Dynamic cohesion measures for object-oriented software

Most of the object-oriented cohesion metrics proposed in the literature define static cohesion at class level. Measurement of object-level dynamic cohesion however gives better insight into the behavioural aspects of the system. In this paper, dynamic cohesion metrics are introduced which provide scope of cohesion measurement up to object level and take into account important and widely used object-oriented features such as inheritance, polymorphism and dynamic binding during measurement. A theoretical framework is introduced before defining the measures and a theoretic validation of the proposed measures is carried out to make them more meaningful. A dynamic analyser tool is developed using aspect-oriented programming (AOP) to perform dynamic analysis of Java applications for the purpose of collecting run-time data for computation of the proposed dynamic cohesion measures. Further, an experiment is carried out for the proposed dynamic cohesion metrics using 20 Java programs and this study shows that the proposed dynamic cohesion metrics are more accurate and useful in comparison to the existing cohesion metrics. Moreover, the proposed dynamic cohesion metrics are validated empirically using source code APIs of Java Development Kit (JDK) and the proposed metrics are found to be better indicators of change-proneness of classes than the existing cohesion metrics.