unused-noqa (RUF100) - false negatives and strange behavior with multiple codes

[DetachHead]

Description

i assume this is an unintentional bug in the logic that allows you to add additional information to noqa comments:

# noqa: D103 this function is self-documenting

but ruff should enforce a space between the code and any additional comments. currently it doesn't seem to do so. it also seems to allow certain completely invalid codes:

# no error even though these aren't valid codes:
def foo(): ... # noqa: D103a
# noqa: Z

interestingly, these are treated as two separate codes:

def foo(): ... # noqa: D103F811

i think it should instead be an error and force the user to separate them with a comma like this:

def foo(): ... # noqa: D103,F811

https://play.ruff.rs/2b3608e9-f97a-459a-960b-f702827ef4e4

[MichaReiser]

Thanks, I think we miss a check for leading_space to be != 0 here

ruff/crates/ruff_linter/src/noqa.rs

Lines 105 to 128 in c39ca8f

	while let Some(code) = Self::lex_code(&text[codes_end + leading_space..]) {
	codes_end += leading_space;
	codes.push(Code {
	code,
	range: TextRange::at(
	TextSize::try_from(codes_end).unwrap(),
	code.text_len(),
	)
	.add(offset),
	});
	codes_end += code.len();
	// Codes can be comma- or whitespace-delimited. Compute the length of the
	// delimiter, but only add it in the next iteration, once we find the next
	// code.
	if let Some(space_between) =
	text[codes_end..].find(|c: char| !(c.is_whitespace() || c == ','))
	{
	leading_space = space_between;
	} else {
	break;
	}
	}

[InSyncWithFoo]

See also #12809 and #14229. I posted a comment at the latter, but perhaps Charlie never saw it.

@MichaReiser Considering what happened in the two aforementioned PRs, this will require a design decision. What do you say?

[MichaReiser]

Can you tell me more about the design decisions you see?

These could also be multiple fixes:

• D103F811: Should not be treated as two rules
• D103a: Handle content after a rule code

[InSyncWithFoo]

@MichaReiser #12809 made it so that D103F811 is treated as one code for both file-level and line-level # noqas. #14229 reverted it for line-level, but not for file-level. The result is now inconsistent:

# This hides the two diagnostics
print(list(sorted()))  # noqa: T201C413

# This doesn't, and the joint code is not reported as invalid.
# ruff: noqa: T201C413
print(list(sorted()))

My question never got an answer, so I assume that it is desired to keep this inconsistency. If you say otherwise, I can work on a fix. The PR will have concrete examples as well as how to deal with those and we can go from there.

[MichaReiser]

Oh, so the problem linked in #14229 is that we incorrectly parsed text coming after a code as a noqa code.

That means we have to define a heuristic of what we treat as codes and what as non-code.

I also just found #12811