ResourceWarning: unclosed file <_io.TextIOWrapper name='/dev/null' mode='r' encoding='UTF-8'>

[blueyed]

I've seen the following on full run of the test suite:

============== 1 failed, 2270 passed, 83 skipped, 13 xfailed in 271.31 seconds ==============
Exception ignored in: <_io.FileIO name='/dev/null' mode='rb' closefd=True>
ResourceWarning: unclosed file <_io.TextIOWrapper name='/dev/null' mode='r' encoding='UTF-8'>
Exception ignored in: <_io.FileIO name=13 mode='rb+' closefd=True>
ResourceWarning: unclosed file <_io.FileIO name=13 mode='rb+' closefd=True>
Exception ignored in: <_io.FileIO name=15 mode='rb+' closefd=True>
ResourceWarning: unclosed file <_io.FileIO name=15 mode='rb+' closefd=True>
Exception ignored in: <_io.FileIO name='/dev/null' mode='rb' closefd=True>
ResourceWarning: unclosed file <_io.TextIOWrapper name='/dev/null' mode='r' encoding='UTF-8'>
Exception ignored in: <_io.FileIO name=19 mode='rb+' closefd=True>
ResourceWarning: unclosed file <_io.FileIO name=19 mode='rb+' closefd=True>
Exception ignored in: <_io.FileIO name=21 mode='rb+' closefd=True>
ResourceWarning: unclosed file <_io.FileIO name=21 mode='rb+' closefd=True>

Related to #3879 in that it would be nice if those would cause CI to fail.

I have not investigated what triggered them.
This was on 8baeb90.

[asottile]

It's going to be hard to turn these into failures -- these are exceptions triggered as the interpreter is shutting down. I think the best we can do is fix them and hope they don't come back.

From what I can tell, the only opening of devnull in the project is in FDCaptureBinary. I suspect ~something is making module-level references to an instance of these (which gets cleaned up at interpreter shutdown).

[RonnyPfannschmidt]

lets just do a gc.collect in session shutdown to see more information, perhaps even after each test (perhaps as part of the lsoff plugin)

[asottile]

A gc.collect isn't necessarily going to catch this. It's likely module globals holding onto these :/

[blueyed]

Is it maybe coming from inner pytest runs?
Or would those show up with the particular tests then only?

[asottile]

mayyybe, but probably much less likely (unless they're complex second or third generation gc objects and the tests that leak them happen to be at the end). I've really only seen the "Exception ignored ..." errors during module teardown though 🤷‍♂️

[blueyed]

=> #5006