Bug #84051 for IO-Async: Testsuite fals if you're running on a machine with sshd running

Tue Mar 19 08:03:31 2013 NEILB [...] cpan.org - Ticket created

Subject:

Testsuite fals if you're running on a machine with sshd running

If you're running with sshd running on its default port of 22, then the testsuite fails: t/51loop-connect.t ........... ok Test died early - listen on port 22 actually succeeded # Tests were run but no plan was declared and done_testing() was not seen. t/52loop-listen.t ............ Dubious, test returned 29 (wstat 7424, 0x1d00) All 13 subtests passed t/53loop-extend.t ............ ok This is like the other bug I reported, where I had to stop httpd to successfully install. Temporarily disabling sshd let me install successfully. Neil

Sun Mar 31 17:15:14 2013 leonerd-cpan [...] leonerd.org.uk - Correspondence added

On Tue Mar 19 08:03:31 2013, NEILB wrote: Show quoted text

> Test died early - listen on port 22 actually succeeded > # Tests were run but no plan was declared and done_testing() was > not seen. > t/52loop-listen.t ............ > Dubious, test returned 29 (wstat 7424, 0x1d00) > All 13 subtests passed

This is most most strange and I do not understand it. First it tests whether listening on the port will fail: | 129 IO::Socket::INET->new( | 130 Type => SOCK_STREAM, | 131 LocalHost => "localhost", | 132 LocalPort => $port, | 133 Listen => 1, | 134 ) and next; It can only fail the unit-test the way you explain, if somehow it fails to listen on the port given the above method, but then later succeeds when IO::Async attempts it. And given that IO::Async basically does theabove as well, I don't get why. Can you strace this test, so I can see what the underlying system calls it makes are? I have a few possible hypotheses, but I'd have to see the strace output to be sure. -- Paul Evans

Sun Mar 31 17:15:17 2013 The RT System itself - Status changed from 'new' to 'open'

Mon Apr 01 05:29:40 2013 neil [...] bowers.com - Correspondence added

Subject:	Re: [rt.cpan.org #84051] Testsuite fals if you're running on a machine with sshd running
Date:	Mon, 1 Apr 2013 10:29:24 +0100
To:	bug-IO-Async [...] rt.cpan.org
From:	Neil Bowers <neil [...] bowers.com>

I'm away from home at the moment -- will give this a go when I get back.

Wed Jan 08 19:31:33 2014 leonerd-cpan [...] leonerd.org.uk - Correspondence added

I think I have finally got to the bottom of this one. It seems to be a disagreement between SO_REUSEADDR being used by IO::Async bind but not the IO::Socket::INET->new() test probe beforehand, but which kicks in -only- if your SSH or Web server are bound listening to -specific- IP addresses but not the catch-all wildcard 0.0.0.0. The fix for the test is just to update either the probe or the real unit test, to ensure they both use the same SO_REUSEADDR setting. -- Paul Evans

Fri Mar 28 13:26:03 2014 leonerd-cpan [...] leonerd.org.uk - Correspondence added

On Wed Jan 08 19:31:33 2014, PEVANS wrote: Show quoted text

> The fix for the test is just to update either the probe or the real > unit test, to ensure they both use the same SO_REUSEADDR setting.

Find patch attached. -- Paul Evans

Subject:

rt84051.patch

=== modified file 't/52loop-listen.t' --- t/52loop-listen.t 2013-09-30 19:51:05 +0000 +++ t/52loop-listen.t 2014-03-28 17:25:25 +0000 @@ -123,6 +123,7 @@ Type => SOCK_STREAM, LocalHost => "localhost", LocalPort => $port, + ReuseAddr => 1, Listen => 1, ) and next;

Fri Mar 28 13:26:04 2014 leonerd-cpan [...] leonerd.org.uk - Status changed from 'open' to 'patched'

Mon Oct 20 14:20:48 2014 leonerd-cpan [...] leonerd.org.uk - Correspondence added

Released -- Paul Evans

Mon Oct 20 14:20:51 2014 leonerd-cpan [...] leonerd.org.uk - Status changed from 'patched' to 'resolved'

Mon Oct 20 14:20:51 2014 leonerd-cpan [...] leonerd.org.uk - Fixed in 0.64 added