Virtual Extension of Meta-models with Facet Tools

Jonathan Pepin

1,2

, Pascal André

, Christian Attiogbé

and Erwan Breton

AeLoS Team LS2N CNRS UMR 6004, University of Nantes, France

Mia-Software - Nantes, 44324 Nantes cedex 3, France

Keywords:

Meta-model, Weaving, Mapping, EMF, Facet, Transformation Tools.

Abstract:

MDE emphasizes the use of models and meta-models to improve the software productivity and some aspects

of the software quality such as maintainability or interoperability. In the software industry, Model Driven

Engineering (MDE) techniques have proven useful not only for developing new software applications but for

re-engineering legacy systems. However the stakeholders have to face costly maintenance operations due to

frequent new standards and upgraded releases of software modules they depend on. Therefore, due to the

limitations of existing techniques, solutions ensuring a better adaptability and ﬂexibility of model evolution

tools are needed. We propose an improved technique of virtual extension of meta-models with Facets that

enables one to modify meta-models already in use without rebuilding completely the software product. This

technique has been implemented and experimented for model alignment and evolution.

1 INTRODUCTION

The importance of modelling in software engineer-

ing and the popularity of the UML notation result

in the dissemination of model-based tools and the

emergence of a modelling language industry based

on the model-driven approach (MDA). The Meta-

Object Facility (MOF) language is then the foun-

dation for an ecosystem of Domain Speciﬁc Lan-

guages (DSL) (Brambilla et al., 2012). Some of them

are standards like BPMN, BMM, SoaML, SysML,

UPDM while others are speciﬁc. Model Driven En-

gineering (MDE) emphasizes the use of models and

meta-models to improve the software productivity

and some aspects of the software quality such as

maintainability or interoperability. MDE techniques

have proven useful not only for developing new

software applications but for re-engineering legacy

systems and dynamically conﬁguring running sys-

tems (Cuadrado et al., 2014).

In the context of software maintenance, includ-

ing model-driven reverse engineering of a legacy

system, we face challenges related to model evo-

lution, model composition, multi-generation models

and multi-layered models.

• Meta-models are essential in MDE since many

model processing are described at the meta-model

level through transformation rules or operations.

However this is also a pitfall since both models

and meta-models evolve separately. In particu-

lar, the modelling languages depend on standards

that evolve continuously and the related models

must also evolve: there is a dependency chain to

maintain. Model evolution is even more complex

when (meta-)models are parts of larger models.

• As mentioned by El Kouhen in (El Kouhen,

2016), MDE promotes the separation of concerns

to deal with the complexity and maintainability of

software design. This current practice implies the

creation of several heterogeneous models using

different notations (and therefore meta-models).

Since the semantics of each individual model is

limited, the consistency and completeness of the

individual models are easier to prove but the difﬁ-

culty is delegated to the composition of these in-

dividual models. Model composition is the sym-

metric paradigm of separation of concerns (Atlee

et al., 2007). It is then necessary to compose mod-

els to reason on the overall designed system for

many purposes such as: checking the global con-

sistency of the models, understanding the inter-

actions between models, generating code, etc. In

particular, we are interested in compositions that

do not interfere with the individual models.

• Multi-generation occurs when different releases

of one model (or meta-model) are maintained e.g.

for different customers or different branches of a

company. This may occur also in software prod-

uct lines or software customization.

Pepin, J., André, P., Attiogbé, C. and Breton, E.

Virtual Extension of Meta-models with Facet Tools.

DOI: 10.5220/0006547100590070

In Proceedings of the 6th International Conference on Model-Driven Engineering and Software Development (MODELSWARD 2018), pages 59-70

ISBN: 978-989-758-283-7

• Multi-layering occurs when we need to maintain

links between models at different levels of ab-

straction (traceability, reﬁnement...). Examples

are the CIM-PIM-PSM stack or the enterprise ar-

chitecture alignment stack (Clark et al., 2011).

A typical scenario for a software provider is to deliver

customised release to some customers but customisa-

tion leads to a tricky maintenance problem when mod-

els and meta-models evolve. This includes the case of

legacy code built using various DSL with various de-

velopment methods at different levels of abstraction

(from implementation code up to business processes).

In order to represent speciﬁc (or customised) aspects

we extend meta-models with new concepts, attributes

and relations. It is always possible to create the new

meta-model with references to the initial meta-model

but any modiﬁcation requires to rebuild the delivery

product. This is also the case when the modelling

language evolves; the new release requires tool adap-

tation.

In all these cases of software maintenance, includ-

ing model-driven reverse engineering of a legacy sys-

tem, it is a primary requirement to have techniques

to build new models without modifying the source

models (to preserve source property), we call it non-

intrusive model mapping. The existing approaches

are not fully satisfying because they are either intru-

sive, volatile or too generic. Paige et al. mentioned

several challenges of evolving models in MDE (Paige

et al., 2016). The work presented in this paper is a

practical contribution to the dependency heterogene-

ity challenge. We focus on model mappings that sup-

port several semantic links, models (or meta-models)

evolution and persistence, while staying non-intrusive

by preserving their parts. More precisely our contri-

bution is threefold:

• A mapping deﬁnition that enables to connect

models without breaking their legacy semantics

and without rebuilding the associated software

tool support.

• A mapping technique which is compliant with the

existing MDA tools. Taking an industrial point

of view, the question is not only to ﬁnd a map-

ping meta-model, which is usually depending on

the context (the models we work with), but also to

implement mapping techniques which are compli-

ant with the existing MDA tools and efﬁcient for

large-scale applications.

