gh-336: support Python 3.13

[Saransh-cpp]

• Add Python 3.13 classifier
• Test on Python 3.13 in CI
• Support for Python 3.13 in noxfile

Closes: #336

[Saransh-cpp]

Version 3.13 was not found in the local cache
Error: The version '3.13' with architecture 'x64' was not found for Ubuntu 22.04.
The list of all available versions can be found here: https://raw.githubusercontent.com/actions/python-versions/main/versions-manifest.json

I am not sure where the cache is coming from. I have tried re-running this multiple times and have also deleted the existing caches from GitHub Actions, but it somehow does not work. I will try re-running this later today, and maybe create an issue in setup-python if it still does not work? (Other pure-python repositories that I work on are not encountering this issue).

cc: @paddyroddy

[paddyroddy]

It will take some time for 3.13 to become available across GitHub

[Saransh-cpp]

Ah, healpix does not have 3.13 binaries on PyPI and we don't allow pip to build it from source -

glass/.github/test-constraints.txt

Line 11 in 1631757

--only-binary healpix

I think this will have to wait till 3.13 wheels of healpix are available on PyPI.

cc: @ntessore

[paddyroddy]

Ah, healpix does not have 3.13 binaries on PyPI and we don't allow pip to build it from source -

We should definitely wait until all dependencies have 3.13 support. It is quite common for the ecosystem to take a while to catch up.

[ntessore]

healpix for cp313 has been uploaded

[paddyroddy]

Just marking as ready to see what happens in CI

[ntessore]

Also built flt wheels for cp312 and cp313. Excluding healpy, are these all of our native extension dependencies done?

[paddyroddy]

Also built flt wheels for cp312 and cp313. Excluding healpy, are these all of our native extension dependencies done?

I think so, hard to tell really unless we comment out healpy

[ntessore]

Still waiting on a new healpy release does make me wonder if our CI wouldn't be better served by using conda ... But this probably causes a lot of issues with tox/nox?

[paddyroddy]

Still waiting on a new healpy release does make me wonder if our CI wouldn't be better served by using conda ... But this probably causes a lot of issues with tox/nox?

Unless I've missed something, conda alone won't sort 3.13 support? It's still on the same version v1.17.3 https://anaconda.org/conda-forge/healpy, so will still need the library fixing.

Some libraries take a long time to support a new release, for example check out numba/numba#9197.

A different question is do we need healpy as well as healpix?

[ntessore]

Unless I've missed something, conda alone won't sort 3.13 support? It's still on the same version v1.17.3 https://anaconda.org/conda-forge/healpy, so will still need the library fixing.

But conda has already provided 3.13 binaries for healpy 1.17.3 – healpy maintainers could add those to PyPI, like I did for healpix and flt.

Some libraries take a long time to support a new release, for example check out numba/numba#9197.

In my experience, that is only numba, which is understandable given what they have to go through.

A different question is do we need healpy as well as healpix?

The healpix package only deals with the HEALPix discretisation as such, but doesn't provide a spherical harmonic transform. Moving away from healpy for the latter is a long-term goal for GLASS, since healpy doesn't work on Windows (and many students come with Windows machines). In fact, I am looking forward to moving towards a generic Array API backend, since that is a very natural point to make that switch.

[paddyroddy]

But conda has already provided 3.13 binaries for healpy 1.17.3 – healpy maintainers could add those to PyPI, like I did for healpix and flt.

Ah yes, I see that now.

In my experience, that is only numba, which is understandable given what they have to go through.

You're very lucky, you've only had delays with numba!

[paddyroddy]

Still waiting on a new healpy release does make me wonder if our CI wouldn't be better served by using conda ... But this probably causes a lot of issues with tox/nox?

I do think we want to use whatever you anticipate end users using. I don't use conda these days in favour of uv.