Class Cohesion Revisited: An Empirical Study on Industrial Systems

The assessment of the changeability of software systems is of major concern for buyers of the large systems found in fast-moving domains such as telecommunications. One way of approaching this problem is to investigate the dependency between the changeability of the software and its design, with the goal of finding design properties that can be used as changeability indicators. In the realm of objectoriented systems, experiments have been conducted showing that coupling between classes is such an indicator. However, class cohesion has not been quantitatively studied in respect to changeability. In this research, we set out to investigate whether low cohesion is correlated to high coupling and thus is a changeability indicator, too. As cohesion metrics, LCC and LCOM were adopted, and for measuring coupling, the Chidamber and Kemerer coupling metrics and variants thereof were used. The data collected from three test systems of industrial size indicate no such correlation. Suspecting that the cohesion metrics adopted for the experiment do not adequately capture the cohesion property, we analyzed manually the classes with lowest cohesion values. We found various reasons why these classes, despite of their low cohesion values, should not be broken into smaller classes. We conclude that cohesion metrics should not only be based on common attribute usage between methods and on method invocation, but also on patterns of interaction between class members and, ultimately, on the functionality of methods