Wrong Signature for Enum with Specific Dataclass Mixin Scenario

[iFailatGitHub]

Note: I was thinking about putting this on pyright but it said standard library problems are more suited here, so I'll give it a shot.

I was trying to code an enum with a dataclass mixin where each value is overridden through the first string variable for easier referencing (rather than a dict from str to enum).

from dataclasses import dataclass
from enum import Enum, auto

@dataclass(frozen=True)
class SampleMixin:
    a_string: str
    a_number: int

class SampleEnum(SampleMixin, Enum):
    def __new__(cls, value: str, *_):
        obj = object.__new__(cls)
        obj._value_ = value

        return obj

    foo = "foo", 1
    bar = "bar", 2

class SampleEnum2(Enum):
    foo = auto()
    bar = auto()

baz = SampleEnum("foo")
qux = SampleEnum2("foo")

baz still gets SampleEnum.foo right, but the function signature is incorrectly identified. Ideally, it should be closer to SampleEnum2's function signature.

I'm open to workarounds at the meantime. The code runs correctly, it's just that the inferred signature I got was wrong.

Code sample in pyright playground.

[lul-cas]

I looked into this a bit.
The runtime behaviour is correct, and the mismatch seems to be in how the typechecker pick the call signature for this pattern.

the dataclass init is used tob uild members at class definition time.
EnumMeta.call is used to look up for existing members at call sites. Type checkers appear to conflate the two and show the dataclass construction signature where the enum lookup signature should be shown instead.

You can workaround it by manual mixin instead of '@DataClass'; 'SampleEnum["foo"]' for name lookup; or
"class SampleEnum(str, Enum)" if the lookup key is always a string.