Enforce empty lines before classes/functions with sticky leading comments.

[yilei]

Description

Enforce empty lines before classes/functions with leading comments:

• Fixes Black doesn't enforce two empty lines before functions on module level when they have a sticky leading comment #246

EmptyLineTracker works on lines that haven't been split yet. When sticky leading comment is used before class/def lines, the tracker must look behind the possible comment lines. So I introduced a new LinesBlock class to keep track of those states.

Checklist - did you ...

• Add an entry in CHANGES.md if necessary?
• Add / update tests if necessary?
• Add new / update outdated documentation?

[github-actions]

diff-shades results comparing this PR (36ae752) to main (575220f). The full diff is available in the logs under the "Generate HTML diff report" step.

╭─────────────────────── Summary ────────────────────────╮
│ 6 projects & 33 files changed / 36 changes [+36/-0]    │
│                                                        │
│ ... out of 2 326 398 lines, 10 700 files & 23 projects │
╰────────────────────────────────────────────────────────╯

Differences found.

What is this? | Workflow run | diff-shades documentation

[yilei]

I pushed a few new commits to address a few unexpected diffs from diff-shades.

I read through the latest diff-shades results again, and they look all expected.

[felix-hilden]

Thanks a ton! Some comments below, and one discussion point: would it make sense to consider the comments to be in the block? Instead of having an outer "semantic comment" I'm not 100 % sure, but that could simplify things so that we can use ordinary newline rules for the whole block. Maybe? 🤔

[yilei]

would it make sense to consider the comments to be in the block? Instead of having an outer "semantic comment" I'm not 100 % sure, but that could simplify things so that we can use ordinary newline rules for the whole block.

This is appealing. But I tried and couldn't yet find a way to actually simplify this by considering the comments to be in the block.

To decide whether a comment belongs to this block, we have to examine what the block is, as well as the newlines in the first leaf prefix. This logic is already part of EmptyLineTracker. So this would require a big surgery to redesign EmptyLineTracker and likely also the LineGenerator.

All the empty line logic is still contained inside the EmptyLineTracker, so I feel this PR's implementation isn't too bad.

[yilei]

PR updated to keep the current behavior for inner functions after code block open. This now only fixes #246.