Re: [Swarm-Support] development priorities (was Re: Membership in Swarm Developmen Group)

[Marcus G. Daniels]

glen e. p. ropella wrote:
> where integrating R with a network protocol enabled tool (like those
> written in Java) only requires an understanding of how to set up those
> other packages to work over, say, http.
> > Like, say, a Swarm model written in Java?
> >     
>
> There is no such thing as a Swarm model written in Java.  A Swarm model
> uses the Swarm scheduler, which is written in Objective-C.
>   
You have a program written in Java.  It imports some Swarm things that 
happen to be implemented using compiled code.  Nothing is stopping you 
from using network protocol enable tools in the usual Java ways.
> "Feasible" is not synonymous with "possible".  It is certainly
> _possible_ to embed Swarm in all sorts of things.  But, it is not
> feasible for most projects because a) they lack the technical expertise
> and/or 
>   
Let me take another tack.  It is easier to embed Swarm in an general 
object-oriented way in another framework because Swarm has a lot of 
ready infrastructure to facilitate it.   That you just feign ignorance 
of this does not to justice to the *relative* ease with which the 
capability could be adapted for direct use by users.   True, it hasn't 
been and thus it isn't obvious how to do it.   If you want to have a 
real technical discussion about how to embed a simulator and the general 
issues involved, you might want to at least not revise history so much.
> 2) other requirements (like running simulations in a jvm started
> from a web browser) conflict with Swarm's architecture.
There is nothing magic about running code in a JVM.  As a practical 
matter, embedding in a .NET sandbox will get you faster startup and a 
larger user base.   Or using Flash.    Yes, it would require a new way 
to express models in order to run them in a browser plugin.  
> > Swarm runs fine
> > inside a Java virtual machine -- it's a DLL and a jar file.
> >     
>
> DLLs don't run inside the jvm.  Saying so won't make it true.
>   
DLLs run in the system allocated address process space of the JVM.  
Along with other DLLs that the JVM may load for its own functioning.  

> > No, the question is what new and interesting things could only be done
> > by starting from scratch.  One is runtime code generation, which would
> > be useful for the performance of evolutionary models (or inferences
> > about possible models).
> >     
>
> No.  That's not the question.  The question is "How does the SDG best
> accomplish its mission?"  The SDG was not formed to do innovative
> computer science or systems engineering.  It was formed to support ABMers.
>   
Swarm phases are a crude way to accomplish runtime code generation, i.e. 
mixes of behaviors.  A more general way to do it is to have a real 
expression evaluator available, once that generates code that is fast 
enough to be comparable to a precompiled ABM simulation.   With this 
capability agents could create new behaviors on-the-fly, say, due to 
mutations or recombinations of their genome.  Swarm doesn't handle that 
well, currently.