In the last few days I dug into the Newton physics-engine a bit deeper because the NewtonCovexHull didn’t suit my needs for concave meshes.
The NewtonTreeCollision works great on static meshes and creates perfect collision on those. But since it’s not possible to create a exact – face by face – collision for dynamic meshes in a way to keep up the performance, I had to alter a different solution for dynamic meshes than NewtonTreeCollision. In order to get the best approximation for concave meshes I finally decided to use CompoundCollsion (well, I guess I really hadn’t many alternativesalternativen).
This CompoundCollision has to consist of an arbitrary count of convex collisions but can result in a huge concave one. Due to the fact that Irrlicht loads MY3D-files in a way that each part (e.g. box, plane or sphere) the format differs is stored its own meshbuffer. So if the map creator just takes care that all used parts are convex he will get a perfect collision. Otherwise if one or two parts are concave it won’t affect the whole CompoundCollision.
Map designers will get a more detailed introduction in what is possible and what might be the restrictions of that solution when the time has come.