Objective-C SWARM at Ubuntu 8.04

Jakson Aquino

Contents

1. Enable the 'universe' repository if it isn't enabled yet. Click here to know how to do it under Ubuntu. If you prefer, you can manually edit the file /etc/apt/sources.list and uncomment the line that allows the installation of software from the 'universe' repository.

2. Then, update your list of available software. That is, open a terminal and type (or copy and paste):

      sudo apt-get update
   

3. Install the software necessary to download and compile swarm:

      sudo apt-get install gobjc gperf libxpm-dev libpng12-dev automake \
           emacs21-nox libhdf5-serial-dev blt-dev autoconf libtool make \
           xfonts-75dpi xfonts-100dpi
   

4. Now, you can download the swarm source code from Paul Johnson's web site:

      wget http://pj.freefaculty.org/Swarm/swarm-2.2.4.tar.gz
      tar -xzf swarm-2.2.4.tar.gz
      cd swarm-2.2.4
   

5. Then, you can compile swarm:

      ./autogen.sh
      ./configure --with-pic --enable-onelib --with-gnu-ld --disable-jar \
         --without-jdkdir --with-tcl=/usr/lib/tcl8.4/ --with-tk=/usr/lib/tk8.4/
      make
      sudo make install
   

   Note: If you are using a version of Ubuntu older than 8.04 you will have to pass different parameters to the configure script.

6. Swarm is already installed. You should now put this line at the end of your ~/.bashrc, because, by default, the Makefiles that come with swarm applications are configured to look for swarm at /usr, but since we installed it at /usr/local, we need to inform the Makefiles about this:

   export SWARMHOME=/usr/local
   

   In addition to putting the above line in your ~/.bashrc, copy and paste it in your terminal too. When you logoff and login again the .bashrc will be re-read and the SWARMHOME environment variable automatically will be active.

7. Finally, you can download, compile and run some examples:

      wget http://ftp.swarm.org/pub/swarm/apps/objc/sdg/swarmapps-objc-2.2-3.tar.gz
      tar -xvzf swarmapps-objc-2.2-3.tar.gz
      cd swarmapps-objc-2.2-3/heatbugs/
      make
      ./heatbugs
   

Debian package

   sudo apt-get install build-essential dpkg-dev debhelper fakeroot
   wget http://pj.freefaculty.org/Swarm/swarm-2.2.4.tar.gz -O swarm_2.2.4.orig.tar.gz
   wget http://jalvesaq.googlepages.com/swarm_2.2.4-1.diff.gz
   wget http://jalvesaq.googlepages.com/swarm_2.2.4-1.dsc
   dpkg-source -x swarm_2.2.4-1.dsc
   cd swarm-2.2.4
   dpkg-buildpackage -rfakeroot -uc -us
   cd ..
   sudo dpkg -i libswarm_2.2.4-1*.deb libswarm-dev_2.2.4-1*.deb

include $(SWARMHOME)/etc/swarm/Makefile.appl

include /usr/share/libswarm/etc/Makefile.appl

   sudo tar -xzf swarm-2.2.4.tar.gz -C /usr/src

Debugging swarm models

Writing ease to debug code

• Avoid the use of id. Use pointers to your classes: instead of declaring something like id agent, declare AgentSwarm *agent. If gcc knowns the class of bug object it will be able to warn you when you do some mistakes. When examining your program within gdb, you be able to fully explore the variable and methods of the agent object.

• Include swarm headers. The inclusion of the headers containing the declarations of the swarm objects that you use will eliminate many distracting warnings from the output of the compilation process.

• Turn gcc more verbose. Sometimes I add the following line to my Makefile when compiling my model:

  OBJCFLAGS += -W -Wpointer-arith -Wreturn-type -Wcast-qual -Wwrite-strings \
  		  -Wswitch -Wshadow -Wcast-align -Wuninitialized \
  		  -Wbad-function-cast -Wchar-subscripts -Winline \
  		  -Wnested-externs -Wimport -Wprotocol -Wlong-long
      

   

  CGDB

  Paul Johnson wrote some good examples of how to use gdb (GDBUsage01, GDBUsage02, and GDBusageExamples). You should look at them, and don't forget of turning off the compiler optimizations before using gdb:

  OBJCFLAGS += -g -O0
  

  Almost always I prefer to use cgdb instead of plain gdb:
  
  

   

  Valgrind

  Valgrind is very useful to catch bugs that neither gcc nor gdb can detect and that are very difficult to find by just reading the code: memory leaks, buffer overruns, use of uninitialized variables, etc...

