RDataFrame Reads Garbage Data instead of File

[davidkowalk]

Check duplicate issues.

• Checked for duplicates

Description

This bug report is the result of this forum thread:

https://root-forum.cern.ch/t/rdataframe-reading-garbage-data/63763

When loading root files created by older versions RDataFrame reads arbitrary data instead of actual file.

Reproducer

The following Root File

using namespace ROOT;

void check_photon_data() {
    RDataFrame df("t", "data/merged_20250610.root");

    auto calculate_photon_pair_mass = [](const RVec<TVector3> Ps, const RVec<double> Es) -> RVec<double> {
        //Calculates the invariant mass of each photon pair in hit
        //Later the hits with only etas can be counted.

        RVec<double> masses;
        for (size_t i = 0; i < Es.size(); ++i) {
                for (size_t j = i + 1; j < Es.size(); ++j) {
                    TLorentzVector p4_1, p4_2;
                    p4_1.SetVect(Ps[i]);
                    p4_1.SetE(Es[i]);
                    p4_2.SetVect(Ps[j]);
                    p4_2.SetE(Es[j]);
                    masses.push_back((p4_1 + p4_2).M());
                }
        }
        return masses;
    };

    auto df_out = df.Define("mass", calculate_photon_pair_mass, {"lP4BGONeutral.fP", "lP4BGONeutral.fE"})
                    .Define("p2",
                        [](const RVec<TVector3>& Ps){
                            RVec<double> p2;
                            for (TVector3 p:Ps) {
                                p2.push_back(p*p);
                            }
                            return p2;
                        },
                        {"lP4BGONeutral.fP"})
                    .Define("E2",
                        [](const RVec<double>& Es) {
                            RVec<double> E2;
                            for (double E:Es) {
                                E2.push_back(E*E);
                            }
                            return E2;
                        },
                        {"lP4BGONeutral.fE"})
                    .Define("nPhotonsP", [](const RVec<TVector3>& Ps){return Ps.size();}, {"lP4BGONeutral.fP"})
                    .Define("nPhotonsE", [](const RVec<double>& Es){return Es.size();}, {"lP4BGONeutral.fE"})
                    ;

    df_out.Display({"mass", "E2", "p2", "lP4Tagger.fE"}, 50) -> Print();

    df.Histo1D("lP4BGONeutral.fE") -> DrawClone();
    df_out.Histo1D("lP4BGONeutral.fE") -> DrawClone();
    df_out.Histo1D({"E2", "E2", 100, 0., 200000.} , "E2") -> DrawClone();

}

ROOT version

ROOT 6.34.08
Built for linuxx8664gcc on May 17 2025, 14:17:01
From tags/6-34-08@6-34-08
With c++ (GCC) 15.1.1 20250425

Installation method

pacman

Operating system

arch linux

Additional context

No response

[dpiparo]

Thanks: now it's clearer!

For a minimal reproducer, let me start from something simplified to the max (or almost).
Here the link to the file you provided.
Then, some code:

// TTree Draw
auto f = std::unique_ptr<TFile>(TFile::Open("output_first1000.root"));
auto t = f->Get<TTree>("t");
t->Draw("lP4Tagger.lP4Tagger.fE>>htemp(128, 750, 3000)");

// RDF
ROOT::RDataFrame df("t", "output_first1000.root");
auto h = df.Histo1D({"h", "h", 128, 750, 3000},"lP4Tagger.fE");
h->SetLineColor(kRed);
h->Draw("Same");

And its result, using the tip of the v6-36-00-patches branch:

Note that TTree seems to agree with what is seen by JSROOT, that can be obtained by clicking on the link above:

There might be more, and this can be seen when extending the range of the histogram to zero: a peak appears.

[ferdymercury]

This shows the issue also:

root -l output_first1000.root -b -e 'auto t = _file0->Get<TTree>("t"); t->SetScanField(0); t->Scan("lP4Tagger.lP4Tagger.fE");' -q > /tmp/t.out
root -l -b -e 'ROOT::RDataFrame df("t", "output_first1000.root"); auto d1 = df.Display("lP4Tagger.fE", 1000); d1->Print();' -q > r.out
// Remove markdown lines by hand in t.out
meld r.out t.out

In entries where leafcounter > 1, the second value is wrongly set to zero by RDF / TTreeReaderArray / TClonesArray.

Maybe related: https://its.cern.ch/jira/browse/ROOT-9250?jql=project%20%3D%20ROOT%20AND%20resolution%20%3D%20Unresolved%20AND%20text%20~%20%22ttreereaderarray%22%20ORDER%20BY%20priority%20DESC%2C%20updated%20DESC

Fyi @pcanal