Re: [swarm-hackers] Swarm in debian/ubuntu

[Paul Johnson]

Scott Christley wrote:
> On Sep 25, 2008, at 12:57 PM, Paul Johnson wrote:
> 
>> I'm curious to know if you will want to call your package "libswarm"
>> rather than "swarm".
> 
> Didn't think about it until you mentioned it, was thinking just  
> "swarm" but there does seem a convention to use "lib" when its a  
> library.  Looking further though we might want to consider different  
> Swarm variations, especially considering the different GUIs and such,  
> for example:
> 
> swarm-core
> swarm-tcl/tk/blt
> swarm-gnustep
> swarm-java
> 

Until now, we've had RPMs for
1. swarm 0bjective C
2. optional JDK addon
3. optional gcc-java addon

This has not been entirely happy because the precise version of the JDK
must be specified, and that caused trouble where some people wanted to
use Sun JDK, others Kaffe, others Blackdown whatever...

Another source of unhappiness for me was that we were trying to compile
this in one shot, so the Swarm configure used --with-jdk= as an option
and then the  RPM packaging was used to sort files needed for Obj-C and
Java.  The Swarm-Obj-C base was really the base of the java version.  I
started building a separate RPM using --without-jdk because I had the
belief/guess/mistaken idea that a Swarm kernel compiled without any of
that language translation stuff would run faster.  From reading the
activity library, I was just guessing that dropping JDK altogether made
programs run faster.  I think I did some testing at one time and it was
a few % better in the runtime.  But that may have been a mistake, if you
tell me on a theoretical level I was wrong, then I'd have to believe you.

If that perception (guess) was accurate, it makes me think we should not
provide a swarm-core at all, but rather we should aim to give each user
just one big package, one for Swarm-Objective C (if possible, it should
support all possible GUI), one for Swarm-Java.  I don't know if the gcj
native code version is useful to anybody.  I compiled some programs with
it and they seemed to be a tad slower than the ordinary Java, but I
think Marcus thought the native code versions might outperform the
virtual machine ones if I knew how to properly optimize Java code.  But
that was, oh, 4 years ago...

pj



-- 
Paul E. Johnson                       email: address@hidden
Dept. of Political Science            http://pj.freefaculty.org
1541 Lilac Lane, Rm 504
University of Kansas                  Office: (785) 864-9086
Lawrence, Kansas 66044-3177           FAX: (785) 864-5700