Does size matter?: a preliminary investigation of the consequences of powerlaws in software

There is increasing evidence that many object-oriented software size metrics are characterised by scale-free, powerlaw distributions. This means programs will have arbitrarily large components, and the size of the largest component will increase as programs' overall size increases. This directly contradicts a crucial assumption of object-oriented design --- that large programs can be build by combining many small components. In this paper, we present a preliminary study of this contradiction. We illustrate the distribution of several size metrics over a corpus of 100 Java systems, and then investigate the largest classes (according to five size and complexity metrics) from one of those systems. We find that, while some large classes may be explained by code-generation or design patterns, most large classes were examples of poor object-oriented design.