On Sun, Apr 09, 2006 at 11:35:08AM +0300, Martin Vermeer wrote: > On Sun, Apr 09, 2006 at 10:36:15AM +0300, Martin Vermeer wrote: > > > > The following can probably be exploited to debug Undo's problems in > > mathed: > > > > 1) Insert a math inset with a 2x2 math array inside. > > 2) Place the cursor in position (2,1), i.e., second row, first column. > > 3) Press Delete. > > > > The second row of the math array will be deleted. > > > > 4) Press Undo (C-z). > > > > You will get the error message > > > > "wrong idx 2, max is 1 at level 3" > > > > which is both right and wrong: Undo has failed to restore the 2x2 > > matrix, where this position would legally exist. > > > > So, Undo doesn't do a proper job here. Repeated undo doesn't help > > either. The second array row is gone and stays gone. > > Correction: the problem is not with undo (wrongly accused ;-/ ) but with > the way the various record methods are used in math. NestInset is not > aware that Delete does some special things inside a math grid. > > The attached fixes the bug at the root cause. > > Does anybody see a problem with this? > > - Martin
I'll commit this to trunk presently. - Martin
Index: math_nestinset.C
===================================================================
--- math_nestinset.C (revision 13612)
+++ math_nestinset.C (working copy)
@@ -621,7 +621,7 @@ void MathNestInset::doDispatch(LCursor &
case LFUN_DELETE_WORD_BACKWARD:
case LFUN_BACKSPACE:
if (cur.pos() == 0)
- // delete whole cell
+ // May affect external cell:
recordUndoInset(cur, Undo::ATOMIC);
else
recordUndo(cur, Undo::ATOMIC);
@@ -630,7 +630,11 @@ void MathNestInset::doDispatch(LCursor &
case LFUN_DELETE_WORD_FORWARD:
case LFUN_DELETE:
- recordUndo(cur);
+ if (cur.pos() == cur.lastpos())
+ // May affect external cell:
+ recordUndoInset(cur, Undo::ATOMIC);
+ else
+ recordUndo(cur, Undo::ATOMIC);
cur.erase();
break;
pgpO61curgtu6.pgp
Description: PGP signature
