5.8.9-to-be on Irix

The latest 5.8.x builds and passes all but one of its tests:
ext/Sys/Syslog/t/syslog.....................# Failed test 'syslog()
should return true: '0''
# at ../ext/Sys/Syslog/t/syslog.t line 108.
FAILED at test 12

but when I run that test by hand:

~/perl-33536/t $ ./perl -T -I../lib ../ext/Sys/Syslog/t/syslog.t \
|grep "not ok"

(note the -T, because syslog.t has that on the shebang line, which perl
doesn't like)

not ok 8 - setlogsock() should return true: '0' # TODO minor bug
not ok 12 - syslog() should return true: '0'
# Failed test 'syslog() should return true: '0''
# at ../ext/Sys/Syslog/t/syslog.t line 108.
not ok 14 - closelog() should return true: '0'
# Failed test 'closelog() should return true: '0''
# at ../ext/Sys/Syslog/t/syslog.t line 113.
not ok 110 - [stream] syslog() should return true: '0'
# Failed test '[stream] syslog() should return true: '0''
# at ../ext/Sys/Syslog/t/syslog.t line 168.
not ok 112 - [stream] syslog() should return true: '0'
# Failed test '[stream] syslog() should return true: '0''
# at ../ext/Sys/Syslog/t/syslog.t line 175.
not ok 114 - [stream] closelog() should return true: '0'
# Failed test '[stream] closelog() should return true: '0''
# at ../ext/Sys/Syslog/t/syslog.t line 184.
# Looks like you failed 5 tests of 199.

--
David Cantrell | Godless Liberal Elitist

Stepped on something soft and wobbly.
Struck a match.
Found it was a dead Chinaman.

-- Sir George Scott

David Cantrell wrote:

Sorry to answer only now. I just catch back on my P5P mails.

What do you mean by "it doesn't like -T"?

This looks like CPAN-RT#30710, for which I have no real idea of the
root of the problem.
» http://rt.cpan.org/Public/Bug/Display.html?id=30710

Could you tell me what was actually written in your syslog?

--
Sébastien Aperghis-Tramoni

Close the world, txEn eht nepO.

~/perl-33536/t $ ./perl -I../lib ../ext/Sys/Syslog/t/syslog.t
"-T" is on the #! line, it must also be used on the command line at
../ext/Sys/Syslog/t/syslog.t line 1.

May 6 15:22:59 6Q:o3k perl: uid 25020 is testing Perl 5.008008
syslog(3) capabilities by connecting to a native socket
May 6 15:22:59 6Q:o3k perl: uid 25020 is testing Perl 5.008008
syslog(3) capabilities by connecting to a native socket, setting a fake
errno: Operation not permitted

--
David Cantrell | Official London Perl Mongers Bad Influence

Safety tip: never strap firearms to a hamster

> ../ext/Sys/Syslog/t/syslog.t ...

FWIW, this works just fine in 5.8.8, which passes all its tests.

--
David Cantrell | Enforcer, South London Linguistic Massive

Suffer the little children to come unto me, as
their buying habits are most easily influenced.
-- Marketroid Jesus

5.8.8 was bundled with Sys::Syslog 0.13, however 5.8.9 wants to be
bundled with 0.18. Or do you mean 5.8.8 runs 0.18 just fine?

I think this calls for someone with bisecting skillz can find out what
change caused the regression.

David

A freshly untarred 5.8.8 source tree passed all its tests.

--
David Cantrell | http://www.cantrell.org.uk/david

Godliness is next to Englishness

Err yes. I must have meant octal. Or something.

"5.8.9 wants 0.18" ? I can't update my copies of blead and maint-5.8

> A freshly untarred 5.8.8 source tree passed all its tests.

So, IIUC, we currently have these results on IRIX:
- 5.8.8 : OK
- 5.8.9-tobe : FAIL ext/Sys/Syslog/t/syslog.t
- blead : FAIL ext/Sys/Syslog/t/syslog.t, ext/Time/Piece/t/
02core.t, lib/Archive/Extract/t/01_Archive-Extract.t

Given 5.8.8 has Sys::Syslog 0.13, and 5.8.9-tobe and blead both have
0.24, could you try the following combinations:
- 5.8.8 with 0.24 (will show if the regression was introduced by
Sys::Syslog)
- 5.8.9-tobe with 0.13 (will show if the regression was introduced
by something else in Perl)

Thanks in advance

OK, in both cases I built a fresh perl (5.8.9-to-be is Nicholas's build
33536), ignored the test failures in 5.8.9-to-be, and intalled to
~/perl-5.8.[whatever]. Then I did the untarred Sys-Syslog 0.13 and 0.24
and for each perl and each dist of the module did:
[path]/perl Makefile.PL
make
make test

0.24 came from the CPAN, 0.13 from the backpan.

