Yann-Gaël Guéhéneuc

Design patterns are important in object-oriented software engineering. They contribute to the quality of programs. They offer design motifs, elegant solutions to recurrent design problems. After implementation, the motifs are disseminated in source code. They are not available directly for maintenance; however they would help in understanding program implementation and design and in ensuring the quality of programs after maintenance. This thesis explores models and algorithms to identify semi-automatically micro-architectures in source code, which are similar to motifs, and to ensure their traceability between implementation and reverse engineering phases. Meta-modeling is used to describe design motifs and Java programs. It leads to characterize certain interclass relations (association, aggregation, and composition) offered by design languages, such as UML, to precise their properties (access type, lifetime, exclusivity, and multiplicity), and to identify them with static and dynamic analyses. It also leads to translate motifs into constraint systems and to identify micro-architectures, which are similar to motifs (complete and distorted forms), by solving constraint satisfaction problems. Explanation-based constraint programming allows guiding the solve interactively and explaining identified micro-architectures. The Ptidej tool suite implements the proposed models and algorithms and is integrated with the Eclipse development environment for Java. It includes the PADL meta-model, derived from the PDL meta-model; Static and dynamic analysis tools, Introspector and Caffeine; A constraint solveur, PtidejSolver, derived from the PaLM explanation-based constraint solver.