Introducing a new predicate network model constructed based on null hypothesis testing for software fault localization

Abstract

The aim of this paper is to introduce a new statistical approach for software fault localization. To this end, a novel weighted predicate tree, P-network, has been introduced. The main contribution of the paper is to consider the behavior of branch statements, namely predicates, together, in failing and passing executions and detect those predicates having different behavior as fault relevant predicates. In order to assess the difference in behaviors of predicates together a null hypothesis testing has been used. The predicates with higher different ratios in failing and passing runs are selected as the nodes of the P-network. By using a BFS method on P-network all faulty predicates could be found. After that, by ranking the faulty predicates we are able to find the most relevant faulty ones, which might help the debugger easily locate the bug. The experiments on Siemens test suite reveal promising results. © 2010 Springer-Verlag Berlin Heidelberg.