Too darned big to test

The increasing size and complexity of software, coupled with concurrency and distributed systems, has made apparent the ineffectiveness of using only handcrafted tests. The misuse of code coverage and avoidance of random testing has exacerbated the problem. We must start again, beginning with good design (including dependency analysis), good static checking (including model property checking), and good unit testing (including good input selection). Code coverage can help select and prioritize tests to make you more efficient, as can the all-pairs technique for controlling the number of configurations. Finally, testers can use models to generate test coverage and good stochastic tests, and to act as test oracles.