Hey folks, thanks for all the review and sorry for the extended delay here, I was out on PTO. I've addressed most of the comments around adding some TODOs and replied to the rest.

CodSpeed Performance Report

Merging #84 will not alter performance

_{Comparing lysnikolaou:freethreading-support (e36539b) with master (4fe87bc)}

Summary

✅ 4 untouched benchmarks

Codecov Report

Attention: Patch coverage is 93.75000% with 1 line in your changes missing coverage. Please review.

Project coverage is 88.99%. Comparing base (4fe87bc) to head (e36539b).
Report is 22 commits behind head on master.

@@            Coverage Diff             @@
##           master      #84      +/-   ##
==========================================
- Coverage   89.02%   88.99%   -0.04%     
==========================================
  Files          18       18              
  Lines         738      745       +7     
  Branches       75       75              
==========================================
+ Hits          657      663       +6     
- Misses         63       64       +1     
  Partials       18       18              

☔ View full report in Codecov by Sentry.
📢 Have feedback on the report? Share it here.

Gentle ping on this. Is there anything I can do to move it forward?

I would suggest waiting for non-alpha Cython release. A Release Candidate is ok, but alpha is not.

It's a compile-time experimental feature on cpython, I don't think anybody is expecting us to have a rock solid supported implementation. It's essentially a preview release at this point, so alpha is fine at this stage (and the only thing we can offer).

If it is experimental, py3.13t should allow failures in build&tests, only py3.13 should use Cython alpha, and all stable builds should use stable Cython.
I worry a little about existing build stability; it would be sad if Cython alpha prevented us from publishing a new propcache release.

The discussion above is to use it only on the 3.13t release. Stable Cython is used on other versions, so this has already been addressed.

@Dreamsorcerer I'm sorry; I've missed that the new config is applied to 3.13t only.
There are no objections from me anymore.

The problem with this is that there's sadly no environment marker for free-threading in the constraint file (though there are plans for it). This creates a conflict between the specified build requirement and the Cython constraints for regular 3.13. Any ideas on how you'd like to proceed here?

This creates a conflict between the specified build requirement and the Cython constraints for regular 3.13. Any ideas on how you'd like to proceed here?

What all other projects that have released cp313t wheels do here is to set the requirements for 3.13 in pyproject.toml and other metadata file for the default CPython build (i.e., 3.0.11 now, not 3.1.0a1) and disable build isolation to produce cp313t wheels, installing Cython alpha/nightly in the wheel build job. I think that's the only good option at the moment, or something equivalent like pointing at a different constraints file (requirements/cython_freethreading.txt ?).

Thanks @rgommers for the pointer! This is what an older verion of this PR did, though my understanding was that the maintainers did not like that. If everyone is okay with this, I can certainly revisit that approach.

a different constraints file (requirements/cython_freethreading.txt ?).

@lysnikolaou I do like this approach. It'd be nice to granularly override it just for 3.13t.

I've update this PR to use the exact same approach as yarl (around building wheels). If someone could approve the workflows, that'd be really helpful.

I'm working on fixing the CI now

Once done, feel free to reach out on matrix or discord any time you need the CI kicked

Thanks @lysnikolaou

I'm doing release now. 🤞 that we don't hit qemu issues again

Manually checked all the workflows. Build looks good. Going to release it out now