   

  Some explantions on the process of building swarm

  If you get any error during the compilation, its necessary to make a distclean. Don't simply fix the problem and type "make" again to continue the compilation because it probably would not work well. It seems that the swarm's configure script improved a lot in the last years, but it is still incomplete. Not all programs required to compile swarm are checked, and we only discover this fact at compile time, after some error message that might be hard to understand:

    installation problem, cannot exec 'cc1obj': No such file or directory
  
    missing -batch --eval '(setq *disable-gui* nil)' -l 
    .../src/defobj/modulemap.elc -f generate-module-map
    
    cp -f ../etc/common.el .../_common.el ...
    missing -batch -f batch-byte-compile .../_common.el
    /bin/sh: missing: command not found
  

  The first message means that the GNU Objective-C compiler isn't installed; the other two mean that emacs isn't installed (but emacs should be required to build swarm only if you want java support). If you are using the official 2.2 version of swarm, you might get the following error:

    mframe.m:1728: internal compiler error: in assign_stack_temp_for_type, at function.c:605
    Please submit a full bug report, with preprocessed source if appropriate.
  

  Then, it will be necessary to comment the line 1728 of libobjc/mframe.m:

  1726: #ifndef __CHECKER__
  1727:         /* Checker-equipped compiler barfs on this on sparc-sun-solaris2.7. */
  1728: //      *(block *) buffer = retframe_block (retframe);
  1729: #else
  1730: #warning Disabling code in mframe for Checker
  1731: #endif
  

  If you have tried to install a Debian package of swarm-2.1, it might be necessary to upgrade some of the listed packages before proceeding, as reported by Jerry Li: Problems installing swarm-2.2 on Ubuntu 5.1. As you can see, we ran the configure script with some parameters. When the script is unable to find a library that you know is already installed, you can use "dpkg -S" and "dpkg -L" to discover the correct path to the library or its headers. Example:

    configure: error: Please specify Tcl header path using --with-tcldir 
    or --with-tclincludedir
    $ dpkg -S tcl.h
    tcl8.4-dev: /usr/include/tcl8.4/tcl-private/generic/tcl.h
    perl-tk: /usr/lib/perl5/Tk/pTk/tcl.h
    tcl8.4-dev: /usr/include/tcl8.4/tcl.h
    $ dpkg -L tcl8.4-dev | grep lib
  

  For versions of Ubuntu before 8.04, the configure parameters were:

     ./configure --disable-jar --without-jdkdir --enable-onefile --with-pic \
          --with-gnu-ld --with-tclincludedir=/usr/include/tcl8.4/ \
  	--with-tclscriptdir=/usr/lib/tcl8.4/ --with-tkscriptdir=/usr/lib/tk8.4/ \
          --with-tcl=/usr/lib/tcl8.4/ --with-tk=/usr/lib/tk8.4/
  

  On amd64, I got the following error:

  /usr/bin/ld: ../../avcall/.libs/libavcall.a(structcpy.o):
  relocation R_X86_64_32 against `a local symbol' can not be 
  used when making a shared object; recompile with -fPIC
  ../../avcall/.libs/libavcall.a(structcpy.o): could not read symbols:
  Bad value collect2: ld returned 1 exit status
  

  And the solution was to pass the parameter "--with-pic" to the configure script. On other architectures, swarm might become slower when built with this parameter. On some architectures, swarm might run faster if you set some compiler flags before running the "configure" script:

    export CFLAGS='-O3 -ffast-math -funroll-all-loops -ftree-vectorize'
    export OBJCFLAGS='-O3 -ffast-math -funroll-all-loops -ftree-vectorize'
    ./configure
  

  and, then, add following line to the Makefiles of your models:

    OBJCFLAGS+=-g
  

  The use of ffast-math flag is controversial. At some webpages we find the information that it may reduce accuracy of float point calculations. Others tell that it actually increases the accuracy. You have to test it yourself to see what happens in your architecture.

   

  ---

  June, 2008