[RF] Segmentation fault in ROOT 6.30 workspace creation about root

[TomasDado]

Hi, I managed to prepare a relatively simple code that reproduces the issue:

void crash() {
  RooStats::HistFactory::Measurement meas("mwe","mwe");
  meas.SetOutputFilePrefix("./");
  meas.SetPOI("signalStrength");
  meas.SetLumi(1);
  meas.AddConstantParam("Lumi");

  RooStats::HistFactory::Channel chan1("channel1");
  chan1.SetData("","");
  RooStats::HistFactory::Sample sample1("sample1");
  sample1.SetHistoName("emu_OS_1b_Ma");
  sample1.SetHistoPath("");
  sample1.SetInputFile("BLIND_separate_tt_Z_29IFB_Apr_29_histos.root");
  sample1.SetNormalizeByTheory(false);
  sample1.AddNormFactor("signalStrength", 1, 0.8, 1.2);
  sample1.AddNormFactor("b", 0.543, 0.5, 0.6);
  sample1.AddOverallSys("Dummy",1,1);

  RooStats::HistFactory::Sample sample2("sample2");
  sample2.SetHistoName("emu_OS_1b_Mb");
  sample2.SetHistoPath("");
  sample2.SetInputFile("BLIND_separate_tt_Z_29IFB_Apr_29_histos.root");
  sample2.SetNormalizeByTheory(false);
  sample2.AddNormFactor("signalStrength", 1, 0.8, 1.2);
  sample2.AddNormFactor("minusONE", -1, -1.1, -0.9);
  sample2.AddNormFactor("b2", 0.2950, 0.0, 1.0);
  sample2.AddOverallSys("Dummy",1,1);

  chan1.AddSample(sample1);
  chan1.AddSample(sample2);

  RooStats::HistFactory::Channel chan2("channel2");
  chan2.SetData("","");
  RooStats::HistFactory::Sample sample3("sample3");
  sample3.SetHistoName("emu_OS_2b_Mc");
  sample3.SetHistoPath("");
  sample3.SetInputFile("BLIND_separate_tt_Z_29IFB_Apr_29_histos.root");
  sample3.SetNormalizeByTheory(false);
  sample2.AddNormFactor("signalStrength", 1, 0.8, 1.2);
  sample3.AddNormFactor("b2", 0.2950, 0.0, 1.0);
  sample3.AddOverallSys("Dummy",1,1);
  chan2.AddSample(sample3);

  meas.AddPreprocessFunction("b2","b*b","b[0.543,0.0,1.0]");
  meas.AddConstantParam("minusONE");
  meas.AddChannel(chan1);
  meas.AddChannel(chan2);
  meas.CollectHistograms();
  RooStats::HistFactory::MakeModelAndMeasurementFast(meas);
}

The code and the input can be found here: /afs/cern.ch/user/t/tdado/public/ROOT_issue_14225

It is probably not the simplest example, but it is realistic (the histograms and the setup come from an actual measurement). Note that using only the first channel does not lead to a crash.

I hope this helps, let me know if I can do more.

[will-cern]

Am skeptical this could ever have worked even in older releases, because looking at the code I see the method "MakeSingleChannelWorkspace" where the measurement object is passed, and therefore it is assuming any parameter in the constant list must be present in the channel???

Tomas do you get the same error message about "cannot find the variable" in 6.28, but it succeeds? It might be blind luck that it did.

If so the fix is a simple one to just skip over said variable if its not found in the channel.

[TomasDado]

Indeed, I get the same error in 6.28.04! I also noticed this error:

[#2] ERROR:HistFactory -- could not corresponding auxiliary measurement  nom_minusONE

But it is amazing that it never crashed in 6.28 even with O(20) runs on different machines/OS/gcc versions