mmm... like in the current z sorting is based on each geometry translation, so we have the glitches.
And if we had the per triangle z sorting, small triangles would work better than big ones.
so, those blender models with a huge amount of triangles could work almost perfectly, as long we use them sparingly (memory, GPU/CPU usage) right?
Despite the boxes are for tests only, if I was going to actually use them, I could subdivide in many small quad geometries to create the boxes 6 walls, and put them all in a BatchNode to speed up, I guess it would work better? I wonder if BatchNodes mesh support skeleton/bones bending btw?
Finally, the pixel thing would be required to be a GPU internal capability because on the CPU it would not work well I guess (too costy)?