Pattern-Based Trace Correlation Techniques to Compare Software Versions

Understanding the behavioural aspects and functional attributes of an ex-isting software system is an important enabler for many software engineering activi-ties including software maintenance and evolution. In this paper, we focus on under-standing the differences between subsequent versions of the same system. This al-lows software engineers to compare the implementation of software features in dif-ferent versions of the same system so as to estimate the effort required to maintain and test new versions. Our approach consists of exercising the features under study, generate the corresponding execution traces, and compare them to uncover similari-ties and differences. We propose in this paper to compare feature traces based on their main behavioural patterns instead of a mere event-to-event mapping. Two trace correlation metrics are also proposed and which vary whether the frequency of the patterns is taken into account or not. We show the effectiveness of our approach by applying it to traces generated from an open source object-oriented system.