> Here are my ideas on bugs: Bugs happen! They're a fact of life,
> omnipresent in all software.
Showstopper bugs should not slip through testing and into release
software.
Correct: _should_ not. That does not mean _will_ not. Mistakes
happen, at least as often as accidents. If testing could be perfected
in this manner, then programs themselves could be perfected.
> Bugs should allways be caught in the testing, but so often they
> aren't.
Minor bugs yes, massive showstoppers no.
How can a bug be known to be a show stopper until it has been found?
For that matter, I seem to recall that it cannot be proven that a
program will _halt_, let alone do something useful, let alone do
something useful without any bugs in any circumstances.
> In the meantime, a subgroup of testers have been created which
> should (hopefully) ensure that things like this cannot take place
> again.
Yes... a good idea. With a shiny new lock on that barn door,
perhaps these horses won't escape a second time and cost GIMPS
hundreds of P90-years of time... I still think it would have been a
good idea to have had this lock from the outset. But it's water
under the bridge...
Actually, a somewhat different "lock" has been on this barn door for
years (since not long after GIMPS started at the latest), in the form
of other programs, implemented independently and able to run on
different hardware. I'm not sure about this bug in particular, but
earlier bugs (in, at least, Prime95's factoring code and many of the
mers package programs) were only discovered by comparing their results
with results of those independent programs.
The new subgroup of testers is still a good idea, of course, but they
will only improve the quality of the releases, not make them bug-free.
Anyone expecting otherwise has quite a bit to learn about programming.
Will
________________________________________________________________
Unsubscribe & list info -- http://www.scruz.net/~luke/signup.htm
