Quoting Vladimir Savic <vladimir@...958...>:
I simply can not live without you inkscapers! :) You're guitar player, aren't you? :)
Keyboardist, actually. More of a composer than a performer though.
I have some (very old) work up online at http://moonbase.rydia.net/mental/music ... Pop and Video Game stuff mostly ... someday soon I intend to put up some more recent work which was done with Rosegarden.
More serious problem is handling of "zoom factor". The more you zoom in the more time it eats to calculate something outside visible canvas area!?
Sort of, yeah. I'm not sure whether to blame our renderer or our canvas widget (a heavily bastardized GnomeCanvas), though.
Every time you move the mouse, it checks which shape the mouse pointer is inside. To perform the actual shape intersection check, it asks the renderer, using the computed shape cached during drawing.
To do this, the renderer just iterates through the shape and counts edges. Unfortunately, if the shape had curves in it, there will be a lot of edges -- beziers are internally divided into tiny straight line segments. The higher the zoom, the more finely divided, the more edges, the longer it takes.
There are really two problems:
1. edge lists are the wrong data structure for this operation: we should be keeping a BSP tree or something which is more efficient for these purposes
2. the current renderer has horrible performance characteristics: our previous renderer was used the same way, but performance was at least tolerable
The reason we abandoned the renderer we inherited from Sodipodi, despite it being tolerably fast, was that it had severe rendering bugs and crashes in corner cases, and was unmaintainable.
The new renderer, which we got as an emergency replacement, renders things properly, but is dog slow and has also proven unmaintainable (simply replacing the guy's open-coded realloc()s with std::vector proved to be a multi-month project).
Really, we need a renderer that is all three of:
1. Fast 2. Correct 3. Maintainable
No "pick any two" comments, please. :) But really I'd settle for just #3, since that means it is still possible to fix the other two.
(Of course that is precisely the approach Cairo is taking -- they got #3, then attacked #2, and are now mainly working on #1.)
-mental