Dynamic Analysis for Integration Faults Localization

Software developers usually integrate third party components to build systems providing distinct functionalities like graph drawing, or persistence capabilities. Developers often use grey-box components: software modules they do not know in details because provided without source code or incomplete specifications or both. Lack of source code and specifications makes the integration of such modules difficult and often causes faults that lead to critical failures if not detected at testing time. Lack of such informations complicates faults localization too. Existing static analysis and debugging techniques rely on source code or specifications, for this reason their applicability is often limited when grey-box components are used. Dynamic analysis techniques do not need such information: they monitor components interfaces, thus being applicable even when this information is missing. Dynamic analysis approaches identify violations of models inferred from data recorded during monitored executions. Unfortunately these techniques suffers from limitations too: they are capable of identifying only specific kinds of faults, their results are often affected by false positives, and they present scalability issues depending on the huge amount of data collected during training or on the identification of many violations during debugging. This paper presents Behaviour Capture and Test (BCT), a dynamic analysis technique that overcomes the limitations of the existing dynamic analysis techniques. BCT uses different kinds of models to localize different types of faults, it prunes false positives, it incrementally builds models to save disk space and guides developers when many models violations are identified. Successful results obtained when applying the technique on injected and real faults are reported in the paper: the considered case studies regard both regression faults and faults depending on rare events sequences not stressed at testing time. ∗Mauro Pezzè is also professor at the University of Lugano, Faculty of Informatics, via Buffi, 13 6900 Lugano (Switzerland).