Some GC types can crash during destruction if a GC runs executes

[pablogsal]

We have been bitten a couple of times already if C extension types that implement the GC interface don't call PyObject_GC_UnTrack in their tp_dealloc slot. The last issue was #95045

Although this is technically not always an error (the tp_free slot untracks the object from the GC), if the GC executes during the deallocator is possible that it will see an object with 0 reference count.

To avoid this in the future, we should add a warning if this happens so we can fix it within CPython

[nascheme]

Although this is technically not always an error (the tp_free slot untracks the object from the GC), if the GC executes during the deallocator is possible that it will see an object with 0 reference count.

Also, perhaps the object structure contains invalid pointers (created while running tp_dealloc) that can be traced by tp_traverse. So this warning is a good idea.