[Rivet] Segmentation fault in InvMassFinalState.cc on SL5

[Frank Siegert]

Andy Buckley, Wednesday 23 September 2009:
> Holger Schulz wrote:
> > I just wanted to inform you that when rivet is compiled on SL5
> > there is a segmentation fault if ZFinders are applied. A backtrace
> > with gdb points to the function "project" in InvMassFinalState.cc
> > Adding a cout in that function "fixed" this problem for me.
> > This happened twice on differently setup environments (DESY Zeuthen
> > and  Humboldt  University Berlin)  that both run SL5.
> > The compiler was in both cases gcc 4.1.2
>
> If I remember correctly, GCC 4.1.x was very buggy, which is why the LHC
> experiments are using GCC 4.3 (in fact, not even the RedHat version of
> 4.3, since that is also buggy). What this says about the suitability of
> SL(C)5 for production use is yours to judge ;) So for production use of
> Rivet I would also recommend explicitly using GCC 4.3 and avoiding 4.1.
>
> On the other hand, the default position should be to try and make Rivet
> work on all major platforms, even when they are a bit weird, like SL5:
> can you try taking a bit of a closer look in the problematic "project"
> method and see if there are any memory accesses that could be to blame?
> That adding couts "fixes" the problem suggests some serious
> instability, possibly from the compiler, possibly from some
> borderline-valid code.

If you are using the python interface (i.e. the normal "rivet" executable) 
it might also have to do with some of the distributed python interface 
files not being up-to-date with the C++ source. Can you try and remove all 
generated files from the pyext directory such that they are regenerated 
using swig? I've seen this cause such indeterministic failures before.

Frank