Automatic diagnosis of software functional faults by means of inferred behavioral models

Software faults have a relevant impact on today economy. Modern software systems are often composed by different third-party components, of which developers have only a partial knowledge because of the absence of source code or complete specifications. The lack of complete knowledge about the integrated components is the principal cause of faults, and makes fault diagnosis difficult as well. The increase of maintenance costs and services unavailability is one of the consequences of difficult faults diagnosis. Different automatic diagnosis techniques have been developed to support developers in fault diagnosis activities. Unfortunately the existing techniques present several limitations: Log analysis techniques often require log specifications which are often not available; Fault localization techniques require source code which is often not available too; Automated debugging techniques detect faults only if their effects are visible from the application state; Anomaly detection techniques force developers to manually inspect and interpret a huge quantity of suspicious behaviors thus resulting not effective in practice. This PhD Thesis presents a framework for the diagnosis of software functional faults that advances the state of the art by: (1) Identifying and correlating different kinds of software misbehaviors, thus effectively describing faults with heterogeneous causes; (2) Automatically filtering false positives, thus permitting developers to inspect only anomalous behaviors useful to debug the application; (3) Identifying the cause effect relationships between multiple misbehaviors, thus simplifying the diagnosis activity; (4) Automatically interpreting software misbehaviors, thus permitting developers to immediately diagnose the fault; (5) Evaluating the solution with multiple third-party and industrial case studies, thus demonstrating the effectiveness of the solution.