Front page | perl.perl5.porters |
Postings from December 2000
On Mon, Dec 18, 2000 at 02:33:57PM +0000, Ton Hospel wrote:
> In article <20001217162924.E97668@plum.flirble.org>,
> Nicholas Clark <nick@ccl4.org> writes:
> > static IV
> > constant(char *name)
> > {
> > errno = 0;
> > switch (*name) {
> > case '_':
> > if (strEQ(name, "_S_IFMT")) /* Yes, on name _S_IFMT return S_IFMT. */
> >
>
> Hey ! what is that evil setting of errno ?
I didn't change that.
> perl -wle 'use Fcntl; $!=1; print O_NDELAY; print $!'
> perl -wle 'use Fcntl; $!=1; print $!'
>
> That is just sooo wrong !
5.004 is the earliest perl I have to hand
$ /usr/bin/perl-5.004 -wle 'use Fcntl; $!=1; print O_NDELAY; print $!'
2048
$ /usr/bin/perl-5.004 -wle 'use Fcntl; $!=1; print $!'
Operation not permitted
I think that it has been this way since 5.000, but I can't easily verify.
Note, the above doesn't express an opinion about wrongness or rightness.
My opinion is to agree that it's wrong. But I am cautious about making
changes that would be visible to the program (however trivial) that it
may have relied upon.
[says he who would like to change the internals of all the maths ops and
numeric conversions to like integers]
Previous threads about errno setting include one where I said:
However, it is irritating when things succeed but set errno. I'm
thinking of op/die_exit.t failing under sfio as an example.
http://www.xray.mpe.mpg.de/mailing-lists/perl5-porters/2000-10/msg00705.
and the conclusion of Hugo was
It would not be in such a minority: it is quite common for applications
to set errno to zero for similar purposes. As has been mentioned, system
library calls do not guarantee to preserve errno just because they are
successful; I see no benefit to perl offering such a guarantee itself -
anyone who needs it can save it easily enough themselves.
http://www.xray.mpe.mpg.de/mailing-lists/perl5-porters/2000-10/msg00726.html
Hugo's position is consistent and well stated. I can't find arguments that
would convincingly undermine it.
Nicholas Clark
Thread Previous
|
Thread Next