Re: [Swarm-Support] Compiling a model on Swarm 2.2 [Pietro Terna's CD version]

[Marcus G. Daniels]

Gary Polhill wrote:

> On Swarm 2.1.1 (Windows 2K) and Swarm 2.2 (Windows XP) I have an issue with 
> message forwarding, which I used in one of my classes (subclassed from Swarm). 
> I have fixed this by reimplementing all the methods and not using forwarding at 
> all, but it would be useful to hear some confirmation that I shouldn't expect 
> message forwarding to work in these environments.
>  
In the source distribution, the file tests/defobj/forwarding.m gives an 
example of how it should work.   I think it has worked no the past.  
Java models rely on it for a variety of things.  There have been 
portability problems with supporting `mframe' code on various 
platforms.  mframe may get replaced with code from GNUstep to avoid 
these problems, but that hasn't been done yet. 

> On Swarm 2.2 (Windows XP) I was calling a class method of a subclass of the 
> Object class before initSwarm() gets called in main.m. I then got a core dump 
> with the error message:
>
> assertion "__objc_class_hash" failed: file "/src/Swarm/swarm/libobjc/class.c" 
> line 130
>
> Moving the class method call to after initSwarm() is called fixed this. It 
> would be useful to have confirmation that it's not a good idea to do any method 
> calls before calling initSwarm() (or one of its variations).
>  
For MacOS X, the initialization sequence of the Objective C runtime was 
modified to have two phases, and this will indeed happen if you do 
Objective C things before initSwarm.

> > From gdb, it was clear that this was choking on a piece of code in the 
> > parameters.scm file that said:
>
> #:winnerCoord <2,2>
>  
You could code your serialization methods to use

 '(2,2)

..that would be more Lispy.

> and (ii) is there any chance the error messages could be a bit more helpful? I 
> spent ages trying to track down this problem!
Some of those aborts are in general purpose subfunctions, so code would 
have to be restructured to return error codes (and all of the callers 
changed to provide context-dependent messages).   I guess if I were to 
work on this code I'd add new serialization code to do XML.  At the time 
the Lisp stuff was written, XML hadn't really caught-on yet.   Also, you 
could try using the HDF5 code.  It tends to have more diagnostic code 
and error messages.  To use it to manage parameter settings, it's 
necessary to use another program like R (or smaller Swarm-based 
programs) to handle reading and writing of the HDF5 files.

Btw, you might want to try the new Cygwin installer for Swarm:

  http://www.santafe.edu/projects/swarm/cygwin/setup.exe

This doesn't put things in /Swarm-2.2 anymore, but in /usr.   The build 
is from newer Swarm code and using newer Cygwin tools.  However, it does 
not yet have Java support.