Trailing comma not removed in type annotation

[fuglede]

Describe the bug

By running black with different line lengths, I can produce inconsistent results that ultimately fail in flake8 (and look strange).

To Reproduce

Consider this relatively silly example (let's call it test.py), with a line of length 81:

from typing import Tuple


def f(a: Tuple[int, int, int, int, int, int, int, int, int, int, int, int, int]):
    return a

Both black and flake8 are happy with this:

$ flake8 test.py
test.py:4:80: E501 line too long (81 > 79 characters)
$ flake8 test.py --max-line-length=88

All is good. Now, run black with -l 70, then without a line length

$ black -l 70 test.py
reformatted test.py
All done! ✨ 🍰 ✨
1 file reformatted.
$ black test.py
reformatted test.py
All done! ✨ 🍰 ✨
1 file reformatted.

Now the file looks different from the starting point:

from typing import Tuple


def f(
    a: Tuple[int, int, int, int, int, int, int, int, int, int, int, int, int,]
):
    return a

Moreover, flake8 doesn't like the trailing comma:

$ flake8 test.py
test.py:5:77: E231 missing whitespace after ','

By ignoring the change of line length in the example above, you can also choose to think of this of an example where black (with no line length specification) causes issues with flake8; I'm not sure if this is a flake8 bug (a la this one) or a black issue. Even then, it seems like black test.py should be equivalent to black -l 70 test.py && black -l test.py.

Environment (please complete the following information):

• Version: black, version 19.10b0
• OS and Python version: Linux/Python 3.7.6

Does this bug also happen on master?

I tried https://black.now.sh/?version=master which fails to convert the intermediate result back to the one-liner when given a line-length of 88. I'm not sure if this is just a separate bug.

[JelleZijlstra]

This no longer reproduces exactly because of the magic trailing comma, but you still get the ,] behavior with black --skip-magic-trailing-comma. Instead, Black should remove the comma.

[JelleZijlstra]

Still reproduces on current main with --skip-magic-trailing-comma.

[vors]

This would be great to fix: we are doing the black --line-length=30 && <some-mechanical-patching> && black -C --line-length=99 and this messes up the diff a great deal.

The mechanical patching part for context is adding suppressions for pyright errors.

[hauntsaninja]

@vors I think I fixed this in #3209 . My guess is Jelle only reproduced the issue because he didn't use --preview, but this is now in the stable style when using --skip-magic-trailing-comma.

https://black.vercel.app/?version=stable&state=_Td6WFoAAATm1rRGAgAhARYAAAB0L-Wj4ADqAH9dAD2IimZxl1N_WlkPinBFoZStZNLAcBn-vRxIYUS7MGdv-aZUpwMHHXlN6ADR5YhOvkX521Cx0wTY5xOdWgxHqd3EFilPK3gBt5WDqzisLFGMbfK6RerlnIRjoO1IXFuQwpYRRMHoEA97YqjmYJ4D6By7mlaQg6zirnODBZfaxwAAADWFLlqSj6CBAAGbAesBAADjtyiJscRn-wIAAAAABFla

Unless you have a different repro?

[vors]

got it, yes that's kind of an old bug lol. I will check and report back.

[vors]

Yes, all fixed. I think we should close this issue.