[ODE] Large trimesh triangles

[Jon Watte]

> I've seen references to this problem on the list several
> times.  Out of curiosity, is there some good reason why
> this should be the case?  It seems to me that collision
> between small geoms and large triangles should be roughly
> equivalent to collision between geoms and planes, which

A plane is very well defined using its normal and a distance. 
The collision test is typically just a dot product along the 
separating axes.

A very large triangle, however, uses more math, and thus has 
more opportunity for screwing up WRT precision.

That being said, if a triangle is size 100, and the collider 
is size 1, I wouldn't expect a lot of trouble. If the triangle 
is size 10,000, I think you would start to see wierd behavior, 
as you're getting into a relative size difference that's hard 
to represent in a 32-bit float.

Cheers,

			/ h+