`filewriter`: Neutron event data (`raw_data_1/detector_1_events`)

[Tom-Willemsen]

The filewriter must be able to write out event-mode data.

This means the contents of the NXevent_data group in raw_data_1/detector_1_events

Questions

• What do we want to do with event_time_bins, which aren't meaningful in streaming system?
• Due to an underlying UDP connection from hardware, the streaming system cannot guarantee that frame N+1 will have a later reference_time than frame N. Do any consumers make assumptions about this?
  • This affects whether we can stream events to file directly, or whether we have to buffer, sort by reference time, and then write to file
• Does mantid have a dependency on event time-of-flights being ordered within a frame? Currently this sort is (optionally) done at event_aggregator level as multiple consumers benefit from improved performance with sorted events.

Potential differences from existing files

• We likely no longer need to apply a random event_time_offset_shift, as the new electronics has a much higher time resolution than the old DAE electronics. See random time offset shift #22 .
• We want to be able to write the veto flags for each frame alongside the event data, to enable downstream consumers to retroactively enable or disable a veto if it chooses:
  • active_vetos: the veto signals that were actually active for a given frame
  • enabled_vetos: the vetoes that were enabled in IBEX at the time of data acquisition
  • By default downstream consumers will want to mask any events where active_vetos & enabled_vetos != 0.
  • If adding this logic is problematic for Mantid, we could (optionally) drop the frames at filewriter level instead - at the cost of reducing flexibility and reducing diagnostics we have access to.

[martyngigg]

For reference, the logic in Mantid to determine whether to apply the randomisation itself will return False if event_time_offset_shift is missing and the program name is not ISISICP.EXE.

If we're pretty confident this can all be dealt with by the streaming side of things then dropping event_time_offset_shift and setting program_name to something else will make Mantid do what we want it to do.

[martyngigg]

For event sorting, it looks like event_index is required to be sorted.

The details on ordering of events in a frame are a little more challenging to understand and need further investigation. It's not clear from the loading code if a sort happens internally on load every time.