• A mapping tool that can be reused by others

and customised to similar model transformation

in practice. At the implementation level, main-

taining a link between model elements can also be

seen as an object relation mapping (ORM) prob-

lem (Ambler, 1997) preserving the link multiplici-

ties and a bi-directional navigation. This problem

needs a robust algorithm engine to maintain the

constraints imposed by the multiplicities.

The paper mainly targets a tool set; it is struc-

tured as follows. In Section 2 we review model map-

ping techniques and motivate our choices. Section 3

overviews EMF Facet and introduces our improve-

ments. Persistence, navigation and testing issues are

discussed in Section 4. New user interface facilities

are presented on the illustrating example in Section 5.

The application ﬁeld is detailed in Section 6 includ-

ing a tutorial, user stories and a return on experience

of larger case studies. Finally, Section 7 summarises

the contribution and draws open perspectives.

2 BACKGROUND AND

REQUIREMENTS

In this section we overview the basic requirements,

deﬁne the core concepts and compare with related

works to set the contribution requirements.

2.1 Background and Core Concepts

A software system sustains several releases that can

even co-exist for different users with different hard-

ware. The deﬁnition of models needs enhancements:

new attributes, new entities, new classiﬁcations, new

links,· · · Thus the real life systems generally handle

more than one model that may co-exist with different

semantics i.e. as deﬁned by their meta-models. To

capture this reality, we need a particular model map-

ping technique that owns the following criteria:

1. Non-intrusiveness: the mapping must not mod-

ify the individual models because they evolve in-

dependently.

2. Semantics: the mapping is not simply a set of

links, it supports a semantic relation to connect

differently the concepts with an equivalence class

of interpretation (an ontology of the concepts and

links).

3. Link Resolution: the mapping techniques must

provide a mechanism to navigate by mapping

links directly from the source model to the target

model and reciprocally.

4. Serialization: the mapping links must be persis-

tent to store the working environment.

These properties come from lessons learned during

model maintenance activities. This was the basis for

ﬁnding an adequate "mapping semantics".

Model mapping is one of the "Model manipulation

and management challenges" of (France and Rumpe,

MODELSWARD 2018 - 6th International Conference on Model-Driven Engineering and Software Development

2007) and in particular to the points (2) and (3) men-

tioned by its authors: (2) maintaining traceability

links among model elements to support model evolu-

tion and round trip engineering and (3) maintaining

consistency among viewpoints. Model mapping tech-

niques are useful for model composition, decomposi-

tion or synchronization (France and Rumpe, 2007). In

fact there are many "similar" operations and our ﬁrst

goal was to ﬁnd the adequate semantics and an asso-

ciated operational technique. Clavreul identiﬁed 88

model composition techniques for different purposes

in his systematic review (Clavreul, 2011). Model

Mapping is a model composition that preserves the

components. It is called the "model-based correspon-

dence" by Clavreul. Considering our requirements,

we reduce the ﬁeld to model mapping and we retain

ﬁve mapping techniques: extension, merging, annota-

tion, weaving and facets.

Model Extension or Merging. In the ﬁrst ap-

proach, the meta-model of one layer is extended

with the concepts of another one. In the second

approach, the meta-models are merged in a single

big model. In both case, the meta-models loose

their consistency and they can hardly evolve (ﬂexi-

bility loss). Clavreul’s mapping language is a merg-

ing approach and the architects must learn a DSL.

We opted for a mechanized approach by providing

a simple tool which applies at both compile and run

time. El Kouhen (El Kouhen, 2016) proposed an

uniﬁed methodology to compose models based on

meta-model extensions. The composition operators

are symmetric (commutative e.g. merge, parallel)

or asymmetric (weaving in the meaning of AOP, se-

quential integration). His work is largely inspired

by (Marchand et al., 2012) who gave a formal seman-

tics for weaving and merging through morphisms of

a category theory. Our approach can be seen as an

ad hoc model composition in their classiﬁcation since

our mapping uses semantic information of the source

models while their approach is a model mapping in

our classiﬁcation. Their mapping approaches enable

the separation of concerns, but the merging and weav-

ing do not preserve the legacy models (and the ex-

isting related tool support) because their result is a

new model. Model merging or extension are intru-

sive techniques because the source models disappear

in the target model.

Model Annotation and Weaving. Model mapping

is close to model weaving as deﬁned by (Jouault et al.,

2010) which was inspired by Aspect Oriented Mod-

elling. "Model weaving operations are performed be-

tween two or more meta-models, or between mod-

els. They aim to specify the links, and their associ-

ated semantics, between elements of source and tar-

get models" (Jouault et al., 2010). Models are woven

by establishing different kinds of links denoting the

semantics of weaving: merge operations, traceabil-

ity links, data translation mappings, text to graphical

representation, etc. Atlas Model Weaver (AMW) in-

cludes a transformation mechanism with ATL

to cre-

ate an automatic weaving. Virtual EMF (Brunelière

and Dupé, 2011) provides a visual assistant to edit

two models from different meta-models and to cre-

ate links between concepts with drag and drop. Un-

fortunately, editors are not supported since the 4.x

versions of Eclipse. The existing tools did not suit

to our requirements but we got inspired by them

to create our own weaving assistant, including im-

provements and new features (see Section 5.3). Di-

donet et al. (Del Fabro and Valduriez, 2007) propose

an approach that uses matching transformations and

weaving models to semi-automate the development of

transformations, which is not our goal. In that case,

weaving can implement transformations but cannot

map models, which is our goal. Our automated map-

pings implementation has been inspired by this work.

