Incomplete resolving of subclass chains in runtime-evaluated-base-classes

[disrupted]

Extracted from #7866 (comment)

Ruff doesn't reliably detect the base class (even within the same file).

given this example

from abc import ABC
from pydantic import BaseModel

class Model(BaseModel):
    x: str


class Empty(BaseModel):
    ...


class Abstract(Empty, ABC):
    inner: Model

Ruff evaluates the following classes and their bases (I simply placed some log statements in this function)

ruff/crates/ruff_linter/src/rules/flake8_type_checking/helpers.rs

Line 42 in a1509df

class_def.bases().iter().any(|base| {

class Model  bases: [Some(["pydantic", "BaseModel"])]
class Abstract  bases: [None, Some(["abc", "ABC"])]

notice the absence of Empty and how Abstract then doesn't get detected as BaseModel.

Therefore, flake8-type-checking rules often result in false positives as it's unable to correctly resolve the base class.