Preprocessing of Metrics Measurement Based on Simplifying Program Structures

In software maintenance, grasping characteristics of software systems by metrics measurement is a basic activity. However, metrics do not always represent characteristics of software systems. For example, Cyclomatic Complexity is a metric counting the number of branches in a given module, and it does not consider its content. One factor that Cyclomatic Complexity becomes large is the presence of repeated structures such as consecutive if-else structures. However, if such a structure is a repetition of simple operations, humans would not recognize a difficulty to understand the source code. In this paper, we propose performing preprocessing for metrics measurement and a methodology of the preprocessing. The proposed preprocessing simplifies repeated structures in source code. By applying the proposed preprocessing to metrics measurement, we can find low-understandability modules more efficiently. Also, we compared results of metrics measurement with and without the proposed preprocessing on open source software systems. As a result, we confirmed that metrics measurement with the proposed preprocessing was more useful to find low-understandability modules than without it.