RUF052: Use of dummy variables is not flagged for loop variables

[MichaReiser]

@AlexWaygood @MichaReiser Searching https://github.com/search?q=repo%3Aastral-sh%2Fruff+RUF052&type=issues does not reveal previous discussions on detecting used dummy variables inside FOR loops and list comprehensions. To avoid extra noise from rushing into opening a new issue, I would like to clarify: Was this rejected somewhere, or was this just not discussed and not implemented yet?

Example:

def no_ruff_warnings():
    my_list = [{"foo": 1}, {"foo": 2}]

    i = 0
    for _ in my_list:
        i += _["foo"]  # A dummy variable is used, but no warnings. 

    for _item in my_list:
        i += _item["foo"]  # A dummy variable is used, but no warnings. 

    [_["foo"] for _ in my_list]  # A dummy variable is used, but no warnings. 

    [_item["foo"] for _item in my_list]  # A dummy variable is used, but no warnings. 

Originally posted by @alessio-locatelli in #14755 (comment)

[MichaReiser]

I don't think this behavior is intentional. It is due to this check:

ruff/crates/ruff_linter/src/rules/ruff/rules/used_dummy_variable.rs

Lines 113 to 131 in 331beca

	// We only emit the lint on variables defined via assignments.
	//
	// ## Why not also emit the lint on function parameters?
	//
	// There isn't universal agreement that leading underscores indicate "unused" parameters
	// in Python (many people use them for "private" parameters), so this would be a lot more
	// controversial than emitting the lint on assignments. Even if it's decided that it's
	// desirable to emit a lint on function parameters with "dummy variable" names, it would
	// possibly have to be a separate rule or we'd have to put it behind a configuration flag,
	// as there's much less community consensus about the issue.
	// See <https://github.com/astral-sh/ruff/issues/14796>.
	//
	// Moreover, autofixing the diagnostic for function parameters is much more troublesome than
	// autofixing the diagnostic for assignments. See:
	// - <https://github.com/astral-sh/ruff/issues/14790>
	// - <https://github.com/astral-sh/ruff/issues/14799>
	if !binding.kind.is_assignment() {
	return;
	}

The comment suggests that the intention was to only exclude parameters, but the way it is implemented excludes many other bindings too.

We should consider making this check more fine grained (behind preview) and have a look at the ecosystem changes.

[mikeleppane]

I can take a look at this.