Thread by @lindsey

Just went on a tirade at a poor unsuspecting student, so now you get it, too!

So, the student was asking whether it's impossible to have a version control system that always automatically resolves merge conflicts, saying that he suspected it was.

The collaborative text editing problem boils down to how to merge conflicting strings. Here's one deterministic merge function: given the two strings, always return the empty string!

This is a terrible idea, but it "works", in the sense that you could implement a CRDT based on this merge function (i.e., the empty string is the top element of the lattice, with every other string one step down from it in the lattice). All replicas would indeed converge!

Or how about last-writer-wins, where "last" means causally last? You could take the causal order, extend it to an total order using a standard approach, and use that to arbitrarily pick a winner out of two concurrent updates. This would also be a deterministic merge.

So the real question is not whether it's possible to automatically and deterministically resolve string merge conflicts (because it is), but rather, how does one do it while having certain desirable properties be true of the result?

What those desirable properties are is an interesting question, and the answer probably depends a lot on the application. Maybe a version control system *could* preserve them. Would you want to use a very domain-specific VCS?

I'm not sure I would. But it's fun to think about!

Anyway, collaborative text editing is interesting. Here's a recent paper I like: dl.acm.org/citation.cfm?i…