Towards a conceptual framework for object oriented software metrics

The development of software metrics for object oriented (OO) languages is receiving increasing attention. We examine the reasons why this is a much more challenging problem than for conventional languages. It seems premature to develop and apply OO metrics while there remains uncertainty not only about the precise definitions of many fundamental quantities and their subsequent impact on derived metrics, but also a lack of qualitative understanding of the structure and behaviour of OO systems. We argue that establishing a standard terminology and data model will help provide a framework for both theoretical and empirical work and increase the chances of early success. One potential benefit is improvement of the ability to perform independent validation of models and metrics. We propose a data model and terminology and illustrate the importance of such definitions by examining the seemingly straightforward concept of the number of methods per class. We discuss the implications of ambiguities in definitions for a suite of metrics which has recently been proposed. Preliminary results from our analysis of industrial systems are presented.