A structured approach to software process modelling

A S uc u ed App oach o So wa e P ocess Modelling1 Xa ie F anch anc [email p o ec ed]. es Uni e si a Poli ecnica de Ca alunya Jo di Gi ona 1-3,08034 Ba celona Ca alonia (Spain) FAX: 34-93-401 7014. Phone: 34-93-401 6965 Abs ac Sys ema ic o mula ion o so wa e p ocess models (SPM) is cu en ly a challenging p oblem in so wa e enginee ing. We p esen he e an app oach o de ine such models ha encou ages: euse o bo h elemen s and models; modula i y and inc emen ali y in model cons uc ion; simplici y and na u ali y o he esul ing model; and a high deg ee o concu ence in hei enac ion. In his pape we ocus on model de ini ion, dis inguishing as usual i s s a ic and dynamic pa s. We de ine he s a ic pa by means o o mally de ined hie a chies in oducing he ca ego ies o elemen s ha ake pa in SPA4 de ini ion. Such hie a chies may be cons uc ed and enla ged acco ding o he equi emen s o any speci ic SPh! We p esen as an example a hie a chy o componen p og amming ha akes in o accoun non- unc ional aspec s o so wa e (e iciency, e c). The dynamic pa l o he SPA4 is de ined by means o p ecedence ela ionships be ween asks ha ake pa in he model. These p ecedence ela ionships a e ep esen ed wi h p ecedence g aphs. De elopmen s a egies a e de ined by encapsula ing new p ecedence ela ionships in modules, ha can be combined and eused. 1. In oduc ion A model o a so wa e de elopmen p ocess [DWK97] (i.e., a so wa e p ocess model) is a desc ip ion o his p ocess exp essed in some p ocess modelling language. The p ocess can be iewed as he execu ion in a sui able o de o a se o asks (e.g., equi emen s elici a ion o module es ing) in ended o de elop some documen s (e.g., speci ica ion o es plan). These asks a e de eloped by some agen s (e.g., people o ha dwa e media) wi h he This wo k has been pa ially suppo ed by he Spanish p ojec TIC97-1158, om he CICYT p og am. Josep M. Rib6 j [email p o ec ed], es Uni e si a de Lleida P. Vic o Siu ana 1,25003 Lleida Ca alonia (Spain) FAX: 34-973-702062. PhLone: 34-973-702000 help o some ools (e.g., edi on o debugge s) and using some esou ces (e.g., da a bases o compu e ne wo ks). Hence, he de ini ion o a so wa e p ocess model mus s a e all he elemen s jus men ioned, and also he way in which his model mus be execu e (enac ed). This idea leads o he no ion o s a ic and dynamic pa s o a model. The s a ic pa is gi en by he desc ip ion o he asks, documen s, agen s, ools and esou ces ha ake pa in he so wa e p ocess model. On he o he hand, he dynamic pa consis s o a desc ip ion o he way in which so wa e is de eloped; so, i mainly ocuses in ques ions like wha and how mus be done o de elop a piece o he model. The sys ema ic desc ip ion o bo h pa s no only helps in unde s anding so wa e de elopmen , bu also makes easible he cons uc ion o sys ems o suppo ing au oma ion o he p ocess up o an accep able le el, cen e ed on he p ocess modelling language. Many di e en s app oaches o such sys ems cu en ly exis ; see [FKN94] o a sumey. Some o hem ha e d awn a special a en ion wi hin he scien i ic communi y, like EPOS [Con95], MERLIN [Jun95, RS971 o SPADE [BNF96], jus o name a ew o hem. Al hough hey suppo a lo o help ul p ope ies in so wa e de elopmen (e.g., p ocess e olu ion, e sioning, concu ency du ing enac ion and coope a ion among asks), hey seem o lack a leas pa ially in suppo ing he ollowing in e es ing ones: Modula i y in model cons uc ion, i.e., he abili y o build a model by combining se e al pa ial models using some ope a o s. A,l hough he e a e some p oposals in his sense ( ema kably [Ch 94]), mos o he epo ed en i onmen s seem no o suppo i . Modula i y a he p ocess model is as impo an as a he p oduc le el, aiding ai building, unde s anding, main aining and eusing so wa e models. 