Subclass Instantiation Distribution

During execution, an objected-oriented program typically creates a large number of objects. This research considers the distribution of those objects that share a common superclass. If this distribution is uniform then all subclasses are equally likely to be instantiated. However, if not, then the lack of uniformity can be exploited by giving preferential treatment to the dominant class (or classes). For example, a tester might spend greater testing resources on the dominant class while an engineer refactoring the code might begin with a more dominant class. An experiment designed to investigate the distribution of subclass instantiations was performed using eight Java programs containing almost half a million lines of code and just over three thousand classes. The results show that outside a few infrequent instances, most distributions are heavily skewed.