I would also appreciate any tips that anyone has on how to properly type annotate highly heterogeneous dictionaries to be used for the purpose of interacting with REST APIs.

I would also appreciate any tips that anyone has on how to properly type annotate highly heterogeneous dictionaries to be used for the purpose of interacting with REST APIs.

https://mypy.readthedocs.io/en/latest/more_types.html#typeddict

The issue itself is valid, and similar to how mypy overreacts to concatenating lists with different types, the latter can be just fixed in typeshed, while this one might require some special-casing. Btw, have you tried giving an explicit annotation to self.d3?

Thanks, typeddict looks very promising. I just tried giving self.d3 an explicit annotation and it appears to not make a difference (same error pops up when I run mypy).

I just came across an interesting variant of this error:

class A:
    foo: typing.Dict[str, typing.Union[int, str]] = {"a": 1}

class B(A):
    # incompatible type "Dict[str, Union[int, str]]"; expected "Mapping[str, str]"
    foo = {"b": "c", **A.foo}

I assume that what's happening is that the type of the RHS of that assignment is being resolved before mypy is aware that B.foo should have the same type as A.foo.

This can be fixed by annotating foo. But I think this is something of a problem for libraries which offer inheritance as part of their interface. If a 3rd party developer inherits from A, they have to annotate foo if they override it.

I've stumbled upon the same issue.

class Question(typing.TypedDict):
    id: str
    body: str

question: Question = {
    'id': 'something',
    'body': 'something else',
}

processed = {
    **question,
    'documents': []
}

Hinting with processed: dict seems to silence the warning. I think this solves the issue, but it would be awesome if mypy could derive the type of processed to be typing.TypedDict('<something...>', id=str, body=str, document=str). I don't know if mypy is capable of doing that, or if this type of inference has precedents, though.

This bug appears to have been fixed. The latest version of mypy doesn't emit an error for any of the above code samples.