Re: 0.8.2 candidate

On 9/27/07, Jean-Sébastien Guay <[email protected]> wrote:
> Jason Perkins said:
> > It is a non-trivial amount of work to untangle
> > that one, but everyone would sure appreciate it if someone took it on.
>
> Jon Watte said:
> > The thing is that the warnings may be helpful when porting to 64-bit,
> > so casting them away would be a disservice.
>
> Ok, so 2 differing opinions.

Well, no not entirely. You can't get fix these warnings by casting --
well, you can *suppress* the warnings by casting, but that doesn't fix
the real issue. The problem is that ODE supports both single- and
double-precision, but OPCODE only uses single. So the trick is to
figure where single-precision is okay, and where double-precision is
needed, and then modify OPCODE to use dReal in the right places.

There was a discussion on the list about this a few (several -- time
flies) months ago, which I think I started. The gist of it was that in
theory it should be possible to use dReal only for positions, and use
single-precision floats for rotations and position offsets (box
widths, sphere radius, etc.). For trimeshes, it is single-precision is
probably sufficient (someone jump in if I've got this wrong) to
represent the mesh vertices; dReal would still be needed to represent
the geom position.

Anyway, that's the 10,000m view of the problem. Volunteers?

Jason