[ODE] Sphere pass through box, mass problem

[Jon Watte (ODE)]

The problem I had was a case where you (as an actor) could push yourself 
into (and through) a box. The problem was that the case that was testing 
for != 0 before dividing, when it really needs to test for < epsilon, 
where epsilon is min-non-denormal-squared.

Cheers,

			/ h+


J. Perkins wrote:
> --- Jon Watte <hplus at mindcontrol.org> wrote:
> 
>>There's a problem where spheres and ccylinders may
>>be able to push through boxes, because there's a division
>>by a denormal number somewhere in the pipe. I submitted
>>a patch for this problem over a year ago, but it was
>>rejected because I had coded the specific floating point
>>constants for denormals, instead of using some other
>>supposedly better method (the specifics of which were
>>never actually submitted to the list).
> 
> 
> I've put this in the patch manager (#1356867). When I get around to
> validating the patches I will put together some tests to demonstrate
> the fix and try to get it into unstable. It would help if I had some
> test code that reliably demonstrated the problem. In the absence of a
> better solution, I for one would prefer a partial fix to no fix at
> all.
> 
> Jason
> 
>