[ODE] Collision response for symetrical tri-meshes

[Flavien Brebion]

Jon Watte (ODE) wrote:

> I'm actually suggesting that you create two trimesh instances that 
> share the vertex data, but use two separate index lists, which can be 
> done with no addition to ODE.
>
> Adding a check as for whether the rotation matrix has a negative 
> determinant for each trimesh collision might incur more of a 
> performance penalty than you might wish for. Perhaps a better option 
> would be to have some kind of flag that specifies whether the trimesh 
> itself is reversed or not.
>
> Or, if the only problem is the contact normals, then you can easily 
> reverse the normals post-collision, when you get the contacts back 
> from dCollide(). That would probably be simplest.

But.. wouldn't that mess up the tri-mesh vs sphere ? that one is working 
fine at the moment with symetrical tri-meshes. If i invert the normals 
or swap the indices, that should only work for box vs tri-mesh... see my 
dilemma ? ODE's behavior unfortunately doesn't seem to be consistent..

F. Brebion