[RF] RooAddPdf still emits unexpected errors about missing normalization sets

[guitargeek]

Describe the bug

If you generate a toy dataset from a model that is a product of sums, you still get the Evaluating RooAddPdf without a defined normalization set warnings, even though the normalization set is defined.

Expected behavior

There should be no such warnings, because the normalization set is defined. When you generate toy data for a given set of observables.

To Reproduce

void testProdOfSum()
{
    RooWorkspace ws{"ws"};
    ws.factory("Gaussian::gs(x[0,10], mu[2, 0, 10], sg[2, 0.1, 10])");
    ws.factory("Exponential::ex(x, lm[-0.1, -10, 0])");
    ws.factory("SUM::gs_ex(f[0.5, 0, 1] * gs, ex)");
    ws.factory("Gaussian::gs_1(x, mu_1[4, 0, 10], sg_1[2, 0.1, 10])");
    ws.factory("PROD::pdf(gs_1, gs_ex)");

    std::unique_ptr<RooDataSet>data{ws.pdf("pdf")->generate(*ws.var("x"), 10000)};
}

Setup

ROOT master on Arch Linux.

Additional context

This issue was observed when running the reproducer code from the related JIRA issues ROOT-9436 and ROOT-10483. If this GitHub issue is fixed, this should also be taken as a opportunity to implement a unit test that covers the linked JIRA issues, which was already fixed in 6.24 but there is yet no unit test that covers it.

There are also two more JIRA issues that might be related to this one:

• https://sft.its.cern.ch/jira/browse/ROOT-9649 (by now, it was checked that it's unrelated)
• https://sft.its.cern.ch/jira/browse/ROOT-10814 (by now, it was checked that it's unrelated)

[guitargeek]

The unit test that covers ROOT-10483 is implemented in #11485.