If one programs with the idea that everything must be right from the
start then testing is superfluous. One might, or might not, actually
do some tests. On the other hand, if testing is seen as a necessity,
required to catch bugs, then one only obtains a program that works
right in the common cases.
This is a religious issue, I'm afraid. There are those who claim (and I
would be one of them) that the problem is with the regression test suite
--- it should be able to test all of the corner cases. (That's what
code coverage tools are all about.)
I do both. I try to write the code to be correct, but I also know that
I'm human, and I make mistakes. When I wrote the mke2fs code, I checked
it and double checked it. But there will always be something that slips
through the cracks. *That's* why I believe testing is a necesity,
because it works as a safety measure in the times when you've screwed
up.
One could make the argument that seat belts in cars are superflous ---
after all, it's much better to simply "drive safely". But I'm paranoid
enough to (a) try to drive safely and (b) still wear seat belts.
- Ted