A software metric for logical errors and integration testing effort

Many software metrics are based on analysis of individual source code modules and do not consider the way that modules are interconnected. This presents a special problem for many current software development project environments that utilize a considerable amount of commercial, off-the-shelf or other reusable software components and devote a considerable amount of time to testing and integrating such components. We describe a new metric called the BVA metric that is based on an assessment of the coupling between program subunits. The metric is based on the testing theory technique known as boundary value analysis. For each parameter or global variable in a program module or subunit, we compute the number of test cases necessary for a "black-box" test of a program subunit based on partitioning that portion of the domain of the subunit that is affected by the parameter. The BVA metric can be computed relatively early in the software life cycle. Experiments in several different languages and both academic and industrial programming environments suggest a close predictive relationship with the density of logical software errors and also with integration and testing effort.