Inconsistent type substitution for `ParamSpec`

[sobolevn]

Bug report

Let's say we have this setup code:

from collections.abc import Callable as abc_Callable
from typing import ParamSpec, TypeVar, Callable
P = ParamSpec('P')
R = TypeVar('R')

This works correctly:

Callable[P, R][0, int]  # correctly raises:
# TypeError: Expected a list of types, an ellipsis, ParamSpec, or Concatenate. Got 0

abc_Callable[P, R][0, int]  # correctly raises:
# TypeError: Expected a list of types, an ellipsis, ParamSpec, or Concatenate. Got 0

We even have a test for this:

cpython/Lib/test/test_typing.py

Lines 2469 to 2477 in 6fb5f7f

	def test_type_subst_error(self):
	Callable = self.Callable
	P = ParamSpec('P')
	T = TypeVar('T')
	pat = "Expected a list of types, an ellipsis, ParamSpec, or Concatenate."
	with self.assertRaisesRegex(TypeError, pat):
	Callable[P, T][0, int]

But, this does not raise for some reason:

>>> Callable[P, int][0]
typing.Callable[[0], int]
>>> abc_Callable[P, int][0]
collections.abc.Callable[[0], int]

>>> Callable[P, int][0, int]
typing.Callable[[0, int], int]
>>> abc_Callable[P, int][0, int]
collections.abc.Callable[[0, int], int]

Somehow the presense of R makes the validation correct. But, when R is removed, we skip the ParamSpec validation.

I would like to work on this if we decide that this is a bug indeed. I remember that we want less runtime checks. But, technically this is a regression in one of them?

CC @JelleZijlstra

[sobolevn]

Ok, looks like I got confused, when we have Callable[P, R][0, int] what really happens is that we want to write Callable[P, R][[0], int] to be the same as Callable[P, int][0]