Minutes of Data Model Meeting
                 =============================

                Wednesday 28-May-1997, 10:30-12:30
                      (NINTH CIRCLE)

Present:
      D.Adams, J.Bantly, F.Bartlett, M.Diesburg, H.Greenlee, Q.Li, 
      L.Lueking, B.May, W.Merritt, L.Paterno, S.Protopopescu, S.Snyder,
      + unkowns

Scribe:   Qizhong Li

Agenda:

    - Summary report from last week's data model working group meeting
                                                          - Wyatt Merritt
    - Discussions on data model converging                - all


The week before this meeting we had a small working group meeting on the
data model and continued the  discussions on the differences between the
proposed data models. So this  meeting starts with that Wyatt summarized
what we had discussed so far and made a status report on the data model.

First Wyatt went  through the  original charge,  which were presented at
3/6/97:  specify and  design the  Event class   relationships and access
methods for Run II reconstruction. Then she listed some conclusions that
have achieved consenses:

    - Subsets of event data (known as  Chunks, KeyedObjects,...) will be
      stored in a flat, strongly typed fashion - not hierachical. 

    - Class designs (and organization  of CVS packages) can - and will -
      be such that you can  link only the code  that describes the event
      subsets which your excutable will actually use. 

    - This  model is   levelizable  and meets  two  major  design goals:
      Testability and Extensibility. 

The things need  more  understanding: 

    - Details   of   association   management,  like    implications for
      persistency mechanism, interactions with versioning 

    - Details of reconstruction code and event data association.

Then the   discussion was  mainly  focused on if  we need  the on-demand
reconstruction as a feature in the model. This is one of the differences
between  Marc/Brent's and Adams's  proposed models.  We had an intensive
discussion on this  topic, most of people in  favor of having an outside
reconstructor rather than having  the generator as a method inside of an
object. i.e. we  do not want to  have the on demand  reconstruction as a
feature in the event model, mainly because of following reasons:

    - Both RECO and L3 want deterministic operations.
    - Either doubles the responsibility or delegates too much to event
      access.
    - Area of usefulness is very limited.
    - It is difficult to know how an physics object was reconstructed if
      most of  analyses do  not use  the objects  reconstructed from the
      production releases of the reconstruction.
    - It is  difficult in  controling of  wasting  computer resources to
      repeat  again  and  again  the  production  farm's  task  for many
      millions events.
    - Avoiding physical coupling could be complicated.
    - ... 
  
We come to the  consenses that the data model  design will start without
having the generators inside of the  event model. Some people feels that
we may consider to impliment this feature sometime later.

Wyatt also pointed out another task: to write out interfaces for as many
event subsets as we can currently think of. After the meeting, we formed
a small group  (Serban, Qizhong,  J.Hobbs and Herb)  to do this task and
will try to report it on the next data model meeting.