Putting assertions in their place

Assertions that are placed at each statement in a program can automatically monitor the internal computations of a program execution. However, the advantages of universal assertions come at a cost. A program with such extensive internal instrumentation will be slower than the same program without the instrumentation. Some of the assertions may be redundant. The task of instrumenting the code with correct assertions at each location is burdensome, and there is no guarantee that the assertions themselves will be correct. We advocate a middle ground between no assertions at all (the most common practice) and the theoretical ideal of assertions at every location. Our compromise is to place assertions only at locations where traditional testing is unlikely to uncover software faults. One type of testability measurement, sensitivity analysis, identifies locations where testing is unlikely to be effective.<>