Inconsistent narrowing for builtins.type and typing.Type

[tyralla]

The following difference is due to Mypy handling typing.Type via TypeType, for which it does not try to create an intersection, and builtins.type as an Instance, for which it does:

# flags: --python-version 3.9 --warn-unreachable

from typing import Type

t1: Type
t2: type
class C: ...

if isinstance(t1, C):
    reveal_type(t1)  # E: Statement is unreachable
if isinstance(t2, C):
    reveal_type(t2)  # N: Revealed type is "__main__.<subclass of "type" and "C">"

I encountered this problem when working on #16330, which affects type checking a sphinx function.

Interestingly, things are consistent when annotating explicitly with Any:

# flags: --python-version 3.9 --warn-unreachable

from typing import Any, Type

t1: Type[Any]
t2: type[Any]
class C: ...

if isinstance(t1, C):
    reveal_type(t1)  # E: Statement is unreachable
if isinstance(t2, C):
    reveal_type(t2)  # E: Statement is unreachable

So, Mypy maybe misunderstands t2: type?

[JelleZijlstra]

See also https://discuss.python.org/t/inconsistencies-between-type-and-type/37404/2.

[tyralla]

So, it's not 100% clear that type should be treated as type[Any]. (If I understand correctly, the alternative would be type[object].) Okay, then I wait to finish #16366 until this decision is final.

[tyralla]

@JelleZijlstra

As the discussion came to a halt and it was only a little work, I completed the pull request. I think it agrees with your and Guido's thinking about how plain type annotations should be understood. Would you like to review it?