[ODE] collisoin_trimesh_box.cpp bug?

[O'dell Hicks]

> > This is because contacts were generated for both polygons with opposite
normal.
> >  One contacts tried to pop the body, but others opposite delayed him.
> > Has fixed this as follows.
> > In "collision_trimesh_box.cpp" after call FetchTriangle(TriMesh,
TriIndex, TLPosition, TLRotation, dv);
> > has inserted check: where is turned normal of the polygon for the centre
of the body.
>
> trimesh<->box collision was rewritten after ODE 0.039... you
> could try one of the snapshots (if you're not already).
>

Oh, is/was this also happening with spheres?  I've noticed that spheres are
more likely to pass through or get stuck in a mesh than cubes of a similar
volume.  These are spheres with a one meter radius moving at the speed of a
walking human.  Physics updates at 60 hz.