A Function Dependency based Approach for Fault Localization with D*

Arpita Dutta, Rajib Mall

Abstract

We present a scheme for hierarchically localizing software faults. First the functions are prioritized based on their suspiciousness of containing a fault. Further, the bug is localized within the suspected functions at the specific statement level. In our approach, a new function dependency graph is proposed, and based on that function prioritization is performed. In order to differentiate between the functions with equal suspiciousness value, function complexity metrics are considered. We proposed two different dependency edge weighting techniques, viz., Distribution Specified Normalization (DSN) method, and Highest Weight Normalization (HWN) method. These techniques help to measure the relevance of an edge in propagating a fault. We use spectrum-based fault localization (SBFL) technique DStar(D∗) to localize the bugs at the statement level. We also extended our approach to localize multiple fault programs. Based on our experimental results, it is observed that using DSN and HWN scoring schemes, there is a reduction of 43.65% and 38.88% of statements examined compared to the well-accepted SBFL technique DStar(D∗) respectively.

Download


Paper Citation