Several methods on built-in types are too strict.

I have identified several methods on built-in types that are too strict and can cause annoying false positives:

• set and frozenset methods (#15470, #15160)
• dict.{get, pop} (#15471)
• list.{count, index, remove} (#15472)
• MutableSet.discard
• Mapping.get
• MutableMapping.pop
• Sequence.{count, index}
• MutableSequence.remove
• probably some more.

It seems like the annotations for these were designed to help detect user errors. For instance, we know that .get on a dict[int, int] will always return the default as the types are incompatible, and so .get was restricted in typeshed to only accept the key type. However, this hurts, even in simple cases like using literals:

from typing import Mapping, Literal

def demo_get(
    d: Mapping[Literal["foo", "bar"], int],
    dynamic_key: str,
) -> None:
    # No overload variant of "get" of "Mapping" matches argument type "str"
    d.get(dynamic_key)  

def demo_sub(
    left: set[Literal["foo", "bar"]],
    right: set[str],
) -> None:
    # Unsupported operand types for - ("set[Literal['foo', 'bar']]" and "set[str]")
    left - right

I believe typeshed is trying to act as a linter here, which shouldn't be its job. Type-checkers and type-informed linters are better at this and can use more complex logic.

For example, in the set difference case set[A] - set[B], the proper test would be to check whether the types have a non-empty intersection or not. The same could be said for Mapping[K, V].get(K2).

Related Topics:

• https://github.com/python/typeshed/issues/8219
• https://github.com/python/typeshed/issues/9155
• https://github.com/python/typeshed/issues/9004

Start by locating the typeshed stubs for the listed dict, list, set, Mapping, Sequence, and Mutable* methods, then read related issues 8219, 9155, and 9004. The scope is not settled because the issue also mentions likely additional methods; done requires an agreed set of affected APIs and consistent annotation changes.