Align line ending comments across multiple lines.

[thomasvl]

Given code like:

let base64Values: [Int] = [
    -1, -1, -1, -1, -1, -1, -1, -1, -1, -1, -1, -1, -1, -1, -1, -1,  // 0x00-0x0f
    -1, -1, -1, -1, -1, -1, -1, -1, -1, -1, -1, 62, -1, 62, -1, 63,  // 0x20-0x2f
    52, 53, 54, 55, 56, 57, 58, 59, 60, 61, -1, -1, -1, -1, -1, -1,  // 0x30-0x3f
    -1, 0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14,  // 0x40-0x4f
    15, 16, 17, 18, 19, 20, 21, 22, 23, 24, 25, -1, -1, -1, -1, 63,  // 0x50-0x5f
    -1, 26, 27, 28, 29, 30, 31, 32, 33, 34, 35, 36, 37, 38, 39, 40,  // 0x60-0x6f
    41, 42, 43, 44, 45, 46, 47, 48, 49, 50, 51, -1, -1, -1, -1, -1,  // 0x70-0x7f
    -1, -1, -1, -1, -1, -1, -1, -1, -1, -1, -1, -1, -1, -1, -1, -1,  // 0x80-0x8f
    -1, -1, -1, -1, -1, -1, -1, -1, -1, -1, -1, -1, -1, -1, -1, -1,  // 0x90-0x9f
    -1, -1, -1, -1, -1, -1, -1, -1, -1, -1, -1, -1, -1, -1, -1, -1,  // 0xa0-0xaf
    -1, -1, -1, -1, -1, -1, -1, -1, -1, -1, -1, -1, -1, -1, -1, -1,  // 0xb0-0xbf
    -1, -1, -1, -1, -1, -1, -1, -1, -1, -1, -1, -1, -1, -1, -1, -1,  // 0xc0-0xcf
    -1, -1, -1, -1, -1, -1, -1, -1, -1, -1, -1, -1, -1, -1, -1, -1,  // 0xd0-0xdf
    -1, -1, -1, -1, -1, -1, -1, -1, -1, -1, -1, -1, -1, -1, -1, -1,  // 0xe0-0xef
    -1, -1, -1, -1, -1, -1, -1, -1, -1, -1, -1, -1, -1, -1, -1, -1,  // 0xf0-0xff
]

It would be nice if swift-format made all the trailing line comments align. i.e. - the one short line should get extra spaces before the comment so all the // aligned. i.e. -

let base64Values: [Int] = [
    ...
    52, 53, 54, 55, 56, 57, 58, 59, 60, 61, -1, -1, -1, -1, -1, -1,  // 0x30-0x3f
    -1, 0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14,            // 0x40-0x4f
    15, 16, 17, 18, 19, 20, 21, 22, 23, 24, 25, -1, -1, -1, -1, 63,  // 0x50-0x5f
    ...
]

It would also apply to things like:

var foo = m.something()  // Start of long comment that continues
                         // to the next line.

and

var foo = m.something(
   mumble: 123,  // Line one.
   x: False)     // Line two.

[allevato]

My gut feeling (without thinking too hard) is that this would require some nontrivial changes to our core algorithm since we make a single linear pass over the token stream. We don't really have a way to look ahead at all the possible line comments to figure out what the spacing would need to be for alignment of later lines, and we don't really have a way to "rewind" either.

Right now the pretty printer is a class, but an interesting idea would be to refactor it into a value type—I don't think there's a fundamental reason we couldn't do that. That would give us the ability to more easily do advanced things like lookahead formatting—have the printer clone itself at certain checkpoints, do some printing up to another checkpoint, then inspect the state of the clone and incorporate that state into the real formatting pass after throwing away the clone.

We'd need to be careful about performance if we did too much lookahead/rewinding, but I think there's some potential there.

[thomasvl]

ps - clang-format does this sorta alignment, which is part of why it seems like it might make sense to support.

[ahoppen]

Synced to Apple’s issue tracker as rdar://136604806