753 1089-6503198 $10.00 0 1998 IEEE Simplici y in bo h he p ocess o model cons uc ion and also he desc ip ion o he esul ing model. This p ope y is no easily achie ed in he sys ems we ha e s udied so a , as we can see in he case s udies appea ing in [FKN94] and also [ABEL97]. Simplici y is a basic p ope y in o de o make hese app oaches use ul o so wa e eams de eloping eal applica ions. Fo malisa ion o he elemen s aking pa in he so wa e p ocess models, and also o he no ion o co ec ness o a model enac ion. The exis ence o such o mal basis would p o ide a well-es ablished ounda ion o eason abou model enac ion. In his pape , we p esen a p ocess model language aimed a suppo ing hese p ope ies. The language is he ke nel o ou PROMENADE app oach (PROcess-o ien ed Modelliza ion and ENAc ion o so wa e DE elopmen s), cu en ly in p og ess. Conce ning he s a ic pa , we desc ibe p ocess elemen s by means o OOZE [AG91] classes, which p o ides a modula and o mal desc ip ion; simplici y is added by p o iding a g aphical no a ion o desc ibe class ela ionships. Abou he dynamic pa , we o mula e ou app oach by es ablishing p ecedence ela ionships be ween asks, and by de ining encapsula ion mechanisms ha enhance modula i y; also, we p o ide he no ion o co ec ness o a so wa e de elopmen wi h espec o a so wa e model. The p oposal elies on a p e ious pape [FR97], which in oduced he basis o he cu en dynamic pa , bu which lacked om a p oposal o he s a ic one. Al hough we can conside his classi ica ion enough o s a so wa e p ocess models de ini ion, we plan o p o ide many de aul hie a chies o di e en de elopmen con ex s. In his pape , we will ake as case s udy a hie a chy o dealing wi h componen p og amming (CP-hie a chy o sho ). New elemen s de ined in e ms o classes can be added a a ce ain place o he hie a chy. The de ini ion o a new elemen in ol es he de ini ion o he a ibu es ha e e y ins ance o he elemen mus possess, he ope a ions ha may be applied o an ins ance o he elemen and he equi emen s ha mus hold in all he ins ances o he elemen (in a ian o an elemen ). Using he hie a chy helps in achie ing eusabili y du ing so wa e p ocess modelling: any pa o he hie a chy is always a ailable in he de ini ion o new models. As men ioned abo e, and since one o ou goals is o p o ide a o mal amewo k o so wa e p ocess de ini ion, i becomes essen ial o speci y o mally he elemen s ha ake pa in model de ini ion. We use he OOZE [AG91] o malism o do so. OOZE combines he widesp ead Z no a ion wi h some s uc u ing mechanisms and, in pa icula , inhe i ance. So, he hie a chy d awn abo e can be in e ed om OOZE classes. We plan o use he hie a chy as an ex e nal language o so wa e p ocess enginee s and o documen a ion pu poses oo. Models desc ibed wi h OOZE ha e a well-de ined o mal meaning, which helps in de e mining he seman ics o so wa e p ocesses. As an example, we a e going o de elop in mo e de ail he pa o he hie a chy conce ning documen s and asks. 2. The S a ic Pa 2.1. Documen s The s a ic pa o he language is de ined upon a hie a chy in eg a ing all he i ems in ol ed in so wa e de elopmen : Documen s, Tasks, Tools, Agen s and Resou ces; also, we include a Domain class de ining some auxilia y concep s (s ing, da e, e c., and also many domain-speci ic ones). These i ems a e encapsula ed and de ined o mally h ough classes o ganised as a hie a chy. These classes ac as classi ica ion c i e ia o o he elemen s, in oducing some a ibu es and ope a ions ha a e inhe i ed by hei hei s. As shown in ig. 1, hey a e in u n hei s o he Type class, which is he oo o he hie a chy. We use a plain a ow o ep esen inhe i ance. Documen Agen Task Tool Resou ce Domain Fig. 1 : The uppe le els o he s a ic hie a chy. The ca ego y o documen s, known as such o being hei s o he Documen class, de ines he a i ac s p oduced du ing so wa e de elopmen . Fo ins ance, some signi ican documen s a e (see ig. 2): SpecDoc, compounded o FspecDoc, ha con ains he unc ional speci ica ion o a componen and NFSpecDoc, o i s non- unc ional speci ica ion; ImplBehDoc, compounded o ImplDoc and BehDoc, o unc ional and non- unc ional pa s o implemen a ions, espec i ely; Tes Doc, o model he es s ha a e o