Software Re-Modularization Based on Structural and Semantic Metrics

The structure of a software system has a major impact on its maintainability. To improve maintainability, software systems are usually organized into subsystems using the constructs of packages or modules. However, during software evolution the structure of the system undergoes continuous modifications, drifting away from its original design, often reducing its quality. In this paper we propose an approach for helping maintainers to improve the quality of software modularization. The proposed approach analyzes the (structural and semantic) relationships between classes in a package identifying chains of strongly related classes. The identified chains are used to define new packages with higher cohesion than the original package. The proposed approach has been empirical evaluated through a case study. The context of the study is represented by an open source system, JHotDraw, and two software systems developed by teams of students at the University of Salerno. The analysis of the results reveals that the proposed approach generates meaningful re-modularization of the studied systems, which can lead to higher quality.