Re: Ada Core Technologies and Ada95 Standards

[bobduff@world.std.com (Robert A Duff)]

In article <828475321.18492@assen.demon.co.uk>,
John McCabe <john@assen.demon.co.uk> wrote:
>As I said before, if I can't prove my software meets all of its
>requirements, my customer will not accept it.

I think different people are using the word "prove" differently.
Clearly, no set of tests can "prove" beyond a doubt that every
combination of inputs is handled correctly, given a moderate-to-complex
program.  I you disagree with that, then I think you're deeply confused.
On the other hand, it is possible to enumerate classes of inputs, and
"prove" (beyond a "reasonable" doubt?) that each such class is handled
correctly.  And it's perfectly reasonable to complain about some
particular test suite, that it doesn't cover a whole class of inputs
that one thinks it ought to.

I suggest that we all use "prove" in a more mathematical sense -- which
rules out any proof by testing.  Even in the mathematical sense, one has
to worry about whether the "proof" is truly correct.

- Bob