The errors we are hitting here are the reason why we didn't do the full shutdown in the past. They only occur when running the full suite in sequence (I'm trying to find the minimal permutation that "works", but it's tricky).

The minimal call I have found so far that reproduces the segfault is

pytest --runtime mono tests/test_method.py tests/test_subclass.py

EDIT: Even further:

pytest --runtime mono -k "test_create_instance"

Code for this test:

EDIT: Latest result:

If the whole subclass suite is run, there is no crash. If the whole test-suite is run without the subclass, there is no crash either. Pretty :)

Just excluding test_create_instance works reproducibly as well.

EDIT: The exact line creating the leak(?) is

@lostmsu This runs through now, but we are not doing the PyObject_GC_Del anymore. Where is the corresponding object creation?

@filmor you would be looking for callers of PythonDerivedType.SetPyObj

I know it is annoying, but can you untangle the fix for the issue from the syntax changes?

Also, I think you might have accidentally changed line endings in the interop .tt file.

BTW, re the original issue: I haven't looked at it yet, but it might have to do with the default tp_dealloc, tp_clear, or (unlikely) tp_traverse being inadequate for derived type instances with their GC untrack "magic". This situation could probably be detected in the finalizer or during shutdown.

Superseded by #1712