Model mapping has also been explored in the con-

text of Domain Speciﬁc Languages (DSL) to deﬁne

new languages from existing ones in a non-invasive

way without re-creating the tool support. Bruneliere

et al. (Brunelière et al., 2015) deﬁne a textual DSL

with extension operators to extend meta-model se-

mantics. It is independent from modelling tooling.

Similarly, Greifenberg et al. propose DSL-speciﬁc

tag language (Greifenberg et al., 2016). Kolovos et

al. (Kolovos et al., 2010) propose decorator extrac-

tion and injection operators based on GMF notes to

ensure the non-invasive property but it requires man-

ual transformations and conﬂicting specializations of

GMF notes may appear. Langer et al. (Langer et al.,

2012) propose EMF proﬁles, a lightweight adaptation

of UML proﬁles to extend meta-models with anno-

tations, constraints and stereotypes. These DSL ex-

tension approaches have in common to work on the

(binary) inheritance relation (one model is more spe-

cialised than another) while we target any kind of

n-ary relations between models such as aggregation,

composition, inheritance, dependency e.g. traceabil-

ity... However they are complementary.

In a previous work (Pepin et al., 2016), we com-

pared different model mapping techniques such as

merging, weaving, and annotation between meta-

models and we showed that they do not support prop-

erly the four above mentioned mandatory criteria for

model maintenance and evolution.

http://eclipse.org/atl/

Virtual Extension of Meta-models with Facet Tools

EMF Facet. In MDA, it’s common to use tools

based on the Eclipse EMF Frameworks. EMF enables

to deﬁne meta-models, load, persist and manipulate

the compliant models. The EMF Facet tool is based

on EMF to extend virtually a meta-model. This solu-

tion answers to the non-intrusiveness and the seman-

tics criteria. However the existing EMF Facet tools

do not fulﬁl all requirements: they have no persis-

tence support for the mapping links (the values are

computed by queries only) and few navigation and

semantics support. However this was the closest ap-

proach conforming to our requirements so we decided

to improve EMF Facet to draw links between models

and manage them in conformance to the deﬁned car-

dinalities.

2.2 Requirements for Meta-model

Extension

We can now revisit the criteria of the beginning of

this section to describe the requirements for a Non-

intrusive but Persistent Model Mapping technique

(NIP-MM).

• Mapping meta-models in a unique meta-model

usually breaks the evolution lifecycle: when the

individual models change, the mapping becomes

inconsistent and the associated tools obsolete. We

need a technique to map meta-models without in-

trusion and without version dependency.

• The weaving and annotation techniques are non-

intrusive but they use only generic links while the

end-user model transformation tools require spe-

ciﬁc information to proceed adequate transforma-

tions according to the meta-model relations and

cardinalities. The mapping technique must in-

clude semantic information for these relations.

• Last, the technique must serialize the mapping

links to support persistence in a way that loading

model mappings enables the navigation between

the original model elements and the new one with-

out disturbing the end-user experience. Behind

these properties we ﬁnd the ascending compati-

bility and the version preservation of existing tool

suites which are industrial concerns. Only the

Facet approach covered these requirements, ex-

cept persistence and navigation facilities which

are not supported.

In summary, we require an equipped Non-intrusive

Model Mapping technique supporting semantic links

and persistence.

3 REVISITING THE FACET

APPROACH

In this section we overview the EMF Facet features,

its interest and limitations and we introduce improve-

ments.

3.1 The Basic EMF Facet

Eclipse EMF Facet is a runtime meta-model exten-

sion framework composed of four parts: Facet, Cus-

tomization, Widgets and Query. The Facet part offers

the possibility to virtually extend (at runtime) exist-

ing meta-models and models. The Customization part

adds UI enhancements on a meta-model. The Widgets

part can be used to apply customizations to model ed-

itors. The Query part enables to compute attribute,

reference and operation value. The queries are written

in Java or OCL. With Facet one can extend models

by adding virtual features to existing models and also

weave models by linking their concepts (Figure 1).

This paper is mainly concerned with the Facet part;

we dealt with customizations when handling speciﬁc

meta-models editors in (Pepin et al., 2016).

Figure 1: Facet and Customization.

A Facet provides a new viewpoint on a model

which is helpful to categorize model elements with

new classiﬁcations, to add information on model el-

ements, to navigate easily between model elements

with new derived links. A facet provides a virtual

mechanism to add new attributes, references or oper-

ations on a model without modifying the initial meta-

model. Several facets can co-exist and be loaded/un-

loaded on demand without re-opening the model in-

stance. Under the hood, the Facet meta-model ex-

tends the meta-class (EClass) from the EMF Ecore

meta-model. Facet applicability is checked by op-

tional conformance rules.

As illustrated in Figure 2, the meta-model of a

Facet may contain FacetAttribute (extend EAttribute

in Ecore), FacetReference (extend EReference in

MODELSWARD 2018 - 6th International Conference on Model-Driven Engineering and Software Development

Figure 2: eFacet class Diagram.

Ecore) and FacetOperation (extend EOperation in

Ecore) which return values based on query evaluation.

Facets are contained in a FacetSet (extend EPackage

in Ecore, and FacetSet can be contained in an another

FacetSet hierarchically.

3.2 Common Use of Facets

Facet is an answer to different use cases in MDA prac-

tice. We illustrate with three situations.

