Use --universal flag with uv pip compile

[CoolCat467]

Follow up to #2958

We might want to try out this new --universal flag (I'm not sure when it was added? but seems useful)

... but that can be a followup. This is good as a drop in replacement.

Originally posted by @A5rocks in #2958 (review)

This pull request is adding the new --universal flag for the uv pip compile runs.

[codecov]

Codecov Report

All modified and coverable lines are covered by tests ✅

Project coverage is 99.63%. Comparing base (0c88bc7) to head (cd7af07).
Report is 115 commits behind head on main.

Additional details and impacted files

@@           Coverage Diff           @@
##             main    #3032   +/-   ##
=======================================
  Coverage   99.63%   99.63%           
=======================================
  Files         120      120           
  Lines       17830    17830           
  Branches     3204     3204           
=======================================
  Hits        17765    17765           
  Misses         46       46           
  Partials       19       19           

[A5rocks]

The changes look good, I can't speak of whether they're accurate/missing things, but definitely something nice. I'm fine merging this despite there not being many docs about it, but once this is documented more we can probably close the issue RE: replacing pip-compile.

[A5rocks]

Nice!

[A5rocks]

I'm not sure this is stable enough to use as is. I ran uv pip compile --universal --no-strip-markers --python-version=3.8 --upgrade test-requirements.in -o test-requirements.txt twice locally and it started duplicating requirements in the requirements file.

See astral-sh/uv#4885

[A5rocks]

Ok so while investigating that I figured out that I can't fix it (the fix would require too much change in uv I think).

However! 1) the reason we have different results is because this wasn't a problem prior to 0.2.19, and 2) a workaround (because we want to stay current) is to just put Sphinx at the top of the requirements file. Though I haven't double checked for 2.

[CoolCat467]

While looking through the documentation, I came across https://github.com/astral-sh/uv-pre-commit, might be interesting to try out using that.

[jakkdl]

While looking through the documentation, I came across https://github.com/astral-sh/uv-pre-commit, might be interesting to try out using that.

I think we want to bump dependencies in pull requests separate from other PRs, so adding a pre-commit check that plays around with requirements sounds like a bad idea to me.

[CoolCat467]

I think there is a case of test flakiness again, the merge run failed on pypy-3.10 with a dtls issue again:

   ______________________ test_gc_before_system_task_starts _______________________
  
      @pytest.mark.filterwarnings("always:unclosed DTLS:ResourceWarning")
      async def test_gc_before_system_task_starts() -> None:
          e = endpoint()
      
  >       with pytest.warns(ResourceWarning):
  E       Failed: DID NOT WARN. No warnings of type (<class 'ResourceWarning'>,) were emitted.
  E        Emitted warnings: [].
  
  /opt/hostedtoolcache/PyPy/3.10.14/x64/lib/pypy3.10/site-packages/trio/_tests/test_dtls.py:757: Failed

I think something similar to this happened recently elsewhere as well, might need to make an issue collecting test flakiness for pypy-3.10

[A5rocks]

Maybe we just have to bump GC counts yet again. (assuming that calls gc_harder)