Cannot assign a type as default value for dataclass attribute annotated with a generic type

[danpascu]

Code sample in pyright playground

from argparse import Namespace
from dataclasses import dataclass
from typing import reveal_type


@dataclass(kw_only=True)
class Application[T]:
    args_type: type[T] = Namespace  # Type "type[Namespace]" is not assignable to type "type[T@Application]"


class Application1[T]:
    args_type: type[T]

    def __init__(self, *, args_type: type[T] = Namespace) -> None:  # no error
        self.args_type = args_type                                  # no error

app = Application()
reveal_type(app)
reveal_type(app.args_type)

app1 = Application1()
reveal_type(app1)
reveal_type(app1.args_type)

In this code sample, I get an error when I assign Namespace as the default value for args_type in the dataclass variant, but the error message is not very helpful.

If I rewrite the class to not be a dataclass and I just define my own __init__ then I do not get any error for specifying Namespace as the default args_type.

Despite the error, the type for app, app1 and the args_type attributes seems to be inferred correctly to be Application[Namespace], Application1[Namespace] and type[Namespace] respectively.

I do not understand if I'm doing something wrong with the dataclass or I hit some bug in pyright.

[erictraut]

See this issue for an explanation.

[danpascu]

Would you say it's safe to ignore this error without having negative consequences when analysing the code involving this dataclass, or do you think it's better to go with the non-dataclass variant?

[erictraut]

If the type variable is used only once in the annotations for your field definitions (like in your example), I think this is safe and can be ignored. If the type variable appears more than once, then there's a potential hole.

[danpascu]

I guess it's better to use the non-dataclass variant then. I have another method like:

    def get_args(self) -> T:
        ...

Thanks for the answer.