• Facet can be used to implement UML derived fea-

tures i.e. attributes or associations that are not im-

plemented but computed from other features (also

called daemons in programming). For example,

the age is computed from the date of birth. Facet

enables one to represent them as attributes or ref-

erences

without storing redundant informations.

• Facet can be used to customise instances of a

meta-model for GUI (user interface) facilities (la-

bels, icons, color...). Facet enables to customize

Eclipse SWT components like trees, arrays, lists,

etc. As an example, Figure 3 shows a software

component classiﬁcation (components and appli-

cations). We modify the GUI by separating com-

ponents and application types and the associated

icons in Figure 4.

• A third case is to add new data in a model; new

features (attributes or references) which are not

computed but stored as standard features.

These examples improve the practice of model

(and meta-model) maintenance and evolution but also

model transformation. New information can be added

References are the way to represent associations at the

implementation level.

Figure 3: Before the GUI modiﬁcation.

Figure 4: After the GUI modiﬁcation.

without breaking the source meta-models and without

requiring a migration to a new meta-model.

This paper is a core contribution to that problem

because the basic Facet does not allow this case. We

call it Non-intrusive but Persistent Model Mapping

technique (NIP-MM).

Limitations. We remind the limitations of the Facet

approach for NIP-MM:

- A new feature of the Facet systematically calls a

query. However the queries cannot access to the

model to map and it is necessary to store the val-

ues. Also queries must be executed during the

model loading; if the model is voluminous, the

computation times can impact the response time.

- New features cannot be valued manually, as the at-

tributes or references of an ordinary meta-model.

- In independent model composition, the mapping

links must be persistent, consequently the values

of the features is to be serializable.

Virtual Extension of Meta-models with Facet Tools

3.3 Facet Improvements

To ﬁx the above limitations, we ﬁrst improve the

meta-model, then we modify the existing Facet man-

ager and serialization mechanisms.

Facet Manager. We improve the meta-model by al-

lowing dynamic structures instead of the static one.

We add FacetAttribute and FacetReference getters

and setters instead of recomputing systematically the

attached query (Figure 5). Since the meta-model con-

straints have an impact on the behaviour of the cur-

rent Facet manager, we upgraded its implementation

to support the new behaviour.

FacetAttributeFacetReference

DerivedTypedElement

EReference

EAttribute

Query

[0..1] fOpposite

[0..1] query

Figure 5: EMF Facet meta-model modiﬁcations.

Technically, enabling the access of values turns

to weakening the multiplicity of FacetAttribute and

FacetReference which extend DerivedTypeElement.

Thus the multiplicity of the query eReference on De-

rivedTypedElement is changed from 1..1 to 0..1.

This feature will permit to get the values of FacetRef-

erence and FacetAttribute without using a query. Fur-

thermore we add a new eReference named fOpposite

to create a reﬂexive reference mechanism like eOp-

posite on EReference in the Ecore meta-model. The

improved EMF Facet enables now to manually extend

and weave models. As a matter of fact, we fulﬁl the

criterion of Section 2.1: the serialization makes the

improved Facet be the most effective approach among

all the mapping techniques of Section 2. We submit-

ted our proposal of extended EMF meta-model as a

contribution to the open-source Eclipse EMF Facet

;

this has been approved.

Serialization Mechanisms. After the modiﬁcation

of the meta-model, we turned to the Java imple-

mentation of the new behaviour of the Facet engine

called FacetManager. At ﬁrst, we supported the two

Available since version 1.0: https://bugs.eclipse.org/

bugs/ show_bug.cgi?id=463898

simplest multiplicities of the bi-directional references

type: one-to-one and many-to-many. But, this was in-

sufﬁcient to cover all the cases so we proceeded with

all the possible cases. In the next section, we present

the cases and their implementation in order to obtain

a complete weaving engine.

4 PERSISTENCE AND

NAVIGATION

At this stage, one can create FacetReference links to

map concepts from different meta-models. In this

section, we describe the mechanism that supports bi-

directional mappings and a two-way navigation.

Mappings are represented here by UML associa-

tions, which are bi-directional by default. An associ-

ation end multiplicity (or cardinality) is an interval of

value represented by a lower and an upper bound. An

association between classes is represented by a set of

links at the instance level.

At the implementation level, maintain a link be-

tween elements is an object relation mapping (ORM)

problem (Ambler, 1997) where the link multiplici-

ties and bi-directional navigation are ensured. This

problem needs a robust algorithm engine to maintain

the constraints imposed by the multiplicities. A (bi-

directional) association is represented by a pair of uni-

directional (one-way) associations as shown in the

Library example of Figure 6. Thus any (instance)

Book Author

l..u

prefacedBy

Book Author

{reverse}

prefaces

lb..ub

prefacedBy

lb..ub

l..u

prefaces

transformation

Figure 6: Bi-directionnal link example.

link modiﬁcation must be propagated to the opposite

link. Depending on the multiplicity values, four cases

are distinguished: one to one (an instance is linked to

one instance only), one to many (an instance can be

linked to several instances), many to one (several in-

stances can be linked to one instance), many to many

(several instances can be linked to several instances).

Note that one to many and many to one cases are

not symmetric because the relation is oriented. Then,

these two cases have different algorithms to preserve

the multiplicity.

The core issue is to preserve the symmetry constraint

(prefacedBy.reverse() = prefaces) of the op-

posite association ends. Updating only one side usu-

MODELSWARD 2018 - 6th International Conference on Model-Driven Engineering and Software Development

ally leads to a symmetry constraint violation. In the

following, we illustrate each case by giving a ﬁgure

and the algorithm we implemented in FacetManager

One to One. In Figure 7, a book is prefaced by one

author and an author prefaces only one book. To

check the multiplicity consistency, the implementa-

tion in the FacetManager involves to keep only one

opposite link during the set operation.

