This is a feature that was incredibly important for our legal market but actually makes a lot of sense for the Web. The idea is that you have two documents which are supposed to be forked variants of the same original, but the changes aren’t neatly stored in revision marks. With lawyers, this is usually because the opposing party is hostile and doesn’t want to make things easier for you. On the Web, this could be because the person you gave the document to didn’t have a revision-savvy editor. So what you basically need to do is run a DIFF on two arbitrary documents, and generate revision marks from that. After that, other issues like merge and review reduce to the easier case.

This comparison is not trivial. In a classic text file DIFF, you just point out which lines differ. But in a real document, you want to show how they differ - what happened. Inferring what happened from what’s left is imprecise, though. It can be very hard to tell if A moved a paragraph and B deleted half of it, or if A left it alone and B moved half and deleted the rest, and so on. We have heuristics that guess but they’re imperfect.

Unfortunately, this will probably be a likely scenario until there is a widely supported revision scheme.