[ODE] ODE 0.50 distribution

[John Miles]

Basically, I have a trimesh terrain that consists of a lot of small
triangles lying in (in this case) a flat plane.  If an object represented by
a relatively-large OBB falls onto the terrain, it tends to sink into the
terrain, sometimes stopping around the halfway point, and other times
falling on through.

You can see an example of this halfway-point sinkage at
http://thegleam.com/test.jpg .  The bottom half of the bow's OBB is below
the trimesh terrain.

With Erwin's code, it behaves fine.  The objects land on the terrain and
settle down rapidly, especially with the new QuickStep solver enabled.

I don't use CCylinder-TriMesh collisions for anything except kinematic
collisions (where I reverse the collision penetration manually to move the
user-controlled object back, without generating any contacts).  This was
pretty heavily-tweaked for use with Erwin's collider, so it's not surprising
that it doesn't work well with the new one without a similar amount of
tweaking and contact-filtering.

Again, if I'm the only one who's noticing this, it's most likely some
artifact of my own implementation.  It works fine when the engine is built
with either Erwin's OBB-TriMesh collider or the Novodex library... just not
with the Croteam collider.

I wouldn't sweat it unless others report similar misbehavior.  That's why I
posted the description of the problem -- to see if it's happening to anyone
else.

-- jm

>     That's very strange men!
>     In my side, Croteam code works very well and I think it's much better
>     than Erwin's. For example, in the case capped cylinder vs trimesh,
>     Erwin's code only generate ONE collision point (which is
>     absolutely not right). Could you please point out the case you
>     found Croteam's code fail?
>