overloads in `if TYPE_CHECKING` shouldn't require implementations

[A5rocks]

Feature

As I explained in the title.

Pitch

Currently this raises an error:

from typing import TYPE_CHECKING, overload

if TYPE_CHECKING:
    @overload  # E: An overloaded function outside a stub file must have an implementation
    def f(x: int) -> str:
        ...
    
    @overload
    def f(x: str) -> bytes:
        ...

I think that if code is in if TYPE_CHECKING then it shouldn't need an implementation though.

[Avasam]

Isn't TYPE_CHECKING meant to be transparent to type-checkers? The above can be done as:

from typing import TYPE_CHECKING, overload

if TYPE_CHECKING:
    @overload
    def f(x: int) -> str: ...
    @overload
    def f(x: str) -> bytes: ...
    def f(x: int | str) -> str |  bytes: ...

mypy doesn't really know either if you meant the above, or if you legitimately misplaced your overloads/forgot the implementation:

from typing import TYPE_CHECKING, overload

if TYPE_CHECKING:
    # Imagine this is an overload that uses types or decorators that would fail at runtime (older python or stub-only symbols)
    @overload
    def f(x: int) -> str: ...
    @overload
    def f(x: str) -> bytes: ...
    
def oops_wongly_placed() -> None:
    return None

def f(x):
    if isinstance(x, int):
        return str(x)
    else:
        return x.encode()

[A5rocks]

I mean maybe it's supposed to be transparent?

I think empty body errors get disabled in there. I think in general it should be treated more like stubs.

[uko3211]

checker.py

According to the comment on that line, it seems the error for this case was intended not to be shown.