On 29 Jul 2001 05:29:51 GMT esteemed Mama Cass Elliot did hold forth thusly:
> > behaves differently depending on data and user event input. There's no
> > way to test with every possible input combination.
>
> Nonsense!
>
> It's perfectly possible to design and implement such a test.
No it isn't.
> You know of all the possible inputs. You know of all the events that a user
> can trigger, you devise a list that has every possible combination, and
> then you go through that list. Yes - it may take more than 5 minutes, but
> it IS *possible*.
No it isn't. Look at just timing bugs. There are too many different combinations
of time lengths between different events. Timing bugs can be very specific to
which particular combination of intervals will cause a particular bug to manifest.
We live in an analog universe. The number of possible timing sequences is
literally infinite and probably incountably so (countable infinities being smaller
than uncountable infinities in mathematical theory).
But even without getting into timing bugs the number of possible inputs is still
infinite. Look at how many ways a web page can be designed. Look at the number
of tags and attributes the number of permutations possible. We are once again
into the land of the infinite.
Back during Reagan's Star Wars project David Parnas (a computer scientist) famously
resigned from a Star Wars advisory panel of the US DOD because in his view the
architecture was not testable in a way that would validate and verify the correctness
of any implementation. The same problem exists for many complex systems.
