The mypyc TypedDict test failures are due to the latest typing_extensions release, not this PR: #15542

According to mypy_primer, this change doesn't affect type check results on a corpus of open source code. ✅

This is quite minor, so I am going to merge this later today or tomorrow unless there are comments/objections.

This seems to cause a regression with dataclasses. In particular None types are striped from Callable types (both arguments and return types) even with --strict-optional set.

from dataclasses import dataclass
from typing import Callable, assert_type

@dataclass
class Description:
    name_fn: Callable[[int | None], str | None]


def f(d: Description) -> None:
    assert_type(d.name_fn, Callable[[int | None], str | None])  # E: \
        # Expression is of type "Callable[[int], str]", not "Callable[[int | None], str | None]"

mypy --strict-optional test.py

Yeah, this is annoying, let me do something to prevent this kind of things once and for all.

@cdce8p See #15571