ensure dask_cuda.__git_commit__ is populated

[jameslamb]

Installing dask-cuda like this:

pip install \
  --extra-index-url=https://pypi.anaconda.org/rapidsai-wheels-nightly/simple/ \
  'dask-cuda==25.4.*,>=0.0.0a0'

The __git_commit__ attribute on the main module isn't populated:

python -c "import dask_cuda; print(dask_cuda.__git_commit__)"

The way this should work is that rapids-build-backend writes a file dask_cuda/GIT_COMMIT which is then read by this code:

dask-cuda/dask_cuda/_version.py

Lines 20 to 28 in 412ef58

	try:
	__git_commit__ = (
	importlib.resources.files(__package__)
	.joinpath("GIT_COMMIT")
	.read_text()
	.strip()
	)
	except FileNotFoundError:
	__git_commit__ = ""

I think that what's happening here is this:

• rapids-build-backend is writing that file
• the file is not being packaged, because this project uses setuptools + a MANIFEST.in, and that MANIFEST.in does not include that file

This proposes the following:

• add GIT_COMMIT to MANIFEST.in
• update RAPIDS-specific pre-commit hooks to their latest versions (not related, but might as well, while we're using a CI run anyway)

Notes for Reviewers

Helpful reference for this... "Controlling files in the distribution" from the setuptools docs: https://setuptools.pypa.io/en/latest/userguide/miscellaneous.html

[copy-pr-bot]

Auto-sync is disabled for draft pull requests in this repository. Workflows must be run manually.

Contributors can view more details about this message here.

[jameslamb]

/ok to test

[TomAugspurger]

Here's a run from a previous PR: https://github.com/rapidsai/dask-cuda/actions/runs/13332958081/job/37252862171. There's no GIT_COMMIT in the build output.

The build output from this PR is at https://github.com/rapidsai/dask-cuda/actions/runs/13439129864/job/37548802662?pr=1453#step:11.

There's a warning about no files matching GIT_COMMIT, which is maybe understandable since it's generated during the build? But we do have dask_cuda/VERSION symlinked to VERSION https://github.com/rapidsai/dask-cuda/blob/branch-25.04/dask_cuda/VERSION.

Anyway, what we really care about is that the built wheel has it, and it looks like it does:

(base) toaugspurger@dgx12:/tmp/a$ curl -LO "https://downloads.rapids.ai/ci/dask-cuda/pull-request/1453/c2cd9ba/dask-cuda_wheel_python_dask-cuda.tar.gz"
...
(base) toaugspurger@dgx12:/tmp/a$ tar xvf dask-cuda_wheel_python_dask-cuda.tar.gz
./
./dask_cuda-25.4.0a23-py3-none-any.whl
(base) toaugspurger@dgx12:/tmp/a$ unzip -l dask_cuda-25.4.0a23-py3-none-any.whl | grep GIT_COMMIT
       41  2025-02-20 15:49   dask_cuda/GIT_COMMIT

[jameslamb]

Yeah thank you for looking carefully at that! I think I can explain what's happening there.

The key point is this from the setuptools docs:

You can think about the build process as two stages: first the sdist will be created and then the wheel will be produced from that sdist.

(docs link)

The first pass through, creating an sdist, is what emits these warnings:

reading manifest template 'MANIFEST.in'
warning: no files found matching 'dask_cuda/GIT_COMMIT'

(build link)

That happens around here in setuptools: https://github.com/pypa/setuptools/blob/ba243756233d0afe944db6e02ddcb70064dcd22c/setuptools/_distutils/command/sdist.py#L342

And at that point, that GIT_COMMIT file doesn't exist because it hasn't been created yet.

But then later on, during wheel-building, those rules in the manifest are executed one at a time.

writing manifest file 'dask_cuda.egg-info/SOURCES.txt'
copying dask_cuda/GIT_COMMIT -> build/lib/dask_cuda
copying dask_cuda/VERSION -> build/lib/dask_cuda

(build link)

And by there, the file exists and gets packaged.

[TomAugspurger]

Looks great, thanks!

[jameslamb]

/merge