[ODE] Trimesh / Box - Zero Length Normal fix

[David Walters]

On 28/10/06, Jason Perkins <starkos at gmail.com> wrote:

> A couple of notes: I only applied the changes directly related to
> GIMPACT support. So the trimesh/box "null normal" fix and the
> autodisable fix were not applied. Those should be submitted and
> considered separately.

I've submitted the box/mesh zero length normal patch to SVN because in
my tests of my own engine and 'test_moving_trimesh' it appeared to
demonstrate that boxes don't get stuck in trimeshes as often.

My own engine is using very large triangle pieces for collision - and
although they do still occur - so this isn't the "silver bullet" for
fixing trimesh / box stuck penetration errors (which are mostly rare
in a balanced simulation anyway) - it does appear to be step in the
right direction.

I have made a small alteration to the patch because it was using a
hard coded epsilon value - this is now dEpsilon, I assume this will
yield correct behaviour for doubles as well.

Regards,
David.