can't narrow a union of type[None]

[ghost]

Bug Report

There's something about type[None] that (understandably) confuses mypy and makes it start producing nonsense results.

While defining a function that takes arguments of type[None | A | B], mypy is able to narrow to type[None|A] (or type[None|B]) but any attempt to narrow beyond that causes no change in type. This makes it impossible to produce an exhaustive if/elif/else that can assert_never in the else case.

To Reproduce

https://mypy-play.net/?mypy=latest&python=3.11&gist=7b278f392c623c63fd5b576c483a0e3c

Expected Behavior

I should be able to write an elif clause that can assert_type(cls, type[None]).
Actual Behavior

Currently there's no way to narrow a type Union[type[None], T] type. (That I found.)

Your Environment

https://mypy-play.net

[erictraut]

I agree this is a bug in mypy, but there were a few bugs in the sample code. Here's an updated sample that should pass type checking.

import types
import typing_extensions as t

NoneType = type(None)

def type_map(cls: type[None | int | str]) -> None:
    if issubclass(cls, int):
        t.assert_type(cls, "type[int]")
        return
    else:
        t.assert_type(cls, type[None | str])

    if issubclass(cls, str):
        t.assert_type(cls, "type[str]")
        return
    else:
        t.assert_type(cls, type[None])

    t.assert_type(cls, "type[None]")
    if issubclass(cls, str):
        t.assert_never(cls)
    else:
        t.assert_type(cls, type[None])

    if cls is None:
        t.assert_never(cls)
    else:
        t.assert_type(cls, type[None])

    if issubclass(cls, type(None)):
        t.assert_type(cls, type[None])
    else:
        t.assert_never(cls)

    if issubclass(cls, NoneType):
        t.assert_type(cls, type[None])
    else:
        t.assert_never(cls)

    if issubclass(cls, types.NoneType):
        t.assert_type(cls, type[None])
    else:
        t.assert_never(cls)

Incidentally, pyright had the same bug as mypy here. It will be fixed in the next release of pyright.

[tyralla]

Thanks for the detailed test case. It helped me to discover three scenarios where Mypy currently fails due to different limitations:

from typing import Type, Union

def f(cls: Type[Union[None, int]]) -> None:
    if issubclass(cls, int):
        reveal_type(cls)  # note: Revealed type is "Type[builtins.int]" (right)
    else:
        reveal_type(cls)  # note: Revealed type is "Union[Type[None], Type[builtins.int]]" (wrong)

from typing import Type, Union

def f(cls: Type[Union[None, int]]) -> None:
    if issubclass(cls, type(None)):
        reveal_type(cls)  # note: Revealed type is "Union[Type[None], Type[builtins.int]]" (wrong)
    else:
        reveal_type(cls)  # note: Revealed type is "Union[Type[None], Type[builtins.int]]" (wrong)

from typing import Type, Union

NoneType_ = type(None)

def f(cls: Type[Union[None, int]]) -> None:
    if issubclass(cls, NoneType_):  # error: Parameterized generics cannot be used with class or instance checks  [misc] (wrong)
                                    # error: Argument 2 to "issubclass" has incompatible type "<typing special form>"; expected "_ClassInfo"  [arg-type] (wrong)
        reveal_type(cls)  # note: Revealed type is "Union[Type[None], Type[builtins.int]]" (wrong)
    else:
        reveal_type(cls)  # note: Revealed type is "Union[Type[None], Type[builtins.int]]" (wrong)

I provided pull request #16315. If accepted by the maintainers, these limitations should be fixed in the next Mypy release, too.

[ghost]

Thanks, all!