We're about to do a 3.1.1 release anyway, so anything that fixes a performance bug can be considered. It seems like the real solution is to get rid of _sane(), but I'm not sure if that's safe in 3.1.x release. Since non-transactional undo will be gone in 3.2, there's no reason to have _sane() or any of the other code that deals with the 'u' status. Jeremy