An empirical investigation into contributory factors of change and fault propensity in large-scale commercial object-oriented software

Abstract

Object-Oriented design and development dominates both commercial and open source software projects. One of the principal goals of object-oriented design is to aid reuse, and hence, reduce future maintenance efforts of software systems. However, the on-going maintenance of large-scale software systems (both changes and faults) continues to be a significant proportion of the lifecycle of the system and the total investment cost. Understanding and thus being able to predict - or even reduce - the impact of the contributing factors of future maintenance efforts of a software system is thus highly beneficial to software practitioners. In this Thesis we empirically study a large, commercial software system with the principal aim to determine the contributing factors to the change and fault propensity over a three-year period. We consider the object-oriented design context of the software, specifically its inheritance characteristics, coupling and cohesion properties, object-oriented design pattern participation, and size. We also explore the effect of refactoring and test classes in the software. Our results show that several aspects of the design context of a class have an impact to the change and fault-proneness of the software. Specifically, we show that classes with high afferent or efferent coupling are more change and fault-prone; we also identify a number of design patterns whose participants tend to have a higher change and fault propensity than non-participants and we identify a range of inheritance characteristics (in terms of depth of inheritance and number of children) that result in an increase to change and fault-proneness. Furthermore we show that refactoring is a commonly occurring maintenance activity, although it is largely limited to simpler types of refactorings. Finally, we provide some insight into the co-evolution of production and test code during refactoring.