[ODE] Sphere pass through box, mass problem

Jon Watte (ODE) hplus-ode at mindcontrol.org
Mon Nov 14 19:03:41 MST 2005


This was one of the complaints about the patch the last time around.

However, dEpsilon is the wrong value for this case. There are at least 
three kinds of epsilons:

1) The minimum quanta you can change by at the value 1.0 -- this would 
be way too big for this particular test.

2) The minimum quanta you can represent without being denormal -- this 
would be too small, because the value gets squared in the end, and you 
don't want to generate denormal numbers after squaring (or infinite 
numbers after division and squaring).

3) The minimum quanta you can represent at all (i e, the smallest 
denormal number) -- this is WAY too small for the current case.

Choosing an appropriate epsilon for each case is an important part of 
robust simulation code design, and using the same, cookie-cutter epsilon 
in all code is exactly the wrong thing to do.

But enough people complained last time that I didn't bother pushing it 
through -- it works in MY copy of ODE, and I've made the patch 
available; that's really all that matters.

Cheers,

		/ h+


Rodrigo Hernandez wrote:
> 
> Use the ODE defined constant dEpsilon for epsilon, otherwise, you'll be 
> introducing inconsistencies in the code.




More information about the ODE mailing list