MFGP-MINER: MAXIMAL FREQUENT GRAPH PATTERN MINING FOR FAULT LOCALIZATION

With the increased workload and difficulty in software maintenance, the researches in automatic debug and software fault localization are more significant to improve software quality. This paper presents a simple framework of software fault localization. Firstly, software execution sequences are collected on the granularity level of basic blocks during the software testing phase. These software execution sequences are mapped as directed software execution graphs. Next, Dynamic BitCode (DBC) data structure is constructed by scanning the graph database just once. In order to discover feature nodes with software faults, this paper proposes MFGP-Miner (maximal frequent graph pattern mining) algorithm to mine maximal frequent graph patterns based on Dynamic BitCode (DBC) data structure. Finally, taking account of the executions set and the executions complementary set, a measure based on Ochiai is designed to calculate the suspicious value of feature nodes. These feature nodes are ranked to help programmers to find faults in descending order according to the suspicious value. Siemens benchmark test suite is used in our experiments, and experimental results display that our approach is both efficient and effective for locating software faults.