pytest is stuck in teardown when using uvloop

[sk-]

First of all, thanks for the all the work put to release v1.0.0. It was really easy to migrate from v0.21. However we found a weird issue when using uvloop (based on the docs).

Using asyncio´s loop we had no issues and test would run normally. However, when using uvloop, tests would get silently stuck at the final teardown. After a lot of debugging I could track the issue to _provide_event_loop, which triggers a ResourceWarning in uvlopp's __uv_walk_close_all_handles_cb.

This warning is caught by pytest's catch_unraisable_exception, so maybe this issue will get resolved with pytest-dev/pytest#12958.

What made the issue hard to understand was the fact that no information was printed out, and that it only fails when warnings are treated as errors. I removed error from filterwarnings and the test passed without problem, but also without printing out the warnings.

At the end we kept errors in filterwarnings and added the entry ignore:unclosed resource <TCPTransport:ResourceWarning.

Below is a debugging capture of the warning producing the issue

System info:
Python version: Python 3.12.9 (main, Feb 12 2025, 15:09:19) [Clang 19.1.6 ] on darwin
Pytest: 8.3.5
Pytest-asyncio: 1.0.0

[seifertm]

Huh, interesting. Thanks for reporting this!

And the feedback on the migration is also much appreciated!

Is there a chance you have a minimal reproducer of the issue? A small piece of code for me to play around with would help a lot.

[sk-]

@seifertm Unfortunately I was not able to narrow down what was triggering the warning as this was only happening for certain tests.