noqa comment stripped due to syntax error elsewhere in file

[stinodego]

Ruff indicates an unused noqa statement, when in fact it's fine. There is a syntax error in the function definition that triggers this, if the * is removed, everything is fine.

def function(*):
    """Veeeeeeeeeeeeeeeeeeeeeeeeeeeeeeeeeeeeeeeeeeeeeeeeeeeeeeeeeeeeeeeeeeeeeeeeeeeeery long."""  # noqa: W505

This is on ruff version 0.0.270.

[charliermarsh]

Thanks, I actually thought we added handling for this case at some point. Will revisit!

[stinodego]

Thanks, I actually thought we added handling for this case at some point. Will revisit!

I know, I have reported something similar before:
#2853

Not sure what caused this one to resurface.

[charliermarsh]

(I believe this regressed when we started generating fixes on every pass, regardless of whether they're applied.)

[charliermarsh]

@MichaReiser - Any opinion on the right behavior here? The issue is that if we hit a syntax error, we fail to identify the W505 violation in the example above, and so the noqa is considered unused and removed.

We could...

1. Avoid autofixing at all if we hit a syntax error.
2. Avoid autofixing RUF100 violations (unused suppression comments) if we hit a syntax error.
3. Avoid raising RUF100 violations if we hit a syntax error.

[MichaReiser]

@charliermarsh My preference is 3. The way I think about this is that rules should only run if their input is correct. What the input is, differs from rule to rule:

• Physical lines: always valid
• logical lines: valid tokens
• AST rules: parsable source tree
• RUF100: All configured rules ran

That's why I think RUF100 should not run if any of the other phases returned an Err.

[stinodego]

For what it's worth, I also believe 3 is the way to go. I don't like all the linting violations popping up in seemingly unrelated places while I'm writing some code.

[charliermarsh]

Sounds good, will fix today.