain known import failures (matplotlib home dir, numba tbbpool, scipy circular import, etc.) that are expected on embedded Python. Rebuilt from actual CI checker output with glob patterns.
Add cupy-cuda12x, nvidia-cuda-nvrtc-cu12, nvidia-cublas-cu12 to RAPIDS runenv.