-As model-based software development becomes increasingly important, the number of models to express various aspects of software at different levels of abstraction raises. Meanwhile evolutionary development and continuous changes demand for explicit dependencies between involved models to facilitate change impact analysis, software comprehension, or coverage and consistency checks. However, there are no comprehensive approaches supporting models, dependencies, changes, and related information throughout the entire software development process. The approach presented in this paper provides a unified and modelspanning concept with a repository for model integration, model versioning, and dependency identification among models utilizing traceability techniques, enhanced with analytic capabilities. The identification is based on a rule set to provide high values for precision and recall. The approach is implemented in a tool called EMFTrace, which is based on Eclipse technology and supports different CASE tools for modeling.
The growing complexity of software is often accompanied by difficulties in implementing changes and by architectural erosion. One possibility to tackle these problems is the use of different models and modeling languages throughout the entire development process to express requirements, design decisions, and dependencies between models, which is commonly referred to as model-based development. The wide use of several modeling languages and CASE tools has led to a variety of models with different scope and levels of abstraction. Integrating the various models and detecting and analyzing dependencies between them has become important.
It is therefore necessary to provide support for software evolution and continuous changes through models at different levels of abstraction and by explicitly expressing dependencies between them. Traceability links are wellsuited to model such dependencies, since they enable further analyses, as change impact analysis, software comprehension, and consistency checks. Moreover, solid tool support is required to cope with the number of modeled entities and dependencies to supply practical applicability and usability.
Existing approaches provide solutions for the
integration of models into centralized repositories, e.g., [
*The research presented in this paper was partly funded by the federal
state Thuringia and the European Regional Development Fund ERDF
through the Thu¨ringer Aufbaubank under grant 2007 FE 9041.
Besides, several different concepts have been proposed for
dependency analyses on models through traceability rules,
e.g., in [
The concept presented in this paper is aimed at creating
an extensible platform for various analyses techniques
required for model-driven engineering, reengineering, and
impact analysis to support software evolution and
continuous changes. It is based on a unified approach for
comprehensive model integration and dependency analysis. In
our approach inter and intra model dependencies are
identified and recorded through traceability links, which are
established by explicitly defined rules. These traceability
rules are combined with information retrieval algorithms to
achieve high precision and recall as well as flexibility.
Special emphasis has also been put on the explicit modeling of
dependency types, since they provide important semantic
information for further analyses. Moreover, our concept
provides a unified treatment of all related models and data
through automated versioning and management of models,
rules, model dependencies, and dependency types within
a centralized repository. We utilize the extensible model
repository EMFStore [
The remainder of this paper is organized as follows. We discuss current development and research regarding model repositories and model dependencies in Section II, whereas Section III introduces our concept and describes our approaches for model integration and dependency analysis. Section IV outlines our main conclusions and future work.
II. STATE OF THE ART
Automated model management and versioning can be achieved by model repositories that therefore assist modelbased development of software. We considered several different repository projects as a base for EMFTrace, including the Eclipse Model Repository1, CDO Model Repository2, EMFStore3 and AndroMDA4. Maturity, supported features, usability, and documentation were the main decisive factors we based our evaluation on. After a careful consideration of all projects, we decided to use the EMF5-based EMFStore as our underlying repository.
Since we focus on model-based development and
dependencies between models, we do not consider
approaches for dependency detection on the level of source
code. Data gained through dependency analyses can either
be stored and maintained in a dependency matrix, as a
graph of traceability links, or as a set of cross-references
[
Most traceability approaches focus either on predefined
rules or on information retrieval. Information retrieval
approaches as proposed in [
In contrast, rule-based approaches require more work to
adopt to new models, but they result in more reliable links.
Rule-based approaches as proposed in [
EMFTrace implements our approach of comprehensive model integration and dependency identification as an extensible platform, which is our basis for further analyses and research. Our main focus is to provide a set of analyses and validation features, which is independent of a certain CASE tool and modeling language, to support engineers, software architects and other stakeholders throughout the 1http://modelrepository.sourceforge.net/project-summary.html 2http://wiki.eclipse.org/CDO 3http://www.eclipse.org/proposals/emf-store/ 4http://www.andromda.org/index.php 5http://www.eclipse.org/modeling/emf/ entire software lifecycle. Based on models and dependencies between them, we want to establish solid tool support for the following activities:
Dependency analysis, Change impact analysis, Goal-oriented decision support for: – Forward engineering and – Reengineering; as well as
We utilize the EMFStore model repository for storing and versioning EMF-based models. Figure 1 provides an overview of the entire approach and illustrates the interaction of EMFTrace with external CASE tools.
To enable the integration and dependency analysis of different models we extended EMFStore by several metamodels as shown in the bottom of Figure 1. We integrated different modeling languages into the repository to span the entire software development process. Our approach supports requirements engineering through URN goal models, URN use case maps, and UML use cases. Furthermore, design models such as class diagrams are provided by UML, which has been fully integrated as well. Additional dependencies between UML and URN models are provided by the integration of OWL ontologies, which supply semantic networks of related terms. To ensure practical applicability, we support several different CASE tools which can be used to create and edit the corresponding models. More details on the integration of the models are given in Section III-B2.
