An Empirical Study of “Removed” Classes in Java Open-Source Systems

Abstract

Coupling is an omni-present and necessary feature of OO systems; ideally, classes with excessive coupling should be either refactored and/or removed from the system. However, a problem that immediately arises is the practical difficulty of effecting the removal of such classes due to the many coupling dependencies they have; it is often easier to leave classes where they are and ‘work around’ the problem. In this paper, we describe empirical coupling and size data of classes removed from multiple versions of four open-source systems. We investigated three related, research questions. First, does the amount of coupling influence the choice of removed class? Second, does class size play a role in that choice? Finally, is there a relationship between the frequency with which a class is changed and its point of removal from a system? Results showed a strong tendency for classes with low ‘fan-in’ and ‘fan-out’ to be candidates for removal. Evidence was also found of class types with high imported package and external call functionality being removed; finally, size, in terms of methods and lines of code did not seem to be a contributing factor to class removal. The research addresses an area that is often overlooked in the study of evolving systems, notably the characteristics and features of classes that disappear from a system.