Skip to content
Snippets Groups Projects
  1. Jun 07, 2018
  2. Jun 05, 2018
  3. May 21, 2018
  4. May 15, 2018
  5. May 14, 2018
    • scott snyder's avatar
      PyDumper+TrkSurfaces: Avoid more undefined behavior. · 4c592dea
      scott snyder authored
      Avoid calling functions that may return null via a reference.
      
      Allow testing whether a CylinderSurface has a bounds object (it can be missing
      depending on how the object was constructed).  Allows avoiding undefined
      behavior sanitizer warnings from event dumping.
      
      Former-commit-id: 6aaed1df488f86e7caf3d14b7b872d9e592e558b
      4c592dea
    • scott snyder's avatar
      TrkEventTPCnv: Suppress ubsan warning. · 25d322b5
      scott snyder authored
      Backport from master:
      
      Suppress undefined behavior sanitizer warning for a deliberate
      cross-cast (related to working around a problem with old EDM versions).
      
      
      
      Former-commit-id: 53c90b4de2af509569681eec1fc0b1c2a36bd65a
      25d322b5
  6. May 08, 2018
  7. May 07, 2018
  8. May 03, 2018
  9. May 01, 2018
  10. Mar 26, 2018
  11. Mar 24, 2018
  12. Mar 12, 2018
    • Will Leight's avatar
      Fix for ATR-16605, round two · a8b711ed
      Will Leight authored
      As discussed in !2959, a bug in the extrapolator was in rare cases causing thousands of scatterers to be added to a track.
      Thanks to Peter Kluit, we believe that we have now tracked this down to a bug in the STEP_Propagator.
      The propagator was getting stuck in a volume and iterating forward in steps of 1-10 um, adding a scatterer each time.
      At the time, we implemented a "fix" that simply aborted the track-finding when more than 1000 scatterers were found on a track.
      Now, the underlying issue should finally be resolved.
      It appears that the problem was occurring when the propagator found two solutions from the starting surface, the first of which was trivial and the second of which gave a negative distance estimate.
      In this case, the propagator would automatically flip direction, bringing it back almost to the starting point, where it would add a scatterer.
      Then the process would repeat thousands of times, until it got far enough away from the start position that there was no second solution, or one with a postive distance estimate.
      By identifying these cases and rejecting their negative-distance second solutions, this problem is avoided.
      Given the rarity of this problem, FT0 tests naturally show no violation, results are available at /afs/cern.ch/work/w/wleight/public/MuonSW/tooManyScatterers/RunTier0Tests.log
      
      
      Former-commit-id: 0c1c0d3b
      a8b711ed
  13. Mar 01, 2018
    • Vincent Pascuzzi's avatar
      Fix use of smart pointers in `Trk::RectangularSegmentation' · e60b7752
      Vincent Pascuzzi authored
      The previous commit to this class incorrectly used smart pointers,
      specifically, their ownership was not released. Moreover, the wrapped
      objects were initialised incorrectly, leading to the same
      `Trk::PlaneSurface' being duplicated, filling the vector with the same
      surface.
      
      Although it appears the owning objects free memory, we play it safe by
      keeping the smart pointers here, as opposed to resorting back to raw
      ones.
      
      
      Former-commit-id: d79f604018172cc9533a35c453d05a589554a346
      e60b7752
  14. Feb 20, 2018
  15. Feb 15, 2018
Loading