That fails, but passes with 0.13. But according to CPAN-testers, 0.24
*passes* with 5.8.8 on some other platforms, so this is Irix-specific.

That passes.

With 0.24 I noticed this:

cc-1174 cc: WARNING File = Syslog.c, Line = 367
The variable "RETVAL" was declared but never referenced.

which might explain the failures.

So given that 0.24 fails with 5.8.8 it looks like the problem is in
Sys::Syslog, and not with perl itself. Tomorrow I'll grovel over the
backpan and see exactly when the bug was introduced.

--
David Cantrell | A machine for turning tea into grumpiness

You know you're getting old when you fancy the
teenager's parent and ignore the teenager
-- Paul M in uknot

0.13 through to 0.18 pass all their tests with 5.8.8 and with
5.8.9-to-be on Irix. 0.19 onwards all fail on both versions.

--
David Cantrell | top google result for "topless karaoke murders"

Feature: an incorrectly implemented bug

> On Fri, May 09, 2008 at 02:26:44PM +0200, Sébastien Aperghis-

Yes, sorry, I'm used to delete old versions in order to improve the
Schwartz factor.

Strange, this file hasn't changed much between 0.18 and 0.19, mostly
by a couple of #ifdef
Does this warning appears in XS_Sys__Syslog_setlogmask_xs()? If not,
could you paste me the part of the C file where it occurs?

> 0.13 through to 0.18 pass all their tests with 5.8.8 and with

Ok, so this means the problem has appeared between 0.18 and 0.19.
Unfortunately, 0.19 was a quite big release because it added Win32
support and numerous bugfixes. Also, this was the time I begun using
a SVN for Sys::Syslog and I made the mistake of importing a WIP
version instead of starting with a clean 0.18 release.

[ two hours of diff'ing and getting lost/distracted by googling later ]

Holy cow. I bet it's related to this change:
» http://svnweb.mongueurs.net/Sys-Syslog/revision/?rev=5

which I just realised was fixed very recently:
» http://svnweb.mongueurs.net/Sys-Syslog/revision/?rev=180
» http://rt.cpan.org/Public/Bug/Display.html?id=34691

So, could you try the version from the SVN and check whether this
fixes the problem?
$ svn co svn://svn.mongueurs.net/Sys-Syslog

Thanks for your hard work. I owe you your favourite drinking for the
next Perl conference :-)

It's here:

XS(XS_Sys__Syslog_setlogmask_xs)
{
dXSARGS;
if (items != 1)
Perl_croak(aTHX_ "Usage: Sys::Syslog::setlogmask_xs(mask)");
{
int mask = (int)SvIV(ST(0));
int RETVAL;
dXSTARG;
#line 129 "Syslog.xs"
setlogmask(mask);
#line 320 "Syslog.c"
}
XSRETURN(1);
}

All tests passed with both 5.8.8 and 5.8.9-to-be. The warning about
RETVAL is still there, and your MANIFEST is out of date - 'make distclean'
says:
Not in MANIFEST: t/cpan-rt-21516.t
Not in MANIFEST: t/cpan-rt-21866.t
Not in MANIFEST: t/cpan-rt-25488.t

Pity I'm not going to be at YAPC::Europe this year :-)

Irregular English:
ladies glow; gentlemen perspire; brutes, oafs and athletes sweat

And actually reading the source once again give me the answer: RETVAL
indeed is unused in setlogmask_xs(), which means its return value is
undefined, but this doesn't matter because setlogmask_xs() is never
called. Currently, it's always the Perl setlogmask() function that's
always used.

Anyway, I've fixed the bug, for the day setlogmask_xs() will be used.
Thanks for poking me on this.

This is normal: these tests really are for me and are not stable
enough to be used in the wild.

I'll attend the London Perl Workshop then ;-)