Mining of Comprehensible State Machine Models for Embedded Software Comprehension

Abstract

Embedded legacy software contains a lot of expert knowledge that has been cumulated over many years. Therefore, it usually provides highly valuable and indispensable functionality. At the same time, it becomes more and more complex to understand and maintain. Mining of understandable models, such as state machines, from such software can greatly support developers in maintenance, evolution and reengineering tasks. Developers need to understand the software in order to evolve it. Existing state machine mining approaches are based on symbolic execution, which means enumeration of all paths. This quickly leads to path explosion problem. One effect of this problem on state machine mining is that the extracted models contain a very high number of states and transitions, and therefore are not useful for human comprehension. This means that additional measures towards comprehensibility of extracted state machines are required. To reach this goal, we introduced user interaction measures that can reduce the complexity of extracted state machines by reducing the number of states and transitions.