Metamodellierung im Sinne der Meta Object Facility (MOF) stellt eine Methode für die strukturelle Definition der abstrakten Syntax von Modellierungssprachen und Modellen im Softwareentwicklungsprozess dar. Um Modellsimulation und dynamische Analysen für metamodellbasierte Sprachen zu unterstützen, fehlt es an einem Kalkül zur operationalen Semantik. In dieser Arbeit wird ausgehend von MOF die Aktionssemantik MActions entwickelt, die die Definition von operationaler Semantik als Verhalten in Metamodellen ermöglicht. Diese Erweiterung geht einher mit der Beschreibung von Laufzeitmodellen sowie Zuständen und Parallelitätseigenschaften, so dass eine Verifikation von dynamischen Eigenschaften möglich wird. Zu diesem Zweck wird mit der Linear Temporal Object Constraint Language (LT-OCL) exemplarisch eine prädikatenlogische Temporallogik entwickelt, die eine metamodellunabhängige Analyse für ausführbare Modelle erlaubt. Dabei ist die Semantik von temporalen Ausdrücken über Zuständsänderungen von (aufgezeichneten) Ausführungsläufen beschrieben, wobei eine Linearisierung parallele Änderungen zusammenführt. Als weiteren Anwendungsfall der dynamischen Analyse untersuchen wir die Relation zum Verhaltensvergleich im Sinne der Bisimulationstheorie. Metamodelle, Aktionssemantik und Temporallogik werden mittels einer erweiterten Abstract State Machine (ASM) formal beschrieben und kommen in zwei Fallstudien zur Anwendung (Timed Automata und C#). / Object-oriented metamodelling as defined by the Meta Object Facility (MOF) provide a means to describe the structure of models and the abstract syntax of modelling languages at various stages in a software development process. However, MOF lacks concepts for the definition of operational semantics and there is no support for dynamic model analysis based on the semantics and abstract states of a language definition. This thesis investigates on extending the metamodelling framework with an action semantics - the MActions - to support the definition of operational semantics in metamodels and enable simulation as well as verification of dynamic properties. For this purpose, runtime models are incorporated with semantics for states, time, and properties of parallelism that allow a generic analysis solely bound to a certain metamodel definition. Furthermore, we develop the Linear Temporal Object Constraint Language (LT-OCL) to perform a dynamic analysis of execution runs based on the executable models. The semantics of this temporal predicate logic is bound to state changes of (recorded) execution traces that are linearizations of parallel changes of the runtimes model. This establishes the link to the theory of bisimulation as a second application case of dynamic analysis. Abstract State Machines (ASM) have been used to formally define the action language in conjunction with metamodels and the temporal logic. As proof of concept of the whole approach, the framework has been implemented and applied to two languages as case studies (namely Timed Automata and C#).

