Bug #45158 for POE-Test-Loops: 30_loops/*/sbk_signal_init.t failure ($SIG{ALRM} not preserved after "use POE")

Mon Apr 20 08:29:06 2009 dolmen [...] cpan.org - Ticket created

Subject:

30_loops/*/sbk_signal_init.t failure ($SIG{ALRM} not preserved after "use POE")

perl 5.8.0 on HP-UX B.11.11 with POE 1.004 See also ticket 19529. $ perl -Iblib/lib t/30_loops/select/sbk_signal_init.t 1..1 not ok 1 - normal SIGALRM dispatched # Failed test (/produits/supervision/agent/cpan/lib/perl5/POE/Test/Loops/sbk_signal_init.pm at line 34) # Looks like you failed 1 tests of 1. $ perl -Iblib/lib t/30_loops/io_poll/sbk_signal_init.t 1..1 not ok 1 - normal SIGALRM dispatched # Failed test (/produits/supervision/agent/cpan/lib/perl5/POE/Test/Loops/sbk_signal_init.pm at line 34) # Looks like you failed 1 tests of 1.

Tue Apr 21 00:17:32 2009 RCAPUTO [...] cpan.org - Correspondence added

Can you reproduce the problem on a version of Perl newer than 5.8.0? Perl 5.8.0 has a number of issues, and I'm hoping that a later release has already resolved this one. If the problem turns out to be a platform-specific issue with HP-UX, then a resolution may take a lot longer. I don't have an HP-UX machine to test against. On Mon Apr 20 08:29:06 2009, DOLMEN wrote: Show quoted text

> perl 5.8.0 on HP-UX B.11.11 with POE 1.004 > See also ticket 19529. > > > > $ perl -Iblib/lib t/30_loops/select/sbk_signal_init.t > 1..1 > not ok 1 - normal SIGALRM dispatched > # Failed test >

(/produits/supervision/agent/cpan/lib/perl5/POE/Test/Loops/sbk_signal_init.pm Show quoted text

> at line 34) > # Looks like you failed 1 tests of 1.

Tue Apr 21 00:17:32 2009 The RT System itself - Status changed from 'new' to 'open'

Tue May 05 09:21:47 2009 dolmen [...] cpan.org - Correspondence added

Le Mar. Avr. 21 00:17:32 2009, RCAPUTO a écrit : Show quoted text

> Can you reproduce the problem on a version of Perl newer than 5.8.0? > Perl 5.8.0 has a number of issues, and I'm hoping that a later release > has already resolved this one.

I'm confirming on another HP-UX machine with perl 5.8.8. Show quoted text

> If the problem turns out to be a platform-specific issue with HP-UX, > then a resolution may take a lot longer. I don't have an HP-UX > machine > to test against.

I will do my best to help you to help me. I can provide any information you need, except a direct access to a machine. Note that I have some issues with RT: I've not found where I can change the preferences to be always mailed when my bugs are updated.

Sat May 30 16:18:04 2009 RCAPUTO [...] cpan.org - Correspondence added

On Tue May 05 09:21:47 2009, DOLMEN wrote: Show quoted text

> Le Mar. Avr. 21 00:17:32 2009, RCAPUTO a écrit :

> > Can you reproduce the problem on a version of Perl newer than 5.8.0? > > Perl 5.8.0 has a number of issues, and I'm hoping that a later release > > has already resolved this one.

> > I'm confirming on another HP-UX machine with perl 5.8.8.

Do you mean that you have confirmed it with HP-UX and Perl 5.8.8? POE should not touch $SIG{ALRM}. If Perl 5.8.8 also exhibits the problem, then I suspect the test itself is at fault. Perhaps try rearranging the code to read: # ... everything before ... my $signal_dispatched = 0; sub dispatch_normal_signal { $signal_dispatched = 1 } BEGIN { $SIG{ALRM} = \&dispatch_normal_signal; } # ... everything after ... Perhaps something is going wrong when $SIG{ALRM} is initialized to refer to a subroutine that does not exist yet? Show quoted text

> Note that I have some issues with RT: I've not found where I can change > the preferences to be always mailed when my bugs are updated.

