Partition Based Differential Testing for Finding Embedded Code Generation Bugs in Simulink

Engineers frequently generate embedded code from Simulink models for control applications. However, target applications using the code could behave unexpectedly, due to the bugs in code generation. In this study, we propose MOPART, the first model partition based differential testing method for code generation testing in Simulink. MOPART uses multiple-way network partitioning to generate diverse bug-triggering Simulink models to thoroughly exercise the code generation process. MOPART then finds bugs by analyzing the outputs of these Simulink models with differential testing. Experiments show that MOPART significantly outperforms existing approaches, which finds 11 confirmed code generation bugs in only two weeks.