Perfect Software and other illusions about testing

is the title of an excellent book by Jerry Weinberg. As usual I'm going to quote from a few pages:
Testing a system is a process of gathering information with the intent that the information could be used for some purpose.
Without a process that includes regular technical reviews, no project will rise above mediocrity, no matter how good its machine-testing process.
At least half your testing costs can be cut before anybody ever runs a test, if only your systems are designed with testability in mind.
"We absolutely need this software in place twenty-four weeks from tomorrow. We need two weeks to staff up and get approvals. Then we'll need four weeks for requirements, four weeks for architecture. four weeks for design, and eight weeks for coding. That adds up to twenty-two weeks, so we'll have two weeks left for testing."
If you have ten kilograms of pure uranium-235 and you add another ten kilograms, you'll have twenty kilograms. But if you do this a few more times, you won't have fifty kilograms, you'll have a nuclear explosion. One plus one doesn't always equal two.
The more bugs you find, the more you're going to find, not the other way around.
The human mind craves meaning. If you feed people a random bit of data, they'll struggle to divine meaning from it - and they'll move from the intake phase to the meaning phase so fast they won't be aware of doing so.
You have to know what you're expecting before you give meaning to a test report, otherwise everything looks or sounds right. That's why I'm a strong advocate of the test-first philosophy, whereby developers write their tests to include expected results before they write a line of code. It's what we did fifty years ago, but the practice was gradually lost when industry trends separated testing from development.
That separation occurred initially because it's psychologically difficult for people to test their own programs. There's still significant risk if you rely on test-first without pair-programming or some other process that casts more than one pair of eyes, and more than one brain, on a program.
If test-first is a good idea, then significance-first is even better. Why? … if you actually perform even an enormous number of tests, you would likely lose the valuable information among all the worthless crud. The number of tests performed should be as small as possible, but no smaller.
"We didn't have any problems until we started testing. We were right on schedule. Testing screwed up everything."
"What the American public wants in the theatre is a tragedy with a happy ending." [William Dean Howells]

No comments:

Post a Comment