MAINT: vendor Tempita in scipy/_build_utils

[rgommers]

• Addresses a part of BUG: generate_f2py mod is called by the wrong interpreter #20535.
• Purpose: make build more robust, stop relying on import Cython but only on invoking the cython executable
• Context for purpose: to make cross compilation better, we want to stop invoking the Python interpreter we're building for. Hence, we should avoid having to use/import things at build time outside of Python's stdlib. We're almost there, only a few places left where we invoke custom_target(..., command: [py3, ...]) in meson.build files. Running Tempita is one of them.

This is an unmodified copy from Cython/Tempita/ as of commit 023d4af35 in Cython's master branch (21 April 2024). Tempita hasn't been maintained independently on PyPI for 10+ years; we've used the Cython version which is semi-public (undocumented) for a while, with the note that we should vendor it if that ever fails. It now failed once (see #20535), and conceptually that makes sense - this is the only place where we don't invoke the cython executable but actually do import Cython. That can fail in a distro setup where Cython is installed for a different Python interpreter than the active one.

License

Opening as a draft PR to clarify the licensing situation of this code first:

• The original Tempita was MIT-licensed, see https://pypi.org/project/Tempita/ (a single declaration, no full copy of the MIT license text in the sdist)
• That code was vendored in Cython in 2011, without making a note of the license: https://github.com/cython/cython/commits/master/Cython/Tempita
• Cython itself is Apache 2.0 licensed (not compatible with SciPy's BSD-3 license). However, there have been no substantial changes to the Tempita code since 2011 it looks like, but rather commits are minor maintenance/style change (running 2to3, pyupgrade, etc.). So I think we should be able still consider this code as MIT-licensed - but that requires confirmation from @scoder I think, as Cython lead dev and author of most of the commits that touched this code.
  • @scoder are you fine with including that Cython/Tempita/ code under the original MIT license, or do you now consider it mixed MIT/Apache licensed?
  • Note that NumPy also vendored a copy until not long ago, and that does have an MIT license file: https://github.com/numpy/numpy/tree/maintenance/1.26.x/tools/npy_tempita. Using that would be the alternative. It is less well maintained than the Cython code, so we'd need to polish it a bit.

TODO:

• Clarify license situation
• Update LICENSES_bundled.txt

[scoder]

IANAL, but I looked through the changes and IMHO none of them would hand any non-trivial chunk of code over to other authors than in the original copy of Tempita that we included and possibly me. Almost all change were trivial adaptations or removals, many of them machine generated.

I hereby grant you the right to use my changes under the original code's MIT license, or the 3-clause BSD license that SciPy currently uses.

[rgommers]

Thanks @scoder, much appreciated!

[rgommers]

I knew the lint CI job was going to fail; python dev.py lint calls the tools/lint.py --diff-against which seems to not honor the exclude files in lint.toml. I'm not planning to reformat this code, so plan to just ignore the failure (it goes away by itself post-merge, at least until someone touches the file again). @lucascolley is this a known issue with the linter perhaps?

[lucascolley]

is this a known issue with the linter perhaps?

Right, I agree with your diagnosis. I can't think of an obvious refactor. It seems we would have to extract the exclude field from lint.toml in lint.py to perform set difference on the list of files. I'm happy to just let this fail.

Alternatively, maybe ruff check --exclude=x.py x.py y.py should only lint y.py.

[lucascolley]

never mind, here's the one-line (okay, two-line) fix!

[lucascolley]

LGTM!

[rgommers]

Thanks for the review @lucascolley. Going to hit the green button here, so I can work on the next step.