Minutes of the off-line meeting,
25th August 1998
Present: P.Binko, T.O'Reilly, E.Van Herwijnen,
A.Jacholkowska, P.Colrain, J.Harvey, P.Mato, N.Zaitsev, M.Cattaneo, J.Saborido,
A.Tsaregorodtsev, M.Frank, T.Nakada, T.Ruf, T.Ketel, F.Filthaut
Minutes: M.Cattaneo
Agenda:
News
Design of Vertex Pile-up Trigger OO software (N.Zaitsev)
Rational Rose customisation (T.O'Reilly)
MCTree Model (P.Binko)
A.O.B.
1. News
Pavel reported on progress on the Iain Last package
The classes now work both on RSPLUS and HPPLUS. It is possible to access
all the banks, and also to modify and write back the modified values.
There is still a problem creating new rows in existing banks, or
creating new banks. It was agreed that this should be fixed as top priority.
The classes do not compile on Digital Unix (Alpha). Since the DUX
compiler is the most stringent, it was felt that getting the classes to compile on Alpha
is a good test of the software. This should be the second highest priority.
Some SICB banks contain references as data members. What should be done
with these?
Timescale: the top priority item can probably be fixed during this week.
2. Design of Vertex Pile-up Trigger OO software (N.Zaitsev)
Nikolai is starting a project to reimplement the pileup trigger
software as a L0TriggerPileup object in order to learn OO methods. The software should
interface to both the "sicb++" environment and the test beam environment - see
his slides (.pdf)
3. Rational Rose customisation (T.O'Reilly)
Thomas described the work he has done since Friday to customise
Rational Rose for the LHCb environment (see his slides .ppt, .ps). The aim is to produce a properties file
that can be used by anyone wanting to generate code from Rose using the LHCb coding
conventions. His starting point was the customisation made by Niko for the RICH. Currently
the code generation produces too many comments (though this was partly fixed after the
meeting). It may be difficult to generate code conforming with all the LHCb conventions.
In the discussion it was requested that a script or a very simple set of
few instructions be provided to invoke Rose in the LHCb configuration. It also seems that
a language sensitive editor is desirable, to highlight the generated code from the
comments. Xemacs is the only such tool known on Unix.
It was felt that Rose on Unix is unpleasant. Andrei would like to try it
out on WNT, Tom will try to insall the free evaluation version for students. It is clear
that there is a licensing problem due to the high price of this tool.
4. MCTree model (P.Binko)
Pavel started to show diagrams of his TransientEvent model,
consisting of a tree-like hierarchy of objects down to the lists of hits,
digitisings etc. There was a lively discussion focusing on his first slide showing
the RawTree. The main objection was that the model is addressing the organisation of the
data and proposes an object structure to support that, but does not identify real objects
(such as RICH detector) that manipulate the data.
Pere proposed a class structure that separates out the detector
structure from the data structure. Pavel felt that this was a different view of the same
thing: his diagram is showing the objects instantiated from Pere's classes - maybe we need
to have a common glossary of OO terms and UML conventions in order to talk the same
language - Tom will work on this.
Markus asked whether starting from the data is the correct
approach. It would be more natural to model behaviour first, for example with use cases
for reconstruction, and then organise the data to support that. What Pavel has shown is
essentially the SICB data model, which is a starting point, to be refined, and allows us
to get started quickly using the existing SICB data, but it is not clear that this is the
correct approach for arriving at a good OO design.
What can we learn from other experiments? Should we look at their
event models? Pavel will invite someone to do this after the CHEP conference.
In order to proceed, the following actions were suggested:
- Try to converge on and document (on the Web) a first simple set of modelling
tools and standards (UML, Rose etc.)
- Put Pavel's MCTree model in Rose, and populate the tree with data. A review
session should be organised with a few experts.
- Also store the complete TransientEvent in Rose to allow further discussion (and
so that it can be added to the minutes...). Pere should do the same for the class
structure that he proposed.
- When Niko returns from his holidays we should work with him to try a functional
approach to the data modelling (i.e. model the data according to the requirements of the
RICH simulation and reconstruction).
5. A.O.B.
The next meeting will be on 1st September at 10:30 in 32-SC-22 as
usual, with the following agenda
LHCb Home | Computing Home | Search
|