be pe o med on some documen (and ha a e compounded o he es code, Tes plan, and he es esul s, kep in A alDoc). The Componen i sel would be ano he example o a composi e documen since i is a p oduc o he p ocess o so wa e de elopmen ha con ains o he kind o documen s as a ibu es. This ca ego y also includes o he a i ac s as Wo kspace, ha keeps he en i onmen o an agen a a gi en ins an , SpecLib (a lib a y o s o ing speci ica ions) and 754 Documen Wo kspace Lib a y Componen Tes Doc Tes Plan A alDoc SpecDoc ImplBehDoc A Fig. 2: The pa o he CP-hie a chy conce ning documen s (only inhe i ance elai ionship is depic ed). ImplLib (ano he one o s o ing implemen a ion documen s). Apa om he inhe i ance ela ionships depic ed in ig. 2, o he kind o ela ionships ( o ins ance, pa -o and consis s-oA which1 a e one in e se o he o he ) apply o documen classes. We will p esen hese ela ionships a he ime hey a e needed. 2.1.1 The Documen class Documen is de ined wi h he ollowing a ibu es: he documen iden i ie , lhe da es o c ea ion and las upda e on ha documen (c ea ion, upda e), he documen s a us (which may be no Comple e, comple e and checked) and, inally, he documen 's owne . Documen is he supe class o all documen ypes and i is speci ied in ig. 3. Class Documen < Type iden i ie : S ing c ea ion, upda e: Da e s a us: S a Doc owne : Agen id?: S ing da ?: Da e own?: Agen ii--------- iden i ie '=id? upda e'=da ? A c ea ion'=da ? s a us'=no Comple e owne '=own? This speci ica ion con ains; some classes (like S ing and Da e) ha a e de ined in he Domain subhie a chy. He ea e , ope a ions o con olled a ibu e modi ica ion and selec ion a e no shown. 2.1.2 The Componen class The Documen subclass Com,ponen is de ined wi h he ollowing a ibu es: he speci ica ion documen (spdoc), ha con ains he unc ional and non- unc ional speci ica ion o ha componen ; and he implemen a ion and beha iou documen (ibdoc), which con ains bo h he componen implemen a ion and he non- unc ional beha iou o ha implemen a ion. No ice ha he a ibu es iden ijie , c ea ion, upda e, s a us and owne a e inhe i ed om he class Documen . The speci ica ion o Componen class in OOZE is gi en in ig. 4. One ema kable aspec o his speci ica ion is he class in a ian which s a es ha he s a us o a componen is checked i and only i he s a us o all he documen s i is coimpounded o a e checked; his p ope y will be usual in compounded documen s. Also, we s a e ha i bo h documen s a e inished, he las upda e o he implemen a ion mus be g ea e o equal han speci ica ion's one. The de ini ion o Componen as a union o a ious documen s b ings up he opic o he exis ence o o he kind o ela ionships be ween classes. A new kind o ela ionship be ween classes may be in oduced in his case: he consis s-o ela ionship. We say ha a class A consis s o classes Cl, ..., Cn i and only i A can be de ined as he Ca esian p oduc o CI, ..., Cn, i.e., A = (CI x C2 x ... x Cn). We iden i y a consis s-o ela ionship be ween he classes SpecDoc, ImplBehDoc and he class Componen (a Componen consis s-o a SpecDoc and a ImplBehDoc). Fig. 3: The class Documen 755 - Class Componen < Documen -- S a e spdoc: SpecDoc ibdoc: ImplBehDoc S a us=checked a (spdoc.s a us=checked A ibdoc.s a us=checked) S a us=checked 3 spdoc.upda e I ibdoc.upda e ( he isola e upda e e e ence is applied o he cu en class, FSpecDoc in his case). The class Tes Doc and i s hei s a e p esen ed nex . -- S a e spec: Speci ica ion limpo : seq S ing es docs: seq Tes DocFS S a us=checked e (V d: Tes DocFS I ddes docs d.a ald.success= ue A d.a ald.upda e < upda e) ~ Fig. 4: The class Componen 2.1.3 The classes o speci ica ions SpecDoc consis s-o he unc ional (FSpecDoc) and non- unc ional speci ica ion (NFSpecDoc) o a componen . I s speci ica ion is simila o he one o Componen and i is no shown. The FSpecDoc documen is de ined wi h he ollowing a ibu es: spec ( he unc ional speci ica ion o a componen exp essed in some o malism); limpo (a lis con aining he unc ional speci ica ion