Plexon2 tests failing again with greater than 25 attempts to open files

[zm711]

Describe the bug

FAILED neo/test/iotest/test_plexon2io.py::TestPlexon2RawIO::test_read_all - OSError: Opening /home/runner/ephy_testing_data/plexon/NC16FPSPKEVT_1m.pl2 failed after 25 attempts.
FAILED neo/test/iotest/test_plexon2io.py::TestPlexon2IO::test_assert_read_neo_object_is_compliant - OSError: Opening /home/runner/ephy_testing_data/plexon/NC16FPSPKEVT_1m.pl2 failed after 25 attempts.
FAILED neo/test/iotest/test_plexon2io.py::TestPlexon2IO::test_create_group_across_segment - OSError: Opening /home/runner/ephy_testing_data/plexon/NC16FPSPKEVT_1m.pl2 failed after 25 attempts.
FAILED neo/test/iotest/test_plexon2io.py::TestPlexon2IO::test_read_with_lazy_is_compliant - OSError: Opening /home/runner/ephy_testing_data/plexon/NC16FPSPKEVT_1m.pl2 failed after 25 attempts.

copied since the action log will be deleted eventually.
https://github.com/NeuralEnsemble/python-neo/actions/runs/15005917152/job/42207865879
log link for now.

Expected behaviour
For our CI to pass :P

Environment:

• OS: Linux + Wine
• Python version 3.9 and 3.12
• Neo version: main
• NumPy version: 1.26 and 2.0

Additional context

This is a know flakiness of running plexon2 through wine on linux. We don't know why but the file fails to open. Normally it works within 25 tries, but suddenly 25 tries is not enough. We can wait a little bit of time and then try increasing the attempts to see if 35 is better, but we don't want the test suite to run forever since the longest part of our suite is installing wine anyway.

@h-mayorquin any thoughts?

[h-mayorquin]

I have no idea. It has not failed for us in a while but it is a really flaky test : /

[zm711]

Seems like it worked based on Luiz's most recent PR. Who knows.....

[h-mayorquin]

Mmm yeah, how strange that is consistently failing for you guys here : /

[cheydrick]

I think I have a fix for this. I don't know why it works, but it allows the tests to successfully run on my Ubuntu PC. I'll submit a PR shortly.

[zm711]

Just to follow up you're using wine? And you're just running our test suite to get it to pass? Or are you running plexon specific testing and can get it to pass?

[cheydrick]

Yes, using wine and the Neo test suite.

On one of my Ubuntu systems I never had issues opening a PL2 file (in the tests or otherwise), but on another I could not open any PL2 file until I realized the symlinks might be confusing the .dll and used .resolve() to convert the file name to an absolute path.

In my head it was a few weeks ago I did this, but now I see it was several months ago. I'm going to re-build my environment from scratch to make sure it all still works as I remember in the tests.

[zm711]

Thanks Chris. We appreciate it.

[cheydrick]

Mysteriously, I can get the plexon2rawio tests to pass once, but subsequent runs (usually) fail. I'm going to see if Nikhil can replicate it.

@nikhilchandra

[zm711]

Also @nikhilchandra, this file seems to be the one you shared for testing some different config options. Not sure if that could be contributing. It is always failing in at least one of our tests, but not always. We could try in your PR @cheydrick to raise the attempts to maybe 30 or 40 and see if that works for the test suite? Ideally we wouldn't want that value to creep up too much. As a non-Linux using person maybe someone else can comment on if Wine is switching to 64-bit or something like that? I think @JuliaSprenger or @h-mayorquin helped set up the wine part of the action so maybe they know a bit better?

[zm711]

One last point. This seems to be failing at our IO level and not at our RawIO level. So maybe what you're seeing @cheydrick is related to the fact that we run our IO tests after we run our rawio tests. Do you want to test running the IO tests by themselves and then run them again?

EDIT: Does the plexon2 DLL release the file handle appropriately? Since we are running this on Windows/Wine the file handle stuff is super finnicky. Windows will not let you touch a file within the same process if the handle has not be released (and even then sometimes it will misbehave). could the DLL be struggling with that for repeated testing? Don't know why that would suddenly become a huge problem so just spitballing.

[cheydrick]

Here's where I'm at:

I can repeatedly get the PL2 IO and raw IO tests to pass using Python 3.12.7 in a virtualenv environment.

I can get the PL2 IO or raw IO tests to pass once, but then subsequently fail, using Python 3.12.7 in a conda environment.

I think Nikhil and I have to dig deeper into the interaction between Wine and the DLL (and the handle behavior as you mentioned) to make any sense of this.

I suspect the reason I thought my small change fixed everything is because four months ago I was using a virtualenv environment!