Closed ansis closed 11 years ago
:+1:
Yeah, this comes up in profiling. Ideally we'd have some sort of persistent RTree with unchanged branches shared between graphs. This paper looks like a promising start: The temporal R-tree G Zimbrão et al.
I haven't look at the paper yet, but we could also just have a single tree store on the history, and just remove and reinsert based on difference. Might be worth trying out quad trees.
Tracking in #753 now.
After panning around a bit, the number of entities gets high, and graph.intersects (linear) gets really slow, taking about 1s each full redraw. Probably a good idea to use a quadtree or rtree. rtree would be perfect, but I'm looking at dropping the dependency because its probably not needed for labelling.