[Hps] Physics-object standardization initiative

[mdiamond at physics.utoronto.ca]

Hi folks,

At the collaboration meeting, it was suggested that we establish some  
standardization for physics-object selections (which would be  
performed prior to analysis cuts). I am volunteering to take the lead  
on this initiative, and would like to discuss it at software&analysis  
meetings starting next week.

I am thinking, we have a few options for where selections can be performed:
- MergeTrackCollections driver in the reco sequence, to eliminate some  
tracks (e.g. using AmbiguityResolver) before even building  
reconstructed particles. Would save computation time throughout the  
rest of the reco sequence. But would make the discarded tracks  
unrecoverable, and rerunning reco on existing files may be impractical.
- In skimming drivers, to create skimmed lcios that everyone would  
use. Would have access to all the saved lcio info while making the  
cuts. But would take considerable time to perform the skims, and extra  
space to store the skims if we also retain the original lcios.
- in the dst maker. Would have access to all the saved lcio info while  
making the cuts, and would reduce the size of the output dsts.  
Discarded stuff could still be recovered by going back to the slcio.
- in the tuple maker. Same advantages as above. We would have to  
ensure the same cuts are made by the dst maker and by the tuple maker  
(although this becomes a moot point if we start making tuples from  
dsts).

We need to gather input from both the BumpHunt and Vertex analyses,  
and from both Tracking and ECal. And there are multiple types of  
objects we should consider standardising, e.g.
- useful tracks
- valid reco particles
- valid vertices
We should also consider whether we want multiple levels of selection,  
e.g. LooseElectron and TightElectron.

I will also start a "standardization" channel on slack, for anyone who  
would like to contribute and/or follow the progress of this initiative.

Cheers,
Miriam