Typing fails when list of validators where one is custom and one is built in

[superosku]

Hi!

I am trying to add multiple validators to a field, which works fine as long as they are both from the attrs library or both written by me. When I mix them up mypy gives me an error. Am I doing something wrong here or is this an issue in the library? Here is a snippet to reproduce the issue:

from typing import Any

from attr import validators, Attribute
import attr


def is_not_empty(instance: Any, attribute: "Attribute[str]", value: str) -> Any:
    if not value.strip():
        raise ValueError("Is not empty")


@attr.s
class MyClass:
    thing1: str = attr.ib(validator=[validators.max_len(10), is_not_empty])  # Mypy complains from this
    thing2: str = attr.ib(validator=[validators.max_len(10)])  # This is fine
    thing3: str = attr.ib(validator=[is_not_empty])  # Fine as well
    thing4: str = attr.ib(validator=[is_not_empty, is_not_empty])  # Even this is fine

This is the error I get from mypy:

snippet.py:14: error: Argument "validator" has incompatible type "list[function]"; expected "Callable[[Any, Attribute[<nothing>], <nothing>], Any] | Sequence[Callable[[Any, Attribute[<nothing>], <nothing>], Any]] | None"  [arg-type]

[Tinche]

It looks like Mypy type inference isn't strong enough.

reveal_type([validators.max_len(10), is_not_empty])  # Revealed type is "builtins.list[builtins.function]"

Funnily enough, if you hold its hand enough, it works:

from __future__ import annotations
from typing import TYPE_CHECKING

if TYPE_CHECKING:
    from attr import _ValidatorType

test: _ValidatorType = is_not_empty

reveal_type([validators.max_len(10), test])  # Revealed type is "builtins.list[def (Any, attr.Attribute[Any], Any) -> Any]"

I couldn't figure out an elegant way of making it work. I'd suggest just sticking a #type: ignore on that line, to be honest.

[superosku]

I did consider importing the _ValidatorType and that it might fix the issue. But I did not want to do it since it is marked as private (The _ prefix). What do you think about making it public for these types of scenarios?

[Tinche]

It wouldn't be the end of the world, but I wish we could come up with a better solution.

[AdrianSosic]

Hi, just wanted to say that I'm running into the same issue, so: +1

[AdrianSosic]

I just noticed that this is not restricted to combinations of built-in and custom validators but also holds for lists of built-in validators. Both the following classes do the same job, but class B is flagged by mypy while A passes as expected:

from attrs import define, field
from attrs.validators import and_, instance_of, min_len


@define
class A:
    name: str = field(validator=and_(instance_of(str), min_len(1)))

@define
class B:
    name: str = field(validator=[instance_of(str), min_len(1)])