Andre Poenitz wrote: > On Wed, Nov 03, 2004 at 01:57:59PM +0100, Alfredo Braunstein wrote: >> The most obvious regression is that the cursor doesn't leave the exact >> vertical center of the screen. > > That was sort of intentional for the time being as it help to fix one > part of the the problem at a time. Right now the main goal was to get > paragraph breaking and cursor movement right. Selecting the correct y > position is fairly orthogonal to this. > > But if you found that it's barely usable work on the other side could > begin... > >> I'll work on removing this limitation (sort of reintroduce the >> anchor thingy - *another* cursor position + y vertical offset). > >> Btw, I didn't appreciate at all that you have *removed* fitCursor and >> friends. The fact that fitCursor isn't used in this version is a >> consequence of the limitations of this approach, but we certainly need it >> in the final version. > > In hindsight you are propably right. It didn't really occur to me that > we sometimes view positions far from the cursor (scrolling with > non-following cursor, I don't do that. From this POV 'fitCursor' is > superfluous as the cursor alwways fits per definition. > > So we should not/cannot use the _cursor_ as primary information but > something like your 'anchor position' (maybe just outer paragraph index > + small y offset instead of a full DocIterator).
Yes, I've arrived to the same conclusion fortunately (and implemented it). One problem that I inmediately had was selections. The whole scheme was too much dependent on the cursor having good coordinates (which was always correct because was always on screen), and this was hidding a small problem: You have moved the selection painting code out of rowpainter and put it between the two drawing steps (after redoParagraph and before paintPar). The problem was that there, y information of paragraphs was not available yet and was needed for drawing the selection. Fortunately, I realised that y information (height and position) for *outer* pars is indeed available after the redoParagraph scan, and can be then be sent to the cache right away. The whole is a bit delicate for my taste... maybe I liked the selection painting in rowpainter was better (what where the disadvantages?). Alfredo
