Annotated[Any, ...]

[orenbenkiki]

Feature

Id like to be able to write:

SeriesOfBool = NewType("SeriesOfBool", Annotated[pandas.Series, "bool"])

def is_series_of_bool(series: pandas.Series) -> TypeGuard[SeriesOfBool]:
    return series.dtype == "bool"

Pitch

I would dearly love to track the data type of the many series values floating around my code. The above seems like a reasonable approach., which seems to work fine if I define ArrayOfBool = NewType("ArrayOfBool", Annotated[numpy.ndarray, "bool"]).

However because pandas has no type annotations, then pandas.Series is essentially Any and trying the above gives the error Argument 2 to NewType(...) must be subclassable (got "Any").

It seems like Annotated[Any, ...] shouldn't be the same as Any - in particular, not all values belong to the annotated type - so this error is over conservative?

[Viicos]

@orenbenkiki pandas has no type annotations but a stubs package is available.
mypy should probably warn you about missing types, you can use the --install-types option.

[orenbenkiki]

@Viicos - Thanks, I'll look into it. The point still stands; it seems that Anotated[Any, ...] shouldn't be the same as Any, but mypy treats them in the same way.

[Viicos]

Any cannot be subclassable, and Annotated doesn't make it subclassable either. This is subject to changes since Python 3.11, as Any is now defined as a class (still, mypy doesn't seem to allow it yet, there might be an issue about this -- Python 3.11 isn't fully supported yet).

Other than that I find your implementation pretty neat, I don't recall seeing this anywhere. Maybe you could even get rid of the NewType here:

SeriesOfBool = Annotated[pandas.Series, "bool"]

def is_series_of_bool(series: pandas.Series) -> TypeGuard[SeriesOfBool]:
    return series.dtype == "bool"

Needs to be tested ;)

[hauntsaninja]

I'm not sure what NewType of Any is meant to do. If you want an opaque new type, you should make a NewType of object.

Annotated[T, ...] is usually exactly the same as T; I don't think it works how you think it works. From PEP 593:

If a library (or tool) encounters a typehint Annotated[T, x] and has no special logic for metadata x, it should ignore it and simply treat the type as T.