Update open process types

[A5rocks]

This should resolve the issue mentioned in Gitter

[codecov]

Codecov Report

All modified and coverable lines are covered by tests ✅

Project coverage is 99.60%. Comparing base (0090581) to head (1688da0).
Report is 6 commits behind head on main.

Additional details and impacted files

@@           Coverage Diff           @@
##             main    #3076   +/-   ##
=======================================
  Coverage   99.60%   99.60%           
=======================================
  Files         121      121           
  Lines       17882    17882           
  Branches     3214     3214           
=======================================
  Hits        17811    17811           
  Misses         50       50           
  Partials       21       21           

Files with missing lines	Coverage Δ
src/trio/_subprocess.py	100.00% <100.00%> (ø)
src/trio/_tools/gen_exports.py	95.93% <100.00%> (ø)

[agronholm]

The updated documentation makes no mention of PathLike – shouldn't that be fixed?

[A5rocks]

I figured it's fine to have more accurate documentation (type hint) and less accurate but clearer (what's written in the docstring). Both are visible in the built docs.

I can make them the same though.

[A5rocks]

Huh https://github.com/python-trio/trio/actions/runs/10649671109/job/29520147803#step:4:208 shouldn't be happening. I'll assume it's transient but it seems like some sort of pip bug probably? Nevermind, it's not transient. Weird.

---

OK it's because @actions/checkout is pulling in the merged version of this PR (I think?), which contains the test-requirements.txt changes I just merged. I'm not clear on why that's leading PyPy to fail. And I cannot reproduce this locally with the same version of PyPy on Windows...

Something I noticed while investigating caching is I'm not sure it's helping much. In fact I think we'd probably be better off without it -- the cache key seems to change every single commit which makes no sense!!

[TeamSpen210]

The cache has been working a few days ago, this run for example. We'd definitely want it if possible. Could the problem be that the Cython build isn't using cache-dependency-path?

[A5rocks]

The cache has been working a few days ago, this run for example. We'd definitely want it if possible. Could the problem be that the Cython build isn't using cache-dependency-path?

It wasn't Cython, it was PyPy on Windows (specifically). It should be using the cache-dependency-path right because it has the same hash (c5182900c6da08ceac54233234be0232412d20dcd40abe0f886b892a46fd3c5e) within the commit for the run I linked. Maybe it's just using the hash from the current branch rather than what @actions/checkout pulls (which is the state were this PR merged)?

---

Here's my current thoughts: the action should only cache pip's cache directory: so there must be something bad there. However, as this run shows, it does actually look at the current test-requirements.txt file, presumably to get the hash.

Also, I'm struggling to understand why the "Setup python" step has a different hash in the cache key than the "Post Setup python" step in this run: https://github.com/python-trio/trio/actions/runs/10649416223/job/29519569640. (which I reran for debug logging and which... failed???)

[TeamSpen210]

It's very mystifying. Looking at the code, the hash should be stored here, and then reloaded here, so it shouldn't be affected if the requirements file was say edited. But I was thinking maybe there could be an issue if we had another action using setup-python internally. Not sure what though.

[CoolCat467]

Wondering, why is type var here called StrOrBytesPath? Wouldn't it be StrOrBytes, or is StrOrBytesPath special somehow?

[agronholm]

Because it's typed as str | bytes | PathLike[str] | PathLike[bytes]?

[CoolCat467]

Reading the docstring currently in place, I wasn't under the impression it supported pathlikes. If you could update that then, that would be great!

[A5rocks]

The docstring atm distinguishes what is possible with shell=True and what isn't, but I think paths are fine either way. I can update the docstring though. (this also appears in the function signature Sphinx makes)

[agronholm]

I would seem to be for the best to mention that, yes. After that, I have no objections.

[CoolCat467]

Looks great, I just have one question in a comment I made.