[match-case]: `TypedDict` interaction with mapping pattern

[randolf-scholz]

It seems match-cases mapping pattern is predestined to work well with TypedDict, but currently the spec is silent regarding TypedDict subtyping with respect to match case: https://typing.python.org/en/latest/spec/typeddict.html#subtyping-with-mapping

At the time of writing, not a single type checker (tested with mypy=1.19.1, pyright=1.1.408, ty=0.0.17, pyrefly=0.53.0 and zuban=0.6.0) can solve the following tests¹, suggesting that some standardization / typing spec guidance may be useful.

from typing import TypedDict, assert_type, assert_never

class D(TypedDict):
    name: str
    value: int

def test_match_key(x: D | int) -> None:
    match x:
        case {"name": _}:
            # runtime equivalent to (isinstance(x, Mapping) and "name" in x)
            assert_type(x, D)
        case _:
            assert_type(x, int)

def test_match_key_and_value(x: D | int) -> None:
    match x:
        case {"name": str()}:
            # runtime equivalent to (isinstance(x, Mapping)
            # and "name" in x and isinstance(x["name"], str))
            assert_type(x, D)
        case _:
            assert_type(x, int)

def test_non_match(x: D | int) -> None:
    match x:
        case {"value": str()}:
            # runtime equivalent to (isinstance(x, Mapping)
            # and "value" in x and isinstance(x["value"], str))
            assert_never(x)
        case _:
            assert_type(x, D | int)

Footnotes

1. technically, it is possible to make a shared subclass of int and collections.abc.Mapping, so the assert_type in the case {...} are debatable, though the inference in the case _ branches should not be affected. ↩

[JelleZijlstra]

I'm not sure we need to say anything in the spec here. Type checkers should reflect the runtime behavior; if they do not, that's an issue that can be handled by improving the type checker.