Automated Extraction of Rich Software Models from Limited System Information

Reverse engineering a software system is challenged by the typically very limited information available about existing systems. Useful reverse engineering tasks include recovering a system's architectural, behavioral, and usage models, which can then be leveraged to answer important questions about a system. For example, using such models to analyze and predict a system's non-functional properties would help to efficiently assess the system's current state, planned adaptations, scalability issues, etc. Existing approaches typically only extract a system's static architecture, omitting the dynamic information that is needed for such analyses. The contribution of this paper is an automated technique that extracts a system's static architecture, behavior, and usage models from very limited, but readily available information: source code and test cases. These models can then be fed into known performance, reliability, and cost prediction techniques. We evaluated our approach for accuracy against systems with already established usage models, and observed that our approach finds the correct, but more detailed usage models. We also analyzed 14 open source software systems spanning over 2 million lines of code to evaluate the scalability of our approach.