Hello,

The simulation program I'm running on a Linux RedHat 9.0 computer using
Geant4.6.2.p01 and the LHEP_PRECO_HP hadronic physics finishes with a
segmentation fault. I then compiled my code and the Geant4 libraries in debug
mode and ran it with the Valgrind memory management debugger to try find out
where the problem is. It turned out that some invalid stuff was done in some
classes of the LHEP_PRECO_HP package.

Then I tried to make my (rather large) code shorter to reproduce the problem in
a simple test case. I ended up with a very short code using only LHEP_PRECO_HP
as the physics list initialised in the main() method (no other specific
PhysicsList classes are needed) and a trivial geometry (a quartz cylinder in a
World filled with air, onto which a 500 MeV proton impinges). Valgrind still
sees some problems in some classes of the LHEP_PRECO_HP package, but yet no
segmentation fault occurs any more (so it looks like the precise outcome
depends on what is defined in the geometry and analysis classes).

Nevertheless, the Valgrind output allows to see precisely where some
uninitialised value(s) are used in the LHEP_PRECO_HP package. Since it is
simpler to use this shorter test program than my original large code, I put it
on ftp://guest:time@ftp.metas.ch/voros/LHEP_PRECO_HP.tar.gz. If you extract the
files from this archive, you will get a directory called LHEP_PRECO_HP of the
same kind as for the Geant4 examples: just cd into it, gmake (the executable
will be called "totabs") and run the code (with Valgrind if you have it in
order to see the problem, otherwise you won't notice any problem).

If you prefer, I also put into this LHEP_PRECO_HP directory a file called
valgrind.out which contains the output of a full Geant4 Valgrind session I ran
on my computer, giving only 2 commands: "/run/beamOn 10" and "exit" after the
vis.mac file was executed automatically by the program. Valgrind reports some
uninitialised value(s) problems in event number 8, as well as some reading at a
previously free'd address in the exit procedure.

This Valgrind output reports the following problems in event number 8:

1) Conditional jump or move depends on uninitialised value(s) in
G4NeutronHPPhotonDist.cc on line 205

     if(repFlag==1)

2) Conditional jump or move depends on uninitialised value(s) in
G4NeutronHPPhotonDist.cc on line 339

     else if(repFlag == 2)

3) Use of uninitialised value of size 8 in G4NeutronHPPhotonDist.hh on line 109

     inline G4double GetLevelEnergy() {return theBaseEnergy;}

4) Conditional jump or move depends on uninitialised value(s) in
G4NeutronHPInelasticCompFS.cc on line 314

     if(theFinalStatePhotons[it]->NeedsCascade())

   and the NeedsCascade() method is defined in G4NeutronHPPhotonDist.hh on line
107 as "inline G4bool NeedsCascade() {return repFlag==2;}"


All the unitialised values "repFlag" (in 1,2,4) and "theBaseEnergy" (in 3)
should have been read in from an "aDataFile" object in method
G4NeutronHPPhotonDist::InitMean(std::ifstream & aDataFile), so it looks like
this method was not called at all since a test on the value of "repFlag" is
performed there, which should have thrown an exception (unless the
uninitialised value of "repFlag" should by chance have been equal to 1 or 2,
which is certainly unlikely). The G4NeutronHPPhotonDist::InitMean(std::ifstream
& aDataFile) should probably have been called in G4NeutronHPInelasticCompFS.cc
on line 119:

     theFinalStatePhotons[it]->InitMean(theData);

on the condition "dataType==12", the "dataType" value being read in the same
file as mentioned above. Since I don't know exactly which file this is and how
to read it, I unfortunately couldn't follow this program flow any futher.

I hope that an expert knowing the way this file reading works can easily find
out where the problem lies.


Finally concerning the problems in the "exit" procedure, I did not examine in
details the Valgrind messages since I thought they might be linked to the
previous errors, and thus it would be better to first solve these ones.
Nevertheless, you can have a look at the "exit" procedure's errors in the
valgrind.out file provided in the tar archive mentioned above.


Best regards,
Sandor