AI for the Win: Improving Spectrum-based Fault Localization

Abstract

A considerable amount of time in software engineering is spent in debugging. In practice, mainly debugging tools which allow for executing a program step-by-step and setting break points are used. This debugging method is however very time consuming and cumbersome. There is a need for tools which undertake the task of narrowing down the most likely fault locations. These tools must complete this task with as little user interaction as possible and the results computed must be beneficial so that such tools appeal to programmers. In order to come up with such tools, we present three variants of the well-known spectrum-based fault localization technique that are enhanced by using methods from Artificial Intelligence. Each of the three combined approaches outperforms the underlying basic method concerning diagnostic accuracy. Hence, the presented approaches support the hypothesis that combining techniques from different areas is beneficial. In addition to the introduction of these techniq