Regression involving overloaded classmethods

[msullivan]

The following fails now:

from typing import Optional, Type, TypeVar, overload, Union

Gid = int

A = TypeVar("A", bound="Assoc")

class Assoc:

    @overload
    @classmethod
    def delete(cls, gid, gid2):
        # type: (Type[A], Gid, Gid) -> Optional[int]
        pass

    @overload  # noqa
    @classmethod
    def delete(cls, gid, gid2=None):
        # type: (Type[A], A, None) -> Optional[int]
        pass

    @classmethod  # noqa
    def delete(cls, gid, gid2=None):
        # type: (Type[A], Union[A, Gid], Optional[Gid]) -> Optional[int]
        pass


def delete_assoc(assoc_type):
    # type: (Type[Assoc]) -> None
    assoc = assoc_type()
    assoc.delete(10, 20)

def delete_assoc2(assoc):
    # type: (Assoc) -> None
    assoc.delete(10, 20)

with errors in the delete_assoc functions saying Invalid self argument "Assoc" to attribute function "delete" with type "Callable[[Type[A], int, int], Optional[int]]".

Looks like this was introduced in #7860.

[ilevkivskyi]

Oh, this one is pretty annoying, I always forget a class method can be a Decorator or an OverloadedFuncDef. Should be easy to fix.