[ODE] problems with trimesh and capsules

[Paul MacKenzie]

Are there outstanding problems with trimesh-capsule collisions?

I tried to follow the scarcity of documentation to set up a simple mesh 
correctly, and it seems to be okay with spheres, boxes and cylinders, 
but with capsules I get either a stack integrity error or else the 
capsule jumps into the mesh and ODE complains about trying to normalize 
a zero vector.

For the stack problem, in _cldTestOneTriangleVSCapsule() in 
collision_trimesh_ccylinder.cpp, there are several local dVector3 
variables used.  However, after a few collision checks, it seems that 
these local variables are being placed within the allocation of the 
global variable gLocalContacts.  When this happens, of course any write 
to the gLocalContacts array stomps on the local variables and I get 
stack errors.  Is gLocalContacts's memory being mysteriously released 
somewhere?

I tried changing gLocalContact to a heap variable with malloc() instead 
of dALLOCA16(), and then I get the zero vector error.

I could be hunting all week on this, but before I do, can anyone tell me 
if their mesh-capsule collisions are working okay?  If so, perhaps I 
have a subtle initialization problem with the trimesh.

Cheers,
Paul