Book Author

0..1 prefaces

prefacedBy 0..1

1984

2001

Asimov

Barjavel

1984

2001

Asimov

Barjavel

✘

set

step 1 step 2

Figure 7: One to one relation.

1 d e l e t e o l d r e f e r e n c e

2 d e l e t e o l d o p p o s i t e r e f e r e n c e

3 c r e a t e new o p p o s i t e r e f e r e n c e

4 c r e a t e new r e f e r e n c e

One to Many. In Figure 8, a book is prefaced by

many authors and author prefaces one book. To check

the multiplicity consistency, the implementation in

the FacetManager involves to replace the opposite

link during the set operation.

Book Author

0..1 prefaces

prefacedBy 0..*

1984

2001

451

Asimov

Barjavel

Clarke

✘

set

1984

2001

451

Asimov

Barjavel

Clarke

set

step 1 step 2

Figure 8: One to many relation.

1 remove o l d r e f e r e n c e from e x i s t i n g s

2 d e l e t e o l d o p p o s i t e r e f e r e n c e

3 c r e a t e new o p p o s i t e r e f e r e n c e

4 add new r e f e r e n c e i n t o e x i s t i n g

Many to One. In Figure 9, a book is prefaced by one

author and author prefaces many books. To check the

multiplicity consistency, the implementation in the

FacetManager involves to delete the old and to add

the new opposite link during the set operation.

1 d e l e t e o l d r e f e r e n c e

2 remove o l d o p p o s i t e r e f e r e n c e from e x i s t i n g s

3 add new o p p o s i t e r e f e r e n c e i n t o e x i s t i n g

4 c r e a t e new r e f e r e n c e

Many to Many. In Figure 10, a book is prefaced by

many authors and author prefaces many books. To

check the multiplicity consistency, the implementa-

tion in the FacetManager involves to delete the old

and to add the new opposite link during the set oper-

ation.

Available since version 1.1: https://bugs.eclipse.org/

bugs/show_bug.cgi?id=510039

Book Author

0..* prefaces

prefacedBy 0..1

1984

2001

451

Asimov

Barjavel

Clarke

set

step 1

step 2

1984

2001

451

Asimov

Barjavel

Clarke

set

✘

Figure 9: Many to one relation.

Book Author

0..* prefaces

prefacedBy 0..*

1984

2001

451

Asimov

Barjavel

Clarke

set

step 1

step 2

1984

2001

451

Asimov

Barjavel

Clarke

✘

set

✘

Figure 10: Many to many relation.

1 remove o l d r e f e r e n c e from e x i s t i n g s

2 remove o l d o p p o s i t e r e f e r e n c e from e x i s t i n g s

3 add new o p p o s i t e r e f e r e n c e i n t o e x i s t i n g

4 add new r e f e r e n c e i n t o e x i s t i n g

Thanks to the automatic management of bi-

directional links, the instance mapping is transparent

for users. Our implementation allows one to weave

the instances without a slave-master semantics. The

user can draw a link in any order: from the source to

the target or reverse.

Testing New Implementations

According to the Test-Driven Development approach,

JUnit tests have been written to check our implemen-

tation and its code coverage before adding it to the

FacetManager. We illustrate this with two simple ex-

amples of the Library. For each multiplicity case,

we set facetReference between different instances of

books and writers, then we check if the reference set-

ting conforms to the reference and its opposite.

Example 1. In the many-to-one case, writer1 and

writer2 write the preface of book1. The test case of

Listing 1 assigns the value, establishes the mapping

and checks the reverse link. It succeeds.

Listing 1: Many-to-one: the direct preface reference

f i n a l F a c e t R e f e r e n c e p r e f a c e = g e t F a c e t R e f (

MANY_TO_ONE, WRITER_EXT , PREFACE ) ;

f i n a l Book b o o k 1 w r i t e r 1 = t h i s . f a c e t M g r .

g e t O r I n v o k e ( w r i t e r 1 , p r e f a c e , Book . c l a s s ) ;

A s s e r t . a s s e r t E q u a l s ( msg ( WRITER1 , BOOK1) , book1 ,

b o o k 1 w r i t e r 1 ) ;

f i n a l Book b o o k 1 w r i t e r 2 = t h i s . f a c e t M g r .

g e t O r I n v o k e ( w r i t e r 2 , p r e f a c e , Book . c l a s s ) ;

Virtual Extension of Meta-models with Facet Tools

A s s e r t . a s s e r t E q u a l s ( msg ( WRITER2 , BOOK1) , book1 ,

b o o k 1 w r i t e r 2 ) ;

Example 2. The test case of Listing 2 checks the

automatic setting of the opposite reference prefaced:

book1 is prefaced by writer1 and writer2. Its exe-

cution also led to success.

Listing 2: Many-to-one opposite: the prefaced reference

f i n a l F a c e t R e f e r e n c e p r e f a c e d = g e t F a c e t R e f (

MANY_TO_ONE, BOOK_EXT, PREFACED) ;

f i n a l L i s t < W r i t e r > w r i t e r s 1 a n d 2 = t h i s . f a c e t M g r .

g e t O r I n v o k e M u l t i V a l u e d ( book1 , p r e f a c e d , W r i t e r

. c l a s s ) ;

A s s e r t . a s s e r t E q u a l s ( msg (BOOK1, " w r i t e r 1 a nd w r i t e r

2 " ) , A r r a y s . a s L i s t ( w r i t e r 1 , w r i t e r 2 ) ,

w r i t e r s 1 a n d 2 ) ;

The above examples are speciﬁc to the Library

case study but generic tests, written at the meta-model

level, can be shared by all applications. Henceforth

EMF Facet enables one to create links between any

models following a deﬁnition at the meta-model level.

