Consistency-based Software Fault Localization with Multiple Observations

Abstract

State-of-the-art implementations of model-based software fault localizationhave strong limitations in terms of their fault model with candidate locations mostly being restricted to a simple subset of the programming language ANSI-C. Moreover, with respect to multiple observations, recent software fault localization algorithms are unable to deliver minimal result sets. However, with the emerge of continuous integration frameworks, etc., it is common that a large numberof observations is available in real world applications.In order to tackle the limitations of existing algorithms, we present a novel approach for software fault localization on ANSI-C, which has the following two major contributions. First, we introduce a well defined fault model for ANSI-C, which addresses also complex features of the language including pointers and arrays. Second, our approach features efficient handling of multiple observations by deploying a recent model-based diagnosis algorithm which guarantees that result sets are minimal with respect to all given observations.We demonstrate significant improvements compared to existing methodson constructed examples as well as on TCAS - the most commonly used benchmark for software fault localization. A full implementation of our approach is available and was used to derive the experimental results.