SLOT: Classes with multiple base classes

[Skylion007]

Description

ruff check --select=SLOT .

ruff --version 0.8.4

class Backend(RandomClass, tuple):
   pass

^ suggests a fixit here. This is the recommended class structure for old versions of python before strenum was introduced. However, enum doesn't have slots defined, so defining a __slots__ wouldn't be that helpful. i also see this bug happen when a class has multiple ancestors, one of which is a tuple.

Actually, nevermind, this error was actually masking another error where enum was not properly imported. Seems odd to suggest a fixit in this situation. In other cases though, I do see the error pop up when only one of the ancestors is a tuple.

[MichaReiser]

Could you explain what's the bug experience? I don't think I understand it from the issue.

Ruff does raise an error about a missing __slots__ even if the base class is known. So warning in the case if any base class is unknown doesn't seem incorrect, because the __slots__ attribute is always required if any parent is tuple.

Playground: https://play.ruff.rs/45b205ae-153c-4a73-8fe6-29995ef0a5c9

[Skylion007]

@MichaReiser To clarify, what is the behavior if a class has two ancestors, and only one of them implements __slots__? My understanding is that the slots in the class itself won't have an effect because it will still need to consult the __dict__ of the first parent, but I could be wrong? Specifically, what happens here:
https://play.ruff.rs/26820797-8f34-4fc4-8c13-eb81bce03d0e

[MichaReiser]

Oh, I see. That's a good point. In that case, I'd say it's not a false positive, at least not when the base class is under the user's control. However, the suggested fix/message is misleading because only adding __slots__ to the current class isn't enough; it also needs to be added to all non-tuple parent classes.

Now, analyzing base classes is something that Ruff struggles with, at least if they're defined in another file. Which is why I'm leaning towards leaving the rule as is for now and revisit it once we have better type inference.