Benchmark on typical data

[brendankenny]

trafficstars

I have some simple benchmarks I sometimes run locally, but they need to be checked in to start solving some performance bottlenecks.

To do this well, we need good example geometry of what people actually need to triangulate, ranging from extremely simple polys (e.g. building footprints), to complicated but well behaved polys, to the self-intersecting messes commonly encountered with geo data.

@bcamper may have a good idea of representative data that's also distributable under the SGI/X11/MIT-like license used here.

[bcamper]

Sorry for the delay -- will get you some representative scene data we use for Tangram shortly. Good to see the project picking back up :)

[brendankenny]

Great!

No worries on the delay. If you had responded earlier, it would have just been you waiting on me then :)

I'm just now getting the automated tests running (and it's just correctness testing in now, no benchmarks yet).

Tangram data will be great. For really big polygons, like the Natural Earth data or the WWF ecoregions, the performance profile is very clear on where all the time is going, but for sets of lots of small polys it's much more of a "100 functions each at 1% of execution time" profile, so it'll be really helpful to have an actual use case to know when improvements are not just theoretical.