Automatic Fault Localization (AFL) is a process to locate faults automatically in software programs. Essentially, an AFL method takes as input a set of test cases including failed test cases, and ranks the statements of a program from the most likely to the least likely to contain a fault. As a result, the efficiency of an AFL method depends on the "quality" of the test cases used to rank statements. More specifically, in order to improve the accuracy of their ranking within test budget constraints, we have to ensure that program statements are executed by a reasonably large number of test cases which provide a coverage as uniform as possible of the input domain. This paper proposes MuTIL, a new statistical test inputs generation method dedicated to AFL, based on constraint solving and mutation testing. Using mutants where the locations of injected faults are known, MuTIL is able to significantly reduce the length of an AFL test suite while retaining its accuracy (i.e., the code size to examine before spotting the fault). In order to address the motivations stated above, the statistical generator objectives are two-fold: 1) each feasible path of the program is activated with the same probability, 2) the sub domain associated to each feasible path is uniformly covered. Using several widely used ranking techniques (i.e., Tarantula, Jaccard, Ochiai), we show on a small but realistic program that a proof-of-concept implementation of MuTIL can generate test sets with significantly better fault localization accuracy than both random testing and adaptive random testing. We also show on the same program that using mutation testing enables a 75% length reduction of the AFL test suite without decrease in accuracy.
展开▼
