Front page | perl.perl5.porters |
Postings from September 2003
I'm catching up on perl mailing lists... Anyway this thread
is the first non-contrived and non-paranoid possible use I've seen for
Acme::Landmine
On Sun, 2003-07-20 at 16:59, Ed Avis wrote:
> On 20 Jul 2003, Kurt Starsinic wrote:
>
> > What should it do this case?
> >
> > use constant N => 0;
> > exit if N == 0;
> > DUMMY:
> > print 1 / N;
> > eval 'goto DUMMY'; # Or "eval <STDIN>"
>
> The '1 / 0' should be an error if it is reached in code. The simplest
> way to do that is to make it a run-time error.
>
> (The above program would exit() on the second line.)
>
> >>#include <stdio.h>
> >>int main(void)
> >>{
> >> const int N = 0;
> >> if (N == 0) exit();
> >> printf("%d\n", 1 / N);
> >> return 0;
> >>}
>
> > gcc 3.2.1 complains. It doesn't even have to have -Wall on, just -O.
>
> True; but it's a warning, not a hard error. Otherwise turning on -O
> would break a previously legal program, and that would be a bug in
> gcc.
>
> > If you wanted to code up a patch to avoid folding "1 / N" if
> >it's NaN, I probably wouldn't complain. However, I consider it a
> >feature that it's a compile time error.
>
> It does make it rather tricky to generate strings of Perl code and run
> them, or to provide tweakable constant parameters which can be
> zero. You have to resort to workarounds like stuffing the constant
> into a variable just so the compiler won't be able to 'optimize' the
> code and break it.
>
> But without a language specification to consult, I can't give any
> hard evidence disagreeing with 'it's not a bug, it's a feature'. I
> will try to write a patch.
>
> Probably 1 / 0 should be turned into an expression that die()s with a
> division by zero error - that way it will still be as efficient as
> possible ;-).
--
David Nicol / If at first you don't succeed, use a bigger hammer.
http://gallaghersmash.com