Stepping statistics mismatches Simulation/Reconstruction
When comparing the number of steps needed for simulating a particle through the detector to reconstructing, there's a huge discrepancy (this makes the KalmanFitter
so much slower than the simulation).
Partly this is due to the missing trigger for smoothing, partly it looks like the accuracy step estimation gets entirely off.
The first can be checked with !682 (closed) which should introduce a possibility to trigger smoothing on last measurement, and then the KF should practically step like the simulation.
The second could be related to #653, and particularly dresses the following code block:
// @HOTFIX to break the RK step trying, @TODO will be replaced
// by proper overstepping mechanism
if (++stepAttempts == 100) {
state.stepping.emergency = true;
// cout some
// step in mm steps, costly but should do
state.stepping.stepSize = state.stepping.navDir * 1_mm;
break;
}
This is still happening numerous times.