CRDTs seem to give the best experience when they correctly model the "intent" of changes. But a diff between two different states of raw text can't convey the

CRDTs seem to give the best experience when they correctly model the "intent" of changes. But a diff between two different states of raw text can't convey the intent of a code change (beyond very simple changes). This is why I think CRDTs haven't caught on for VCSes and I'm not sure they _could_ without some kind of structured editing.