documen s ha mus be impo ed in o de o comple e his one); and es docs (which con ains a lis wi h all he es cases o he e i ica ion o he unc ional speci ica ion documen along wi h he esul s o each es ). We a e no choosing he e a pa icula o malism o he speci ica ion; di e en componen s may be speci ied in a di e en way. Speci ica ion s yles may come in o exis ence jus de ining hei cha ac e isa ion by means o new OOZE classes decla ed as hei o he speci ica ion one. The speci ica ion o FSpecDoc documen in OOZE is gi en in ig. 5. The class in a ian es ablishes ha FSpecDoc will be conside ed o be checked only a e all es s planed o be un on i ha e inished success ully and hey ha e been execu ed wi h he cu en e sion o he speci ica ion, which is checked using he las upda e da e The class NFSpecDoc is de ined in a simila way. Also, implemen a ions wo k he same way as speci ica ions do and a e no shown he e. 2.1.4 The Tes Doc class This is he class ha pe o ms some kind o es (including bo h he es code and he es esul s) on he di e en documen s (namely FSpecDoc, NFSpecDoc, ImplDoc and BehDoc). We conside a di e en kind o Tes Doc class o each class o documen (i.e. Tes DocFS o es ing FSpecDoc classes; Tes DocNFS o es ing NFSpecDoc classes ...). Hence we need o enla ge he ype hie a chy o ig. 2 by making hese new classes hei s o Tes Doc. Le us p esen , as example, he class Tes DocFS. The a ibu es o his class a e he ollowing: es eddoc ( he documen which is being es ed), es pl ( he es code) and a uld ( he documen con aining he esul o such es ). Fig. 6 con ains a speci ica ion o his class wi h he usual class in a ian in ol ing s a us and da es. Tes classes in oduce ano he kind o ela ionship: is- es ed-in. We say, o ins ance, ha a FSpecDoc is- es ed-in a Tes DocFS. Unlike he ones p esen ed up o now, his is a use -de ined ela ionship, local jus o a pa o he hie a chy. These new ela ionships may be la e used in he dynamic pa o he model. On he o he hand, he consis s-o ela ionship may also be applied he e since each kind o Tes Doc class consis s-o a Tes Plan and an A ulDoc. 756 - Class Tes DocFS < Tes Doc - S a e - es eddoc: FSpecDoc es pl: Tes F'lanFS a ald: A alDocFS S a us=check.ed e ( es pl.s a us=checked A a ald.s a us=checked) S a us=checked = es phpda e I a ald.upda e ..ini and o he ope a ions Fig. 6: The class Tes DocFS 2.1.5 The classes ai lib a ies The class Lib a y is mean o s o e de eloped documen s. Hence i will only con ain documen s wi h a checked s a us. In he de aul CP-hie a chy, we conside jus wo kinds o lib a ies (al hough i can be wo hy o de ine o he ones): SpecLib, o s o e speci ica ion documen s and ImplLib, o s o e implemen a ion documen s; no e ha es s a e pa o hese documen s. Ano he kind o ela ionship be ween classes ises wi h lib a ies: is-s o ed-in. Fo ins ance, a SpecDoc is-s o ed- in a SpecLib, while a ImplDoc is-s o ed-in an ImplLib. We s o e in he co esponding lib a y he SpecDoc as a whole (i is no allowed o s o e only he FSpecDoc o he NFSpecDoc o a gi en componen ). Fig. 7 shows he speci ica ion o SpecLib a y class. The wo class in a ian s s a e, espec i ely, ha all he documen s con ained in he lib a y a e in a checked s a us and ha a lib a y is sel -con ained (i.e. all he documen s impo ed by a documen s o ed in he lib a y mus be also s o ed in he lib a y). No e also ha he in a ian allows he s o ed e sion o he speci ica ion no o be he las one. 2.1.6 The Wo kspuce class The Wo kspace class ep esen s a documen eposi o y compounded o hose documen s ha a e isible o an agen a a gi en ins an . This includes some documen s aken om some lib a y and some o he documen s which a e being cons imc ed. The ela ionship be ween classes is-s o ed-in may also apply he e. Bu in his case i is no compulso y o s o e in he wo kspace he pai o speci ica ion (o implemen a ion) documen s, because documen s may be incomple e in he wo kspace. The OOZE speci ica ion is s aigh o wa d and WI: do no include i he e. Class SpecLib < Lib a y S