Improving test coverage measurement for reused software

Test coverage adequacy measures provide a widely used stopping criterion. Engineering of modern software-intensive systems emphasizes reuse. In the case that a program uses reused code or third-party components in a context that is different from the original one, some of their entities (e.g. branches) might never be exercised, thus producing a code coverage level far from full and not meaningful anymore as a stopping rule for the program at hand. We introduce a new coverage criterion, called "Relevant Coverage", that in each testing context in which a code is reused calculates coverage measures over the set of relevant entities for that context. We provide an approach for identifying relevant entities using dynamic symbolic execution. The introduced coverage adequacy criterion is assessed in an exploratory study against traditional coverage in terms of test suite size reduction factor, cost-effectiveness ratio and rate of fault detection. The results of our study showed that relevant coverage can considerably reduce the test suite size while preserving a high cost-effectiveness ratio with respect to the traditional approach.