[ODE] ODE crashs? stack overflow?

[Geoff Carlton]

Another cause I've found is conflicting contacts.  For example two 
contacts at the same position but with opposite directions can screw up 
the object.  I used to see this quite a bit and had to destroy objects 
occasionally when they became invalid.  However several months ago I put 
in code that examines an object's contacts and merges/skips these 
contacts into something sensible for the solver, and I've almost never 
seen them now.

Geoff


Aras Pranckevicius wrote:

>>I have used ODE for my game developement. Everything seems to be fine.
>>However, when the game runs for a few minutes , it would use crash because
>>of stack overflow. It is a pretty random case and usually I find that some
>>of my object's velocity is -1#QN.
>>    
>>
>
>Sometimes I observe this as well. In my situation, it mostly affects
>very light objects that have their mass distribution
>long-thin-box-like. It goes like this: after some contact, the object
>obtains some angular velocity (starts spinning), and for some reason
>sometimes the spinning only accelerates, first to high numbers, then
>to infinities, then to IND/NAN. Then at some point linear velocity
>also degenerates, then position, and then everything breaks :)
>
>Applying some damping (eg. air drag) to angular and linear velocities
>does help a bit. Also, I do explicitly clamp the angular velocity to
>some limit.
>
>I'm not sure why this behavior exists, now I attribute that to single
>precision floating point, quickstep solver and pretty high timestep I
>use (too busy to investigate... clamping angular velocity did the
>trick :))
>
>  
>