Inconsistent behaviour of Optional in regards to Type[T]

[septatrix]

🐛 Bug Report

Optional[str] is valid as an argument to cast, however invalid as a value of a variable of type Type.

To Reproduce

from typing import *

reveal_type(cast)

a = cast(7, True)

b = cast(Optional[int], a)
reveal_type(b)

c: Dict[str, Type]
c = {'foo': Optional[str]}
reveal_type(c)

d = {'bar': Optional[str]}
reveal_type(d)

$ mypy opt.py	
opt.py:3: note: Revealed type is 'Overload(def [_T] (typ: Type[_T`-1], val: Any) -> _T`-1, def (typ: builtins.str, val: Any) -> Any)'
opt.py:5: error: Invalid type: try using Literal[7] instead?
opt.py:8: note: Revealed type is 'Union[builtins.int, None]'
opt.py:11: error: Dict entry 0 has incompatible type "str": "object"; expected "str": "Type[Any]"
opt.py:12: note: Revealed type is 'builtins.dict[builtins.str, Type[Any]]'
opt.py:15: note: Revealed type is 'builtins.dict[builtins.str*, builtins.object*]'
Found 2 errors in 1 file (checked 1 source file)

Expected Behavior

The assignment to c passes type checking.

Actual Behavior

Optional[str] is apparently recognized as builtins.object and the assignment to c get recognized as an invalid assignment. However the call to cast (which itself does get typechecked) passes validation but also expects a Type.

Your Environment

• Mypy version used: mypy 0.782
• Mypy command-line flags: none
• Mypy configuration options from mypy.ini (and other config files): nothing
• Python version used: Python 3.8.2
• Operating system and version: Ubuntu 20.04.1 LTS

[gvanrossum]

Quick answer: Type expects a class, not other forms of type -- in particular, Union is not allowed.

If you have more questions around this, please ask on Gitter.

[septatrix]

Yes but if this is the case I do not understand why using Optional[int] as the first argument to cast does not generate an error.

[septatrix]

But why does this work for cast then? Is this an error in the typeshed?

[gvanrossum]

Please ask on Gitter.

[septatrix]

For anyone else stumbling across this:
The issue was due to python/typeshed#4810 and what I was trying to is not yet possible AFAIK until we have something like #9773.