Pythia8Generator produces different results with multi-threading

[andiwand]

I experimented with this a bit and it seems like as soon as we use a single Pythia8Generator instance twice, which happens with nhard > 1 or npileup > 1 with addPythia8, the output will be randomly different.

My best guess is that Pythia8 has some internal state and starts to cross-talk over events if one generator is called from multiple threads.

This would also mean that skipping events will not produce the same results as actually processing the events.

Until this is fixed we should stick to single threading and not skip events if reproducibility is important.

[paulgessinger]

You can supply the random number sequence but maybe that's not enough to make reproducible results?