I've been working on my own text editor (which I'm apparently not the only one doing that here), and my plan is to switch to using a single B-tree, where the leaf nodes, besides holding the text (and also a color list for basic syntax highlighting), also a list of "pointer" objects, which would hold the carets/extra highlighting references/whatnot. Means two objects per selection (separate start & end pointer), but only one tree to have to scan through.