Performance issues with "Snap to path intersections"

[ScislaC]

When "Snap to path intersections" is enabled, performance can take a huge hit. To make it worse, even if snapping is disabled on a global level, it still affects performance if that option was enabled.

Tested with 0.48.3.1 and trunk r11935 on Ubuntu 12.10. Try to move the objects with the mouse in the attached file. Simply turning off the "Snap to path intersections" option will make it perform as expected.

If this gets fixed and a backport is possible, we should do it.

[su_v (suv-lp)]

Reproduced with Inkscape 0.47, 0.48.x and current trunk (r11936) on OS X 10.7.4.

Importance 'High' -> if global snapping is turned off, no snapping should occur at all.

[Diederik van Lierop (mail-diedenrezi)]

This has been improved in rev. #11937. It is line 1199 in sh-shape.cpp

  cs = self_crossings(*path_it)

which is the culprit. It was even called twice! I fixed that, and also made sure that it is no longer being called when not needed (determined by the status of the snap toolbar).

So it can still be slow for complex paths, but at least the delay for the selection to start moving has halved. It should start moving at some point though, just try to be patient ;-)

Please feel free to apply these changes to the v0.48.4 branch, should you have a checkout of that branch lying around somewhere....

[ScislaC (scislac)]

it's an improvement, working on backport

[Diederik van Lierop (mail-diedenrezi)]

Thanks!

The things we could do to improve this even further are:
1) speed-up the 2geom routines
2) calculate the intersections pre-emptively, while inkscape is idle,
and cache them
3) calculate the intersections while dragging (not before), in a
seperate thread

Each of them is a whole different ball-game unfortunately

[ScislaC (scislac)]

not all could be backported as a lot of changes have been made since 0.48, however, the most crucial bits did backport so it performs the same as trunk. (backported in r9934 of 0.48.x branch)

[Diederik van Lierop (mail-diedenrezi)]

Sorry, didn't see that coming! If I had known than I would have taken
care off this myself.

Thanks for your help though!

[su_v (suv-lp)]

@Diederik - if time permits, could you please take a look at bug #1093739, and close that report if this change in 0.48.4 wrt the default snap behavior is essential to address the performance issue reported here (bug #1087470)?