MBorne

I think that precision lost is a matter of serialization. We failed to keep "exact computation" between PostGIS function calls because : - geometries are "GSERIALIZED" between each function call...

@vmora If we use a single kernel, we should use Epeck. Provided points are not created (situation where Epeck is required), the price should be a factor 2 compared to...

> why not an un-templated Geometry (variant) with both epick and epeck cases ? The idea is that there is enough dynamic dispatch to manage in the C++ with the...

@vmora It's possible to do approximate computation with Epeck kernel : See this example from Sébastien : https://github.com/Oslandia/SFCGAL/blob/sprint/doc/devnotes/cgal-approximate.md (should be "just" twice slower than pure double operations)