a e SI: seq SpecDoc - Vd: SpecDoc I d E SI 0 d.s a us=checked Vd: SpecDoc I d E sl 0 (Vd': SpecDoc 1 cl ' E d.limpo d' E SI) .. ini and o he ope a ions; Fig. 7: The class SpecLib a y 2.2. Tasks A ask ep esen s an ac ion ha mus be pe o med in he p ocess o so wa e de elopmen . I may be a composi e ac ion, which, in u n, will be decomposed in mo e simple asks called sub asks, o an a omic one. Since he so wa e p ocess model we p opose is mos ly ask-o ien ed, ask elemen s ha e a majo impo ance in i . We de ine asks as classes in he ype- hie a chy. Tasks a e speci ied by means o OOZE classes which a ibu es ep esen he pa ame e s o he ask. An addi ional pa ame e keeps ack, a enac ion ime, o he ask's sub asks ha ha e been execu ed. The class is also p o ided wi h wo me hods, named espec i ely begin and end ha a e called a he s a ing and end ins an s o he ask execu ion. Bo h me hods pe o m e e y hing needed o keep he consis ency o he ask (e.g. begin pu s he ask s a us o ache, ini alizes some ask pa ame e s, e c.; end calcula es he success condi ion o he ask, pu s he ask s a us o comple e, e c.). The issue o how o ge he unc ionali y o he ask (i.e. in which way we desc ibe he ac ions o be unde aken in o de o ge he ask goals) is he main ma e o he dynamic pa o he model, de eloped in he nex sec ion. We p esen in ig. 8 and 9 wo exemples o ask speci ica ion: he class Task which ac s as supe class o all asks, and Tes FSpec, which pe o ms he es o a unc ional speci ica ion docu nen ) wi h espec o some es plan. The pa ame e s o his las ask a e he FSpecDoc (see ig. 5) we wan o es and he Tes Doc (in his case Tes DocFs, see ig. 6) used o pe o m his es . A he beginning, he link be ween he speci ica ion documen and he es s is c ea ed. A he end, he success condi ion o he ask, success I, e alua es o ue i all he single es s which is compounded o ha e been execu ed 757 success ully. We call SingleTes FSpec he class o asks ha pe o ms a single es on a unc ional speci ica ion documen ; we conside ha each o hese asks includes a es plan, a es esul and success condi ion. Finally we s a e ha he es esul s s o ed in he e alua ion documen a e exac ly hose p oduced by he applica ion o SingleTes FSpec asks on he ac ual ins ance o FSpecDoc. sb?: P Task s a us’=idle sub asks ’ =sb? - Class Task < Type s a us’= comple e -- S a e s a us: S a Task success: Boo1 sub asks: P Task execu ed: seq Task an execu ed c sub asks s a us = comple e an execu ed = sub asks Fig. 8: The class Task 3. The Dynamic Pa The dynamic pa o he model s a es (1) wha mus be done du ing model enac ion (i.e. wha asks a e o be execu ed), and (2) wha cons ain s a e o be applied in such enac ion (i.e. wha p ecedences in ask execu ionmus be sa is ied). We ely on ask decomposi ion in o de o s a e wha a ask mus do (i.e. wha sub asks a e in ol ed in i s execu ion) and we de ine p ecedence ela ionships be ween asks in o de o es ablish p ecedence cons ain s in ol ing ask enac ion - Class Tes FSpec < Task - S a e - specld: FSpecDoc : es : Tes DocFs -- begin specdoc?: FSpecDoc es p?: Tes PlanFS specld’ = specdoc? es ’. es pl= es p? es ’. es eddoc= specdoc? specld’.s a us=comple e __ end - success’= (Vp: Code I p E es . es p1.specbody (3 pa : SingleTes FSpec pa E execu ed A pa .plan=p A pa .success= ue)) V : Resul E es .a ald’.l esul s e (3 pa : SingleTes FSpec I pa E execu ed ( pa . esul = A pa .success= ue)) specld’.s a us=checked e success’= ue Fig. 9: The class Tes FSpec 3.1. P ecedence ela ionships P ecedence ela ionships be ween asks s a e he equi emen s ha mus be sa is ied in o de o be able o s a he execu ion o a ask. These equi emen s a e es ablished in e ms o he asks whose execu ion mus ha e inished success ully in o de o s a he execu ion o a gi en ask. Mo e p ecisely, we say ha he e is a p ecedence ela ionship om ask A o ask B (A 3 B) i a equi emen needed in o de o ini ia e ask B is ha ask A has been comple ed success ully (i.e. wi h success condi ion e alua ing o ue). 