[ODE] Curious finite friction crash bug.

[RDG O'Sullivan]

I started using ODE a few weeks back - I'm putting together an engine 
for a group project for university - and I'm really impressed. However,
I've come up against a crash bug (an access violation) which seems to 
occur almost randomly during worldStep (or worldStepFast1) depending 
upon the masses of objects, the friction at contact joints and the 
surface.bounce parameter. The only fix I've found so far is to use 
surface.mu = dInfinity, which is only really a temporary solution - but
the crash never comes up under these conditions, so it definitely isn't
any of my code causing the problem.

I haven't got source debugging up for ODE in MSVC yet, but the crash 
occurs in dWorldStep -> dWorldStep -> dInternalStepIsland -> 
dInternalStepIsland_x2 -> dSolveLCP -> dMessage -> dMessage -> ... and 
then it slips into the system libraries and crashes.

My contact joints are built identically to those in the test programs, 
modified solely to use finite friction, and cleared after each step. 
There are no other joints in the program. The system only crashes after
creating new contact joints under certain conditions which I haven't 
figured out yet.

I'm trying to isolate the problem and produce as simple a program as 
possible which will reliably generate the bug... however, I've got a 
lot of other things to do so it's kind of a low priority. Has anyone 
encountered anything like this before?

-randomnine-