Windows 3.13.7 build regression `ModuleNotFoundError: No module named '_posixsubprocess'`

[burgholzer]

We are happily using uv and the Python distributions shipped through this project in all of our CI.
Over at https://github.com/cda-tum/mqt-problemsolver we noticed a regression in the latest release that affects our CI. More specifically, it affects Python 3.13 on Windows.

• Successful build a week ago downloading cpython-3.13.6-windows-x86_64-none

• First failing build two days ago dowloading cpython-3.13.7-windows-x86_64-none

The precise error is

path = 'D:\\a\\mqt-problemsolver\\mqt-problemsolver\\.nox\\tests-3-13\\Scripts\\python.exe'
args = ['D:\\a\\mqt-problemsolver\\mqt-problemsolver\\.nox\\tests-3-13\\Scripts\\python.exe', '-c', 'from multiprocessing.resource_tracker import main;main(15)']
passfds = [12, 15]

    def spawnv_passfds(path, args, passfds):
>       import _posixsubprocess
E       ModuleNotFoundError: No module named '_posixsubprocess'

The same test suite runs perfectly fine on Ubuntu 24.04, macos-13, macos-14, as well as on Windows with Python 3.9-3.12.

How to reproduce:

git clone https://github.com/cda-tum/mqt-problemsolver
cd mqt-problemsolver
uv run --python 3.13.7 pytest

Changing to

uv run --python 3.13.6 pytest

makes the test pass.

Anything apparent to you that could be the cause for this?
Thanks for your help!

[xmatthias]

I ran into exactly the same problem this morning, with CI passing yesterday until 3.13.7 was released (seemingly, as the last test passing was with 3.13.6, and downgrading to it fixes it, too).

My problem so far is - i couldn't reproduce it "standalone" (without my project's tests) - as in - my standalone actions either fail on both 3.13.6 and 3.13.7, or work on both (so far, trying with python -c "import _posixsubprocess" as well as python -c "from multiprocessing.resource_tracker import main" - which seems to be the culprit).

[zanieb]

The only suspicious change off the top of my head is #756 but I would be surprised if that was it?

[zanieb]

(Thank you for the reports!)

[zanieb]

Have you confirmed this does not reproduce on the official Windows binaries? (from python.org)

@geofft just linked me to python/cpython#137738

[geofft]

I wonder if this is python/cpython#131788 / python/cpython#137738 which already has some reports that the backport is causing issues.

[zanieb]

Yeah this looks like an upstream bug

See also

• Issues in resource tracker for Python 3.13.7: ModuleNotFoundError: No module named '_posixsubprocess' on Windows and warnings Cannot register X for automatic cleanup: unknown resource type Y on posix joblib/loky#459

[burgholzer]

Have you confirmed this does not reproduce on the official Windows binaries? (from python.org)

I personally haven't, sorry for that. Would have probably led me to the other issues.

Maybe it's still worth to keep this issue open until the upstream issue is resolved and a new version is out? Feel free to close it though if you feel like that's the better plan.

[zanieb]

It's fine to keep this open. We'll track a fix upstream in python/cpython#138031

[lesteve]

FYI: a fix for this issue has been released yesterday in loky 3.5.6 and joblib 1.5.2 (that vendors loky 3.5.6), see joblib/loky#459 and joblib/loky#461 for more details.

[zanieb]

I don't think there's more for us to do here. Thanks!