Quality assurance with dynamic meta modeling

One way to deal with the complexity of today’s software systems is model-driven development (MDD), where the target software system is first modeled on a very abstract level in a platform-independent way (e.g., by using UML use cases), and then—step by step—refined. The final, platform-specific model contains enough information to serve as input for code generation of the target system. MDD has several benefits: For instance, the (usually visual) modeling languages allow for better communication with stakeholders, which is particularly true when using domain-specific languages (DSLs), i.e., languages containing concepts of the problem domain. Another advantage is that the modeler’s task is simplified by the small complexity of getting from one to the next abstraction level, some steps of which are even applied using automatic model transformations. MDD is most beneficial if the modeling languages in use have a well-defined syntax and semantics. This is often true for the syntax and static semantics part, e.g. by using MOF metamodeling techniques as suggested by the Object Management Group. For the behavioral semantics, the situation is usually worse. For instance, the UML has a MOF-based syntax definition, but its behavioral semantics is defined with natural language, leaving room for ambiguities. The same is true for many DSLs. One reason for this is that semantics specification for behavioral modeling languages is a difficult task. This is where Dynamic Meta Modeling (DMM) comes into play. DMM is a semantics specification technique targeted at MOFbased modeling languages, where a language’s behavior is defined by means of graphical operational rules which change runtime models. The DMM approach has first been suggested by Engels et al. in 2000 [63]; Hausmann has then defined the DMM language on a conceptual level within his PhD thesis in 2006 [96]. Consequently, the next step was to bring the existing DMM concepts alive, and then to apply them to different modeling languages, making use of the lessons learned to improve the DMM concepts as well as the DMM tooling. The result of this process is the DMM++ method, which is presented within this thesis. Our contributions are three-fold: First, and according to our experiences with the DMM language, we have introduced new concepts such as refinement by means of rule overriding, and we have strengthened existing concepts such as the dealing with universal quantified structures or attributes. Second, we have developed a test-driven process for semantics specification: A set of test models is created, and their expected behavior is fixed. Then, the DMM rules are created incrementally, finally resulting in a DMM ruleset realizing at least the expected behavior of the test models. Additionally, we have defined a set of coverage criteria for DMM rulesets which allow to measure the quality of a set of test models. Third, we have shown how functional as well as non-functional requirements can be formulated against models and their DMM specifications. The former is achieved by providing a visual language for formulating temporal logic properties, which are then verified with model checking techniques, and by allowing for visual debugging of models failing a requirement. For the latter, the modeler can add performance information to models and analyze their performance properties, e.g. average throughput.