add free-threading support

[kumaraditya303]

This adds free-threading support by adding freethreading_compatible=True in loop.pyx and also enables CI and wheels for 3.14t.

cc @fantix

[Vizonex]

Don’t mind by me asking but do you have the time to look into winloop as well? Or will I have to mirror everything and guess with blind faith?

[achimnol]

👀 Also looking forward to MagicStack/asyncpg having free-threading support...

[Vizonex]

👀 Also looking forward to MagicStack/asyncpg having free-threading support...

Same

[kumaraditya303]

Don’t mind by me asking but do you have the time to look into winloop as well? Or will I have to mirror everything and guess with blind faith?

There isn't anything to guess with blind faith here, you can follow the guide at https://py-free-threading.github.io/ci/ to add free-threading CI to projects you wish. I have never looked at winloop before so I'll not have time for that atm.

[1st1]

Would be nice to add a test (should be possible with uv) that it does actually work cc @fantix

[kumaraditya303]

Would be nice to add a test (should be possible with uv) that it does actually work cc

The existing tests pass on free-threading 3.14t CI which I added as part of this PR (the failing one test is a flaky test), do you want a test where uvloops are run in parallel threads?

[fantix]

I added a race-condition test, which is confirmed to fail without freethreading_compatible=True. @1st1

Many thanks to @kumaraditya303 for putting this altogether!

[Vizonex]

@kumaraditya303 I'm going to start mirroring these changes going on. No promises of when I'll finish unless It gets done in a rational amount of time. If needed I will open a uv virtual environment with the free-threading settings and version 3.14 set and testing locally.

[Vizonex]

Good News: I have successfully mirrored the changes from here, winloop 0.3.0 will release soon since I now obey how version numbers actually work and since the __init__.py script for winloop significantly changed. Thanks for your help again.

Bad News: uvloop also seems there's some rounding bugs with macos from the tests that I've seen I am also question why my library has the same problem. But maybe if we find a way to fix it I'll try doing the same with winloop's end. Rounding has always been a bit of a problem for a while (not sure why exactly).

[fantix]

Yeah, it's definitely getting more flaky to run tests, and there're a few warnings to take care of. Let's release for Python 3.14 first, and get back to those as follow-ups.

[Vizonex]

Yeah, it's definitely getting more flaky to run tests, and there're a few warnings to take care of. Let's release for Python 3.14 first, and get back to those as follow-ups.

Good idea.