As mentioned in Section II our approach utilizes
rulebased traceability for dependency identification among
models. To enable a unified treatment of all models and
related data, we integrated additional metamodels into
EMFStore to maintain and store traceability links and
rules directly in the repository. Our traceability metamodel
provides explicit support for modeling a hierarchy of
dependency types to ensure the proper categorization of
detected dependencies. Dependency types can be created
and maintained by users in the repository and aggregated
in special catalogs, and are therefore subject of model
versioning as well. The metamodel for traceability rules
offers a similar concept, which enables users to create
rules directly in the repository and group them together in
catalogs, to support fast exchange and easy maintenance.
The design of our rules is based on similar approaches as
proposed in [
1) Dependency Concept: We store dependency information as traceability links to benefit from their rich semantics. Our traceability metamodel supports binary and n-ary traceability links. Each traceability link can be enhanced with additional information such as the type of a dependency, design decisions, design alternatives, and ! ! !
! $)'3>&)!#*+),8! #*+),-".(/'01)-203)&4'()! OP!Q5G%! KP!2=7*&3!
KP!Q5G%! OP!".7*&3! ! 6)1&)889*0! %)83901!! -!@K%I!-! ! 5)='039(!L)E! -!MLG!-! I&*3N1N! ! assessments of the confidence of the link performed by users.
Moreover, we are able to combine transitively related traceability links, i.e., links that share common model elements as their source or target elements, into traces. These traces store chains of links to span design decisions and multi-level dependency relations. The example shown in Figure 2 illustrates the transition from a URN goal to its architectural realization through UML models. Furthermore, EMFTrace provides automated validation features for traceability links and traces, to ensure the integrity of stored dependency information. Corrupt or outdated links will be erased while broken traces are either split into smaller sub-traces or removed as well, if they contain less than two links.
"#$%!
&'(!)$'(!
)#*+#,(,-!
.$/0$1(!
2) Rule Concept: Following the general design
proposed in [
Element Defini+on What models are affected by the rule? Query How are they related? Result Defini+on What should be done with them?
In contrast to existing approaches as [
B. Architecture and Realization
1) Architecture: Since EMFTrace is our evolving basis
for many different research activities related to
modeldriven engineering, its architecure must support the
addition of new concepts and features. Our entire tool is
based on Eclipse technology and consists of several
plugins that provide the core functionality, a user interface,
and our adapted metamodels. EMFStore provides a server
(the repository) and a client to operate with the repository,
which is why we integrated our additional features into
the client to provide a unified view and usability. We
designed several additional components that integrate new
features into the EMFStore client, such as an interpreter
for traceability rules as shown in the bottom right corner
of Figure 1. Our architecture is extensible and allows
for easy integration of new components into EMFTrace
to introduce new features, i.e., new analysis components.
All plug-ins and components were implemented and tested
with Java to ensure the platform independent applicability
of our tool [
2) Model Integration: Using a modeling language with EMFTrace requires the integration of its metamodel into EMFStore. The metamodel must be converted into an EMF-based Ecore model, which can either be modeled with EMF or generated from an XSD file. Once an Ecore model has been generated, it must be adapted to EMFStore’s metamodel, which provides additional attributes to facilitate the management through EMFStore. We performed these steps for the metamodels of UML, URN, and OWL.
Since the adaptation to EMFStore’s metamodel introduces new attributes to the metamodels, an adaptation of models (metamodel instances) is required as well, to add these attributes while importing models from CASE tools. We perform the integration and adaptation as illustrated by the arrows in the middle of Figure 1 by applying XSLT templates on models and encapsulated the entire import and export procedures in new components (Model-Exchange-Interface), which have been integrated into EMFTrace. Therefore, XML-based models exported by CASE tools and XSLT provide the core technology for our integration approach.
As CASE tools (see top of Figure 1) we currently support the Eclipse UML2Tools (UML), Visual Paradigm (UML, BPMN), the Eclipse-based jUCMnav (URN), and Prote´ge´ (OWL) with our XSLT templates. Additional modeling languages and their respective CASE tools such as BPEL can be added to EMFTrace at any time through adding an EMF-based metamodel to the underlying EMFStore repository and providing further XSLT templates.
3) Extensions: One addition which has already been
integrated into EMFTrace is the tool EMFfit [
We presented an approach and a prototype tool EMFTrace for model integration and dependency identification between models of the entire software development process. We support different artifacts ranging from requirements models to design models which are managed by a unifying repository. Our tool provides comprehensive and automated dependency identification through rulebased traceability to ensure decent precision of results, enhanced with information retrieval techniques. EMFTrace facilitates the use of semantically important link types and is capable of storing, versioning, and maintaining traceability links. Standard modeling languages, such as UML, URN, and OWL, are integrated into the repository and several CASE tools are supported.
Further work will focus on several issues: additional modeling languages such as U2TP and feature models shall be integrated into EMFTrace along with the appropriate metamodels and XSLT templates, allowing for additional CASE tools to be used in conjunction with our tool. New models enable new analysis capabilities and require new rules or existing rules to be refined. Our rule concept should be enhanced to support consistency checks of models stored in the repository. The combination of dependency identification with consistency analysis provides the basic features to perform change impact analysis on models through new components and rules. To improve the usability of our tool, model synchronization with CASE tools and methods of change recognition for maintenance are considered as further enhancements. Drawing more benefits from discovered dependency relations requires sophisticated visualization of thereby created traceability links to enable the user to navigate on hierarchical chains of dependencies and to trace models efficiently. Finding appropriate means of dependency visualization is therefore one important research question related to EMFTrace.