TrackClustersMapper only considers the first event

Reported by @vlisovsk:

Good morning, I would like to report an issue with the TupleToolDEDX spotted first by @jnicolin. In her case, some (1 permille) jobs failed on the grid due to this issue, and here I attach a minimal(-ish) reproducible example.

If the attached script is run with lb-run -c best DaVinci/v46r4 gaudirun.py Lbellell_debug.py, one gets

tuple_eeLine                                                  FATAL Tool 'TupleToolDEDX' acting on particle 'L1' returned a failure status.
tuple_eeLine                                                  FATAL Tool 'TupleToolDEDX' acting on particle 'L2' returned a failure status.
tuple_eeLine                                                  FATAL Failed to fill a candidate. Please fix your options.
tuple_eeLine                                                  FATAL ### NOTE : This is a new behaviour since bug https://its.cern.ch/jira/browse/LHCBPS-690
tuple_eeLine                                                  FATAL ### NOTE : If you think this is wrong, report to lhcb-davinci@cern.ch

However, if one sets DaVinci().SkipEvents = 1 the crash does not happen and the code runs smoothly. This happens only on certain (very few) MDSTs, two of them are referenced at the end of my example script. I also link there one MDST which is "normal" where the crash does never happen.

The underlying issue is that the TrackClustersMapper tool has a cache (m_streamsDone) which remembers that the first event didn't have any Velo clusters. The cache is never reset so it effecitvely remembers what happened in all later events:

• master: https://gitlab.cern.ch/lhcb/LHCb/-/blob/0f1e65a9e8e575871598188c3a200cac7603de9f/Event/EventPacker/src/component/TrackClustersMapper.cpp#L136-154
• run2-patches: https://gitlab.cern.ch/lhcb/LHCb/-/blob/289be64a6a404f89ff9f98e0a12f9faeaa06e311/Event/EventPacker/src/component/TrackClustersMapper.cpp#L70-90

This code is in both run2-patches and master.