[Rivet] Rivetgun and Sherpa 1.1.1

[Holger Schulz]

Andy Buckley wrote:
> Holger Schulz wrote:
>   
>> Hi!
>>
>>
>> Having installed the latest svn versions of rivet and agile yields the 
>> following
>> error message after calling "rivetgun -h" if the only generator 
>> installed is Sherpa 1.1.1 :
>>
>>         rivetgun: symbol lookup error: 
>> /home/holger/generators/sherpa/1.1.1/lib/SHERPA-MC/libAnalysisTools.so.0: 
>> undefined symbol: 
>> _ZN6ATOOLS15Getter_FunctionIN8ANALYSIS15Analysis_ObjectENS1_15Argument_MatrixESt4lessISsEED2Ev
>>
>> If in addition any other generator is installed too, the 'error message' 
>> is just 'segmentation fault'.
>>     
>
> Is this problem still there? 
This problem has vanished, but still event generation doesn't work.
Calling for instance
          rivetgun -g Sherpa -p PATH=LEP91 -P lep1.params
results in a normal exist on the first call.
Then I build the libraries with ./makelibs and trying to run
the same line as above yields a segmentation fault.

Running rivetgun having set a certain parameter to a certain value like
       
          rivetgun -g Sherpa -p PATH=LEP91 -P lep1.params -p 
"ALPHAS(MZ)=0.1"

yields a lot more output, instead of "Segmentation fault" I get:

*** glibc detected *** rivetgun: free(): invalid pointer: 0x08e14700 ***
======= Backtrace: =========
/lib/libc.so.6[0x433b38b6]
/lib/libc.so.6(cfree+0x90)[0x433b6e00]
/usr/lib/libstdc++.so.6(_ZdlPv+0x21)[0x43f31691]
/users/eeh/hschulz/Software/lib/libHepMC.so(_ZN5HepMC8GenEventD1Ev+0x12d)[0xafeda811]
/users/eeh/hschulz/Software/lib/libHepMC.so(_ZN5HepMC8GenEventaSERKS0_+0x3c)[0xafedc098]
/mnt/eeh/hschulz/Software/lib/libAGILeSherpa.so(_ZN5AGILe6Sherpa9fillEventERN5HepMC8GenEventE+0x2c)[0xae9e74e0]
/mnt/eeh/hschulz/Software/lib/libAGILeSherpa.so(_ZN5AGILe6Sherpa9makeEventERN5HepMC8GenEventE+0x45)[0xae9e754d]
rivetgun(_ZN5Rivet8generateERNS_13ConfigurationERNS_3LogE+0x182b)[0x8053e27]
rivetgun(main+0x3a3)[0x80500ef]
/lib/libc.so.6(__libc_start_main+0xdc)[0x43362dec]
rivetgun[0x804fcc1]

and a lentghy memory map...

Holger

> There's something odd about Sherpa's shared
> libraries, so I re-wrote the AGILe dynamic loader to only test that the
> library files exist on the first run, and only call "dlopen" for the
> particular generator that's being loaded: this might have had an
> influence on this sort of error.
>
> I didn't see the error myself, but I do get a segfault from HepMC on the
> *second* Sherpa event: looks to me like HepMC isn't handling loop
> structures in its graphs properly when deleting, and Sherpa is the only
> generator currently highlighting this problem.
>
> Andy
>
>