Use TypeIs for is_typeddict

[max-muoto]

Use TypeIs for is_typeddict, which will allow code like this to be type-checked properly, similarly to: #11929:

def f(X: type[Any]) -> None:
    if is_typeddict(X):
        X.__required_keys__
        reveal_type(X.__required_keys__)  # E: Revealed type is 'frozenset[str]'

[max-muoto]

The only change from mypy primer fixes a false positive in Pydantic:

    if is_typeddict(tp):
        return tp.__orig_bases__  # type: ignore

(This is valid, and will type check correctly now, on 3.12+)

[AlexWaygood]

The only change from mypy primer fixes a false positive in Pydantic:

The mypy primer hit is saying that the type: ignore was previously unused but, with this change, would now be used. That seems like it's introducing a false positive rather than taking one away

[max-muoto]

The only change from mypy primer fixes a false positive in Pydantic:

The mypy primer hit is saying that the type: ignore was previously unused but, with this change, would now be used. That seems like it's introducing a false positive rather than taking one away

I'm fairly certain it's saying it's no longer necessary? From using Pyright, at least, it'll throw a warning if one isn't being used, but perhaps I'm misinterpreting.

Edit: One reason why that might be is because __orig_bases__ isn't present on TypeDict before 3.12, and before Any was getting used. So I don't think it's problematic.

[max-muoto]

    # __orig_bases__ sometimes exists on <3.12, but not consistently,
    # so we only add it to the stub on 3.12+
    if sys.version_info >= (3, 12):
        __orig_bases__: ClassVar[tuple[Any, ...]]

[AlexWaygood]

Mypy primer results are a diff. If a line is coloured in red with a minus sign at the beginning, it means that mypy emitted that message with the typeshed main branch, but no longer emits that message with the PR branch.

[max-muoto]

Mypy primer results are a diff. If a line is coloured in red with a minus sign at the beginning, it means that mypy emitted that message with the typeshed main branch, but no longer emits that message with the PR branch.

I see, makes sense!

[AlexWaygood]

But yeah, as you say, the new error seems like a true positive rather than a false positive — it shows mypy applying type narrowing correctly!

[AlexWaygood]

Could you apply the same change to typing_extensions.is_typeddict? The two need to be kept in sync

[max-muoto]

Could you apply the same change to typing_extensions.is_typeddict? The two need to be kept in sync

Will do!

[max-muoto]

Could you apply the same change to typing_extensions.is_typeddict? The two need to be kept in sync

dd309b5

[github-actions]

Diff from mypy_primer, showing the effect of this PR on open source code:

pydantic (https://github.com/samuelcolvin/pydantic)
- pydantic/_internal/_decorators.py:285: error: Unused "type: ignore" comment  [unused-ignore]

[AlexWaygood]

Advantages of doing this:

• It allows type checkers to apply type narrowing when they see is_typeddict being used, which means that users can safely access class-level attributes such as __required_keys__ and __optional_keys__ on TypedDict classes.

Disadvantages of doing this:

• It's a bit weird for type checkers to narrow to a fictional type. typing._TypedDict doesn't actually ever appear in a TypedDict class's mro; it's a class factory rather than a base class:

  >>> from typing import TypedDict
  >>> class Foo(TypedDict): pass
  ... 
  >>> Foo.__mro__
  (<class '__main__.Foo'>, <class 'dict'>, <class 'object'>)

• The only real-world case where mypy_primer highlights this making a difference regards __orig_bases__. We already have a type-safe helper function for regarding that attribute (types.get_original_bases, backported as typing_extensions.get_original_bases), so I'm not sure I see that as a convincing use case.

Overall, I think I'm -0 on this. But curious about what other maintainers think.

[JelleZijlstra]

I would oppose adding this. _TypedDict is not a real type and type checkers do not have to use it; type checkers that don't would likely get confused.