I (finally!) had a look at this. If I understand it correctly, the core change is the change from partition_after_prev_or_before_next to Source.cmt_placement. I have some remarks on the implementation itself but it might a bit early to go in too much detail.

The or-pattern changes seem to be a net improvement to me! And the rules seems much simpler to understand.

It's a bit early to talk integration, but if you can continue improving that on the side and have net wins like the or-patterns I think this can balance the large diff this is going to have. We can also gate that behind a feature flag if the outcome is not super clear.

I'm closing this one because:

• it diverged from master too much and it's painful to maintain
• the implementation is even more difficult to maintain than master's
• I will explore the approach of attaching the comments only once, just after the parsing, instead of after each iteration until we reach a fixpoint, for that I will work on having ast passes (and one of these passes will attach the comments in the ast)

I will explore the approach of attaching the comments only once, just after the parsing, instead of after each iteration until we reach a fixpoint, for that I will work on having ast passes (and one of these passes will attach the comments in the ast)

You'll still need to make sure you reach a fix point with the comment placement.