758 / Tes Fspec(sp. spec) Tes NFspec(sp.n spec) 1 / S o e(speclib, sp) IFig. 10: A possible p ecedence g aph o de eloping speci ica ions. We can ep esen p ecedence ela ionships be ween asks by means o p ecedence g aphs, being hei nodes asks, and hei edges p ecedences. Fig. 10 p esen s a p ecedence g aph o de eloping a speci ica ion wi h unc ional and non- unc ional pa s. We use asks o de ining he ope a ions o he componen , o c ea e bo h pa s, o design es s o hem, o ca ying he es s ou and o s o ing he wo pa s in he speci ica ion lib a y as a whole. No e ha asks appea pa ame e ised, using he a ibu es in oduced in he in ol ed classes. This example g aph is a de aul one in PROMENADE, and i could be in e ed om some ela ionships s a ed a he s a ic le el, mainly ha ing o do wi h da es and success condi ions. I is impo an o no ice ha by desc ibing asks using p ecedence ela ionships we s a e all he in e ac ions ha mus be obse ed be ween asks du ing model enac ion. Apa om hose in e ac ions, he p ocess engine is ee o selec any o he execu ion o de among no ela ed asks. This imp o es he concu ency o model enac ion. 3.2. De elopmen s a egies Gi en he modelisa ion o p ecedence ela ionships using g aphs, we can conside a de elopmen s a egy as a se o new edges binding nodes o hese g aphs. Some imes, edges will ela e asks (nodes) in he same g aph, o say hings like “ he unc ional speci ica ion o a componen mus be de eloped be o e he non- unc ional one”; howe e , in he gene al case, edges will in ol e asks appea ing in g aphs bound o di e en modules, as in “i is necessa y o speci y all he componen s impo ed by a componen M be o e any implemen a ion o A4 is buil ”. Se s o ela ed ules a e encapsula ed in s a egy modules. Fo ins ance, we show in ig. 11 h ee di e en s a egy modules ha add edges o he g aph p esen ed in ig. IO. The i s one o ces he inaliza ion o he unc ional speci ica ion be o e s a ing he non- unc ional one. The second oine implemen s he idea o bo om-up speci ica ion, saying ha impo ed speci ica ions mus be inished be o e s a ing new ones. Las , a new s a egy can be s a ed jus by combining he p e ious ones. s a egy FUNCTIONAL-BEFORE-NON-FUNCTIONAL sp: SpecDoc Fspeci y(sp. spec) -> N speci y(sp.n spec) end module s a egy BOTTOM-UP-SPECIFICATION sp, Z: SpecDoc o all Z in sp. spec.limpo : Fspeci y(Z. spec) -> Fspeci y(sp. spec), NFspeci y(Z.n spec) -> NFspeci y(sp.n spec) NFspeci y(sp.n spec) end module combines BOTTOM-UP-SPECIFICATION, FUNCTIONAL~BEFORE~I’JO“CTI0NAL end module Fig. 11 : Some s a egies o speci ica ion de elopmen . 3.3 Modula p ocess cons uc ion In PROMENADE, model de ini ion is in ended o allow so wa e p ocess model conslmc ion in a modula and inc emen al way. One pa o his modula cons uc ion 759 o models elies on he eusabili y and ex ensibili y o he hie a chy con aining he s a ic elemen s o he model. The o he , and mo e undamen al pa , deals wi h he p ocess o ask desc ip ion. Tasks play he ole o uling p ocess de elopmen by appliying s a egies. Cons uc ing a so wa e p ocess model in a modula way consis s in selec ing wi hin a s a egy lib a y hose ones wi h he equi ed unc ionali y and combining hem wi h some sui able p ecedence ela ionships in o de o build a desc ip ion g aph o he model. This will de ine he model s a egy. Following his p ocess, i is possible o combine many pa ial models o o m he inal one. These combina ion may be o wo kinds. On he one hand, we can build p ecedence g aphs o a subse o so wa e documen s (speci ica ions, lib a ies, wo king con ex , e c.) by adding new p ecedences o an ini ial g aph, by joining wo g aphs wi h he same nodes, e c. On he o he hand, we can jus pu oge he some o hese g aphs o ob ain a new one co e ing mo e documen s (i.e., dealing wi h mo e so wa e de elopmen s ages); o ins ance, we can pu oge he he g aph o ig. 10 wi h o he conce ning implemen a ion cons uc ion, o ob ain a g aph co e ing he whole componen de elopmen p ocess. The esul ing g aphs, i con enien , may be in u n, s o ed in a