Handling High-Level Model Changes Using Search Based Software Engineering
Usman Mansoor · Deep Blue (University of Michigan) · 2017
Model-Driven Engineering (MDE) considers models as first-class artifacts during the software lifecycle. The number of available tools, techniques, and approaches for MDE is increasing as its use gains traction in driving quality, and controlling cost in evolution of large software systems. Software models, defined as code abstractions, are iteratively refined, restructured, and evolved. This is due to many reasons such as fixing defects in design, reflecting changes in requirements, and modifying a design to enhance existing features. In this work, we focus on four main problems related to the evolution of software models: 1) the detection of applied model changes, 2) merging parallel evolved models, 3) detection of design defects in merged model, and 4) the recommendation of new changes to fix defects in software models. Regarding the first contribution, a-posteriori multi-objective change detection approach has been proposed for evolved models. The changes are expressed in terms of atomic and composite refactoring operations. The majority of existing approaches detects atomic changes but do not adequately address composite changes which mask atomic operations in intermediate models. For the second contribution, several approaches exist to construct a merged model by incorporating all non-conflicting operations of evolved models. Conflicts arise when the application of one operation disables the applicability of another one. The essence of the problem is to identify and prioritize conflicting operations based on importance and context – a gap in existing approaches. This work proposes a multi-objective formulation of model merging that aims to maximize the number of successfully applied merged operations. For the third and fourth contributions, the majority of existing works focuses on refactoring at source code level, and does not exploit the benefits of software design optimization at model level. However, refactoring at model level is inherently more challenging due to difficulty in assessing the potential impact on structural and behavioral features of the software system. This requires analysis of class and activity diagrams to appraise the overall system quality, feasibility, and inter-diagram consistency. This work focuses on designing, implementing, and evaluating a multi-objective refactoring framework for detection and fixing of design defects in software models.