gh-128632: fix segfault on nested __classdict__ type param

[tom-pytel]

Add reserved name check in symtable.c:symtable_visit_type_param_bound_or_default() for __class__ and __classdict__ to prevent segfault when __classdict__ is used.

• Issue: NULL ptr deref in _PyCode_ConstantKey when compiling code #128632

[bedevere-app]

Most changes to Python require a NEWS entry. Add one using the blurb_it web app or the blurb command-line tool.

If this change has little impact on Python users, wait for a maintainer to apply the skip news label instead.

[JelleZijlstra]

Could you also test the following related cases:

• Using a generic function or type alias instead of a class with these names for the type parameters
• Using the names __classcell__ and __classdictcell__, which appear internally in the implementation.

[tom-pytel]

Added func and alias tests. Didn't add __classcell__ or __classdictcell__ tests as those are not excluded and don't cause problems currently. Do you want to add them to the forbidden list?

>>> class A:
...     class B[__classcell__]: print(type(__classcell__))
...     
<class 'typing.TypeVar'>
>>> class A:
...     class B[__classdictcell__]: print(type(__classdictcell__))
...     
<class 'typing.TypeVar'>

[JelleZijlstra]

Thanks for adding the tests! I want the *cell* names to also be tested, because they're also somewhat likely to cause problems, even if they don't currently crash.

[tom-pytel]

So I interpret 'somewhat likely to cause problems' to mean yes disallow these names (and test them).

[JelleZijlstra]

Since they work fine now we should not disallow them, especially because we'd want to backport this change. However, we should have a test to make sure future changes don't introduce crashes with these names.

[tom-pytel]

Names re-allowed and tests added. Also re-allowed __class__ as that doesn't crash either even though it is special.

[tom-pytel]

Just a reminder, according to @ZeroIntensity this needs backporting to 3.13 and 3.12.

[ZeroIntensity]

I'm going to yield to Jelle. This seems complex enough that backporting could be too risky.

[tom-pytel]

I'm going to yield to Jelle. This seems complex enough that backporting could be too risky.

Had a look and doesn't look too bad, same functions are present to fix.