Formatter undocumented deviation: slice formatting

[JonathanPlasse]

Black formatting

def f():
    for event_index in range(header.number_non_empty_events):
        section_header_data = struct.unpack(
            JSON_BINARY_PARAMETERS.fmt_section_header,
            byte_array[
                byte_begin_index
                + byte_step_index * event_index : byte_begin_index
                + byte_step_index * (event_index + 1)
            ],
        )

Ruff formatting

def f():
    for event_index in range(header.number_non_empty_events):
        section_header_data = struct.unpack(
            JSON_BINARY_PARAMETERS.fmt_section_header,
            byte_array[
                byte_begin_index + byte_step_index * event_index : byte_begin_index
                + byte_step_index * (event_index + 1)
            ],
        )

Use Ruff 0.0.289 with line length to 100

[zanieb]

Do you have an ideal formatting for this case? I find both of these hard to reason about.

[JonathanPlasse]

Maybe this

def f():
    for event_index in range(header.number_non_empty_events):
        section_header_data = struct.unpack(
            JSON_BINARY_PARAMETERS.fmt_section_header,
            byte_array[
                byte_begin_index + byte_step_index * event_index
                : byte_begin_index + byte_step_index * (event_index + 1)
            ],
        )

[MichaReiser]

Thanks for reporting this. The deviation here is rooted in the fact that:

• Black breaks a line by all binary operators first as if the left and right side of the : operator are a single expression
• Ruff considered the left and right two independent expressions and split them individually

Here's an example where Black splits before all + operators even though it wouldn't be necessary to make the left side fit:

def f():
    for event_index in range(header.number_non_empty_events):
        section_header_data = struct.unpack(
            JSON_BINARY_PARAMETERS.fmt_section_header,
            byte_array[
                byte_begin_index
                + byte_step_index : byte_begin_index
                + byte_step_index
                + (event_index + 1)
            ],
        )

# Ruff
def f():
    for event_index in range(header.number_non_empty_events):
        section_header_data = struct.unpack(
            JSON_BINARY_PARAMETERS.fmt_section_header,
            byte_array[
                byte_begin_index + byte_step_index : byte_begin_index
                + byte_step_index
                + (event_index + 1)
            ],
        )

I generally prefer Ruff's approach of treating them as separate expressions but we should probably add a softline break before the : operator similar to other operands (e.g. in binary expressions) and potentially indent the right side.

Prettier's formatting looks similar to what @JonathanPlasse proposes (I had to adjust the syntax slightly to get the JS equivalent)

for (event_index in range(header.number_non_empty_events)) {
  section_header_data = struct.unpack(
    JSON_BINARY_PARAMETERS.fmt_section_header,
    {
      [byte_begin_index + byte_step_index * event_index]:
        byte_begin_index + byte_step_index * (event_index + 1),
    },
  );
}

[MichaReiser]

We decided not to implement the deviation for now and instead focus on Black compatibility because:

• It eases migration for Black users
• Having more deviation makes it harder to assess whether a deviation is a bug or due to som other deviation. The fewer deviations, the easier it is to identify the root cause and improve overall black compatibility

We plan to re-visit the formatting eventually

[charliermarsh]

We're not planning to change this for the Beta. We want to re-open the possibility of changing to match @JonathanPlasse's suggested style once we have better tooling to evaluate the impact (e.g., ruff-shades).

[MichaReiser]

We may want to consider #8897 when making a style decision to make sure our handling is consistent.