On the Relationship between Program Evolution and Fault-Proneness: An Empirical Study

Abstract

Abstract-Over the years, many researchers have studied the evolution and maintenance of object-oriented source code in order to understand the possibly costly erosion of the software. However, many studies thus far did not link the evolution of classes to faults. Since (1) some classes evolve independently, other classes have to do it together with others (co-evolution); and (2) not all classes are meant to last forever, but some are meant for experimentation or to try out an idea that was then dropped or modified. In this paper, we group classes based on their evolution to infer their lifetime models and coevolution trends. Then, we link each group's evolution to faults. We create phylogenetic trees showing the evolutionary history of programs and we use such trees to facilitate spotting the program code decay. We perform an empirical study, on three open-source programs: ArgoUML, JFreechart, and XercesJ, to examine the relation between the evolution of object-oriented source code at class level and fault-proneness. Our results indicate that (1) classes having a specific lifetime model are significantly less fault-prone than other classes and (2) faults fixed by maintaining co-evolved classes are significantly more frequent than faults fixed using not co-evolved classes.

Anti-pattern Mutations and Fault-proneness

“…Indeed, during the evolution of the system, several classes can be renamed or divided into two files with different names. To mitigate such cases, we identified class renamings and class structural changes using a previous approach ADvISE [26] [27]. ADvISE identifies class renamings using the structurebased and the text-based metrics, e.g., their common methods, attribute types, and relationships, which assess the similarities between original and renamed classes.…”