Irrlicht Engine

Ok, took a little longer, but it was worth waiting
I finally found some images which really took LOOOONG to load (or better, to create the scene node) so I implemented the per-material mesh buffer approach initially mentioned by myself in this thread. Now, loading is quite fast even with lots of faces (it was the bounding box calculation performed for each face ). I've also implemented a way to avoid the 16bit indices problem of Irrlicht by creating a new mesh buffer if the old one is full. so as long as no mesh has more than 65535 points the complete scene will be shown correctly.
Also fixed some material and normals problems, moved the normal calculation into the parser (so normals are calculated only once now) and probably some other good things
Ah, another change to Irrlicht headers: The subString method in irrString.h must be declared const (for the same reason as toEuler)

Another major update: Groups and figures (non-animated, though) are working. Groups are fully supported, figures are correctly read, but some placements of objects seem to be wrong, although all calculations should be fine. The objects are currently placed as additional scene nodes, which is definitely not the best way - I know that! But it was a quick hack based on the former object loading. Even the skeleton is built from several test scene nodes
I guess I'll need some good tutorial on the way skeletal animations should be built in Irrlicht as I don't get a sense out of the directX confusion. Any easy guideline of which data structures are involved and should be accessible/ are used?

I simply would expose a transformation matrix and a parent per joint, that would be enough to do animations. (BTW: going to rewrite that skeletal animation stuff anyway soon.)