[ruff] Detect mutable defaults in field calls (RUF008)

[eureka928]

Summary

Resolves #16495.

RUF008 previously only caught bare mutable defaults like mutable_default: list[int] = [] in dataclass attributes, but missed mutable defaults wrapped in field(default=...) calls. For example, the following was not flagged:

@define
class A:
    mutable_default: list[int] = attrs.field(default=[])

This PR modifies mutable_dataclass_default() to look inside recognized dataclass field calls (dataclasses.field(), attrs.field(), attr.ib(), attr.attrib()) and check the default keyword argument for mutability.

The approach mirrors how RUF009 already uses is_dataclass_field() to identify field calls — RUF008 now reuses the same helper to extract and inspect the default keyword argument. No duplicate diagnostics are introduced since RUF009 already skips field calls entirely.

Test Plan

cargo nextest run -p ruff_linter and cargo clippy.

Added test cases covering:

• Positive: field(default=[]), attrs.field(default={}), attr.ib(default=[]), attr.attrib(default=set()), field(default=dict())
• Negative: factory=list, default=(), default="hello", default=1, default=KNOWINGLY_MUTABLE_DEFAULT, ClassVar annotation with field(default=[])

[eureka928]

Hi @ntBre would you review my first PR?
Thank you

[astral-sh-bot]

ruff-ecosystem results

Linter (stable)

✅ ecosystem check detected no linter changes.

Linter (preview)

✅ ecosystem check detected no linter changes.

[ntBre]

Nice! This is looking good.

The main thing we need to do is make this a preview change. See my inline comment for a link to other example preview functions. Other than that, I just a couple of nits!

[ntBre]

I think we need to make this a preview change and also mention that here:

Suggested change

	/// This rule also detects mutable defaults passed via the `default` keyword
	/// In [preview](https://docs.astral.sh/ruff/preview/) this rule also detects mutable defaults passed via the `default` keyword

[ntBre]

I think this might be slightly nicer as a match to avoid these dangling elses:

        let value_to_check = match &**value {
            Expr::Call(ast::ExprCall {
                func, arguments, ..
            }) if is_dataclass_field(func, checker.semantic(), dataclass_kind) => {
                arguments.find_argument_value("default", 0)
            }
            value => Some(value),
        };

I would probably just shadow value instead of using value_to_check too.

[eureka928]

I updated this and pushed!
Would you review once more? this would be good to merge 😉

[ntBre]

Oh, and thank you for handling dataclasses as well! I thought from the issue that this was only about attrs, but the stdlib dataclasses were affected too.

[ntBre]

This still has not been gated behind a preview check, and my suggestion to add comments to the test cases was for every new test case, not that single line.

Please make sure you've addressed all of the feedback before pinging us for an additional review.

[ntBre]

This comment is not correct. This line emits a diagnostic.

[eureka928]

Can you review my last commit if that looks good to you?

[eureka928]

Thank you for merging this!