False positive when overriding an async method

[JelleZijlstra]

from typing import Any, Callable, Awaitable

AnyFunction = Callable[..., Awaitable[Any]]

def retry() -> Callable[[AnyFunction], AnyFunction]:
    def decorator(fn: AnyFunction) -> AnyFunction:
        return fn
    return decorator

class B:
    async def run(self) -> bool:
        return True

class C(B):
    @retry()
    async def run(self) -> bool:
        return True

with master produces

$ /home/jelle/mypy/venv/bin/python -m mypy untitled.py 
untitled.py:16: error: Signature of "run" incompatible with supertype "B"

This is a regression since 0.600.

[ilevkivskyi]

OK, this is an actual type error, you should either update the AnyFunction to Callable[..., Coroutine[None, None, Any]] or use a type variable with current type as a bound. I would recommend the second one, I always use it myself and it gives tighter types, with AnyFunction = TypeVar('AnyFunction', bound=Callable[..., Awaitable[Any]]) the code passes.

This is a bit verbose, but I think @gvanrossum proposed that we should actually infer the most precise types for async defs. Possible solution is to define an alias like NativeCoro = Coroutine[None, None, T].

[JelleZijlstra]

Thanks for clarifying! So the issue is that in the base class run is now inferred as (roughly) Callable[..., Coroutine], and in the child class it is the more general type Callable[..., Awaitable].

I feel like this should perhaps be allowed (just like we allow overriding a function with a different Callable), but it doesn't seem like a big deal.

[hauntsaninja]

I think mypy is behaving as expected