typing icon indicating copy to clipboard operation
typing copied to clipboard

How to alias `Annotated` to a subscriptable type

Open FFY00 opened this issue 5 years ago • 3 comments

Hi, I want to create a type alias that simply wraps a tuple into Annotated.

PEP 593

I want to be able to write

SpecialReturn[int, bool, str]

And get

Annotated[Tuple[int, bool, str], 'special return!']

Among other things, I've tried:

class SpecialReturn():
    def __class_getitem__(self, params: Tuple[Any, ...]) -> Tuple[Any, ...]:
        return typing.cast(
            Tuple[Any, ...],
            _typing.Annotated[
                typing.Tuple.__getitem__(params),
                'multiple_return',
            ]
        )

Which mypy is happy with the declaration, but not using it:

import typing

from typing import Annotated, Any, Tuple


class SpecialReturn():
    def __class_getitem__(self, params: Tuple[Any, ...]) -> Tuple[Any, ...]:
        return typing.cast(
            Tuple[Any, ...],
            Annotated[
                typing.Tuple.__getitem__(params),
                'multiple_return',
            ]
        )


def test() -> SpecialReturn[int, int]:
    return 1, 2
test8.py:17: error: "SpecialReturn" expects no type arguments, but 2 given
test8.py:18: error: Incompatible return value type (got "Tuple[int, int]", expected "SpecialReturn")

Which happens because I guess mypy does not evaluate the type annotation, it just treats SpecialReturn as a separate type. I am not sure what is the reasoning behind this, but I assume there it makes sense in some way(?).

So, is there any alternative? How do achieve this? I essentially want to alias Tuple and be able to find out at runtime if the type is a normal tuple or my special one.

My first jab at this, and what I think makes most sense to a Python developer is the following

from typing import Tuple


class SpecialReturn(Tuple):
    pass


def test() -> SpecialReturn[int, int]:
    return 1, 2

But mypy can't handle it either:

test.py:4: error: Missing type parameters for generic type "Tuple"
test.py:8: error: "SpecialReturn" expects no type arguments, but 2 given
test.py:9: error: Incompatible return value type (got "Tuple[int, int]", expected "SpecialReturn")

I was thinking maybe something like this was possible with Annotated, but it seems not.

FFY00 avatar Jan 19 '21 15:01 FFY00

I have also tried using generics in the past, but they also don't seem suitable, as I want the fields that go into the type to be anything that Tuple supports.

FFY00 avatar Jan 19 '21 15:01 FFY00

__class_getitem__ is a dynamic thing; I think static checkers will never evaluate it. In the current state Python typing, I don't see a way to achieve what you want.

In fact, you need to parametrize a tuple, which is variadic, so you need to create yourself a variadic type, and that's not possible for now. You can take a look at #193, but this is a complicated issue, opened 4 years ago, still without resolution.

A common solution of this variadic issue is to declare one type by "variation", like this:

SpecialReturn = Annotated[T, ...]
SpecialReturn1 = SpecialReturn[tuple[T1]]
SpecialReturn2 = SpecialReturn[tuple[T1, T2]]
SpecialReturn3 = SpecialReturn[tuple[T1, T2, T3]]
SpecialReturn4 = SpecialReturn[tuple[T1, T2, T3, T4]]
… # do you need more than 4

But I agree it's not so nice.

wyfo avatar Jan 21 '21 20:01 wyfo

@FFY00 @wyfo I noticed you referenced python/typing#193 on variadic generics in this thread. Heads up that we've been working on a draft of a PEP for this in PEP 646. If this is something you still care about, take a read and let us know any feedback in this thread in typing-sig. Thanks!

mrahtz avatar Feb 20 '21 14:02 mrahtz