Konzeptionelle Modellierung von Plugin-Systemen mit Petrinetzen

The complexity of modern software systems increases. Among other reasons this is due to the goal of software engineers to satisfy actual and future needs of many different users with one single system. In order to cope with the complexity of such systems it is a common solution in real-world software systems to integrate a plug-in system. A plug-in system enables users to adapt a slim core system to their individual needs after deployment. A user just has to add or remove pre-fabricated plug-ins in a simple configuration process. To further increase the comfort of the configuration process it becomes more and more common to implement dynamically reconfigurable plug-in systems. While plug-in systems are rather common to encounter in the real world, researchers brought rather little attention to the topic until now. Therefore, the goal of this thesis is to define, model, and implement essential concepts of plug-in systems. An additional challenge comes from the observation that many modern software systems – and thus, plug-in systems as well – are inherently concurrent systems. The possibilities and risks that may be deduced from this observation are caught within the modelling process. Some paradigms and techniques within the area of software engineering are already used to describe plug-in systems. These include the paradigm of component-based software engineering and techniques to describe software architectures or design patterns. However, the concepts formulated within these approaches show some weaknesses when it comes to the modelling of dynamic changes or concurrency. Therefore, they are extended by the paradigm of agent-oriented software engineering and the technique of modelling with reference nets. Both approaches have not been used within the context of plug-in systems up to now. The broad spectrum of real-world plug-in systems is outlined along some complex example systems, namely the Mozilla application suite and the Eclipse platform. These two large examples are complemented by some more systems. From this spectrum, essential properties of plug-in systems are drawn. These are simple (dynamic) reconfigurability, strong encapsulation, and the concept of intended extensibility. The conceptual model should satisfy further requirements. It should be able to define plug-in life-cycles, cope with multiple layers of management, and allow concurrency within the system. Furthermore, the conceptual model needs an intuitive visual representation with precise semantics. The concept of a plug-in is defined based on these characteristics and related to the concepts of objects, modules, components and agents. To enhance the definitions, the plug-in concepts are modelled based on reference nets along the principle of consecutive refinement (“implementation by specification”). The model covers all requirements mentioned above, does not restrict concurrency more than necessary and is supported by precise operational semantics. A comparison with the manifold characteristics of real-world plug-in systems illustrates the suitability of the model. Furthermore, the conceptual results are transferred to the Petri net tool Renew by implementing several prototypes.