GitHub Actions get stuck when using fs fixture and importing torch

[mniebisch]

Hey,

I'm using pyfakefs for a project. While writing tests locally, everything worked like a charm.
At some point, I pushed my code and applied it to GitHub Action. During the test phase of my workflow, the tests just got stuck, and the workflow didn't terminate.

I tracked the problem down to a conflict between the fs fixture and the import of torch.
If torch is imported, all tests are executed until a test uses the fs fixture for the first time.

I reproduced the problem. You can find it here https://github.com/mniebisch/check_pyfakefs_github_action/runs/7429852799?check_suite_focus=true

Additionally, here's a screenshot showing that everything works just fine locally.

I hope this isn't a known issue. My Google search didn't turn up any help. If so, I'm sorry :)

[mrbean-bremen]

Thanks for the report! No, I don't know of this issue, and I have no idea what causes this so far.
I will have a closer look after I fix a couple of regression bugs that came up recently, so this may take a bit...

[mrbean-bremen]

@mniebisch - Can you check if it is fixed with master?

[mniebisch]

@mrbean-bremen - I checked with pyfakefs 3.6.3. Unfortunately, it doesn't work.

[mrbean-bremen]

@mniebisch - it is not integrated into a release yet - can you please check with master?
e.g.

pip install git+https://github.com/jmcgeheeiv/pyfakefs.git@master

[mniebisch]

@mrbean-bremen Ah, my bad :D Seems like I need a coffee. Your fix works like a charm. Thanks a lot for the fast support :) What caused the problem

[mrbean-bremen]

That's good! I'm not completely sure what caused it as I could not reproduce it locally - I just found the commit and the specific changes that caused this in CI. It seems that accessing __name__ in non-module objects did not behave as expected (e.g. instead of raising an exception it caused the test to hang). I reintroduced the module check using inspect, that I had removed as a performance optimization some time ago. The performance seems not to be affected anyway, at least for newer Python versions.
I will check if I can resolve another open issue and probably make another patch release afterwards.