Abstract

This thesis shows the results of an empirical experiment that used three industrial case studies to investigate the effect of coverage testing on software quality. Moreover, the experiment investigated the relative costs of coverage testing.

Several test suites were created that achieved 5 different coverage percentages for the metrics statement, branch and basis path coverage. Test suites consisted mostly of test cases that were created based on the test-driven development methodology.

We show that a positive correlation is likely to exist for each metric between the coverage percentage that is achieved by a test suite and the number of faults that are found. In most cases, increasing the coverage percentage for any of the tested metrics results in a higher number of faults found. We also show that despite using coverage metrics, even with high coverage percentages, a significant portion of faults is left undetected.

Moreover, we show that increasing the coverage percentage influences the costs by a linear increase. The costs were calculated using the number of test cases, number of tested functions and the total tested cyclomatic complexity.

Additional Resources

1. The paper