lib a y o u u e euse. 3.4. Co ec ness conce ns Using he p ecedence g aphs and also he success condi ions s a ed in he s a ic pa o he model, i is possible o o mula e some co ec ness condi ions, bo h conce ning he model i sel and also conce ning a pa icula de elopmen p ocess wi h espec o a model. This issue has been ou lined in [FR97] and has been e ined by inco po a ing he idea o h ee dimensional g aphs, in which an axis co esponds o p ecedence ela ionships and he o he o ask decomposi ion. Ano he poin conce ning he co ec ness is he ma ching be ween he s a ic and he dynamic pa s o he model. Ob iously, p ecedence ela ionships be ween asks and decomposi ion o elemen s in o smalle pa s mus ag ee. We a e cu en ly wo king on he cha ac e iza ion o his ma ching. hink a e no cu en ly o ally co e ed in he ield. The PROMENADE app oach plays a pa in a mo e ambi ious sys em called ComE' oLab [FBBR97] de ined o suppo many di e en aspec s o componen p og amming, In ou p oposal, he language consis s o s a ic and dynamic pa s. Conce ning he s a ic pa , we use a hie a chy o in oduce so wa e p ocess elemen s (documen s, asks, e c.), which a e encapsula ed using OOZE classes and hus p o ided o a clea seman ics. Wi h espec o he dynamic pa , we use p ecedence g aphs as he unde lying model o ask enac ion o de ing, and we allow he de ini ion o de elopmen s a egies using modules ha can be eused and combined. We can classi y and e alua e he adequacy o ou app oach wi h espec o he aspec s p oposed by Con adi and o he s in [CLJ91]. We pu a s a (*) on hose issues s ill no co e ed bu jus planned o: Basic p ocess model appa a us: P ecedence ela ionships be ween asks modelled by p ecedence g aphs. S a ic pa co e ed by class hie a chies. Co e age o p ocess en i ies: p oduc s, ac i i ies, ools, agen s, oles and esou ces. Ou app oach is, howe e , ac i i y-o ien ed. * Co e age o so wa e p ocess li e-cycle: All s eps can be modelled in ou app oach, including speci ica ion o so wa e a ibu es (non- unc ional speci ica ion). Task s uc u ing: We p esen a ully ask s uc u ing by means o ask abs ac ion. Type s uc u ing: Yes, in an objec -o ien ed way. We de ine a de aul hie a chy o ypes ha may be ex ended on demand. (*) Cus omiza ion and e olu ion o he p ocess model: I will be allowed by using me a- ypes. *Concu ence: Suppo ed by he ac ha jus p ecedence ela ionships a oid concu en enac ion o asks. (*) Con igu a ion managemen : A usual e sioning sys em will be p o ided. We hink ha he mos in e es ing poin s o ou app oach a e: Moaulu i y/ eusabiZi yli y: PROMENADE o e s he possibili y o eusing agmen s o exis ing models unde an objec -o ien ed app oach. Hence, in o de o build a new model i is possible o euse asks, documen s, oles and any o he elemen p e iously gene a ed o cons uc some o he model, which makes he p ocess o model cons uc ion much mo e simple. PROMENADE sha es his ea u e wi h some o he sys ems like EPOS and E3. 4. Conclusions and u u e wo k We ha e p esen ed he p ocess modelling language o he PROMENADE app oach. This language add esses o many in e es ing p ope ies a he p ocess le el which we 760 On he o he hand, some well-known sys ems like SPADE, ADELE o IMERLIN lack his p ope y o , a leas , i is no shown explici ly how o ge i in he e ised li e a u e. Wha is new in PROMENADE wi h espec o he e ised sys ems is i s explici abili y o cons uc ing in a modula manne new models adding some s a egies o exis ing ones. These s a egies a e encapsula ed in wha we call s a egy modules, which a e p esen ed b ie ly in sec ion 3.2 and in mo e de ail in [FR97]. Simplici y/comp, ehensibili y; PROMENADE seems o acili a e he gene a ion o so wa e p ocess models (SPMs) in an in ui i e and qui e simple way by means o a g aphical ep esen a ion ( ha will be ansla ed in o a o mally de ined language, which is cu en ly being de ined). This g aphical ep esen a ion is based on de ining hie a chies o en i ies ( o he s a ic pa ) and new asks by means o s a ing he p ecedence ela