Merging Models with the Epsilon Merging Language - A Decade Later

Merging Models with theEpsilon Merging Language

– A Decade Later

Dimitris Kolovos, Richard Paige, and Fiona Polack

Department of Computer ScienceUniversity of York

[email protected]

D. Kolovos et. al. – ACM/IEEE MoDELS 2016 October 6, 2016 – Slide 1/23

Ambition (in 2004)

Develop a hybrid textual language for model mergingIn the same way that ATL targeted M2M, OCL targeted modelvalidation etc.

Desirable featuresExpressivenessMetamodel and modelling-technology independenceSupport for merging heterogeneous models


First Steps (2004–2005)

Started with a version of the OCL grammar (ANTLR) andextended it with merging-specific syntaxWrote an interpreter for the ASTIntroduced an abstraction layer for modelling technologies

Implemented the abstraction layer for the NetBeans MetaDataRepository (MDR) – MOF 1.4

Implemented the first version of EML’s Eclipse-baseddevelopment tools


Heterogeneous Model Merging?






Merging Process

Focused on 2-way mergingSteps

1 Establish correspondences between elements in the two models2 Merge matching elements (optional)3 Transform non-matching elements (optional)


Matching Elements

NotesWe establish correspondences based on national insurancenumber matchingList!Person means: type “Person” from model “List”

Inspired by ATL

rule MatchPersonWithTaxPayermatch p : List!Personwith tp : HMRC!TaxPayer {

compare: (p.country = "UK") and (p.taxId = tp.nin)

}


Merging Elements

NotesWe merge matched Persons with their correspondingTaxPayers and produce (notification) Letters as outputWe don’t want to bother those whose incomes match

rule DiscrepancyLettermerge p : List!Personwith tp : HMRC!TaxPayerinto l : Outbox!Letter {

guard: tp.declaredIncome <> p.income

l.address = tp.address;l.text = "Dear " + p.name + ", ";if (p.income > tp.declaredIncome) {l.text += "you are doomed.";

}else { l.text += "we love you."; }

}


Transforming Unmatched Elements

NotesUnmatched Person elements from the UK means non-existingnational insurance numbers (i.e. fraud!)

rule FraudLettertransform p : List!Personinto l : Outbox!Letter {

guard: p.country == "UK"

l.address = p.address;l.text = "You are doomed beyond imagination.";

}


Modularisation

An EML program consists ofMatching rulesMerging rulesTransformation rules

These are three different languages!. . . which share a common OCL-based imperative expressionlanguage, model connectivity facilities etc.


Modularisation

An EML program consists ofMatching rulesMerging rulesTransformation rules

These are three four different languages!. . . which share model connectivity facilities etc.


Modularisation (2006)


Modularisation (2006)


Eclipse Modeling Project

Epsilon joined the Eclipse Modeling GMT incubator in 2006Graduated (v1.0) in 2012

Currently in v1.3 (v1.4 approved yesterday)Substantial boost to adoption

≈ 8,000 forum posts since 2008In open-source projects and in industryTo teach modelling and MDE courseshttp://eclipse.org/epsilon/users


http://eclipse.org/epsilon/users

Epsilon Validation Language (EVL)

Key capabilitiesHuman-readable messagesConstraint dependenciesQuick fixes

context List!Person {

constraint ValidNin {

guard : self.country = "UK"

check : HMRC!TaxPayer.all.exists(tp|self.taxId = tp.nin)

message : "Unknown National Insurance Number: " +self.taxId

}}


Epsilon Validation Language (EVL)

context List!Person {

constraint MatchingIncome {

// Won’t be evaluated if ValidNin is not satisfiedguard : self.satisfies("ValidNin")

check {var taxPayer = HMRC!TaxPayer.all.selectOne(tp|self.taxId = tp.nin);

return taxPayer.declaredIncome = self.income;}

message : "Mismatching income: " + self.income+ "<>" + taxPayer.declaredIncome

}

}


Epsilon Generation Language (EGL)

Key capabilitiesTemplate-based M2T languageSeparate sub-language for rule-based template coordinationSupport for hand-written code in generated files

rule EClass2DBClasstransform c : EClass {

guard : c.isAnnotatedAs("db")

template : "eclass2dbclass.egl"

target : c.name + ".java"}


Epsilon Generation Language (EGL)

[%for (f in c.eStructuralFeatures.select(f|f.isMany)) { %]protected List<[%=f.eType.getJavaBoxedName()%]>[%=sf.name%] = null;

[%}%][%for (r in c.eReferences.select(r|not r.isMany)) { %]protected [%=r.eType.getJavaName()%] [%=r.name%] = null;[%}%]

[%if (c.isAnnotatedAs("customize")){%][% out.setContentType("Java"); %]// protected region custom-fields-and-methods on begin// protected region custom-fields-and-methods end[%}%]


Epsilon Pattern Language (EPL)

Key capabilitiesTextual language for model pattern matchingIn-place transformation capabilitiesMatches exposed as a “virtual” model

patterns → types, matches → elements, roles → properties

pattern PublicFieldclass : ClassDeclaration,

field : FieldDeclaration from: class.bodyDeclarations,

setter : MethodDeclaration from: class.bodyDeclarationsguard: setter.name = "set" + field.getName(),

getter : MethodDeclaration from: class.bodyDeclarationsguard : (getter.name = "get" + field.getName()

or getter.name = "is" + field.getName()) { }


Epsilon Pattern Language (EPL)

context Patterns!PublicField {

constraint GetterAndFieldSameType {check : self.getter.returnType.type =

self.field.type.type

message : "The getter of " + self.class.name +"." + self.field.fragments.at(0).name +" does not have the same type as" +" the field itself"

}

}


Epsilon (2004 - 2016)


Epsilon (2004 - 2016)


Epsilon (2004 - 2016)


Reflection

EML was the first language tailored for model mergingThe 2006 MoDELS paper made the case for non-trivialmerging of heterogeneous modelsEML was the seed from which Epsilon eventually grew

EML is too verbose for merging homogeneous models. . . although homogeneous merging strategies can be triviallygenerated using HOTs

EML turned out to be one of the least popular languages inEpsilon

Little practical need for complex model merging?


Looking Ahead

IncrementalityPrototype for EGL → Robust implementation across Epsilon

User experienceCode completion, refactoring etc.

IntegrationEnterprise Architect, MagicDraw etc.

Optimisatione.g. technology-specific implementations of select() to avoidnaive iteration and use indexes instead
