[ On Saturday, May 5, 2001 at 03:42:21 (+0200), Matthias Buelow wrote: ]
> Subject: Re: bin/12838: new expr(1) is totally broken
>
> Seriously, he is right.. if nothing else, computer science education
> should've indoctrinated into each of us at least some common sense
> regarding tests, i.e., any program should (ideally) have a full
> regression test suite available, for a) verifying that it works
> exactly like the last version for a subset of input for which the
> old version works correctly, and b) a thoroughly crafted set of tests
> (often cannot be exhaustive of course) which probes the perimeters
> of the algorithm as to catch as many new errors as possible (you of course
> cannot show the absence of errors, as Dijkstra, who, to our general
> amusement, prowls lectures in bare-socks, has hammered into our collective
> brains already some time ago.)

Couldn't have said it better myself!  :-)

But of course such test suites don't write themselves....

The only even close-to-comprehensive test suites for `expr' conformance
that I'm aware of are proprietary....

At least there are the beginnings of some tests in
src/regress/bin/expr/Makefile now....

-- 
							Greg A. Woods

+1 416 218-0098      VE3TCP      <gwoods@acm.org>     <woods@robohack.ca>
Planix, Inc. <woods@planix.com>;   Secrets of the Weird <woods@weird.com>