Abstract
This paper presents multiple empirical experiments that investigate the impact of fault quantity and fault type on statistical, coverage-based fault localization techniques and fault-localization interference. Fault-localization interference is a phenomenon revealed in earlier studies of coverage-based fault localization that causes faults to obstruct, or interfere, with other faults’ ability to be localized. Previously, it had been asserted that a fault-localization technique’s effectiveness was negatively correlated to the quantity of faults in the program. To investigate these beliefs, we conducted an experiment on six programs consisting of more than 72,000 multiple-fault versions. Our data suggests that the impact of multiple faults exerts a significant, but slight influence on fault-localization effectiveness. In addition, faults were categorized according to four existing fault-taxonomies and found no correlation between fault type and fault-localization interference. In general, even in the presence of many faults, at least one fault was found by fault localization with similar effectiveness. Additionally, our data exhibits that fault-localization interference is prevalent and exerts a meaningful influence that may cause a fault’s localizability to vary greatly. Because almost all real-world software contains multiple faults, these results affect the practical use and understanding of statistical fault-localization techniques.
