Where do these .coverage.hostname.number.number files come from?

[danijar]

I cant remember changing anything but recently Pytest with Pytest-cov started creating a lot of file in the working directory in addition to the expected .coverage file.

.coverage.danijar-xps13-arch.18737.903083
.coverage.danijar-xps13-arch.19190.316378
.coverage.danijar-xps13-arch.19192.651456
.coverage.danijar-xps13-arch.19197.538345
.coverage.danijar-xps13-arch.19199.138180
.coverage.danijar-xps13-arch.19212.482496
.coverage.danijar-xps13-arch.19213.812768
.coverage.danijar-xps13-arch.19301.688885
...

Where do there files come from?

[ionelmc]

They appear if you use subprocesses or xdist.

[ionelmc]

They don't get removed at the end of the testrun?!

[danijar]

Ah, that's a useful bit of information. Yes, I use subprocess and on error, I sys.exit() so they probably cannot be cleaned up. Do you have an idea? Here is the code inside setup.py.

[ionelmc]

I mean, if you use subprocesses in your tests, then pytest-cov will automatically start measuring coverage in those processes (if they are python of course). That's what is making those files (besides using xdist).

[danijar]

Oh, I don't use subprocess in my tests, just in my setup.py to run pytest.

[ionelmc]

Does it reproduce without the setup.py shim?

[danijar]

Just tried that. The files are still created when I just run python -m pytest --cov=layered test or py.test --cov=layered test. Pytest version 2.8.5 and Pytest-cov version 2.2.0.

[ionelmc]

What's your coverage version?

[danijar]

Coverage is version 4.0.3. Pip says all of the three packages are up-to-date. Update: I just noticed messages of this form in the Pytest report:

Coverage.py warning: Couldn't read data from '/home/danijar/repository/layered/.coverage.danijar-xps13-arch.22012.752324': CoverageException: Doesn't seem to be a coverage.py data file

[ionelmc]

Is there a repository that reproduce the issue?

[danijar]

Yes, cloning danijar/layered and running python3 setup.py test reproduces the problem.

[iAnanich]

Have the same issue when using asyncio with ProcessPoolExecutor.

At the testing, it is a lot of .coveage.<..> files, but 2 of them always not get removed after tests complete.

python 3.5.3, coverage==4.4.1 pytest==3.0.6 pytest-cov==2.5.1

[ionelmc]

So I've looked at @danijar code and I bet @iAnanich has the same problem: unclean multiprocessing pool shutdown. You should have a pool.join() or similar before the test using the pool is done.

[danijar]

Thanks for looking into this!

[drorata]

I have a similar issue when testing a piece of code which uses GridSearchCV. Under the hood, joblib.Parallel is used and I guess/understand that it somehow messes up the files generated.

Any idea how to resolve this issue?

[ionelmc]

@danijar pls open up a separate issue and give some details.