An Input-Aware Method of Trace-Based Fault Diagnosis

Execution trace indicates which part of the software statements are involved in the execution, which is an important diagnosis basis of the several existed diagnosis methods. However, the features of the input of the trace have been ignored in these researches, which may lead to a wrong or oversized result and impact the efficiency of the diagnosis directly. In this paper, we introduce the trace input analysis into diagnosis, and propose an input- aware method, which can select the statements that need to be examined further of small amount. Then, the suspiciousness of each selected statement is calculated according to the number of passed traces that execute the statement. These statements with higher suspiciousness should be checked earlier. An experimental study is performed for several programs, together with another two trace-based diagnosis methods. The results show that our method can improve the accuracy and efficiency of the diagnosis.