bpo-44553: Correct failure in tp_new for the union object

[pablogsal]

https://bugs.python.org/issue44553

[pablogsal]

Before this PR:

❯ ./python -m test test_types -m test_or_type_operator_with_genericalias
0:00:00 load avg: 0.92 [1/1] test_types
Objects/object.c:1915: _Py_ForgetReference: Assertion failed: invalid object chain
Enable tracemalloc to get the memory block allocation traceback

object address  : 0x7ffff671f110
object refcount : 0
object type     : 0x555555a84dc0
object type name: tuple
object repr     : <refcnt 0 at 0x7ffff671f110>

Fatal Python error: _PyObject_AssertFailed: _PyObject_AssertFailed

after this PR:

❯ ./python -m test test_types -m test_or_type_operator_with_genericalias
0:00:00 load avg: 0.75 Run tests sequentially
0:00:00 load avg: 0.75 [1/1] test_types

== Tests result: SUCCESS ==

1 test OK.

Total duration: 224 ms
Tests result: SUCCESS

[pablogsal]

I'm landing this as soon as the tests pass to unblock the buildbots

[pablogsal]

@Fidget-Spinner For next occasions, the problem with this is that PyObject_GC_Del cannot be called like that because it overrides a bunch of cleanups like the call to _Py_ForgetReference and other possible handling.

[Fidget-Spinner]

Sorry bout that. I'll keep that in mind in the future.

BTW, does this mean this should be changed as well? https://github.com/python/cpython/blob/main/Objects/genericaliasobject.c#L654
I'm confused as to why it doesn't break the buildbots, or maybe because none of our tests make the object setup fail so PyObject_GC_Del is never called?

[pablogsal]

Yes, that has exactly the same problem. I assume is not being reaches by the tests and that's why we are not seeing corruption.

[miss-islington]

Thanks @pablogsal for the PR 🌮🎉.. I'm working now to backport this PR to: 3.10.
🐍🍒⛏🤖

[bedevere-bot]

GH-27009 is a backport of this pull request to the 3.10 branch.