Reverse Engineering Systems to Identify Flaws and Understand Behaviour

Accurate system models are applicable to many software engineering tasks. Despite their utility, models are often neglected during development. It is therefore desirable to reverse engineer them from existing systems. One way to do this is to record traces of the system and infer a model by generalising from this behaviour.

Unfortunately, the models inferred by current techniques often cannot represent how the data values associated with each action affect system behaviour. This raises the following questions. What kind of model do we need in order to show the interplay between behaviour and data? How can we infer such models from system traces? How can we infer functions to relate input data with subsequent outputs? How can we use our models once they have been inferred?

To answer these questions, the first contribution of this thesis is a new model definition designed to show the relationship between data and behaviour. Secondly, I present a technique to infer such models from system traces, and define a preprocessing step to infer functions that relate system inputs and outputs. I then empirically evaluate the models produced by my technique and compare them to those produced by a state-of-the-art tool. Finally, I show how the inferred models can be used to analyse properties of the systems they represent.

The results show that my technique infers models which are more accurate and intuitive than the current state of the art. My tool can also handle circumstances where the output of a system depends on data values not present in the traces, and can identify situations where the result of particular actions depends on specific data values. The models inferred by my tool can be used by existing verification tools to prove and refute properties of the underlying systems.