Evolution of cluster design in the Instrumented Flux

Evolution of cluster design in the Instrumented Flux Return (IFR) of Ba. Bar Luca Lista INFN, Sezione di Napoli for the Ba. Bar Computing Group Luca Lista - CHEP '98

The Instrumented Flux Return of Ba. Bar • 18 -19 layers of Resistive Plate Chambers (RPC) • 2 layers of cylindrical RPC inside the coil • Digital strip readout, 2 -3 cm pitch Luca Lista - CHEP '98

Ifr basic reconstruction information • The Ifr reconstruction provides the following basic information strip 1 D cluster Luca Lista - CHEP '98 3 D cluster

Clustering in the IFR • The IFR identifies m, p and K 0 L • The reconstruction information in the IFR is made of clusters of strips which are “close” in space • Details of the pattern information give particle ID discriminating power m cluster • Estimate of the impact point provides a determination of the flight direction of the K 0 L p cluster Luca Lista - CHEP '98

2 D and 3 D Clustering • Projective geometry in the planar RPC’s ® 2 D reconstruction in each projection • 3 D association of the two projections Ifr 2 DCluster Luca Lista - CHEP '98

Early class design • The basic class structure Ifr 3 DCluster 1 2 Ifr 2 DCluster 1 1. . n Ifr 1 DCluster • The initial interface: Ifr 3 DCluster center. Of. Gravity () : Point cov. Matrix () : Bbr. Error direction (from : Point) : Vector 1 number. Of. Hits () : int number. Of. Strips () : int first. Layer () : int last. Layer () : int hit. Layers () : int 2 1 1. . n Ifr 1 DCluster number. Of. Strips( ) Luca Lista - CHEP '98 Ifr 2 DCluster number. Of. Hits () : int number. Of. Strips () : int first. Layer () : int last. Layer () : int hit. Layer () : int

Composite clusters • Cluster segments associated to the same charged track are combined together Ifr 3 DComposite Ifr 3 DCluster center. Of. Gravity () : Point cov. Matrix () : Bbr. Error direction (from : Point) : Vector number. Of. Hits () : int number. Of. Strips () : int first. Layer () : int last. Layer () : int hit. Layers () : int 1. . n Luca Lista - CHEP '98 center. Of. Gravity () : Point cov. Matrix () : Bbr. Error direction (from : Point) : Vector number. Of. Hits () : int number. Of. Strips () : int first. Layer () : int last. Layer () : int hit. Layers () : int

Polimorphic clusters • The user shouldn’t care about the detail of the cluster substructure: Ifr. Abs 3 D center. Of. Gravity () : Point cov. Matrix () : Bbr. Error direction (from : Point) : Vector number. Of. Hits () : int number. Of. Strips () : int first. Layer () : int last. Layer () : int hit. Layers () : int – just use Ifr. Abs 3 D • All clusters implement the same interface. Ifr 3 DComposite center. Of. Gravity () : Point cov. Matrix () : Bbr. Error direction (from : Point) : Vector number. Of. Hits () : int number. Of. Strips () : int first. Layer () : int last. Layer () : int hit. Layers () : int Luca Lista - CHEP '98 Ifr 3 DCluster 1. . n center. Of. Gravity () : Point cov. Matrix () : Bbr. Error direction (from : Point) : Vector number. Of. Hits () : int number. Of. Strips () : int first. Layer () : int last. Layer () : int hit. Layers () : int

Inner RPC clusters • The inner RPC readout layout changed during software evolution: – stereo readout strips • 2 D clustering is no longer appropriate • Composite clusters can’t be done of homogeneous elements Luca Lista - CHEP '98

Evolution of the cluster model • Cluster information is intrinsically complex • It is not possible to predict now all the needed functions • The extension of the class functions requires changing the base class interface • Extension to new cluster types? Luca Lista - CHEP '98 Ifr. Abs 3 D () : Ifr. Abs 3 D (a : const Ifr. Abs 3 D&) : Ifr. Abs 3 D position () : Hep. Point ~Ifr. Abs 3 D () number. Of. Digis () : int number. Of. Strips () : int cov. Matrix () : Bbr. Error pol. Cov. Matrix () : Bbr. Error $pol. Cov. Matrix ( : Bbr. Error&, : Hep. Point&) : Bbr. Error first. Layer () : int last. Layer () : int hit. Layers. I () : int hits. In. Layer (n : int) : int strips. In. Layer (n : int) : int max. Missed () : int section. Code () : Ifr: : Section. Code component ( : int = 0) : Ifr 3 DCluster* number. Of. Components () : int number. Of. Components ( : Ifr: : Sector. Code) : int number. Of. Components ( : Ifr: : Section. Code) : int get. Composite () : Ifr 3 DComposite* has. Barrel () : bool has. End. Cap () : bool has. Fwd. Cap () : bool has. Bwd. Cap () : bool has. Inner () : bool operator == ( : const Ifr. Abs 3 D&) : bool number () : unsigned int $reset. Counter () : void save. Guts ( : RWvostream&) : void restore. Guts ( : RWvistream&) : void Ifr 3 DComposite 1. . n Ifr 3 DCluster

