On the Conceptual Cohesion of Co-Change Clusters

Abstract

The analysis of co-change clusters as an alternative software decomposition can provide insights on different perspectives of modularity. But the usual approach using coarse-grained entities does not provide relevant information, like the conceptual cohesion of the modular abstractions that emerge from co-change clusters. This work presents a novel approach to analyze the conceptual cohesion of the source-code associated with co-change clusters of fine-grained entities. We obtain from the change history information found in version control systems. We describe the use of our approach to analyze six well established and currently active open-source projects from different domains and one of the most relevant systems of the Brazilian Government for the financial domain. The results show that co-change clusters offer a new perspective on the code based on groups with high conceptual cohesion between its entities (up to 69% more than the original package decomposition), and, thus, are suited to detect concepts pervaded on codebases, opening new possibilities of comprehension of source-code by means of the concepts embodied in the co-change clusters.