No abstract

Abstract. An important demand in Model-Driven Development is the simple and efficient expression of model patterns. Current approaches tend to distinguish the language they use to express patterns from the one for modelling. Consequently, productivity is reduced by dealing with a distinct new language, and new intermediate steps are introduced in order to support pattern-matching. In this paper we propose a framework for expressing patterns as model-snippets. We present how modelsnippets are specified upon concepts in a given domain (meta-model), and how we perform pattern-matching with model-snippets, whatever the meta-model. We also provide an implementation which is well integrated with existing technologies, such as Eclipse Modelling Framework.

The lack of a formal semantics for UML-RT makes it inadequate for rigourous system development, especially if the preservation of behaviour is a major concern when applying well-known model transformations, like refactorings and refinements. In this paper, we propose a semantics for UML-RT active classes through a mapping into Circus, a specification language that combines CSP, Z and specification statements. As a consequence of the translation, we are able to prove that model transformations preserve both static and dynamic behaviour, using refinement laws and a relational semantics of Circus, based on the Unifying Theories of Programming.

In previous work we presented a CSP-based systematic approach that fosters the rigorous design of component-based development (CBD). Our approach is strictly defined in terms of composition rules, which are the only permitted way to compose components. These rules guarantee the preservation of properties (particularly deadlock-freedom) by construction in component composition. Nevertheless, their application is allowed only under certain conditions whose verification via model checking turned out impracticable even for some simple designs, and particularly those involving cyclic topologies. In this paper, we address the performance of the analysis and present a significantly more efficient alternative to the verification of the rule side conditions, which are improved by carrying out partial verification on component metadata throughout component compositions and by using behavioural patterns. The use of metadata, together with behavioural patterns, demands new composition rules, which allow previous exponential time verifications to be carried out now in linear time. Two case studies (the classical dining philosophers, also used as a running example, and an industrial version of the leadership election algorithm) are presented to illustrate and validate the overall approach.

No abstract