check_type does not check required fields when mixing required and non-required fields in TypedDict

[ksaaskil]

Hi, thanks for the awesome library! A recent fix for an issue (#94) added support for totality in TypedDict, thanks for that! I'm trying to mix required and non-required fields in TypedDict as instructed in mypy documentation:

class MovieBase(TypedDict):
    name: str
    year: int

class Movie(MovieBase, total=False):
    based_on: str

This definition should make name and year required fields in Movie.

However, check_type does not seem to complain if those are missing:

from typeguard import check_type

movie = Movie()   # Missing `name` and `year` , type-checker complains
check_type("movie", movie, Movie)  # Does not raise error but it should?

Any idea what might be wrong here? Thanks!

[agronholm]

How am I going to extract the information about MovieBase at run time? It doesn't show up in the MRO or the list of base classes of Movie.

[ksaaskil]

I have no idea how that should be implemented 🙂 If it cannot be, maybe we could add a note to the documentation?

[agronholm]

Sounds reasonable.

[antonagestam]

Note that this situation is addressed in Python 3.9 with dunder attributes for __required_keys__ and __optional_keys__. Would you consider reopening this? :)

[agronholm]

Sure.

[agronholm]

It seems the new patch was botched:

>>> DummyDict = TypedDict('DummyDict', {'x': int, 'y': str}, total=False)
>>> DummyDict.__required_keys__
frozenset({'x', 'y'})

It's ignoring the total argument.

[agronholm]

I've submitted a fix against CPython: python/cpython#22736

If I add support for __required_keys__, then it will behave wrong with the 3.5 compatible syntax in Python 3.9.0. I'm not sure what to do.