How much testing is enough?

We've already established that untested code is broken code. But how can we tell how well our code is tested? How do we know how much of our code is actually being tested and how much is broken? The first question is the more important one, but it's hard to answer. Even if we know we have tested every line of code in our application, we do not know that we have tested it properly. For example, if we write a stats test that only checks what happens when we provide a list of integers, it may still fail spectacularly if used on a list of floats or strings or self-made objects. The onus of designing complete test suites still lies with the programmer.

The second question—how much of our code is actually being tested—is easy to verify. Code coverage is essentially an estimate of the number of lines of code that are executed by a program. If we know that number and the number of lines that are in the program, we can get an estimate of what percentage...