Regression in support for idiomatic TypedDict discrimination? #10517
Description
#10378 removed support for the common usage of "in" to discriminate between TypedDicts, as in
class ResponseOne(TypedDict):
one: str
class ResponseTwo(TypedDict):
two: str
class ErrorResponse(TypedDict):
error: str
Response = ResponseOne | ResponseTwo | ErrorResponse
def do_something_that_expects_response_one(response: Response) -> None:
if "one" not in response:
raise RuntimeError(f"Unexpected response: {response}")
value = response["one"]
print(f"Got what we expected: {value}")As of this latest release, this now errors with e.g. "one" is not a defined key in "ResponseTwo"
In the absence of any approved PEP to indicate a TypedDict is closed, I believe this has made it impossible to express this idiomatic code? (Outside of hand-crafting TypeGuard functions for every single type.) Using PEP-728, for instance, would as I understand it result in type annotations that cannot be used by mypy.
mypy currently supports marking these discrimination types as @final to achieve this result, but pyright decided to simply allow the idiomatic option in the past. If this decision is being reversed, could we reintroduce the ability to "close" these TypedDicts with @final until such time as a PEP has been agreed and integrated into other type-checkers?