python3-Cython: update to 3.0.1.

[ahesford]

This is a big update that reflects several years of extensive work, and needs some testing to be sure it doesn't break any Void packages.

Testing the changes

• I tested the changes in this PR: in progress

The following packages need to be fixed to build with new Cython:

• gbinder-python
• jack_mixer
• libimobiledevice
• memray
• pyliblo
• python3-cypari2
• python3-cysignals
• python3-efl
• python3-fpylll
• python3-grpcio
• python3-h5py
• python3-jq
• python3-logbook
• python3-lxml
• python3-mpi4py
• python3-pandas-msgpack
• python3-peewee
• python3-pplpy
• python3-pyFFTW
• python3-pygame
• python3-scikit-learn
• python3-scipy
• python3-shapely
• python3-tables
• python3-uvloop
• python3-yaml
• sagemath
• urh
• vapoursynth
• vidcutter

[tornaria]

Related: sagemath/sage#29863

I'll have a look into (at least) my pkgs: cypari2, cysignals, fpylll, pplpy, sagemath. I can also have a look into scipy as it's needed for sagemath.

Question: could we have cython2 and cython3 coexist for a while?

[ahesford]

The two versions of Cython can't coexist without a lot of pain because they will conflict in the Python package namespace. We can just hold this update until everything is resolved.

[icp1994]

pyyaml realeased v6.0.1 which blocks building with Cython 3.0.0

[ahesford]

Yeah, I'll bump this and patch around it if necessary.

[tornaria]

scipy needs patching to build with pybind 2.11 (or maybe --skip-dependency-check since all the patches are about relaxing dependencies (cython, numpy, and now pybind11).

Note that in their main branch they have relaxed the upper bound for cython and pybind11 (they still fix numpy to a particular version). See: https://github.com/scipy/scipy/blob/main/pyproject.toml

[tornaria]

Suggested change

	+ "language_level": 2
	+ "language_level": 2,

My bad, that's what I get for hand-editing the patch after checking... This should fix it.

[ahesford]

@tornaria I decided it wasn't worth trying to patch around all of these packages and added a new python3-Cython0.29 which conflicts with python3-Cython and offers the "legacy" Cython 0.29.36. The legacy package can be a build-time dependency for the ~30 packages that have yet to support Cython 3. Only two packages have run-time dependencies on Cython: urh and sagemath. I have moved both the build-time and run-time dependencies of urh to python3-Cython0.29 for now. How should I handle sagemath? We can do the same if you think that is the best course of action.

[tornaria]

@tornaria I decided it wasn't worth trying to patch around all of these packages and added a new python3-Cython0.29 which conflicts with python3-Cython and offers the "legacy" Cython 0.29.36. The legacy package can be a build-time dependency for the ~30 packages that have yet to support Cython 3. Only two packages have run-time dependencies on Cython: urh and sagemath. I have moved both the build-time and run-time dependencies of urh to python3-Cython0.29 for now. How should I handle sagemath? We can do the same if you think that is the best course of action.

I think we have a way forward with sagemath (sagemath/sage#36109 is almost done with review and sagemath/sage#36110 is small). I think it might be possible to have sagemath running on cython 3 within the week.

[ahesford]

With python3-pplpy built with python3-Cython0.29 (which is its current state in the repository), all sagemath tests in the --long suite pass (with the exception of some new warnings emitted by Pythran), so this is good to go; merged as cb4c843.

[tornaria]

With python3-pplpy built with python3-Cython0.29 (which is its current state in the repository), all sagemath tests in the --long suite pass (with the exception of some new warnings emitted by Pythran), so this is good to go; merged as cb4c843.

sagemath/sage#36109 was merged a few hours after you merged this, and that is the only "long" part of the move to cython 3.

The actual patch needed on top of that to be able to use cython 3 is just 6 lines (plus 19 lines in doctests due to some changed strings). See #45887 for details.