Section 5 presents additional tools to handle the links

at the instance level.

5 END-USER RUNNING THE

MAPPING

The mapping API and the tests have been presented

in Section 3 and Section 4. In this section, we il-

lustrate the implemented industrial tools associated to

the mapping framework with the Library example.

We extend Book and Writer meta-classes by cre-

ating new FacetSet for each link mapping cases con-

taining all facets which deﬁne the new virtual ref-

erences with the speciﬁc multiplicity: prefaces and

prefaced. A mapping process includes the follow-

ing activities: facet mapping deﬁnition, instance valu-

ation, instance linking, mapping navigation and eval-

uation.

5.1 Mapping Deﬁnition

Each facet deﬁnes at least a name and the meta-class

type. A facet optionally extends an existing Facet.

The facet refers to the original meta-class by its ab-

solute Universal Resource Identiﬁer (URI). The facet

deﬁnes three kinds of features: FacetAttribute, Face-

tReference and FacetOperation. We can create as

many new features as required. A FacetReference

deﬁnes at least a name, a multiplicity, a type and an

opposite reference if the association is bidirectional.

The type is a meta-class from any Ecore reachable

meta-model or another predeﬁned FacetReference in

the current FacetSet. A FacetAttribute deﬁnes at least

a name, a multiplicity and the meta-class type as in

FacetReference.

Example. Assume a Book meta-model describing

the books and a Writer describing authors. In Fig-

ure 11, we create four FacetSets, one for each map-

ping multiplicity: many to many (MToM), many to

one (MToO), one to many (OToM), and one to one

(OToO). For each case, a new reference preface links

’writers to books’ and the opposite reference pref-

aced links ’books to writers’. The multiplicity differs

through the upper and lower bound. In the example

of Figure 11, -1 represents the ’many’ upper bound

value. The purpose of this new deﬁnition is to extend

the existing classiﬁcation of a library of books and

authors to obtain an enriched catalogue.

Figure 11: FacetSet deﬁnition example.

5.2 Setting Links by Property Value

Property editors set the value of the attributes and ref-

erences that will be used by queries.

Example. To experiment the different multiplicity

links, we create a library with books and writers. We

apply a speciﬁc FacetSet at a time: MToM, MToO,

OToM, or OToO. In Figure 12 the FacetSet OToM

is applied, the book "2001 Space Odyssey" have

two prefaced writers "Isaac Asimov" and "Arthur C.

Clarke". The property ﬁeld varies according to the

multiplicity: a pop-up menu selector (one) or a dou-

ble selector (many). We can check that the opposite

link is correctly set in Figure 13: "Isaac Asimov" pref-

ace "2001 Space Odyssey", and "Arthur C. Clarke"

preface "2001 Space Odyssey".

MODELSWARD 2018 - 6th International Conference on Model-Driven Engineering and Software Development

Figure 12: Editing multi-valued reference.

Figure 13: Editing mono-valued

references.

5.3 Setting Links by Model Weaver

Setting many links between instances with the prop-

erty values is quite fastidious and editors are really

useless in this case. We developed a speciﬁc weaving

editor with multiple views (drag-and-drop).

Example. On the right part of Figure 14, an outline

displays the different models to weave i.e. a ﬁrst

model ’library with books’ and a second model

’library with writer’. On the left, a speciﬁc view

organizes the weaving result by facets. The MToM

facet is loaded. This design allows us to drag and

drop elements from right to left to link elements by

references corresponding to the FacetSet deﬁnition.

In this example, we drag and drop two writers

"Ray Bradbury" and "George Orwell" on reference

prefaced of the book "Fahrenheit 451".

Figure 14: Model weaver editor.

5.4 Exploring Links with OCL Query

Using a model mapping by FacetReferences makes

possible the navigation through the models, from one

FacetReference to another. The TreeEditor is com-

patible with EMF Facet and enables one to browse

hierarchically the models by the FacetReference, but

it is still not very convenient; architects would like to

get some statistics to measure which elements from

models are mapped or not. Since EMF offers OCL ex-

pressions parser on Ecore models, we have produced

an engine extension in order to make Facet compati-

ble with virtual features and used as property in OCL

statements. This extension is also integrated in a plug-

in to EMF Facet project.

Example. In Figure 15, we use the previous model

weaving result. We open the model ’library with

books’ and write an OCL query on the console

to know all writers who have prefaced the books:

Figure 15: OCL Facet Console.

Virtual Extension of Meta-models with Facet Tools

"self.books.prefaced". The query writing is assisted

with auto-completion, and the result appears on the

console.

Section 6 draws experimentations of the above

tools and wizards.

6 EXPERIMENTATIONS

The above tools and wizards (and others) have

been integrated in the EMF Facet open-source

project. The improved Facet tooling is avail-

able at the open-source Eclipse EMF Facet

https://wiki.eclipse.org/EMF_Facet.

6.1 Starter Kit Example

EMF Facet remains a framework with components to

be integrated by developers in their applications. This

Eclipse project has no introductory examples to show

how to integrate the components to make an applica-

tion, to handle models compatible with facet extensi-

bility or to create a facet set (or facet customization)

to extend an existing meta-model.

