[ODE] Normal generation on Box Box collision, Integration Error

[Scott Johnson]

Question #1
I have been looking at the OBB vs OBB collision code in geom.cpp.  It
generates a normal based on the separating axis that penetrates the least.
I can envision a problem when two OBBs happen to be axis aligned.  If you
picture them in 2D colliding at a corner, then you can't generate the
correct normal without a translation velocity.  The correct normal is from
the face that is hit first and you can't make that determination without a
velocity.

Can't this cause problems in stacking boxes when accuracy is important?

Question #2

Does anyone know a reference that explains in more detail what Russell is
talking about when he says that explicit integrators cause error by adding
energy and implicit integrators drain energy?  I am referring two section
11.2 My Simply Rotating Bodies are Unstable!

Scott Johnson
milkboy@austin.rr.com