I'm sorry, but I don't know rt.cpan.org well enough to say if that is an option. Please ask rt-cpan-admin@bestpractical.com

Tue Jun 02 09:08:12 2009 dolmen [...] cpan.org - Correspondence added

Le Sam. Mai. 30 16:18:04 2009, RCAPUTO a écrit : Show quoted text

> Do you mean that you have confirmed it with HP-UX and Perl 5.8.8?

Yes. Show quoted text

> POE should not touch $SIG{ALRM}. If Perl 5.8.8 also exhibits the > problem, then I suspect the test itself is at fault. Perhaps try > rearranging the code to read: > > # ... everything before ... > > my $signal_dispatched = 0; > sub dispatch_normal_signal { $signal_dispatched = 1 } > BEGIN { > $SIG{ALRM} = \&dispatch_normal_signal; > } > > # ... everything after ...

It looks like that SIGALRM dispatching is not working event without POE involved, as I tried without the "use POE" and it fails too. The perldoc for 'alarm' says: It is usually a mistake to intermix "alarm" and "sleep" calls. ("sleep" may be internally implemented in your system with "alarm") Also the HP-UX sleep(3C) says: http://www.docs.hp.com/en/B2355-90695/sleep.3C.html If a SIGALRM is generated for a multi-threaded process, it may not be delivered to a thread currently in sleep(). See sigwait(2) man page for details. In a multi-threaded process delivery of a SIGALRM to a thread in sleep() simply causes sleep() to return without invoking the SIGALRM handler. So we can consider that this test is not reliable/portable. The test should be skipped if alarm+sleep is not reliable.

Tue Jun 02 11:00:20 2009 dolmen [...] cpan.org - Queue changed from POE to POE-Test-Loops

Tue Jun 02 11:01:28 2009 dolmen [...] cpan.org - Broken in 1.005 added

Fri Jul 24 11:36:41 2009 RCAPUTO [...] cpan.org - Correspondence added

Does this patch fix it? =================================================================== --- lib/POE/Test/Loops/sbk_signal_init.pm (revision 2596) +++ lib/POE/Test/Loops/sbk_signal_init.pm (working copy) @@ -29,7 +29,7 @@ use POE; alarm(1); -sleep 5; +select(undef, undef, undef, 5); ok($signal_dispatched, "normal SIGALRM dispatched");

Sat Jul 25 13:37:39 2009 RCAPUTO [...] cpan.org - Correspondence added

I committed the previous patch as revision 2599. I'll be releasing 1.021 shortly.

Sat Jul 25 13:37:40 2009 RCAPUTO [...] cpan.org - Status changed from 'open' to 'resolved'

Tue Jul 28 05:58:38 2009 dolmen [...] cpan.org - Correspondence added

Le Ven. Jul. 24 11:36:41 2009, RCAPUTO a écrit : Show quoted text

> Does this patch fix it?

Yes, I verified that the patch fixes the problem on HP-UX (POE::Test::Loop 1.005 patched and POE 1.006). However, could you add a comment in the source code to point to this bug? Using "select" instead of "sleep" is quite strange and deserves an explanation (not for me, but for your followers).

Tue Jul 28 05:58:39 2009 The RT System itself - Status changed from 'resolved' to 'open'

Tue Jul 28 06:08:22 2009 dolmen [...] cpan.org - Correspondence added

Le Mar. Jul. 28 02:58:38 2009, DOLMEN a écrit : Show quoted text

> Le Ven. Jul. 24 11:36:41 2009, RCAPUTO a écrit :

> > Does this patch fix it?

> > Yes, I verified that the patch fixes the problem on HP-UX > (POE::Test::Loop 1.005 patched and POE 1.006).

Verified on HP-UX (POE::Test::Loop 1.021 and POE 1.006) Show quoted text

> However, could you add a comment in the source code to point to this > bug? Using "select" instead of "sleep" is quite strange and deserves an > explanation (not for me, but for your followers).

Comment in 1.021 is perfect.

Tue Jul 28 11:56:23 2009 dolmen [...] cpan.org - Status changed from 'open' to 'resolved'

Tue Jul 28 11:56:23 2009 dolmen [...] cpan.org - Broken in 1.020 added

Tue Jul 28 11:56:23 2009 dolmen [...] cpan.org - Fixed in 1.021 added