Formatter: Pragma comments

[MichaReiser]

See #6670 for the proposal for the details:

• Measuring LineSuffix when formatting to include trailing comments #5630
• Include trailing end-of-line comments in the measured line width #6771
• Exclude pragma comments from measured line width #6772

[davidszotten]

do you have any examples of this happening? (how did you find them)

[MichaReiser]

do you have any examples of this happening? (how did you find them)

I haven't looked closely into it but there's some related Black logic:

https://github.com/psf/black/blob/2fd9d8b339e1e2e1b93956c6d68b2b358b3fc29d/src/black/lines.py#L227-L294

and

https://github.com/psf/black/blob/268dcb677ce80331e0ef104d8335c2ada32872fb/src/black/comments.py#L324-L335

I don't know if this is a valid noqa placement, but splitting by the arguments could change what it annotates

def test(a, b, # noqa: RUFXXX
	c, d): pass

# Formatted
def test(
	a, 
	b, # noqa: RUFXXX
	c, 
	d
): pass

This could cause RUFXXX to trigger now if it suppressed a or test and not b. Playground

[roshanjrajan-zip]

Hi! This is one of the problems we have in our repo. However, we use # pyright: ignore instead of # type: ignore. It would be great to have this as dynamic or something to support different use cases. Please lmk if there is anything I can do to provide more info or any help - psf/black#3661

[MichaReiser]

Relevant lines in the Black formatter (I think)

https://github.com/psf/black/blob/2fd9d8b339e1e2e1b93956c6d68b2b358b3fc29d/src/black/lines.py#L227-L294

[charliermarsh]

My one-second reaction is that I think we probably do a reasonably good job of avoiding the uncollapsible comment case since we don't really collapse comments (comments always expand the parent), but I don't think we handle the unsplittable line case at all.

[MichaReiser]

Some more relevant issues:

• flake8 #noqa comments get moved a place where they're ignored psf/black#195
• Unstable formatting with "# type: ignore" comment psf/black#1061

[MichaReiser]

@roshanjrajan-zip The example you brought up on the Black issue is

some_long_variable_name = some_long_dict_name[0] # pyright: ignore[reportGeneralTypeIssues]

How would you preferred formatting look like?

Would the following work for you?

some_long_variable_name = ( # pyright: ignore[reportGeneralTypeIssues]
	some_long_dict_name[0] 
)
# Or
some_long_variable_name = some_long_dict_name[  # pyright: ignore[reportGeneralTypeIssues]
	0
] 

It would still require manual intervention (you have to move the comment to the opening parentheses), but it would allow you to suppress the issue.

[MichaReiser]

Black issue and PR that added the special type: ignore handling

• Please don't break long lines if a '# type: ignore' takes it over the max length psf/black#997
• Don't break long lines when type: ignore is present psf/black#1040

[MichaReiser]

See #6670

[roshanjrajan-zip]

Hi @MichaReiser,

Definitely the first one is much better imo and my team agrees! Thanks so much for your work on the proposal. Taking a look at it!

some_long_variable_name = ( # pyright: ignore[reportGeneralTypeIssues]
	some_long_dict_name[0] 
)