El propósito de esta Tesis es llevar a cabo un proceso en el que se aplique la práctica de la integración continua en un desarrollo de software dirigido por modelos de forma eficiente, mediante el cual los desarrollos de software puedan beneficiarse conjuntamente de las mejoras y ventajas que proporcionan la aproximación de desarrollo de la ingeniería dirigida por modelos y la práctica de la integración continua. La aproximación de la ingeniería dirigida por modelos es el último salto natural de la ingeniería del software en cuanto a la búsqueda de métodos de desarrollo que elevan el nivel de abstracción hasta el punto en el que los expertos de un dominio de conocimiento, ajenos al mundo informático, son capaces de guiar y cambiar la lógica de los sistemas informáticos. La práctica de la integración continua es una recomendación de las principales metodologías de desarrollo, que tiene como objetivo la realización de integraciones automáticas del software en etapas tempranas del desarrollo, ofreciendo ventajas como la reducción del riesgo intrínseco que, dado su carácter temporal y único, tienen todos los proyectos. Con la unión de la ingeniería dirigida por modelos y de la práctica de la integración continua se busca ofrecer, a los equipos de desarrollo que trabajan utilizando algún tipo de iniciativa de la ingeniería dirigida por modelos, la posibilidad de integrar de forma continua y distribuida sus desarrollos. Al mismo tiempo, los clientes, verdaderos expertos del dominio de conocimiento en su ámbito de negocio, se benefician del aumento del nivel de abstracción de las técnicas de desarrollo para que ellos mismos, y de forma transparente, sean capaces de modificar su propio sistema informático sin la ayuda de personal técnico ajeno a su negocio, ahorrando así tiempo y costes. Para cumplir con el objetivo de esta Tesis doctoral se construye un prototipo que salva los impedimentos actuales que no permiten la unión entre estos dos nuevos activos de la ingeniería del software. Los principales problemas encontrados están relacionados con la selección de una iniciativa de desarrollo apropiada, los sistemas de control de versiones especialmente adaptados para trabajar con modelos, la generación incremental de artefactos a partir de modelos y la adaptación a las herramientas actuales de integración continua de forma optimizada. La separación del trabajo realizado en diferentes bloques permite ofrecer soluciones de forma tanto aislada como en conjunto, dando lugar a un trabajo iterativo e incremental de comienzo a fin. Para analizar las ventajas que ofrece la propuesta de este trabajo frente a otras posibilidades de desarrollo, se realiza una evaluación mediante la creación de diferentes casos de prueba en los que la medición de diferentes parámetros ofrecen una estimación numérica de las ventajas reales obtenidas. El análisis descriptivo, el contraste de hipótesis y las técnicas de regresión permiten una mejor interpretación de los resultados. Finalmente, se define el proceso, objetivo último de este trabajo, mediante la respuesta a diferentes preguntas planteadas, que facilitan su comprensión y entendimiento. / The purpose of this Thesis is to create a process in which the continuous integration practice can be applied to a model-driven software development in an e ective way, through which software developments can bene t jointly and simultaneously from the improvements and advantages provided by the model-driven engineering development approach and the continuous integration practice. The model-driven engineering approach is the last natural step of software engineering in the search for development approaches that raise the level of abstraction to the point that experts in a domain of knowledge, outside the computer world, are able to guide and change the logic of computer systems. The continuous integration practice is a recommendation of the most widely accepted development methodologies that aims to carry out automatic software integrations in early stages of development, o ering bene ts such as reducing the inherent risk that, given its unique nature, every project has. By merging the model-driven engineering and the continuous integration practice, the aim is to provide to development teams that work using some kind of model-driven engineering initiative, the possibility to integrate their developments in a continuous and distributed way. At the same time, customers, the real experts in the domain of knowledge in their eld of business, can bene t from the increased level of abstraction in developing techniques. Thus, they, in a transparent manner, are able to modify their own computer system without the help of external technical sta , so saving time and costs. To meet the objective of this Thesis, a prototype which saves all the current constraints that do not allow the union between these two new tools of software engineering is build. The main problems found were related to the selection of an appropriate development initiative, the version control systems specially adapted to working with models, the incremental generation of artifacts from models, and the optimized adaptation to existing continuous integration tools. The separation of work in di erent blocks can provide solutions, both in isolation or in conjunction, resulting in an iterative and incremental work from beginning to end. To analyze the bene ts of the proposal in this work compared to other development possibilities, an evaluation is performed by creating di erent test cases in which the measurement of di erent parameters can give a numerical estimate of the real bene ts obtained. The descriptive analysis, the hypothesis testing, and regression techniques allow a better interpretation of results. Finally, the process, the main objective of this work, is de ned by answering various questions posed to facilitate its comprehension and understanding.

