Skip to content

TrackParticleCreator: Go around the issue of ForceUpdate discussed in ATR-19773

Christos Anastopoulos requested to merge ATLAS-EGamma/athena:TrackParticleCreator_ATLAS_THREAD_SAFETY_ATR-19773 into master

One of the main thread safety issues of the TrackParticleCreator was the ForceUpdate Summary where one forces a const Trk::Track to be updated.

i.e this call

if (m_forceTrackSummaryUpdate) {
        // Do we really need to update const Trk::Track* track?
        Trk::Track& nonConstTrack = const_cast<Trk::Track&>(*track);
        m_trackSummaryTool->updateTrack(nonConstTrack);
      }

This option (at least directly ...) is used only in legacy Trigger code, but according to ATR-19773 although this code is legacy, people want to retain it with the same semantics, as the code needs to run in NoMT as reference or so (@jmasik can correct me but this is my take away from ATR-19773).

If we ever remove it or there is no need to run like this, we can get rid also of the need of a noMT version.

To avoid duplicating too much code hide the non-MT part inside a virtual call. The standard version does not do anything here, no const_cast --> thread safe

The NoMT version does the const_cast above and I marked it at large as ATLAS_NO_THREAD_SAFE. Legacy clients that rely on updating a const Trk::Track taken from store should be using the NoMT version and can run only in legacy mode i.e no-MT ... (hope the names make it a bit explicit ...)

Mentioning @oda , @jmasik

Edited by Abhishek Nag

Merge request reports