Inconsistent name mangling in TypedDict in function and class forms

[sobolevn]

Bug report

Let's say that you have a dict like {"__key": 1} and you want to type it.

You can write:

>>> import typing
>>> class A(typing.TypedDict):
...     __key: int

>>> A.__mutable_keys__
frozenset({'_A__key'})

and:

>>> B = typing.TypedDict("B", [("__key", int)])
>>> B.__mutable_keys__
frozenset({'__key'})

Note that A mangles __key as a regular name. While B does not.

I guess that it is expected, but!
Docs (https://docs.python.org/3/library/typing.html#typing.TypedDict) does not say anything about this behavior. We only mention that functional form should be used for invalid identifiers. But, __key is a valid indentifier.
We don't have explicit tests for this either.
And Typing Spec does not mention this as well: https://typing.readthedocs.io/en/latest/spec/typeddict.html

So, what we can do:

• Do not mangle names in this case (hard and problematic: it can break existing stuff)
• Document and test current behavior (my vote)

Please, share your thoughts on this. And I willing to send a PR with the fix.

Linked PRs

• gh-129567: Add a note to typing.TypedDict docs about name mangling #130233
• [3.13] gh-129567: Add a note to typing.TypedDict docs about name mangling (GH-130233) #130841
• [3.12] gh-129567: Add a note to typing.TypedDict docs about name mangling (GH-130233) #130842

[picnixz]

I'm +1 for documenting this and linking to the section about mangling so that users know how to write a mangling/demangling helper if needs arise (if they don't want to use the functional syntax)

[ZeroIntensity]

This looks done :)