RooMCStudy::generateAndFit for zero event samples

[mnomura]

Hello RooFitters,

I am trying to perform some MC tests to check if we can recover a specified amount of injected signal. As a part of this test, we try to generate toy samples with zero events, where we would expect a delta function around zero, but we obtain a distribution shown below. Is this the expected behavior?
Screenshot 2018-09-21 13.48.50.pdf (31.0 KB)

  RooWorkspace* r = new RooWorkspace("r");
  r->factory().process("N_landau[17.1525631, 0., 50.]");
  r->factory().process("RooLandau::landau(mjj[0., 100.], landau_mean[40.],"
                       "landau_width[10.])");
  RooExtendPdf* extended_landau = new RooExtendPdf(
      "extended_landau", "", *r->pdf("landau"), *r->var("N_landau"));
  r->import(*extended_landau);
  r->var("landau_mean")->setConstant();
  r->var("landau_width")->setConstant();
  extended_landau_mc_study = new RooMCStudy(*extended_landau,
                                            *r->var("mjj"),
                                            Binned(kTRUE),
                                            Extended(),
                                            FitOptions(Save(kTRUE)));

    extended_landau_mc_study->generateAndFit(
        1000, 0, kTRUE);