ionships ha mus hold be ween some o he asks ( o he dynamic one). On he o he hand, once he SPM has been cons uc ed, p ecedence g aphs make i qui e comp ehensible. P ecedence g aphs (i.e. diag ams which depic s he p ecedence ela ionships be ween ac i i ies) a e c ucial in o de o unde s and he whole p ocess o so wa e de elopmen . Cu iously enough, mos o he P ocess-cen e ed So ih a e Enginee ing En i onmen s (PSEEs) we ha e explo ed do no use his kind o diag ams. Two aspec s ha also help in he achie emen o simplici y a e he objec -o ien ed app oach we unde ake and he high le el cons uc s we p o ide o ou sys em. Unlike o he PSEEs, like SPADE o APPLiA, we do no equi e he so wa e enginee o explain how he SPM will be enac ed (which is usually a c ip ic ma e ) bu ins ead, wha mus be done in o de o de elop so wa e, which is clea ly wha he so wa e enginee knows (and wha heishe is in e es ed in modelling). Al hough he simplici y p ope y should be a e y impo an one o PSEEs, mos o hem ail (a leas o some ex en ) in achie ing i 2. SPADE and APPLiA a e di ec ly enac able PSEEs ha ge a ema kable pe o mance in model enac ion. Since hey achie e ha pe o mance on he basis o a low le el cons uc ion, we hink ha SPM w i en in hese sys ems (specially in SPADE) a e di icul o w i e, di icul o ead and di icul o unde s and. MERLIN and EPOS a e high le el PSEEs, bu in ou opinion, i is di icul o ge wi h hem ully comp ehensible models: he esul ing model Al hough in some cases i is possible ha he complexi y o he model is a consequence o he inhe en complexi y o he p oblems ha a e o be sol ed. in MERLIN is wo king-con ex -o ien ed which makes i di icul o g asp he whole de elopmen p ocess model. In EPOS he ask sequence o be execu ed in he esul ing SPM is no ob ious amd he ad an age o using an AI planne is no clea . Finally, E3 p o ides simplici y on he basis o a wide ange o diag ams (including p ecedence ones), an objec -o ien ed app oach and high le el cons uc s. This leads lo a e y comp ehensible model bu wi h qui e simplis ic cons uc s. Di ec enac abili y: The e is a usual co ela ion be ween he le el o he cons uc s o e ed by he sys em in o de o gene a e a SPM and he enac abili y o he esul ing model. Fo ins ance, di ec ly enac able sys ems like APPLIA and SPADE o e qui e low le el cons uc s. On he o he haind, sys ems like E3 and MERLIN, by a mo e high-le el sys ems, canno be enac ed di ec ly; hey need a ansla ion in o a lowe le el language. PROMENADE achie es a lbalance be ween he le el o he language o c ea e he model (which is clea ly high le el) and he SPM enac abilii y, since a PROMENADE SPM is di ec ly enac able by he applica ion o he algo i hm desc ibed in [FR97]. Concu en model enac ion: No only o e s PROMENADE an enac able SPM bu also a e y na u al concu en model based on a imul iagen app oach (each agen is esponsible o execu ing some asks) wi h he es ic ions imposed by he p ecedence ela ionships be ween asks. We a e cun en ly wo king on hese enac ion aspec s. An o e iew o hem may be ound in [FR97]. Concu en models o e ed by some o he PSEEs need synch oniza ion be ween asks ( he case o APPLIA since i uses Ada asks). In o he cases (as SPADE) he concu en model is based on ,a o mal mechanism (Pe i ne s). This leads o a e y e icien bu low le el app oach. In some o he cases o no di ec ly enac able models (MERLIN, E3), he concu ence o he inal enac able model is no epo ed. I is impo an o say ha , unlike o he sys ems like ADELE o MERLIN, we do no deal, o he momen , wi h concu en accesses o documen s, which is a e y impo an esea ch a ea. Fo mali y: This is one o he sho comings o mos PSEEs, o only a ew o hem a e cons uc ed on he basis o a o mal app oach. One o he goals o PROMENADE is o de ine a SPM es ablished on some o mal basis. This will allow he comple e unde s anding o he speci ica ion o he elemen s (classes and ela ions) ha ake pa in model de ini ion; he igo ous de ini ion o he mechanism o model enac ion; he es ablishmen o he co ec ness o a 761