[CI] Run stubtest with Python 3.13

[srittau]

No description provided.

[srittau]

Initial analysis of the failures:

• cffi (Linux, Windows): cffi 1.16 - which our stubs target - doesn't support Python 3.13, was fixed in 1.17
  [cffi] Bump to 1.17.* #13653
  [cffi] Fix a few argument types for C code #13661
• corus (Linux): Python 3.13 not supported, last commit 2 years ago -> disabled stubtest
• fpdf2 (Linux): [fpdf2] Make fpdf.fonts.Glyph a dataclass #13645
• greenlet (Linux): [greenlet] Fix for Python 3.13 #13648
• hdbcli (Linux): Fix various argument of extension modules #13651
• humanfriendly (Linux): Last commit 4 years ago, doesn't support Python 3.13 -> disabling
• mysqlclient (Linux): Fix various argument of extension modules #13651
• passlib (Linux): [passlib] Fix for Python 3.13 #13649
• psycopg2 (Linux): Fix various argument of extension modules #13651
• pygit2 (MacOS, Windows, Linux): "fatal error: 'git2.h' file not found."
  scheduled for removal in May (Remove pygit2 (May 2025) #12787) -> disabling stubtest
• pytz (Linux): Make BaseTzInfo inherited methods pos-only #13650
• regex (Linux): Fix various argument of extension modules #13651
• seaborn (Linux): Add TypedDict.__{readonly,mutable}_keys__ #13646
• setuptools (Windows): Add TypedDict.__{readonly,mutable}_keys__ #13646
• simplejson (Linux): Fix various argument of extension modules #13651
• tensorflow (Linux): tensorflow 2.19 doesn't support Python 3.13, disabling for now
  (see also It doesn't support on python3.13 tensorflow/tensorflow#78774)
• tqdm (Linux): tensorflow 2.19 doesn't support Python 3.13, removing it from stubtest for now
  (see also It doesn't support on python3.13 tensorflow/tensorflow#78774)
• tree-sitter-languages (Linux): unmaintained and no wheels for 3.13 -> disabling
• WebOb (Linux): [WebOb] Add cgi_FieldStorage.make_file on Python 3.13+ #13654

[AlexWaygood]

* [ ]  fpdf2 (Linux): "fpdf.fonts.Glyph.__replace__ is not present in stub" -> needs investigation

I suspect this is due to mypy not synthesizing a __replace__ method for a namedtuple or dataclass on Python 3.13+. It's a py313+ feature. So it would have to be fixed in mypy (if it hasn't been already on mypy master).

* [ ]  seaborn (Linux): Multiple: "X.__{readonly,mutable}_keys__ is not present in stub" -> needs investigation

* [ ]  setuptools (Windows): Multiple: "X.__{readonly,mutable}_keys__ is not present in stub" -> needs investigationbob.compat.cgi_FieldStorage.make_file is not present in stub" -> needs investigation

Similarly, I suspect this is due to mypy not synthesizing __readonly__ and __mutable_keys__ ClassVar attributes for TypedDict classes? (Possibly a typing_extensions-only thing right now? Can't remember...)

[srittau]

I suspect this is due to mypy not synthesizing a __replace__ method for a namedtuple or dataclass on Python 3.13+. It's a py313+ feature. So it would have to be fixed in mypy (if it hasn't been already on mypy master).

In the case of fpdf2 it's a dataclass in the implementation, but not the stubs. Will fix.

[srittau]

pynput error currently in CI is unrelated: 1.8.1 was released today: #13647

[srittau]

I think that's it. Once the linked PRs are merged into main and merged back into this branch, this PR should be ready to go.

[AlexWaygood]

It's a bit weird that the "this is unmaintained" note was PR'd by the author of the competing package and merged without comment: grantjenks/py-tree-sitter-languages#84. But it's true that it can't be installed on Python 3.13 currently and hasn't had any significant updates since February 2024.

The stubs were added by @Akuli in #8548 and IIRC he's contributed to the upstream package as well

[srittau]

I haven't found out which package installs crypt.py, but I have had it happen once locally, and the same issue seems to happen in CI. (In this particular run. Previously, it complained that passlib.hosts.host_context was spurious. 🤷‍♂️)

[srittau]

A quick review about the not so fortunate changes that were necessary.

Don't support Python 3.13, but I expect support in the future:

• corus
• tensorflow
• tqdm: Module tqdm.keras only due to tensorflow

Disabled stubtest for these for now. I see this as less critical, as any update to the stubs would probably mean updating the stubs to the next version anyway, which will hopefully support Python 3.13, which means that we'll be able to re-enable stubtest.

Don't support Python 3.13 and are unmaintained:

• humanfriendly
• tree-sitter-languages

Finally, our version of pygit2 doesn't support Python 3.13, but newer version have added stubs, so our pygit2 stubs are scheduled for removal anyway.

[AlexWaygood]

I'm okay with this as long as we open an issue reminding us to revert the corus, tensorflow and tqdm changes once they support Python 3.13. Thanks!

[srittau]

I'm okay with this as long as we open an issue reminding us to revert the corus, tensorflow and tqdm changes once they support Python 3.13. Thanks!

My idea was to add a warning to stubsabot's created PR when it updates a stubs package where stubtest is skipped.

[AlexWaygood]

My idea was to add a warning to stubsabot's created PR when it updates a stubs package where stubtest is skipped.

It actually already does that: #13460 (comment)

Maybe that's enough. I don't feel strongly; we can go without the issue if you don't think it's necessary :-)

[srittau]

I fear that an issue just gets lost among the ~200 issues we have. And then rediscovered a few years down the line when it's not relevant anymore. I think the stubsabot warning will be more effective.