[ODE] Serialization?

[Adam D. Moss]

Marc Rochel wrote:
> It could be possible that there are hashmaps inside of ode-lib. To 
> create the equations for the next time step, all elements of this 
> hashmap have to be iterated. But since hashmap elements don't have 
> a defined order in which they are iterated, serialization from outside 
> of ode-internals would prevent me from saving the practical existing 
> order of elements in these hashmap. So after unserialization by adding 
> the elements by the ode-interface, the order could be different.

Yeah, I've been worrying about this lately when I looked again
into replacing ODE's use of libc's PRNG.  I couldn't come to a
conclusion either way (I would totally expect ODE's various
hashing and/or self-optimizing structures to not offer
determinism after depopulation+repopulation, but there have been
various users over the last year or two who I seem to remember
swearing blind that they were getting deterministic
behaviour from ODE, perhaps simply under very controlled
circumstances).

I'm becoming quite inclined to agree with the idea that fighting
for determinism is a very, very tough battle unless your whole system
is built for it (extending beyond ODE and the application to one's
choice of scripting language, even graphics API etc).

--Adam
-- 
Adam D. Moss   . ,,^^   adam at gimp.org   http://www.foxbox.org/   co:3