@Nirmos: I think this is not a problem anymore?

Things are still broken. For a minimal test case, see for example https://sv.wikipedia.org/w/index.php?diff=46200983

The .diff-deletedline is further down than .diff-addedline, despite the article only containing one line, and that's the line being changed. It didn't use to be like this, but it's been like this for quite some time now.

In T164795#5411230, @Nirmos wrote:

Things are still broken. For a minimal test case, see for example https://sv.wikipedia.org/w/index.php?diff=46200983

The .diff-deletedline is further down than .diff-addedline, despite the article only containing one line, and that's the line being changed. It didn't use to be like this, but it's been like this for quite some time now.

This is a pretty clear and reproducible simple thing, that's why I put it into a separate ticket: T230432: Changes in lines with only one word are not detected as change. It seems to be a regression from the wikidiff2 algorithm changes.

The original issue described in this ticket describes a different problem. The example diff [1] above looks a bit confusing due to the new line that was removed, while also edits in the paragraphs were done. - When lines are removed or added the diff algorithm does not align these changes very well in some cases.

That's a problem that exists for some time now and also has nothing to do with the changes in wikidiff2 done in the last three years. The ability to detected changed and moved paragraphs improves that situation even slightly. We tried to fix the alignment while working on wikidiff2 but the problem turned out to be quite complex T184531: Treat empty lines as part of the previous paragraph ( we even used the above example for that )

[1]https://en.wikipedia.org/w/index.php?title=Honeyguide&diff=prev&oldid=779422017

@Nirmos if you're fine with it, I would change the ticket description to reflect the bug I'm describing above. - I did not really find a ticket for that so far so it might be good to use this one.

Absolutely, I'm very happy that you're trying to sort things out! 👍

Example reported at T7072

In T7072#3297458, @Zazpot wrote:

As discussed here, another example of this bug is at https://en.wikipedia.org/w/index.php?title=MalwareTech&diff=782456060&oldid=782449642 . In that diff, both the left and right columns have hunks that begin "Following his work on the WannaCry ransomware attack in 2017" and that are almost identical (edit distance: 3) but that have been aligned with other hunks instead of with each other, making it very hard to spot what has changed between them. (To spare you searching, it is "he's" to "he has".)

I expect the solution to this bug will involve matching paragraphs according to minimum edit distance, with a fallback algorithm in case two or more paragraphs are equal edit distances away.