Abstract for CHEP98
David Adams
20may98 1550

TRF++: an object-oriented framework for finding tracks

D. L. Adams, S. Kulik, B. Kulik, N. Graf

TRF++ provides an extensible, object-oriented framework for finding and
fitting tracks in particle physics detectors.  The concepts natural to
the problem are mapped onto abstract classes.  These include surfaces,
layers, clusters, hits and tracks.  There are also abstract classes
to describe track propagation, fitting, finding and filtering.  The base
part of TRF++ provides concrete subclasses where they are independent of
surface type while TRF++ extensions provide implementations for particular
surface types.  Present extensions include cylinders with axis along z,
planes perpendicular to z and planes parallel to z.

The track-finding part of TRF++ is based on a road-following algorithm
and the one concrete implementation uses a Kalman filter update.  This
allows for easy integration of interactions such as multiple scattering
and, most importantly, can be very CPU-efficient.  The latter is essential
so that TRF++ can be applied to modern hadron collider experiments with
their high data rates and high track densities.

TRF++ is implemented in C++ because of its wide availability, its strong
support for object-oriented concepts and its inherent CPU-efficiency.
It is intended that TRF++ conform to the new ANSI/ISO C++ standard.
It has been successfully compiled and run using the KAI compiler[1]
under IRIX6.

TRF++ is organized as a collection of packages with acyclic dependencies
so that users may limit their use (and dependency) to the relevant pieces.
Unneeded extensions or the track-finding part may be omitted.
The software within packages is organized into components consisting of
header, implementation and test files.  The implementation is compiled
and put into the package library.  The test is a main program which
tests the code for the component.  Typically a component contains one
class and associated free functions.  This organization is very close to
that described by Lakos[2].

The instructions for building the software are contained in CTEST[3]
input files such as COMPONENTS listing the components for a particular
package.  This avoids the need to duplicate instructions in different
packages and should ease integration into another environment.

The software has gone through two major design/implementation cycles.
The D0 experiment plans to make use of TRF++ for its run II offline
reconstruction.  At present, it is being used to find Monte Carlo
tracks in the D0 central scintillating fiber and silicon strip detectors.

A description of TRF++ may be found at 
http://www.bonner.rice.edu/adams/trf++.

References
----------

1. See http://www.kai.com/C_plus_plus.

2. J. Lakos, "Large Scale C++ Software Design", Addison-Wesley, 1996.

3. CTEST is described at http://www.bonner.rice.edu/adams/ctest and
   in a separate submission to this conference.