Closed rmartin16 closed 4 weeks ago
The Windows failure definitely isn't transient; it's persisted through multiple re-runs, include full reruns after a rebase.
In a desparate search for anything that could be causing this, I noticed that pytest-xdist is installed; I removed it, and the problem went away. I don't know why xdist would be the problem, other than the fact that multi-threading a test suite could cause some interesting cleanup side effects.
Tagging @rmartin16 in case you have strong opinions to the contrary; my inclination is to drop xdist for the test suite if it's going to be more stable.
my inclination is to drop xdist for the test suite if it's going to be more stable.
Would tox -e py-fast
still work?
Moreover, though, what if you roll back the recent version bump for pytest-xdist? They reworked how process forking works in the latest version.
my inclination is to drop xdist for the test suite if it's going to be more stable.
Would
tox -e py-fast
still work?
I guess not - so there'd need to be other cleanups.
Moreover, though, what if you roll back the recent version bump for pytest-xdist? They reworked how process forking works in the latest version.
Well that'd do it... I'll try and see.
Moreover, though, what if you roll back the recent version bump for pytest-xdist? They reworked how process forking works in the latest version.
Well that'd do it... I'll try and see.
Looks like it doesn't - 3.5.0 generates the same error.
So - my inclination is to drop xdist. While parallel support is nice to have, it doesn't yield that much of a performance benefit; plus, dropping the extra dependency is one less cause of error, and one less dependency to keep up to date. Thoughts?
Moreover, though, what if you roll back the recent version bump for pytest-xdist? They reworked how process forking works in the latest version.
Well that'd do it... I'll try and see.
Looks like it doesn't - 3.5.0 generates the same error.
well, my guess is coverage
is doing something weird....at least, if my previous investigation showing this unclosed file is coming from coverage
is correct.
So - my inclination is to drop xdist. While parallel support is nice to have, it doesn't yield that much of a performance benefit; plus, dropping the extra dependency is one less cause of error, and one less dependency to keep up to date. Thoughts?
I do normally use tox -e py-fast
...since it is about 3x faster (although, tox -e py
only takes 18s for me...).
One more plea perhaps...what if we just don't install pytest-xdist
for 3.13? And if the problem still exists when 3.13 is released, we drop it completely then.
One more plea perhaps...what if we just don't install
pytest-xdist
for 3.13? And if the problem still exists when 3.13 is released, we drop it completely then.
I can live with that for now. It's also possible that this might be fixed in 3.13-b2, or with a coverage bump at some point; if we get close to 3.13-final and this is still lingering, we can re-evaluate then.
@rmartin16 Looks like the py3.13 exclusion on xdist has worked; If you're happy, feel free to merge.
Changes
license
field is specified inpyproject.toml
Notes
PR Checklist: