Compilation of Sherpa under MAC OS

[Frank Siegert]

Sherpa does not compile on all common versions/architectures of Mac OS. The problem is related to linking with Fortran libraries for the internal PDF or Lund routines.

[Frank Siegert]

I have improved the Fortran linking such that I would hope it works more reliably on a MacOS. From my testing it seems to compile fine on 10.4/PowerPC and 10.5/Intel now, but there remain problems:

• Without autoreconf'ing on the Mac itself before the installation, it prints at runtime lots of error messages and eventually crashes.
• If autoreconf'ed before installation, it starts running normally, but crashes as soon as something has to be written out to file (typically the process libraries).

Can Mac users please test with the attached tarball, and write me their experiences? Ideally you would do two tests:

1. without autoreconf:
   • tar xzf Sherpa-beta-1.1.3.tar.gz
   • cd SHERPA-MC-1.1.3
   • ./configure
   • make -j2
   • make install
   • cd SHERPA/Run/LHC
   • ../Sherpa
2. with autoreconf:
   • tar xzf Sherpa-beta-1.1.3.tar.gz
   • cd SHERPA-MC-1.1.3
   • autoreconf -fi
   • ./configure
   • make -j2
   • make install
   • cd SHERPA/Run/LHC
   • ../Sherpa

[Frank Siegert]

Replying to fsiegert:

Can Mac users please test with the attached tarball, and write me their experiences?

I just noticed that I can't attach the tarball here, because it is too big. So I have uploaded it to

​http://www.hepforge.org/archive/sherpa/Sherpa-beta-1.1.3.tar.gz

[jmonk]

random number seeds for event with NaN weight

[anonymous]

Hi Frank,

In the first case if compiled ok for me but there was immediately a problem with the makelibs script. It exits with the error:

./makelibs: line 73: syntax error near unexpected token `done' ./makelibs: line 73: `done'

I can't see what is wrong myself, but if I substitute the makelibs from version 1.1.2 modified by replacing libtoolize->glibtoolize that I know works then I get my libraries. Maybe instead of using the sequence of calls libtoolize, aclocal, automake, autoconf it might work better if you used autoreconf instead - I believe that does the same thing and seems to know about the difference between liibtoolize/glibtoolize (on 10.5).

When I come to run it it initialises and does the integration, however it throws an exception when it starts generating events. It seems to be generating a NaN weight:

Exception_Handler::SignalHandler: Signal (11) caught.

Cannot continue.

Exception_Handler::Terminate(): Pre-crash status saved to '/Users/jmonk/Downloads/sherpa/SHERPA-MC-1.1.3/SHERPA/Run/LHC/Status__Tue_Sep__2_14-03-27_2008'. Cluster_Partons_Base::WriteOutWeights(): Weight statistics {

Misclusterings: 0 <w>_{2-jet} = 3.8109e+71 +- nan ( -2.14748e+07 % ) <w>_{3-jet} = 0.870299 +- 4.07357e-05 ( 0 % )

I put the status directory to this ticket if you need the seeds for that event

[Frank Siegert]

Replying to anonymous:

In the first case if compiled ok for me but there was immediately a problem with the makelibs script. It exits with the error:

Ah, my mistake. I forgot to close the "else if" in line 57 by "fi". I didn't notice, because on my PC it doesn't run into that case (and on Mac I didn't get that far). I have uploaded a new tarball, such that others don't have the same problem.

When I come to run it it initialises and does the integration, however it throws an exception when it starts generating events.

Interesting. Can you try the following things:

• Generate only 1 (2, 5, 10, ...) events, to see whether it is a specific event where it crashes (or look at the "number of generated events" in the exit message).
• Run the crash with debug output but fragmentation disabled (to make it shorter), i.e. "OUTPUT=15 FRAGMENTATION=Off" on the command line. Pipe everything into a logfile, and attach it here.

It seems to be generating a NaN weight:

<w>_{2-jet} = 3.8109e+71 +- nan ( -2.14748e+07 % ) <w>_{3-jet} = 0.870299 +- 4.07357e-05 ( 0 % )

This NaN might also appear because no(t enough) events have been generated yet, but of course could also be a NaN in the event somewhere. The debug output should tell us.

Thanks for the feedback... Frank

[Frank Siegert]

The nan above was indeed a real problem, and I have fixed this now.

This means, all problems I am aware of on MacOS have been fixed, and I have been able to install and run the updated tarball on MacOS 10.4 PowerPC and MacOS 10.5 Intel without any problems. That's why I am now asking you to test it, I have updated the tarball at

​http://www.hepforge.org/archive/sherpa/Sherpa-beta-1.1.3.tar.gz

All you should do is untar it and run "TOOLS/makeinstall -c" and then test e.g. "cd SHERPA/Run/LHC; ../Sherpa".

Additionally I started a list of all the little things which one can do wrong when installing (Sherpa) on MacOS, such that you can check for these first if (or rather "when") you run into installation problems:

​http://projects.hepforge.org/sherpa/dokuwiki/documentation/installation/macos

Thanks in advance for testing and giving me feedback. I am going to mark this bug as "fixed" now, because I think (hope!) that all remaining problems are specific to somebody's installation. That's why I would ask you to file new separate bugs for all remaining problems.

Cheers, Frank

[Kyle Cranmer]

Hello,

This seems to work for me, thank you.

[Stefan Hoeche]

Milestone old deleted