Build for Linux AArch64 and other miscellaneous cleanups

[agriyakhetarpal]

This PR closes #288; it adds the new ubuntu-22.04-arm runner to build Linux AArch64/arm64 wheels. SSE2 and AVX2 x86-specific SIMD optimizations have been disabled at build time. This is in line with Spack downstream: spack/spack#40754. Additionally, the minimum macOS deployment target has been bumped from 10.9 to 10.13 (High Sierra), as CPython supports 10.13+ post 3.12.6, which is used by cibuildwheel's GitHub Action.

I've moved some code in setup.py to dedicated functions for now, but I agree with the direction in #464 wherein moving to a more modern build system in comparison to setuptools has been suggested; it sounds like a better long-term solution.

Additional context: https://github.blog/changelog/2025-01-16-linux-arm64-hosted-runners-now-available-for-free-in-public-repositories-public-preview/

TODO:

• Unit tests and/or doctests in docstrings
• Tests pass locally
• Docstrings and API docs for any new/modified user-facing classes and functions
• Changes documented in docs/release.rst
• Docs build locally
• GitHub Actions CI passes
• Test coverage to 100% (Codecov passes)

[joshmoore]

Thanks! Launched.

[codecov]

Codecov Report

All modified and coverable lines are covered by tests ✅

Project coverage is 99.96%. Comparing base (a2bdbe5) to head (912c207).

Additional details and impacted files

@@           Coverage Diff           @@
##             main     #710   +/-   ##
=======================================
  Coverage   99.96%   99.96%           
=======================================
  Files          63       63           
  Lines        2754     2754           
=======================================
  Hits         2753     2753           
  Misses          1        1           

[joshmoore]

Did not find Miniforge3-Linux-arm64.sh latest in cache

Perhaps see conda-incubator/setup-miniconda#385

[agriyakhetarpal]

Thanks! That should be fixed with 6b05be5.

Also, I'm sorry – I didn't notice that there is an open PR for this: #315. I should have checked! I noticed a few prior attempts when scrolling through #288, but I didn't see that PR as it wasn't linked to the issue 🙈

I would be happy to go forward with whatever is suggested.

[agriyakhetarpal]

In the meantime, I added a CHANGELOG entry for this change via 82a4565. I am happy to preserve authorship from the previous PRs or close my PR; either option is acceptable to me.

[joshmoore]

E   ImportError: /home/runner/work/numcodecs/numcodecs/numcodecs/blosc.cpython-311-aarch64-linux-gnu.so:
undefined symbol: pthread_atfork

[agriyakhetarpal]

Ah, thanks – I'll have to take a look at that; I don't immediately see why that happens. The only reference to pthread_atfork is in https://github.com/Blosc/c-blosc/blob/dcf6813d46f9fcbf7af4395b5536f886dfc76dd0/blosc/blosc.c#L2225, but https://anaconda.org/conda-forge/blosc/ and https://anaconda.org/conda-forge/python-blosc/ successfully ship AArch64, so maybe there's something to uncover there. The recommendation is to switch to blosc-2, though: Blosc/c-blosc#268, xref: #413, and that might require supporting either both or dropping blosc-1 because of the lack of forward compatibility.

[agriyakhetarpal]

I guess we aren't linking pthread properly, so the symbol isn't resolved.

[agriyakhetarpal]

@joshmoore, the tests should be fixed now. They ran on my fork and passed (see logs); the failure sign there is just because the Codecov token won't be set up in my fork's settings. Could you please relaunch the workflows here?

[joshmoore]

Launched.

[agriyakhetarpal]

Thanks! :) The Ubuntu-3.11 failure looks like a sporadic one.

[basnijholt]

Thanks @agriyakhetarpal, I am excited for this to land!

When launching the devcontainer of pipefunc building numcodecs takes like 99% of the time, even though I am installing ≈200 dependencies.

Hope this PR will have the momentum to get merged (unlike #315 which seems inactive).