No errors for keyword arguments in class statement

[JelleZijlstra]

Example:

from typing import TypedDict

class TD(TypedDict, a=3):
    a: int

https://pyright-play.net/?code=GYJw9gtgBALgngBwJYDsDmUkQWEMoAqiApgCYAiSAxjAFC1UA2AhgM6uHkAURCZlNADRRmAXgDMASgBctKPJHTMKGEA

This fails at runtime, and type checkers should reject it too. (cf. python/mypy#16438)

[erictraut]

This is currently by design. Pyright doesn't ever report extra keyword arguments passed in a class declaration. So this issue isn't specific to TypedDict.

I'm going to change this from a bug to an enhancement request and consider it in that light. I think it's a reasonable feature to add, assuming it doesn't produce too many false positives in typical code bases.

[JelleZijlstra]

Thanks, I didn't know that. For normal classes, keyword arguments may be valid, depending on the metaclass and __init_subclass__ in base classes. It would be good if pyright could check keyword arguments for correctness, but it's not simple. For TypedDict classes, however, any keyword arguments are always invalid.

[erictraut]

Currently, pyright does check __init_subclass__ if it's defined somewhere in the class hierarchy other than the object class. The object.__init_subclass__ method is ignored. I don't recall why I did it this way. It may have been because of TypedDict, which is a special form and doesn't define an __init_subclass__ or a metaclass.

[erictraut]

The reason pyright is not flagging the error for a TypedDict is because of a faulty type definition in typing.pyi and typing_extensions.pyi. I suspect the same is true for mypy. Both of these stubs define _TypedDict with a metaclass ABCMeta (which is a lie — the actual runtime metaclass is _TypedDictMeta). ABCMeta's __new__ method accepts any keyword arguments, which is why you're not seeing an error in your code sample above.

I propose that we fix this by modifying the definition of _TypedDict as follows.

@type_check_only
class _TypedDictMeta(type):
    def __new__(
        __mcls: type[typing_extensions.Self],
        __name: str,
        __bases: tuple[type, ...],
        __namespace: dict[str, Any],
        *,
        total: bool = False,
    ) -> typing_extensions.Self: ...

@type_check_only
class _TypedDict(Mapping[str, object], metaclass=_TypedDictMeta):
    ...

This not only addresses the bug and more closely reflects the actual runtime class hierarchy, but it also has the benefit of providing better in-editor support for the total parameter.

Thoughts?

[JelleZijlstra]

Good catch, that sounds like a good idea!

Type checkers may still need a special case to prohibit a custom metaclass= on TypedDict, but your suggestion should automatically prohibit other keyword arguments.

[erictraut]

I've improved pyright's handling of __init_subclass__ to handle the case where there is no custom metaclass with a __new__ method and no __init_subclass__ that overrides object.__init_subclass__. I've also improved the error messages for __init_subclass__ in general. This set of changes still doesn't result in an error for the code in the original sample above, but that will be fixed once the _TypedDict definition is updated in typeshed.

[erictraut]

This is addressed in pyright 1.1.336, which I just published. It will also be included in a future release of pylance.