On Mon, Jun 16, 2003 at 11:38:52AM +0200, Andre Poenitz wrote: > I have been able to reproduce the 'diagonal lines in table' drawing > issue (which was one of two for which 'remove update()' got thumbs > down) with 1.3.x.
That was one of the two reasons you gave. I saw massive horkage of almost everything table-related. > So this is nothing introduced by the 'remove update' patch but some old > issue that has been somehow covered so far (probably by the excessive > updates and redraws). Covered means the user can't see it normally. > [The argument goes like this: > > A: > 1. update() does not work as seen by the 1.3.x diagonal lines > 2. nobody understands it anyway > 3. update() is the core reason for a few owner_ back-pointerage > 4. ... which makes the core a mess as such lead to non-trivial > copy and assignment in insets. > > B: > 1. metrics()/draw() conceptually works in theory > 2. there are no known technical limitations > 3. it works in practice as shown in math > 4. it has recently simplified some code in insets/* > > A, B -> let update() die even if there is an intermediate set back > like the mentioned drawing problems.] My argument goes like this: 1. do it on a branch, get it working with no regressions 2. merge it I know I will annoy you by saying this. But the last thing 1.4 can afford is massive breakage of code nobody understands and will take at least 4 months to fix up. Undo is really a drop in the ocean compared to the complexity of the update() stuff. I would *love* to see update() die but we cannot afford to do such dangerous work on the trunk, not now. Just my small change ... john
