[ODE] New release?

[Fredrik Sandstrom]

On Thu, 2006-12-21 at 20:59, Jason Perkins wrote:
> On 12/21/06, erwin at erwincoumans.com <erwin at erwincoumans.com> wrote:
> > Indeed. Other libraries like Bullet and SOLID have the same issue, and
> > require a #define to enable double precision. Usually an application needs
> > either one or the other, not both at the same time.
> 
> I was the one who originally suggested the idea. In my application
> (and I doubt I'm alone in this) I don't care about the physics
> precision, I always use floats. For some uses of my application, I
> want dReal to be float, for others double. dReal bleeds up into
> anything that touches the physics code, which IMHO is annoying.
> 
Ah, I see. So what you want is to add functions that retrieve and set
data in the format of your choice and then if needed converts to the
underlying engine format? This should be easily done by just adding f/d
-functions that converts the data if needed and then call the original
function. I don't see a reason to change the inner workings of ODE to
make this more practical to use.
Since you don't care of the physics precision you should run it in
floats and then convert your double data to floats the few times you
have to use doubles in your application.

> The other part of that conversation was that dReal is used
> /everywhere/ in ODE, even in places where double precision doesn't
> matter, like quaternions and box lengths. It would be nice if the
> precision were only used where it mattered. Especially in the
> collision detection code where, for example, it took a long time to
> get all of the OPCODE float/double issues worked out.
> 
I agree that it might not be needed for box dimensions but I fail to see
why rotational precision of bodies doesn't affect the precision and
stability of the simulation. It would be unfortunate to have vital parts
of the engine running in a lower precision, one missplaced double to
float conversion could (in theory at least) remove all benefits of using
double precision mode.

> Jason
> _______________________________________________
> ODE mailing list
> ODE at q12.org
> http://q12.org/mailman/listinfo/ode
-- 
Fredrik Sandström
Oryx Simulations AB
tel: +46 90 154815