My initial thoughts on a solution are to create a map module, MapCppDataSelection or something, which sits in a chain:

InputCppRoot -> MapCppDataSelection -> OutputCppRoot

i.e. it would take in a maus_output ROOT file containing the full dataset, and output a new maus_output file containing only the desired reconstruction events.

The mapper itself could be implemented in a number of ways - initially I may try the most direct route of editing each Spill in place, removing those events which fail the cuts from the recon events vector, and then passing the resultant Spill and Data object up to the output module to create the final ROOT file.

An alternative would be to add a kill / keep flag to the recon event class, setting this in the mapper, then modifying the output modules to only write those events with the flag set to keep (think this boils down to the same thing, just with the output module doing the actual pruning of the events).

Just curious, where did the request come from?

Request originates with Ken. We would like to be able to study tracker performance for those events which have, for example, one and only one spacepoint in TOF1 and TOF2. A purely offline script is not sufficient as we need to extract pattern recognition data which are not (nor should be) written into the standard MAUS output file (specifically all the road cut residuals, whether that trial track is accepted or not).

First iteration of this now working in lp:~ajdobbs/maus/data-selection.

I noticed when I was writing the code that InputCppRoot still uses JSON conversion, which we need to fix. Will create a separate issue.

I have done some more work on this. I have added a boolean field to the Data class called _should_write, with the corresponding get and set methods. I have also implemented an EventSelection mapper which will set this. Lastly I have updated OutputCppRoot to check _should_write before writing data to the output ROOT file.

The current road block is that the _should_write field gets reset to true in between the EventSelection mapper and OutputCppRoot. The only intervening module is ReducePyDoNothing. I have tried switching to a cpp reducer instead but that in turn is blocked by issue #1868. The reason for the reset is currently unknown.

Latest work on Launchpad at lp:~ajdobbs/maus/event-selection