Towards using coupling measures to guide black-box integration testing in component-based systems

In component-based software development, integration testing is a crucial step in verifying the composite behaviour of a system. However, very few formally or empirically validated approaches are available for systematically testing if components have been successfully integrated. In practice, integration testing of component-based systems is usually performed in a time- and resource-limited context, which further increases the demand for effective test selection strategies. In this work, we therefore analyse the relationship between different component and interface coupling measures found in literature and the distribution of failures found during integration testing of an automotive system. By investigating the correlation for each measure at two architectural levels, we discuss its usefulness to guide integration testing at the software component level as well as for the hardware component level where coupling is measured among multiple electronic control units (ECUs) of a vehicle. Our results indicate that there is a positive correlation between coupling measures and failure-proneness at both architectural level for all tested measures. However, at the hardware component level, all measures achieved a significantly higher correlation when compared to the software-level correlation. Consequently, we conclude that prioritizing testing of highly coupled components and interfaces is a valid approach for systematic integration testing, as coupling proved to be a valid indicator for failure-proneness.