Re: test-getaddrinfo failures in Fedora Rawhide

[Dmitry V. Levin]

On Tue, Apr 23, 2013 at 10:54:03AM +0100, Pádraig Brady wrote:
> On 04/23/2013 10:44 AM, Dmitry V. Levin wrote:
> > On Tue, Apr 23, 2013 at 10:11:46AM +0100, Pádraig Brady wrote:
> >> On 04/23/2013 08:32 AM, Richard W.M. Jones wrote:
> >>>
> >>> Not looked into this in any detail yet, but just a note that the
> >>> getaddrinfo test is now failing in Fedora Rawhide.  This failure
> >>> started to happen within the last week.
> >>>
> >>>   system error: Connection refused
> >>>   system error: Connection refused
> >>>   system error: Connection refused
> >>>   system error: Connection refused
> >>>   FAIL: test-getaddrinfo
> >>>
> >>> Because this is a connection problem, it's remotely possible this has
> >>> happened because of a change in Koji (the Fedora build system) which
> >>> has no network access.
> >>>
> >>> Full build log:
> >>>   http://kojipkgs.fedoraproject.org//work/tasks/180/5290180/build.log
> >>>
> >>> Versions of things:
> >>>   http://kojipkgs.fedoraproject.org//work/tasks/180/5290180/root.log
> >>>
> >>> The version of gnulib is about a month old.  I will try a newer one
> >>> shortly.
> >>
> >> I don't think the test should fail on any system error actually.
> > 
> > This is not quite an "any system error", but rather a glibc bug:
> > http://sourceware.org/bugzilla/show_bug.cgi?id=15339
> > The fix was submitted to libc-alpha and is waiting for review:
> > http://sourceware.org/ml/libc-alpha/2013-04/msg00248.html
> 
> Thanks for the info.
> 
> I'm still thinking the gnulib patch should be applied though,

Since the regression in glibc was detected by this gnulib test,
I'm not sure ignoring all EAI_SYSTEM errors would be a good idea.

> as you wouldn't want to indicate a gnulib issue if you ran
> out of file descriptors on the system for example?

Lets skip the test in case of EAI_SYSTEM && (EMFILE || ENFILE) then.


-- 
ldv

pgp61SDstMDZJ.pgp
Description: PGP signature