Errors when using type parameter defaults

[BaconPancakes]

Best expressed through code examples, with questions in comments.

Code sample in pyright playground

from typing import Any

class Bar[T=None]:
    ...

class AnotherBar[D]:
    ...

class MyClass[D, T=None]:

    def foo(self) -> type[Bar[T]]:
        return Bar # type error:  "type[Bar[None]]" is not assignable to "type[Bar[T@MyClass]]"

    def another_foo(self) -> type[Bar[T]]:
        return Bar[T]  # no type error, but this returns a Generic alias, not a type object
        # Why is this allowed? How can I enforce a type object return?

    def yet_another_foo(self) -> type[AnotherBar[D]]:
        return AnotherBar # no type error when no type default is used

    def workaround(self) -> type[Bar[Any]]:
        return Bar # no error, but slight loss in type specification

[erictraut]

If you provide an explicit specialization for a generic class (e.g. Bar[T] or Bar[Any]), then the type checker will honor that specialization. If you don't provide an explicit specialization, then the type checker will assume the default value for the argument. If no such default is specified, the default becomes Any.

The fact that Bar[T] or Bar[Any] is represented as a GenericAlias at runtime is an implementation detail of the type system in Python. For all intents and purposes it acts like the original class object. For example, reveal_type(list[int]) will reveal type[list[int]. It's not GenericAlias. You won't see GenericAlias appear anywhere within the Python typing spec.

[BaconPancakes]

Thanks for your responses.

So to be clear, in method 3, yet_another_foo is actually annotated as AnotherBar[Any] implictly? I thought pyright treated lack of specialization as Unknown?

[erictraut]

The typing spec says that AnotherBar (with no type argument) should be interpreted as AnotherBar[Any]. From the perspective of the typing spec, it's correct to say Any here. Unknown is the way that pyright spells "implicit Any", so you're correct to point out that Unknown is what you'll see if you're using pyright.