Formatter undocumented deviation: No wrap in long return

[henribru]

Black:

def foo():
    if foo:
        return (
            aaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaa.bbbbbbbbbbbbbbbbbbbbbbbbbbbbbbbbbbbbbbbbbbbbbbb
        )

Ruff:

def foo():
    if foo:
        return aaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaa.bbbbbbbbbbbbbbbbbbbbbbbbbbbbbbbbbbbbbbbbbbbbbbb

[MichaReiser]

I'm slightly confused. Black generally tries to avoid parentheses if the parenthesized content also exceeds the line length. We added support for this behavior in #6817 (and refined it later). Now, it seems that this logic doesn't (always?) apply for attribute chains?

[charliermarsh]

I don't quite understand that either, I wonder if it's intentional.

[njzjz]

I got the same behavior with assert.

[MichaReiser]

I played with this further and it is correct. Black always wraps attributes in parentheses, even if it doesn't make them fit.

Black

# fits parenthesized
______ = (
    aaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaa.bbbbbbbbbbbbbbbbbbbbbbbbbbvvvvvvvvvvvv
)
return (
    aaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaa.bbbbbbbbbbbbbbbbbbbb
)

# exceeds parenthesized
______ = (
    aaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaa.bbbbbbbbbbbbbbbbbbbbbbbbbbvvvvvvvvvvvvvvvvv
)
return (
    aaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaa.bbbbbbbbbbbbbbbbbbbbbbb
)


def doesnt_exceed_line_length_when_parenthesizing():
    ______ = (
        aaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaa.bbbbbbbbbbbbbbbbbbbbbbbbbbvvvvvvvv
    )
    return (
        aaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaa.bbbbbbbbbbbbbbbb
    )


def exceeds_line_length_parenthesized():
    ______ = (
        aaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaa.bbbbbbbbbbbbbbbbbbbbbbbbbbvvvvvvvvvvvvvvvvvv
    )
    return (
        aaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaa.bbbbbbbbbbbbbbbbbbbbbbbbbb
    )

Ruff would remove the parentheses for the "exceeds line length" cases. I believe the same is true for call chains.

Changing our code to use the normal Multiline layout for attribute chains in itself isn't too hard. However, BestFit and `Multiline don't just differ in when they set parentheses, they also use different breaking precedences:

• BestFitParentheses: Breaks right first. It first breaks the value (right) before parenthesizing the target (left). Similar to BestFitting
• Multiline: Breaks the left first: It breaks the target (left) before parenthesizing the value (right). Normal group sequence.

Changing the precedence between breaking the target or value has the result that:

state[
    "event_queue_longpoll_timeout_seconds"
] = settings.EVENT_QUEUE_LONGPOLL_TIMEOUT_SECONDS

gets reformatted to

state["event_queue_longpoll_timeout_seconds"] = (
    settings.EVENT_QUEUE_LONGPOLL_TIMEOUT_SECONDS
)

which matches Black's preview style, at least for attributes. I believe implementing the preview style requires changing BestFit to be right to left too, but that seems straightforward.

Changing the precedence regresses our similarity index significantly and I don't know of a good way to make our parenthesizing left to right. The way to achieve this is by:

• Either use BestFitting which will be slower and requires another MaybeParenthesize layout which I would prefer to aovid.
• Change the group hierarchy so that the target expression's group encloses the value. This is challenging because the left could be any expression.

I don't think it's worth spending time fixing this now, considering that implementing preview style "solves" the problem and it seems rare enough (I didn't find any usages in our baseline projects). Arguably, Ruff's behavior is also more consistent because it uses it for all expressions that are "unsplittable" whereas Black only uses it for some unsplittable expressions.

@charliermarsh I propose that we document this as know and intentional deviation.

[MichaReiser]

I opened an issue on the Black repository to see if omitting parentheses is intentional. psf/black#4001

[MichaReiser]

Black considers aligning with ruff's style.

[MichaReiser]

I recommend us to keep it as for now and see if Black changes their style.

[henribru]

Can it get added to the documented deviations in the meantime?

Edit: I see now that you already suggested that earlier.

[henribru]

@MichaReiser This is in a return type and not a return value, but is this the same issue?

https://black.vercel.app/?version=stable&state=_Td6WFoAAATm1rRGAgAhARYAAAB0L-Wj4AORAahdAD2IimZxl1N_WlbvK5V8VwGvpky2hHvxJKDBSYISDIkXZGJG4SJMgx6_11S4gUOnTDWjHwJaiM6dlpIbDBLvtd7rfffNEd3HafefdTpwBZWZsH2D652XvLSeuqzfBgpUW8YBK4WelZ6dWGMxHiSwGu2chnbe3hJfPqsGS4WucjxLTHCHhdIxlqwYohxVBltlO12woMOHr1NP6DVUc5HBxFOIKjgjZbilMZn-7-E1XWXCYLkoStiN4kGXio0D_kEOEYgY75dedJRg1cOuhcbG-GVyX8IRNbluxBAwSUGnl6RgPCh_ifBOYvI9kFWBVXB5zOFIysanReiU28yBqcIToJ_cstX_N4sw12ftCNgzy9bOwu4KyvMd0Y59g5XqhSmuoIYKChIENIVJh319IdJH9L5MqaH3UwWcwzCSiNQDGuMOZBxT2UX-zmAazCS-7sLFQ4P77cdKdjhFuEzr-ujdHBakv9G9elRwKG2IVIJXSEVXP6RF6aYg0nhgk9rAUFqYsNQjf6wSpDFqrW-SGITiAO2unHRfDZe2DFfkGs_SRpDO3Jpm6KpowVYAJhy94zLR2lsAAcQDkgcAAAXNihyxxGf7AgAAAAAEWVo=

https://play.ruff.rs/88f317d4-7f5a-4d59-8404-a2b117c47caa?secondary=Format

[MichaReiser]

@henribru yes this seems to be the same issue. Changing the return type to an identifier (remove all dots) makes Black match Ruff's formatting

https://black.vercel.app/?version=stable&state=_Td6WFoAAATm1rRGAgAhARYAAAB0L-Wj4AONAaNdAD2IimZxl1N_WlbvK5V8VwGvpky2hHvxJKDBSYISDIkXZGJG4SJMgx6_11S4gUOnTDWjHwJaiM6dlpIbDBLvtd7rfffNEd3HafefdTpwBZWZsH2D652XvLSeuqzfBgpUW8YBK4WelZ6dWGMxHiSwGu2chnbe3hJfPqsGS4WucjxLTHCHhdIxlqwYohxVBltlO12woMOHr1NP6DVUc5HBxFOIKjgjZbilMZn-7-E1XWXCYLkoStiN4kGXio0D_kEOEYgY75dedJRg1cOuhcbG-GVyX8IRNbluxBAwSUGnl6RgPCh_ifBOYvI9kFWBVXB5zOFIysanReiU28yBqcIToJ_cstX_N4sw12ftCNgzy9bOwu4KyvMd0Y59g5XqhSmuoIYKChIENIVJh319IdJH9L5MqaH3UwWcwzCSiNQDGuMOZBxT2UX-zmAazCS-7sLFQ4P77cPOFJOUUipdc1UEHCM3XhBwE05-vHjsZevtZNgzB30fWy286CplsUtQ3fVAA-HF6kXbP--S2KEDJfhJi2ewDi4v_VnBJOZXP3iUkqYLeyDRAADRqnwBGWR3AAABvwOOBwAAINigZ7HEZ_sCAAAAAARZWg==