pypy and pypy3 Exception warnings upon closing

[nicoddemus]

We get a lot of ignored exceptions when the interpreter closes:

ResourceWarning: unclosed file <_io.FileIO fd=1114 mode='rb' closefd=True>
Exception ignored in: <_io.FileIO fd=1115 mode='rb' closefd=True>

At first glance it seems the capture plugin is not closing files properly.

[blueyed]

That are quite a lot - have you checked older logs? I do not remember seeing that many, i.e. it might be a recent regression.

[nicoddemus]

have you checked older logs

I haven't... indeed I don't recall seeing so many

[blueyed]

Builds on master/features also started to fail: https://travis-ci.org/pytest-dev/pytest/branches

[blueyed]

https://travis-ci.org/pytest-dev/pytest/builds/537998957 (master-into-features PR) was OK, let's see how its merge comes out.

[blueyed]

As for the CI failures I suspect this is related to the lastest pluggy maybe?! (at least some failures I've looked at are related to entrypoint-related tests). /cc @asottile

[asottile]

yes, master failures are expected

releasing 4.6 or merging this will fix master: #5325

[pierreglaser]

FYI, I am getting recent failures on loky using pypy with pytest >= 4.6. The root cause seems to be unclosed file descriptors, so it seems related to this issue.

[asottile]

I found the root cause of this by the way:

• issue: https://bitbucket.org/pypy/pypy/issues/3021/ioopen-directory-leaks-a-file-descriptor
• PR: https://bitbucket.org/pypy/pypy/pull-requests/647/fix-leak-of-file-descriptor-with-_iofileio/diff

the same thing is affecting pre-commit's CI as well :S

[blueyed]

Awesome.

Can we remove pypy3 until there is some progress / fix in this regard?
Nobody is paying attention to real failures there anyway, and it slows down CI in general, especially now with it being run last due to allowed-failures.

[nicoddemus]

Can we remove pypy3 until there is some progress / fix in this regard?

I agree. We should test it again when a new pypy3 release comes out.