`filewriter`: Neutron histogram data (`raw_data_1/detector_1`)

该 issue 未指定任何文件或测试；首先定位现有的事件模式 filewriter，以及针对 raw_data_1/detector_1 的 NeXuS 处理。明确受支持的直方图/事件组合和对流式直方图的预期，然后验证所选方案是否会生成所需的直方图模式 .nxs 输出。

The file-writing infrastructure must be able to write out histogram-mode data (.nxs files without an NXevent_data group).

This refers to the NeXuS groups under raw_data_1/detector_1.

The issue is intentionally phrased as

file-writing infrastructure

because this does not necessarily have to be done in the same process as the event-mode filewriter (though having it in the filewriter would be more performant and potentially more flexible in future).

Questions

• Support streamed histograms (hs00) at some point in future? i.e. a scheme where a monitor emits a histogram every N frames, where N could be 1 in the extreme case. This then lets a run be sliced a bit like event mode, but only on N-frame boundaries.
  • This might be needed for SANDALS2 if monitor event-rates are too high for pure-event mode.
  • Need to define how we would write this kind of data to .nxs.

Approaches

• This could be an option in the filewriter process itself (preferred). It would then need various options to cope with:
  • event files only
  • combined histogram+event files
  • separate histogram and event files
• This could be implemented as a separate process which reads in an event-mode .nxs file and outputs a histogram-mode .nxs file.
  • But that then makes it harder to support all combinations of files we may need to generate, e.g. event-only, histogram-only, event+histogram in the same file, event-detectors with histogram-monitors, etc.
  • This may also have a performance disadvantage as all of the events need to be written by one process, then read back in to another process before histograms can be generated. For large files this file-loading cost may be significant, especially for beamlines that ultimately only care about the histogram files.

Potential differences from existing files

None known at this time.