[ODE] PSCL implemented in Rocket

Jon Watte hplus-ode at mindcontrol.org
Wed Jul 21 23:39:07 MST 2004


> split window view.  To be quite honest, I gave up with ODE right away
> because it was so unstable it would stack overflow with just a one
> hundred block brick wall and I didn't have a lot of patience to deal
> with that nonsense when I had access to better and more stable physics
> engines.

To be quite honest, this was documented in the documentation for 
dWorldStep(), with the suggestion that dWorldStepFast didn't suffer 
from the problem :-)

These days, there's the preferred dWorldQuickStep(), which is more 
stable than StepFast(), yet has the same linear growth in scope. 
I'm running 100-box stacks using 32-bit floating point, and it stays 
pretty stable (although I have to bump up the default body 
hibernation parameters to get them to hibernate in that 
configuration). This is with 100 steps/second; YMMV.

I'm writing this, because it feels like what you're saying is 
"I had a problem and didn't have the time to delve into it deeper" 
but what you're actually saying is "I claim ODE is unstable". There 
is a BIG difference between the two statements, and I'd hope that 
you're honest enough to be clear about what you mean, because other 
people will read your comments with no context -- if you really 
mean "ODE is unstable" then I have some counter-proof to offer ;-)


>   virtual SHAPEGUID SHP_CreateShape(const char *name) = 0;


Stylistically, this is a member function, living inside a class, 
living inside a name space (I hope?). Adding decorations on top 
of the already sufficient scoping adds a little more clutter than 
I'd think warranted.


Cheers,

			/ h+




More information about the ODE mailing list