To illustrate the usability and availability

of the contribution of this paper, we pushed

some source codes and model examples in a

public and open source Starter Kit project at:

https://gitlab.com/jpepin/EmfFacet_StarterKit/.

Eclipse users can build the Update Site to install

easily the example in Eclipse installation (compatible

with the Kepler and more recent versions).

1. In Eclipse ’Help > Install New Software’, enter

the update site url https://gitlab.com/jpepin/

EmfFacet_StarterKit/raw/master/in.jpep.emf.

facet.starterkit.updatesite/.

2. As illustrated by Figure 17, select EMF Facet

Starter Kit Feature and next...

Figure 17: EMF Facet Starter Kit.

3. After the plugin installation, create the example

project with ’File > New > Example’ and

4. choose Models Example from the EMF Facet

Starter Kit category.

The tiny example is given in Figure 16. Two mod-

els are proposed: human and city.

Three testing scenarios are proposed, as follows.

Intra Model Facet Extension. You can open and

compare the ﬁles example.human and exam-

ple.mset. This mset is opened with an editor com-

patible with Facet and the facet deﬁnitions are en-

abled at the beginning. You can see through the

Properties view, that the facet extension adds a

new attribute ’Surname’ and a new bidirectional

reference ’parent’ / ’children’ of John. The values

of the two new features can be assigned and stored

Figure 16: EMF Facet starter kit models example.

MODELSWARD 2018 - 6th International Conference on Model-Driven Engineering and Software Development

in the data ﬁle.

Inter Models Facet Extension. The human and city

models are clearly independent. This example

shows a new facet extension to add bi-directional

reference ’live’ / ’habitants’ between the models.

Similarly to the ’intra’ scenario, the values can be

updated with the Properties view, and saved with

result serialized in data.efs.

Customization. By default the human meta-model

does not distinguish the gender man and woman.

To add the visual distinction of Figure 16, open

the ﬁle example.mset, press the button ’Load / Un-

load Customizations’ from the tool bar and

choose Family Custo deﬁnition. Now, men and

women are displayed with different preﬁx label

and icon.

If you are software developer, the sources of

the example can be download from the EMF Facet

Starter Kit Git repository. It demonstrates how to im-

plement an editor with facet compatibility integration:

extended property view, load / unload facet and cus-

tomization buttons, set / load facet serialization ﬁle.

Facet and customization deﬁnitions for meta-models

Human and City are available too for example and

can be widen. Two examples of query are in the cus-

tomization deﬁnition: one in Java, the other in OCL.

This tiny example shows that the facet persistence

and navigation contribution is already available in the

Eclipse open source project for developers to create

simple tools without complex conﬁguration for end-

user. Moreover, the example demonstrates that EMF

Facet is adapted to different use cases: to add at-

tributes and association in a model; to link two differ-

ent and independent models; or to customize the user

interface. This example can be adapted and extended

for largest and real use cases.

6.2 User Stories

We illustrate here two situations of software main-

tenance and reverse-engineering where the non-

intrusive but persistent model mapping is pertinent.

In legacy modernization, the architects need to

establish the as-is state of the information systems

(from business processes to deployed applications).

We need to enrich the models with additional in-

formation without modifying the meta-models just

like source code annotations: deprecated, renaming...

These information are useful to develop the to-be state

of the next generation information systems but must

not interfere with the existing programs. Conversely

to code annotation, the improved Facets enables to en-

rich with more than one layer (one layer by aspect we

are interested in).

When standard meta-models (UML, BPMN....)

do not ﬁt the requirements of one organisation, one

can customize the meta-models and create a family

of extensions that is consistent with the standard re-

leases, as far as a new release of one standard did

not impact too many concepts (like the transition from

UML 1.x to UML 2.x). The maintenance of the cus-

tomizations follow their own life cycles.

6.3 Large Scale Experimentations

The Library example was a toy example for illustrat-

ing this paper. We experimented our improved Facet

in the context of Business-IT alignment in Enterprise

Architecture. This context covers a wider ﬁeld than

the scope of this paper. We brieﬂy report here this

experimentation but the reader will ﬁnd more details

in (Pepin et al., 2016).

The general problem was to align models from

different points of view, in the context of Informa-

tion System maintenance. We deﬁned different meta-

models (business process, functional, application) and

we implemented techniques to feed the corresponding

models from legacy information (source code, data

and models when they exist).

The mapping support, as deﬁned in this paper, was

the core technique to establish the alignment links be-

tween the models. We experimented real size case

studies provided by insurance companies with het-

erogeneous information supports: lots of java source

ﬁles, MEGA repositories, databases,... The experi-

ments showed that big mappings are hardly manage-

able by humans and tool assistance is mandatory. Our

tool support handled efﬁciently big models. Even in

bulky case with manual mapping, our Weaving Editor

(cf. ﬁgure 14) provides an optimized user interface

with search engine to ﬁnd concepts and highlight con-

cepts matching. However additional tool is needed to

visualize big mappings, to evaluate the mapping prop-

erties (consistency, completeness) and quality (mis-

alignment, evolution). Our Facet query engine is a

ﬁrst step to reach this goal.

7 CONCLUSION

We proposed an improved technique of virtual exten-

sion of meta-models that uses Facets. It enables one

to modify meta-models already in use by software,

without rebuilding completely the legacy tool support.

The extension takes account of the multiplicity of as-

sociations and considers two-way references between

the involved entities.

