[RF] Creating snapshots blows up the size of RooWorkspace to easily reach 1GB buffer limit

[guitargeek]

Check duplicate issues.

• Checked for duplicates

Description

Creating snapshots of variable values clones all the relevant RooRealVar objects instead of simply storing values. This blows up the RooWorkspace size, reaching easily the 1GB buffer limit.

If the buffer limit can't be increased, we have to look at alternatives to store the information in the RooWorkspace with less redundancy.

Reproducer

std::size_t nVars = 1000000;
std::size_t nSnapshots = 3;

RooArgSet args;

for (std::size_t i = 0; i < nVars; ++i) {
    std::string name = "var_" + std::to_string(i);
    args.addOwned(std::make_unique<RooRealVar>(name.c_str(), name.c_str(), i));
}

RooWorkspace ws;
std::cout << "importing..." << std::endl;
ws.import(args, RooFit::Silence());

std::cout << "snapshotting..." << std::endl;
for (std::size_t i = 0; i < nSnapshots; ++i) {
   ws.saveSnapshot("snapshot_" + std::to_string(i), ws.allVars(), true);
}

ws.writeToFile("repro_ws.root");

Output:

root [0]
Processing repro.C...
importing...
snapshotting...
Error in <TBufferFile::WriteByteCount>: bytecount too large (more than 1073741822)
root [1]

ROOT version

any

Installation method

any

Operating system

any

Additional context

This is a problem for current ATLAS Higgs combinations.

[guitargeek]

Problem was again brought up on the ROOT forum:
https://root-forum.cern.ch/t/workspace-combiantion-error-in-tbufferfile-checkcount-buffer-offset-too-large-larger-than-1073741822/63468

[Superharz]

This also seems to be a problem for the analysis of my group.