Model-driven software development (MDSD) and component-based software development are both paradigms for reducing complexity and for increasing abstraction and reuse in software development. In this thesis, we aim at combining the advantages of each by introducing methods from component-based development into MDSD. In MDSD, all artefacts that describe a software system are regarded as models of the system and are treated as the central development artefacts. To obtain a system implementation from such models, they are transformed and integrated until implementation code can be generated from them. Models in MDSD can have very different forms: they can be documents, diagrams, or textual specifications defined in different modelling languages. Integrating these models of different formats and abstraction in a consistent way is a central challenge in MDSD. We propose to tackle this challenge by explicitly separating the tasks of defining model components and composing model components, which is also known as distinguishing programming-in-the-small and programming-in-the-large. That is, we promote a separation of models into models for modelling-in-the-small (models that are components) and models for modelling-in-the-large (models that describe compositions of model components). To perform such component-based modelling, we introduce two architectural styles for developing systems with component-based MDSD (CB-MDSD). For CB-MDSD, we require a universal composition technique that can handle models defined in arbitrary modelling languages. A technique that can handle arbitrary textual languages is universal invasive software composition for code fragment composition. We extend this technique to universal invasive software composition for graph fragments (U-ISC/Graph) which can handle arbitrary models, including graphical and textual ones, as components. Such components are called graph fragments, because we treat each model as a typed graph and support reuse of partial models. To put the composition technique into practice, we developed the tool Reuseware that implements U-ISC/Graph. The tool is based on the Eclipse Modelling Framework and can therefore be integrated into existing MDSD development environments based on the framework. To evaluate the applicability of CB-MDSD, we realised for each of our two architectural styles a model-driven architecture with Reuseware. The first style, which we name ModelSoC, is based on the component-based development paradigm of multi-dimensional separation of concerns. The architecture we realised with that style shows how a system that involves multiple modelling languages can be developed with CB-MDSD. The second style, which we name ModelHiC, is based on hierarchical composition. With this style, we developed abstraction and reuse support for a large modelling language for telecommunication networks that implements the Common Information Model industry standard.

Model-driven software development (MDSD) and component-based software development are both paradigms for reducing complexity and for increasing abstraction and reuse in software development. In this thesis, we aim at combining the advantages of each by introducing methods from component-based development into MDSD. In MDSD, all artefacts that describe a software system are regarded as models of the system and are treated as the central development artefacts. To obtain a system implementation from such models, they are transformed and integrated until implementation code can be generated from them. Models in MDSD can have very different forms: they can be documents, diagrams, or textual specifications defined in different modelling languages. Integrating these models of different formats and abstraction in a consistent way is a central challenge in MDSD. We propose to tackle this challenge by explicitly separating the tasks of defining model components and composing model components, which is also known as distinguishing programming-in-the-small and programming-in-the-large. That is, we promote a separation of models into models for modelling-in-the-small (models that are components) and models for modelling-in-the-large (models that describe compositions of model components). To perform such component-based modelling, we introduce two architectural styles for developing systems with component-based MDSD (CB-MDSD). For CB-MDSD, we require a universal composition technique that can handle models defined in arbitrary modelling languages. A technique that can handle arbitrary textual languages is universal invasive software composition for code fragment composition. We extend this technique to universal invasive software composition for graph fragments (U-ISC/Graph) which can handle arbitrary models, including graphical and textual ones, as components. Such components are called graph fragments, because we treat each model as a typed graph and support reuse of partial models. To put the composition technique into practice, we developed the tool Reuseware that implements U-ISC/Graph. The tool is based on the Eclipse Modelling Framework and can therefore be integrated into existing MDSD development environments based on the framework. To evaluate the applicability of CB-MDSD, we realised for each of our two architectural styles a model-driven architecture with Reuseware. The first style, which we name ModelSoC, is based on the component-based development paradigm of multi-dimensional separation of concerns. The architecture we realised with that style shows how a system that involves multiple modelling languages can be developed with CB-MDSD. The second style, which we name ModelHiC, is based on hierarchical composition. With this style, we developed abstraction and reuse support for a large modelling language for telecommunication networks that implements the Common Information Model industry standard.