Discrimination of Class Inheritance Hierarchies – A Vector Approach

Abstract

Numerous inheritance metrics have been proposed and studied in the literature with a view to understand the effect of inheritance on software performance and maintainability. These metrics are meant to depict the inheritance structures of classes and related issues. However, in spite of a large number of inheritance metrics introduced by researchers, there is no standard set of metrics that could discriminate the class hierarchies to decipher or predict the change-proneness, defect-proneness of classes or issues that could effectively address maintainability, testability and reusability of class hierarchies. In fact, very different hierarchical structures lead to the same values of some standard inheritance metrics, resulting in lack of discrimination anomaly (LDA). In an effort to address this problem, three specific metrics have been studied from the point of view of providing an insight into inheritance patterns present in the software systems and their effect on maintainability. Empirical analysis shows that different class hierarchies can be distinguished using the trio – average depth of inheritance, specialization ratio and reuse ratio.