Finding related events for specification mining

Abstract

In contemporary software development practice, programmers reuse components by invoking their APIs to construct large systems. These APIs often involve constraints on the temporal order of method calls. For the example of the file usage, a programmer should first open a file, then read and/or write its content, and at last close it. Trying to read a closed file will cause exceptions to be thrown. Such constraints are often represented as a finite state machine (FSM) with a set of related events (typically method calls) as its alphabet. A set of events are related if interactions among them possibly obey some meaningful temporal specifications. In recent years, various specification mining techniques have been developed to automatically mine API specifications from API client programs [1]. A typical API specification miner conceptually has three steps. First, it decide which events are related. Second, different interactions among related events (which are sub-traces and sample strings of the specification FSM) are extracted either from source code of client programs or from their execution traces. Third, extracted interactions are passed to customized or off-the-shelf FSM learners which generalize these sample sub-traces to recover the specification FSM.