[Rivet] Jimmy stops with CODE 300

[Andy Buckley]

On 27/05/10 18:46, Holger Schulz wrote:
> On 27/05/10 16:27, Andy Buckley wrote:
>> On 27/05/10 11:00, Jonathan Butterworth wrote:
>>   
>>> Hiya,
>>>
>>> Andy Buckley wrote:
>>>
>>>     
>>>> Got to event 78887, with the .3 JIMMY build, before getting the fatal
>>>> 300 message and abort. Damn, some more digging required...
>>>>        
>>> Mmm. So this looks like you reduced the problem but didn't elimiate it?
>>> It might be that running in this minbias mode with this tune,
>>> herwig+jimmy actually (rarely?) exceeds 10000 particles? If the fortran
>>> herwig and jimmy is compiled without bounds checks this could have
>>> unpredictable results.
>>>
>>> (of course good fortran should protect against such things and herwig
>>> and jimmy may do but I am not 100% sure).
>>>
>>> If I had a working set up now with herwig and jimmy compiled myself, I
>>> would insert write statements in to see what NHEP and MAXER are at the
>>> end of each event...
>>>      
>> Think I've got it: despite trying to linking against the ".3" builds,
>> the NMXHEP size was always 4000! Digging a bit, it turns out that we
>> were relying on Genser's symlink to Herwig with the version name being a
>> variant of "6510", i.e. "6510.3" in
>> /afs/cern.ch/sw/lcg/external/MCGenerators/herwig/... but these symlinks
>> hadn't been updated for the builds with larger HEPEVT. So JIMMY was
>> expecting NMXHEP = 10000 but we were accidentally getting a HERWIG with
>> NMXHEP = 4000.
>>
>> I've changed some bits of the generator loader so that we don't get the
>> wrong symlink any more, when using Genser's builds. For private biulds
>> of the generators, make sure you are using the ".3" tarballs of both
>> HERWIG and JIMMY. Note that the change in the loader means that the
>> command-line name for HERWIG now ends in "6.510" rather than "6510".
>>
>> Holger, can you try this out? Hopefully problem solved, but you might
>> well need to rebuild your local versions of HW and JIMMY from the Genser
>> ".3" tarballs.
>>    
> Sweet, seems to work! I got past 40000 events now then I got
> impatient, will try to generate 1M events know. Thanks a lot
> guys!

I ran 100k successfully before committing. Think it's good.

Andy

PS. For the big scan run that you're doing, make sure you use a fairly
mainstream PDF, i.e. something from LHAPDF rather than the HERWIG
default! Just in case... ;)

-- 
Dr Andy Buckley
SUPA Advanced Research Fellow
Particle Physics Experiment Group, University of Edinburgh