bpo-46730: Fix refleak and tighten NULL checks

[tiran]

PyType_GetQualName returns a new reference.

Signed-off-by: Christian Heimes christian@python.org

https://bugs.python.org/issue46730

[sweeneyde]

I'll try to be more cautious about ref-leaks in the future. Thanks for catching this!

[bedevere-bot]

🤖 New build scheduled with the buildbot fleet by @markshannon for commit a22802e 🤖

If you want to schedule another build, you need to add the ":hammer: test-with-buildbots" label again.

[tiran]

@sweeneyde The buildbots have not finished verifying this PR yet. Next time please wait until all buildbots are successful.

[sweeneyde]

Ah, my apologies, I was just looking at the refleak bots. I will wait next time. Thank you for the help!

[Alex-Blade]

Oh, apologies, should have checked that

[geryogam]

@tiran In which case could you have the last error message "AttributeError: property has no getter" that you introduced in the else clause? In other words, when does PyType_GetQualName(Py_TYPE(obj)) return NULL in C?

I can’t get that error message by making hasattr(type(obj), "__qualname__") return False in Python:

classM(type): def__getattribute__(self, name): raiseAttributeErrorclassA(metaclass=M): passobj=A() hasattr(type(obj), "__qualname__") # returns Falseproperty().__get__(obj) # raises AttributeError: property of 'A' object has no getter

It seems that property().__get__(obj) uses vars(type)["__qualname__"].__get__(type(obj)) (which returns 'A') and not type(obj).__qualname__ (which raises AttributeError) to get the __qualname__ attribute of type(obj).

So is your else clause dead code?