Iterative solver (again) [Was: Re: [ODE] Russ' plans for ODE?]

[Anton]

> A> Internally, global timestep is subdivided into several smaller chunks
(but
> A> not too much; it was about 3 subdivisions for the box stack with 10ms
> A> simulation time step), but no collision detection is performed during
these
> A> substeps, only existing contacts are used.
>
>    Yes! That is one obvious gain. But it would cause some pain also :
>    we might have problem with fast objects. (But we always should
>    treat fast objects differently! Should we?)

Well, in Novodex fast objects don't seem to get any special treatment
(ideally, though, fast objects should use continious collision detection; on
the other hand, if an object already has contacts, it's unlikely to move
'too fast'). In fact, I think they might even entirely skip collision
detection between pairs of objects that are sufficiently simple and have
enough contacts between them already.