Problems spotted during the development • The cluster class interface growth rate was more than linear with time • Code which used deep information of the cluster pattern tended to “understand” which subclass type it was using (dynamic_cast) • Dependencies on cluster subtypes rather than base class increased • Problems to effectively evolve the design Luca Lista - CHEP '98

Visitor pattern • Each function of the cluster classes is implemented in a separate class for all cluster types – all inherit from Ifr. Cluster. Visitor abstract interface • Extending the functionality does not require any change to classe interfaces – just add a new visitor class • Dependency on cluster subclasses is concentrated in the visitor classes • The design is, according to an OO principle, – open to extensions – closed to class interface changes Luca Lista - CHEP '98

1. . n Visitor pattern class diagram Ifr. Abs 3 D T Ifr. Cluster. Visitor accept (Ifr. Cluster. Visitor<T>) : T operate (const Ifr 3 DCluster*) : T operate (const Ifr. Inner 3 DCluster*) : T operate (const Ifr 3 DComposite*) : T operate (const Ifr 2 DCluster*) : T Ifr 3 DComposite accept (const Ifr. Cluster. Visitor<T>& v) : T Ifr. Cluster. Visitor{Hep 3 Vector} Ifr. Inner 3 DCluster accept (const Ifr. Cluster. Visitor<T>& v) : T Ifr 3 DCluster accept (const Ifr. Cluster. Visitor<T>& v) : T 2 Ifr 2 DCluster accept (const Ifr. Cluster. Visitor<T>& v) : T Ifr. Vst. Center. Of. Gravity operate (const Ifr 3 DCluster*) : Hep 3 Vector operate (const Ifr. Inner 3 DCluster*) : Hep 3 Vector operate (const Ifr 3 DComposite*) : Hep 3 Vector operate (const Ifr 2 DCluster*) : Hep 3 Vector { return v. operate(this); } Luca Lista - CHEP '98

Visitors features • New visitors can be added as new subclasses of Ifr. Cluster. Visitor<my. Type> with no other changes • New cluster types can be added without mayor design changes (e. g. : Ifr 3 DFast. Cluster) – but all the visitors subclasses should be updated to work on the new cluster subclass • Details can be handled internally: • – calibrations, alignments The code which uses Ifr. Abs 3 D and visitors appears to be much more robust than before – a couple of unsolved bugs have been understood and fixed during the migration Luca Lista - CHEP '98

Visitors features (cont. ) • Recursive action on Ifr 3 DComposite components matches the composite pattern – Ifr 3 DComposite has n Ifr. Abs 3 D’s • A visitor subclass handles a built-in cache mechanism implemented with a hash dictionary. – a visitor “knows” the cluster he has already “visited” – the cache is invalidated at: • destruction of every cluster for the deleted cluster • at the end of every event (to prevent the accumulation of possible memory leaks) Luca Lista - CHEP '98

Available Visitors • Visitors are singleton/multiton • Some of the available visitors are: • Ifr. Vst. Center. Of. Gravity • Ifr. Vst. Cov. Matrix • Ifr. Vst. First. Layer – first | last. Barrel • Ifr. Vst. Has. Sector – forward | backward | barrel • Ifr. Vst. Hit. Layers • Ifr. Vst. Hits. In. Layer – n = 1, . . . , 19 • Ifr. Vst. Max. Missed • Ifr. Vst. Number. Of. Strips • . . . Luca Lista - CHEP '98

Visitors Usage • How to use visitors: Ifr. Abs 3 D* ifr; // get a cluster from somewhere // compute center of gravity Hep. Point c = ifr->accept( Ifr. Vst. Center. Ofgravity: : instance() ); // has hits in the barrel? bool barrel = ifr->accept(Ifr. Vst. Has. Sector: : instance (Ifr. Vst. Has. Sector: : barrel) ); // get number of hits in each layer int n[19], i; for (i = 0; i < 19; i++) n[i] = ifr->accept( Ifr. Vst. Hits. In. Layer: : instance(i+1) ); // compute max. number of missed layers int max. Miss = ifr->accept( Ifr. Vst. Max. Missed: : instance() ); Luca Lista - CHEP '98

References • On-line Ifr documentation: – http: //www 1. na. infn. it/wsubnucl/accel/Ba. Bar/Reco/6. 7. 3/ see Ifr. Data and Ifr. Visitor packages for class diagram and C++ code • Original proposal document: – http: //www 1. na. infn. it/wsubnucl/accel/Ba. Bar/Reco/visitors. html Luca Lista - CHEP '98