Thoughts on improving B902? about flake8-bugbear HOT 5 OPEN

See https://github.com/PyCQA/pep8-naming#options for an alternate implementation that can leave a lot of these decisions up to the user.

from flake8-bugbear.

In general I am in favor of making this less noisy. If we can detect that the user does something reasonable, we shouldn't warn. So I'm open to pull requests to that effect.

As for "subclassing type", I am already making a check like you're describing. There are passing tests to that effect. Maybe what you're seeing is this: Flake8 plugins are rather dumb, they don't see the full MRO of a class. So my fix only helps in the case where "type" is explicitly listed as a base on the class which we're analyzing now.

from flake8-bugbear.

don't see the full MRO

That makes sense. Integrating with mypy for type analysis sounds like a horrible pain (at least in its present incarnation).

less noisy

How do you feel about false negatives? I.e. if we have __new__ and the first arg is metacls, can we just assume good intent, even if type is not one of the explicit bases?

I think that unconditionally accepting the underscore-prefix version of the canonical variable sounds safe. Do you?

As far as subcls / submetacls — I would just add those to the list of canonical variables for __init_subclass__, rather than (opinionatedly) requiring them. Agreed?

from flake8-bugbear.

All of your proposed changes sound good to me.

from flake8-bugbear.

Here is another case where B902 is flagged, where it should not (IMO):
@abstractclassmethod should allow/expect the 'cls' argument name in the same way as @classmethod does.

from abc import ABC, abstractclassmethod

class Foo:
    @abstractclassmethod
    def bar(cls, baz):  # noqa: B902
        """..."""

from flake8-bugbear.