hatch fmt not placing noqa comments in the correct location

[benjunmun]

Running hatch fmt can move noqa comments to the wrong line when line-splitting. This caused me additional confusion because a second run of the formatter removed the now-extraneous noqa, and started showing me errors again.

Here is a minimal reproducer, used in a clean hatch new project. The first run of hatch fmt will reformat and return no other errors, but will split the noqa to the wrong line. A second run of hatch fmt will then strip the "invalid" noqa and start showing errors again.

def main():
    some_list = [1, 2, 3]
    another_list = []

    if True:
        if True:
            for item in some_list:
                another_list.append(generate_item( # noqa: PERF401 # Personal preference
                    item=item,
                    stuff=1))

def generate_item(item, stuff):
    return item + stuff

Personally, the biggest usability issue for me was that the first fmt run returned no errors - I usually assume the code would be stable after a run of fmt with no errors.

Hatch version: 1.14.0

[Grub4K]

hatch fmt uses ruff as backend, running first the linter and then the formatter. Similar behavior can be reproduced using plain ruff:

$ ruff version
ruff 0.11.0

$ ruff check --select PERF401 .
All checks passed!

$ ruff format .
1 file reformatted, 4 files left unchanged

$ ruff check --select PERF401 .
testing/src/testing/__main__.py:11:17: PERF401 Use `list.extend` to create a transformed list
   |
 9 |           if True:
10 |               for item in some_list:
11 | /                 another_list.append(
12 | |                     generate_item(  # noqa: PERF401 # Personal preference
13 | |                         item=item, stuff=1
14 | |                     )
15 | |                 )
   | |_________________^ PERF401
   |
   = help: Replace for loop with list.extend

Found 1 error.

I think this is rather a ruff problem and not much a hatch problem. The reason is that the formatter breaks the line, moves the comment and then it no longer targets the correct line.

I do not think there is a good solution to this problem from the hatch side.