gh-102615: Fix type vars substitution of collections.abc.Callable and custom generics with ParamSpec

[sobolevn]

I would prefer to create two PRs to fix #102615

1. This one that fixes a bug with Callable and custom generics with ParamSpec
2. gh-102615: Use list instead of tuple in repr of paramspec #102637 to change the repr

• Issue: Representation of ParamSpecs at runtime compared to Callable #102615

[Fidget-Spinner]

I encountered another bug, the __repr__ needs to be updated as well:

>>> ca.Callable[[int, P], int][P][int]
collections.abc.Callable[[int, (<class 'int'>,)], int]

EDIT: THIS MIGHT NOT BE A BUG, AS ITS AN INVALID USE. PLEASE IGNORE THIS @sobolevn

[sobolevn]

@Fidget-Spinner I would like to move all repr related problems to another PR if that possible :)

I've only touched params substitution here.

[sobolevn]

Wait a moment, this is not about repr, this is an __args__ containing a tuple 😒

>>> ca.Callable[[int, P], int][P].__args__
(<class 'int'>, ~P, <class 'int'>)
>>> ca.Callable[[int, P], int][P][int].__args__
(<class 'int'>, (<class 'int'>,), <class 'int'>)

[sobolevn]

Funny thing, I just removed this code from _collections_abc.py as an experiment:

if (len(self.__parameters__) == 1
                and _is_param_expr(self.__parameters__[0])
                and item and not _is_param_expr(item[0])):
            item = (item,)

All tests still pass :)

🤯

[AlexWaygood]

I think the runtime is correct in raising TypeError for these situations, actually: see #102615 (comment)

Maybe instead, we should focus on improving the error message, so that users get more specific feedback on why the substitution is being disallowed? Though if it adds too much complexity to the code, I probably wouldn't bother with it.

[sobolevn]

I am going to close this for now, I will add new tests in other PRs, because this one has a big unrelated history.