My personal criticism

[Martin Hinsch]

Hi everybody

These are my personal experiences with installing SWARM:

We tried to install it on two different machines, a PC with 
Linux and a Sparc station with solaris. On the Linux machine
we tried to use the Linux binary distribution and failed.
After several frustrating hours I downloaded the sources and
had no further problems.
Then we went on to the sparc. On this station there was nearly
no standard software installed (due to a recent system crash), so
we had to build it from the scratch up. As long as it was for tcl/tk
and objctcl everything worked perfectly. The first problem was to
persuade blt to use the locally installed tcl/tk 7.5/4.1 instead of
the systemwide versions 8.0/.. After having done this we went on to
xpm. This happened to be really difficult. After having run imake to
configure the makefiles, we tried to compile it with gcc which didn`t
work. The README recommended to use the native compiler of the machine
which wasn`t available. After some really ugly makefile-hacking the 
solution appeared to be to use gcc but compile a static rather than a
shared library. 
Well, nonetheless SWARM now runs without problems on both machines.

Now some criticism (which is absolutely my personal opinion
and in several points certainly a matter of taste):

- Why is it necessary to use this whole bunch of different libraries
which are not even especially fast or easy to use?
Wouldn`t it make sense to use for example gtk instead of tcl/tk? It`s
pretty
fast and already has an objective c interface. And it`s ONE library
instead
of four.

- Concerning the design of SWARM:
Although it`s easy to build useful simulations in a short time using
swarm, the design of the "programmer`s interface" to me seems to lack
a central idea. I think it looks somehow patchy.
Also in my opinion the whole thing is far to monolithic. As a programmer
I would prefer a design where you have several seperate parts from which
you
can choose which ones to use in your program (isn't that one of the
clues of
object oriented programming?). It should be much more modularized.
There are certainly projects which use all the powerful mechanisms swarm 
offers, but often I think it would be useful to have a lean, fast
framework 
wich just offers basic functionality.

- In my opinion swarm is to slow.
I really like Objective C, it's very simple and very well designed.
But on the other hand it is quite slow. As long as you are doing simu-
lations with hundreds of agents there is no problem but when you are
getting
to thousands or tens of thousands of agents (which I think is a
reasonable
number if you for example want to simulate evolutionary dynamics of real
animal populations) every method call makes a difference.
So, why not use C++ or (as there, as far as I can see it, already is
some
really non-standard hacking included) even design an own specially
suited
OO approach.


Well that`s it for now. Maybe some of you consider my thoughts useful.

Greetings
Martin Hinsch


                  ==================================
   Swarm-Support is for discussion of the technical details of the day
   to day usage of Swarm.  For list administration needs (esp.
   [un]subscribing), please send a message to <address@hidden>
   with "help" in the body of the message.
                  ==================================