I am currently the test manager for an enterprise system. We test every functional requirement. In attributes with a fixed number of values, we test each variation and check for expected results. Tests are mapped to one or more requirements and vice-versa. For each test misstep, I write a test deficiency 'report' that includes the test number to aid in retesting and I track them. I must be able to show status, by requirement and by test. Not easy/cheap/quick, but surprises are get increasingly rare every time we deploy.
You can pay this cost up front, or you will pay at the end (and may not get done). Upfront is cheaper, by far. Thank you, Mark Snyder -----Original Message----- > Because it is virtually impossible for *everyone* to check > *everything* you want to absolve MS from checking even the simplest > things. You sure do know how to exaggerate. As I'm sure you well know, that is not at all what I said. What I said was, "You can't test everything," which is rather different from "You don't need to test anything". You have to pick and choose what you test. The Freescale chip is a complex device that does a LOT of stuff (see http://tinyurl.com/y78x3d). If I were responsible for testing it, and I didn't have sufficient time to test every bit of it, date arithmetic would not leap to the top of my to-do list. I'd spend whatever time I had available on more complicated stuff that has a higher probability of being wrong. ************************************************************************* ** List info, subscription management, list rules, archives, privacy ** ** policy, calmness, a member map, and more at http://www.cguys.org/ ** *************************************************************************
