Co-change clustering

Abstract

Modularity is a key concept to embrace when designing complex software systems. Nonetheless, modular decomposition is still a challenge after decades of research on new techniques for software modularization. One reason is that modularization might not be viewed with single lens due to the multiple facets that a software must dealwith. Research in programming languages still tries to define new modularization mechanisms to deal with these different facets, such as aspects and features. Additionally, the traditional modular structure defined by the package hierarchy suffers from the dominant decomposition problem and it is widely accepted that alternative formsof modularization are necessary to increase developers productivity. In order to contribute with a solution to this problem, in this thesis we propose a novel technique to assess package modularity based on co-change clusters, which are highly inter-related source code files considering co-change relations. The co-change clusters are classifiedin six patterns regarding their projection to the package structure: Encapsulated, Well- Confined, Crosscutting, Black-Sheep, Octopus, and Squid. We evaluated our technique in three different fronts: (i) a quantitative analysis on four real-world systems, (ii) a qualitative analysis on six systems implemented in two languages to reveal developersperception of co-change clusters, (iii) a large scale study in a corpus of 133 GitHub projects implemented in six programming languages. We concluded that Encapsulated Clusters are often viewed as healthy designs and that Crosscutting Clusters tend to be associated to design anomalies when they represent tangled concerns. Octopus Clusters have a significant statistical association with ripple effect but they are normally associated to expected class distributions, which are not easy to implement in an encapsulated way.