Fix B020 false positive for attribute iterables

[gaoflow]

Fixes #521.

Summary

• stop B020 from treating the base name of a plain attribute iterable like smoother.smoothers as the iterable being reassigned
• keep method-call iterable cases such as values.items() covered
• add a regression case for attribute access on the loop target name

Tests

• uv run --python 3.13 --with '.[dev]' pytest tests/test_bugbear.py -k B020 -q
• uv run --python 3.13 --with '.[dev]' pytest tests/test_bugbear.py -q
• uv run --python 3.13 --with flake8 --with attrs flake8 bugbear.py tests/test_bugbear.py
• uv run --python 3.13 --with mypy --with types-flake8 --with attrs --with flake8 mypy bugbear.py
• git diff --check

[cooperlees]

I'm not sure we want to fix this, open to other perspectives tho ... and I'll go with majority ...

[cooperlees]

Suggested change

	for smoother in smoother.smoothers:
	print(smoother)
	for smoother_loop in smoother.smoothers:
	print(smoother_loop)
	smoother = smother_loop

I argue not doing this is a good practice even tho it's technically not overwriting what your iterating ...

I am a fan of being explicit even at the cost of more assignments etc.

[Copilot]

Pull request overview

This PR fixes a B020 false positive where the loop target name appears as the base of an attribute-access iterable (e.g., for smoother in smoother.smoothers:), while preserving detection for method-call iterables (e.g., values.items()), and adds a regression test to cover the reported case from #521.

Changes:

• Update B020 iterable name collection to ignore plain attribute access while still traversing method-call bases.
• Add an eval-file regression case ensuring for smoother in smoother.smoothers: does not trigger B020.

Reviewed changes

Copilot reviewed 2 out of 2 changed files in this pull request and generated 1 comment.

File	Description
bugbear.py	Adjusts B020NameFinder traversal to skip ast.Attribute nodes and special-case method calls so existing B020 coverage (like values.items()) remains.
tests/eval_files/b020.py	Adds a regression example for attribute access on the loop target name that should not emit B020.

---

💡 Add Copilot custom instructions for smarter, more guided reviews. Learn how to get started.

[gaoflow]

Thanks for looking. My reasoning for treating this as a false positive is that B020 is most useful when the loop target overwrites the iterable object itself, e.g. the warning protects for value in values: and receiver-style calls like for key in values.keys().

For plain attribute access, for smoother in smoother.smoothers: reads the iterable before the loop target is rebound, and the rebound name does not change the object that supplied .smoothers. That makes it closer to a normal attribute traversal than to the self-iteration bug B020 is trying to catch.

The patch keeps the existing method-call behavior (values.items(), etc.) because those still indicate iterating over the same named object. But I understand if the project prefers the more conservative warning despite the false positive.