Black 22.1.0 strips some comments

[henryiii]

Describe the bug

Black will now occasionally strip comments (including type: ignore comments).

To Reproduce

For example, take this code:

if hasattr(view, "sum_of_weights"):
    return np.divide(  # type: ignore[no-any-return]
        view.variance,  # type: ignore[union-attr]
        view.sum_of_weights,  # type: ignore[union-attr]
        out=np.full(view.sum_of_weights.shape, np.nan),  # type: ignore[union-attr]
        where=view.sum_of_weights > 1,  # type: ignore[union-attr]
        where=view.sum_of_weights**2 > view.sum_of_weights_squared,  # type: ignore[union-attr]
    )

And run it with these arguments:

$ black file.py

The resulting Python code is:

if hasattr(view, "sum_of_weights"):
    return np.divide(  # type: ignore[no-any-return]
        view.variance,  # type: ignore[union-attr]
        view.sum_of_weights,  # type: ignore[union-attr]
        out=np.full(view.sum_of_weights.shape, np.nan),  # type: ignore[union-attr]
        where=view.sum_of_weights > 1,  # type: ignore[union-attr]
        where=view.sum_of_weights**2 > view.sum_of_weights_squared,
    )

Expected behavior

Not to strip a type ignore comment!!!

(Personally, I think all trailing comments should be ignored from all line length calculations, etc, since they are often turning off linting or type checking for the line they are on. But it clearly shouldn't just vanish!)

Environment

• Black's version: main and 22.1.0
• OS and Python version: Any (macOS, pre-commit.ci's linux, and whatever the web app runs on)

Additional context

Pretty self contained. :) I had noticed a similar error on upgrading to 22.1.0, but it went away due to other changes, and I never worked it back down, but I'm hoping it was the same one.

[ichard26]

Hmm, is there a reason why we ignore type comments in the AST @JelleZijlstra? this would be an excellent reason to stop doing this as this is problematic.

[henryiii]

Simpler example:

return np.divide(
    where=view.sum_of_weights_of_weight_long**2 > view.sum_of_weights_squared,  # type: ignore
)

I think it has to be in a function call, and it has to have a ** in the same line, and has to be longer than the line length. Changing ** to * causes this stripping not to happen.

I think type: ignore comments are ignored because they can never be moved to the next line. Though that's also true for # noqa:*, # pylint disable=*, etc.

[ichard26]

Ah this is probably my fault, I suspect the power op hugging code I added in GH-2726 fails to copy over the Leaf prefixes which is where comments are stored. I'm currently not at home though so I won't be able to fix until later.

Is this worth a hotfix release? The damage is bad but it could not be common enough to warrant a hotfix.