Virtual Extension of Meta-models with Facet Tools

• A mapping model We experimented existing tech-

niques for model composition and we propose an

improved version of virtual extension of meta-

models that uses Facets and enables one to mod-

ify meta-models already in use without rebuilding

the software product. The extension is called vir-

tual because it does not directly impact the initial

meta-models.

• A mapping technique which is compliant with the

existing MDA tools. EMF enables to deﬁne meta-

models, load, persist and manipulate the compli-

ant models. The EMF Facet tool is based on EMF

to extend virtually a meta-model. This solution is

non intrusive, it supports link semantics but it does

not have a mechanism for persistence (the values

are calculated by queries) and an adequate tech-

nique and tools for mapping models. We have im-

plemented this technique which is now integrated

to the open-source Eclipse EMF Facet project.

• A mapping tool We overcome this limitation by

improving the EMF Facet technique, by modi-

fying the meta-model and implementing several

tools to manage model mapping in practice. Our

implementation preserves the link multiplicities

and a bi-directional navigation.

The proposed technique has been implemented, then

experimented on various case studies and integrated

in the open-source Eclipse EMF Facet project. We

have then contributed to solve an important model and

software evolution issue.

The next step will provide more assistance to the

user; we target the implementation of heuristics to

propose a list of possible model mappings to the mod-

eller: he can then choose the desired ones. These

heuristics will depend on the nature and the seman-

tics of the mappings. For example, when mapping

two releases of the same model, it is usually easier to

detect equality mapping. In speciﬁc cases, one can

detect patterns or naming conventions.

REFERENCES

Ambler, S. W. (1997). Building Object Applications That

Work: Your Step-by-Step Handbook for Developing

Robust Systems with Object Technology. Cambridge

University Press.

Atlee, J. M., France, R., Georg, G., Moreira, A., Rumpe,

B., and Zschaler, S. (2007). Modeling in software en-

gineering. In Companion to the Proceedings of the

29th International Conference on Software Engineer-

ing, ICSE COMPANION ’07, pages 113–114, Wash-

ington, DC, USA. IEEE Computer Society.

Brambilla, M., Cabot, J., and Wimmer, M. (2012). Model-

Driven Software Engineering in Practice. Morgan &

Claypool Publishers.

Brunelière, H., García, J., Desfray, P., Khelladi, D. E.,

Hebig, R., Bendraou, R., and Cabot, J. (2015). On

lightweight metamodel extension to support modeling

tools agility. In Modelling Foundations and Applica-

tions - 11th European Conference, ECMFA.

Brunelière, H. and Dupé, G. (2011). Virtual EMF - trans-

parent composition, weaving and linking of models.

In EclipseCon Europe 2011.

Clark, T., Barn, B. S., and Oussena, S. (2011). Leap: A

precise lightweight framework for enterprise architec-

ture. In Proceedings of the 4th India Software En-

gineering Conference, ISEC ’11, pages 85–94, New

York, NY, USA. ACM.

Clavreul, M. (2011). Model and Metamodel Composition:

Separation of Mapping and Interpretation for Unify-

ing Existing Model Composition Techniques. PhD the-

sis, Université Rennes 1.

Cuadrado, J. S., Izquierdo, J. L. C., and Molina, J. G.

(2014). Applying model-driven engineering in small

software enterprises. Sci. Comput. Program.

Del Fabro, M. D. and Valduriez, P. (2007). Semi-automatic

model integration using matching transformations and

weaving models. In Proceedings of the 2007 ACM

symposium on Applied computing.

El Kouhen, A. (2016). Panorama : A Uniﬁed Framework

for Model Composition. In 15th International Confer-

ence on Modularity, malaga, Spain. MODULARITY

2016.

France, R. and Rumpe, B. (2007). Model-driven develop-

ment of complex software: A research roadmap. In

2007 Future of Software Engineering, FOSE ’07.

Greifenberg, T., Look, M., Roidl, S., and Rumpe, B. (2016).

Engineering tagging languages for dsls. CoRR.

Jouault, F., Vanhooff, B., Bruneliere, H., Doux, G., Berbers,

Y., and Bezivin, J. (2010). Inter-DSL Coordina-

tion Support by Combining Megamodeling and Model

Weaving. In Proceedings of the SAC 2010.

Kolovos, D. S., Rose, L. M., Drivalos Matragkas, N., Paige,

R. F., Polack, F. A. C., and Fernandes, K. J. (2010).

Constructing and navigating non-invasive model dec-

orations. In Tratt, L. and Gogolla, M., editors, The-

ory and Practice of Model Transformations: Third In-

ternational Conference, ICMT 2010, Malaga, Spain,

June 28-July 2, 2010. Proceedings.

Langer, P., Wieland, K., Wimmer, M., and Cabot, J. (2012).

EMF proﬁles: A lightweight extension approach for

EMF models. Journal of Object Technology.

Marchand, J. Y., Combemale, B., and Baudry, B. (2012).

A categorical model of model merging and weaving.

In Proceedings of the 4th International Workshop on

Modeling in Software Engineering, MiSE ’12.

Paige, R. F., Matragkas, N., and Rose, L. M. (2016). Evolv-

ing models in model-driven engineering: State-of-the-

art and future challenges. Journal of Systems and Soft-

ware.

Pepin, J., André, P., Attiogbé, C., and Breton, E. (2016). Us-

ing ontologies for enterprise architecture integration

and analysis. CSIMQ, 9.

MODELSWARD 2018 - 6th International Conference on Model-Driven Engineering and Software Development