exploring prior art of merging techniques

[joyously]

I've read in the Git manual that it doesn't make guesses when merging. If it's not obvious, it is a conflict.
And yet, there are (perhaps old) stories of Git messing up:
When a Clean Merge is Wrong
Is it possible for Git merging to make a mistake without detecting a conflict?

So I assume there is some magic code in jj to be able to do all the rebases and rewriting that it does. Wasn't there a fairly recent comment about maybe needing more information for the more complex cases? I know Darcs and Pijul can compute the opposite of a change and that helps somehow (I'm not familiar with the algorithm even).

I was reading about AutoMerge, which uses CRDT for collaborative editing, and their docs state the rules it uses for its basic types to merge. I wonder if this is rigorous enough for code.

[amiryal]

With regards to the question:

Wasn't there a fairly recent comment about maybe needing more information for the more complex cases?

I did mention it in passing in #3109 (comment) (look for “side note” and “pathological cases”). That comment includes a couple more links to such pathological cases with extensive commentary:
https://tahoe-lafs.org/~zooko/badmerge/simple.html
https://tahoe-lafs.org/~zooko/badmerge/concrete-good-semantics.html

[amiryal]

When a Clean Merge is Wrong

Thanks for sharing this one!