Automatic test set generation for event-driven systems in the absence of specifications combining testing with model inference

The growing dependency of human activities on software technologies is leading to the need for designing more and more accurate testing techniques to ensure the quality and reliability of software components. A recent literature review of software testing methodologies reveals that several new approaches, which differ in the way test inputs are generated to efficiently explore systems behaviour, have been proposed. This paper is concerned with the challenge of automatically generating test input sets for Event-Driven Systems (EDS) for which neither source code nor specifications are available, therefore we propose an innovative fully automatic testing with model learning technique. It basically involves active learning to automatically infer a behavioural model of the System Under Test (SUT) using tests as queries, generates further tests based on the learned model to systematically explore unseen parts of the subject system, and makes use of passive learning to refine the current model hypothesis as soon as an inconsistency is found with the observed behaviour. Our passive learning algorithm uses the basic steps of Evidence-Driven State Merging (EDSM) and introduces an effective heuristic for choosing the pair of states to merge to obtain the target machine. Finally, the effectiveness of the proposed testing technique is demonstrated within the context of event-based functional testing of Android Graphical User Interface (GUI) applications and compared with that of existing baseline approaches.