Toward Understanding Information Models of Fault Localization: Elaborate is Not Always Better

Fault localization defines information models from raw runtime information as the input, depicting program behaviors for supporting localization algorithms. It is natural that an elaborate information model is desirable and likely to improve the effectiveness of fault localization, because it typically depicts subtle and more program runtime behaviors. In fact, much work on fault localization assumes and exploits this intuition. However, there has been no large-scale study to confirm or refute this folklore. This paper fills this void—indeed, an animated debate on this topic has led to this work. We present a large-scale empirical study for a deeper understanding of the impact of information models on fault localization. Specifically, our study evaluates four representative information models and reveals that an elaborate information model has no strong correlation with localization effectiveness. Furthermore, based on the results, we analyze and suggest the directions of designing information models for fault localization on an extensively studied topic.