startin
A Delaunay triangulator where the input are 2.5D points, the DT is computed in 2D but the elevation of the vertices are kept. This is used mostly for the modelling of terrains.
The construction algorithm used is an incremental insertion based on flips, and the data structure is a cheap implementation of the star-based structure defined in Blandford et al. (2003), cheap because the link of each vertex is stored a simple array (Vec
) and not in an optimised blob like they did.
It results in a pretty fast library (comparison will come at some point), but it uses more space than the optimised one.
The deletion of a vertex is also possible. The algorithm implemented is a modification of the one of Mostafavi, Gold, and Dakowicz (2003). The ears are filled by flipping, so it's in theory more robust. I have also extended the algorithm to allow the deletion of vertices on the boundary of the convex hull. The algorithm is sub-optimal, but in practice the number of neighbours of a given vertex in a DT is only 6, so it doesn't really matter.
Robust arithmetic for the geometric predicates are used (Shewchuk's predicates), so the library is robust and shouldn't crash (touch wood).
I made this in Rust because I wanted to learn Rust.
But if you prefer Python, I made bindings (very beta): https://github.com/hugoledoux/startin_python/
Usage
extern crate startin;