crash in MergedGenParticleProducer

[arizzi]

we have a crash here:

#5 0x00007f6d6c89183a in MergedGenParticleProducer::isPhotonFromPrunedHadron(pat::PackedGenParticle const&) const () from /cvmfs/cms.cern.ch/slc6_amd64_gcc630/cms/cmssw/CMSSW_9_4_0/lib/slc6_amd64_gcc630/pluginGeneratorInterfaceRivetInterface_plugins.so

you can reproduce with (on lxplus) ~arizzi/public/test92X_NANO.py on 940 release + nanoaod master branch

@intrepid42

[arizzi]

the issue is here:
https://github.com/cms-sw/cmssw/blob/b5ad3f4c7b4987d89bfb52258f4e2a973df6ca29/GeneratorInterface/RivetInterface/plugins/MergedGenParticleProducer.cc#L136
as mother(0) is not checked for validity

then it crashes somewhere in produce() of
https://github.com/cms-sw/cmssw/blob/a1a699103d53ed00ae87f2a2578dd7e4b6bd0b5b/GeneratorInterface/RivetInterface/plugins/GenParticles2HepMCConverter.cc

[arizzi]

it looks like the sample has a fancy particle listing with only a few particles and perhaps some assumptions of the GenParticles2HepMCConverter do not hold. @perrozzi

[mseidel42]

I can confirm, fixing the problem in MergedGenParticleProducer (see below) leads to a crash in this line of GenParticles2HepMCConverter:

  HepMC::GenVertex* vertex1 = new HepMC::GenVertex(FourVector(parton1->vertex()));

Using John's latest patch (cms-sw#21349) does not solve the problem. Does this sample even have incoming protons stored??

@jhgoh Do you have any idea? This is the input file: root://cmsxrootd.hep.wisc.edu//store/mc/RunIISummer17MiniAOD/GGToEE_Pt-50_Elastic_13TeV-lpair/MINIAODSIM/92X_upgrade2017_realistic_v10-v1/100000/5A440CEE-DCA5-E711-9424-0242AC11001A.root

[jhgoh]

Yes I can reproduce crash even after applying PR21349.
I was asking a parton to have |eta| > 100 (just a random large number), but in this generator, this selection was too tight to catch them.

# pdgId eta pz
22 -6.86314 -103.904
22 7.69135 31.1981
2212 -10.9831 -6396.1
-11 -0.100614 -5.08589
11 -1.42446 -98.8182
2212 13.0257 6468.8
11 -1.10236 -67.6201

I guess releasing eta cut will fix the problem. let me try.

[jhgoh]

OK, two separate problems as pointed by @arizzi

1. missing guard in MergedGenParticleProducer

diff --git a/GeneratorInterface/RivetInterface/plugins/MergedGenParticleProducer.cc b/GeneratorInterface/RivetInterface/plugins/MergedGenParticleProducer.cc
index 42cf610af0e..895e93465f4 100644
--- a/GeneratorInterface/RivetInterface/plugins/MergedGenParticleProducer.cc
+++ b/GeneratorInterface/RivetInterface/plugins/MergedGenParticleProducer.cc
@@ -133,6 +133,7 @@ void MergedGenParticleProducer::produce(edm::Event& event, const edm::EventSetup
 
 bool MergedGenParticleProducer::isPhotonFromPrunedHadron(const pat::PackedGenParticle& pk) const
 {
+  if ( pk.mother(0) == nullptr ) return false;
   HepPDT::ParticleID motherid(pk.mother(0)->pdgId());
   return
     ( pk.pdgId() == 22 // We care about photons for lepton dressing here

2. Tight eta cut on incident particle.
   set_beam_particles() had to be commented out to avoid "Event beams mismatch"

diff --git a/GeneratorInterface/RivetInterface/plugins/GenParticles2HepMCConverter.cc b/GeneratorInterface/RivetInterface/plugins/GenParticles2HepMCConverter.cc
index c7fe72ef4c8..d65a9f9a3f0 100644
--- a/GeneratorInterface/RivetInterface/plugins/GenParticles2HepMCConverter.cc
+++ b/GeneratorInterface/RivetInterface/plugins/GenParticles2HepMCConverter.cc
@@ -113,7 +113,7 @@ void GenParticles2HepMCConverter::produce(edm::Event& event, const edm::EventSet
     genCandToHepMCMap[p] = hepmc_particle;
 
     // Find incident proton pair
-    if ( p->pdgId() == 2212 and std::abs(p->eta()) > 100 and std::abs(p->pz()) > 1000 ) {
+    if ( p->mother(0) == nullptr and std::abs(p->eta()) > 5 and std::abs(p->pz()) > 1000 ) {
       if ( !parton1 and p->pz() > 0 ) {
         parton1 = p;
         hepmc_parton1 = hepmc_particle;
@@ -132,7 +132,7 @@ void GenParticles2HepMCConverter::produce(edm::Event& event, const edm::EventSet
   hepmc_event->add_vertex(vertex2);
   vertex1->add_particle_in(hepmc_parton1);
   vertex2->add_particle_in(hepmc_parton2);
-  hepmc_event->set_beam_particles(hepmc_parton1, hepmc_parton2);
+  //hepmc_event->set_beam_particles(hepmc_parton1, hepmc_parton2);

Since the PR is not closed yet, I can apply changes above if this is reasonable.

[mseidel42]

Hi John, thanks a lot!

+  if ( pk.mother(0) == nullptr ) return false;

I think this will miss out on orphaned photons where no mother information is stored. Please see mseidel42@b49a38c for my solution.

[arizzi]

can you guys push the fixes in cms-sw/cmssw both for master and 94X branches?

[jhgoh]

@intrepid42 I can cherry-pick this one, if it is not already in the master branch.
@arizzi I'ml pushing the change now (at least for the 2nd point to GenParticles2HepMCConverter)

[mseidel42]

I can cherry-pick this one, if it is not already in the master branch.

Yes, please, I made the patch this morning :)