[ODE] Pushing through collisions

[GARY VANSICKLE]

[snip]
>     I had used the difference between the flat cylinder and sphere to
>     determine what parts of cylinder (side, top, bottom) collided with
>     sphere.
>     So, I wonder if that's a bad practice?

Not a bad practice... for interference detection.  Interference detection
and resolution (which is what's implemented in ODE among others) is not the
same thing as collision detection and resolution.  Collision detection has
to take the previous state of the system into account somehow.