Regression testing of bug-fixes with AI techniques

Abstract

Regression testing is to check whether a new software version fails to preserve specification properties that its previous versions have preserved. It is needed when new version has been filed for fixing a previously reported bug. We present artificial intelligence (AI) techniques to enhance the accuracy of test selection in such a setting. The core technique of our regression testing is an adaptive fitness function that learns how to select test cases based on their similarity to the bug-revealing test cases. The evaluation of similarity between two test cases is in turn based on feature variables extracted with the LCS (longest common subsequences) algorithm. We implemented our techniques with a symbolic simulator of the models of the software under test. We also show how to take the capabilities of the simulator into consideration to learn for better testing performance. Experiment report shows the potentials of applying AI techniques to software verification. © 2011 Springer-Verlag Berlin Heidelberg.