Re: [PATCH] nptl: Avoid expected SIGALRM in most tests [BZ #20432]

[Florian Weimer <fweimer at redhat dot com>]

On 03/15/2017 03:20 PM, Andreas Schwab wrote:
> On Mär 15 2017, Florian Weimer <fweimer@redhat.com> wrote:
>
> > On 03/15/2017 02:15 PM, Andreas Schwab wrote:
> > > On Dez 15 2016, Florian Weimer <fweimer@redhat.com> wrote:
> > >
> > > > The test previously called exit from the signal handler.  Why did it work
> > > > before?
> > >
> > > That only "works" because _IO_cleanup is not thread-safe.
> >
> > Sorry, I don't see the test failure.  Based on the comments in
> > libio/genops.c, I think there is an expectation that even with locked
> > stdout, exit does not block.
>
> POSIX does not guarantee this.
>
> http://austingroupbugs.net/view.php?id=611

Right, but I think the intent of the upstream libio implementation is to 
support this (i.e., no deadlock).

> > I may be missing something, but in that case, please elaborate.
>
> You are not really missing anything, since you are not using this patch.
>
> https://build.opensuse.org/package/view_file/Base:System/glibc/fix-locking-in-_IO_cleanup.patch?expand=1

I think you should disable this test in the patch, or check in the test 
that the deadlock happens.

Thanks,
Florian