Martin Vermeer wrote: > This bug is apparently not related to the patch. Oldfashioned \frac > crashes too. It is again a case of "cursor pos in mid-air": create a > \frac, type a character into the numerator, then into the denominator > (leaving cur.pos() = 1), then move out of the fraction to the right. > Then, repeatedly undo. > > At some point you will get a cursor at pos=1, in a cell having only 0 > character positions.
Maybe we should disallow direct write access to DocIterator::pit() and DocIterator::idx() and provide setter methods that reset pos() if needed. Or would this be a performance problem? > Is the FIXME in undo.h related to this? > * FIXME: We need something to record undo in partial grids for mathed. > * Right now we use recordUndoInset if more than one cell is changed, > * but that puts the cursor in front of the inset after undo. We would > need * something like > * recordUndoGrid(LCursor & cur, Undo::undo_kind kind, idx_type from, > idx_type to); * and store the cell information in class Undo. Not directly. The reason for this FIXME is the fact that the cursor jumps out of a grid inset after undo, but that is simply an UI issue, the structure in memory is correct. Maybe we need recordUndoInset instead of recordUndo at more places? > I have a backtrace but it is not useful. It looks the same as in earlier > Are all these in bugzilla? IIRC not. Georg
