Scooby upwards of version 0.5.7 segfaults Python

[Huite]

Coming from: https://github.com/pyvista/pyvista/issues/1961

I get an immediate segfault when importing scooby, versions 0.5.8, 0.5.9, and 0.5.10.

I'll try and see if I can narrow it down a little more.

System information, for completeness (collected with 0.5.7 ^^):

--------------------------------------------------------------------------------
  Date: Thu Dec 23 16:28:20 2021 W. Europe Standard Time

                OS : Windows
            CPU(s) : 8
           Machine : AMD64
      Architecture : 64bit
               RAM : 31.9 GiB
       Environment : Python

  Python 3.8.12 | packaged by conda-forge | (default, Oct 12 2021, 21:22:46)
  [MSC v.1916 64 bit (AMD64)]

             numpy : 1.20.3
             scipy : 1.7.3
           IPython : 7.30.1
        matplotlib : 3.5.1
            scooby : 0.5.7
--------------------------------------------------------------------------------

[Huite]

0d1a3fd523d6e191b74874ba1e83757994ddf736 introduces the following lines in scooby.knowledge:

if psutil:
    # Code by https://stackoverflow.com/a/35291824/10504481
    my_path = str(Path(__file__).resolve())
    best_match = ""
    FILESYSTEM_TYPE = ""
    for part in psutil.disk_partitions():
        if my_path.startswith(part.mountpoint) and len(best_match) < len(part.mountpoint):
            FILESYSTEM_TYPE = part.fstype
            best_match = part.mountpoint
else:
    FILESYSTEM_TYPE = False

Turns out this is sufficient to segfault:

import psutil
psutil.disk_partitions()

I'll try some other Python versions on Windows as well.

[Huite]

Python 3.9.9 also crashes for me.

Downgrading psutil from 5.8.0 to 5.7.3 solves the issue for me.

[Huite]

Opened an issue on psutil repo as well:

https://github.com/giampaolo/psutil/issues/2038

[banesullivan]

Thank you for the thorough investigation and opening the upstream issue!

This seems very problematic and I'd like to get a patch in for this ASAP.

I may just add a check for windows after v5.7 to avoid calling the function that seg faults. Do you think that'd be okay?

[akaszynski]

I may just add a check for windows after v5.7 to avoid calling the function that seg faults. Do you think that'd be okay?

That's probably our best bet until this is fixed in psutil.

[Huite]

I just came home from the office, figured I'd have another look: I no longer get the segfault.

Looks like it's due to having a certain USB plugged in: https://github.com/giampaolo/psutil/issues/1953

PR is already merged: https://github.com/giampaolo/psutil/pull/1962

EDIT: on the other hand: https://github.com/giampaolo/psutil/issues/2003 Avoiding disk_partitions on Windows seems like a good idea for now.