Classifying test suite effectiveness via model inference and ROBBDs

Abstract

Deciding whether a given test suite is effective enough is certainly a challenging task. Focusing on a software program’s functionality, we propose in this paper a new method that leverages Boolean functions as abstract reasoning format. That is, we use machine learning in order to infer a special binary decision diagram from the considered test suite and extract a total variable order, if possible. Intuitively, if an ROBDD derived from the Boolean functions representing the program under test’s specification actually coincides with that of the test suite (using the same variable order), we conclude that the test suite is effective enough. That is, any program that passes such a test suite should clearly show the desired input-output behavior. In our paper, we provide the corresponding algorithms of our approach and their respective proofs. Our first experimental results illustrate our approach’s practicality and viability.