Re: [PATCH] Ignore "yes: standard output: Broken pipe" errors

[Eryu Guan <eguan@redhat.com>]

On Wed, Jul 26, 2017 at 10:22:13AM +0200, Jan Kara wrote:
> On Wed 26-07-17 15:47:19, Eryu Guan wrote:
> > On Tue, Jul 25, 2017 at 04:33:20PM +0200, Jan Kara wrote:
> > > Recently, some tests started failing because they had
> > > 
> > > yes: standard output: Broken pipe
> > > 
> > > in their output. Fix the problem by discarding errors from yes(1)
> > > program.
> > > 
> > > Signed-off-by: Jan Kara <jack@suse.cz>
> > > ---
> > >  common/rc         | 12 ++++++------
> > >  tests/generic/081 |  2 +-
> > >  tests/generic/108 |  2 +-
> > >  3 files changed, 8 insertions(+), 8 deletions(-)
> > 
> > I'm wondering which case failed for you and what's the diff output and
> > .full file? And what change caused the failure for you? Because I didn't
> > see any failure on my RHEL7 box nor Fedora rawhide box. I compiled
> > latest yes from coreutils git repo and didn't hit the error either.
> > 
> > So I suspect that it might be other commands in the pipeline that failed
> > and caused the broken pipe.
> 
> I've started seeing failures after I've updated the testing VM to openSUSE
> Leap 42.2. And yes(1) failing with this error is "normal" if SIGPIPE is
> ignored - just try ( trap "" PIPE; yes | exit; ) and observe the error.
> So it is just that in the new VM something makes yes(1) ignore SIGPIPE and
> honestly I'm not sure what that is even though I've looked for a while. But
> regardless of that, ignoring stderr from yes(1) seems like a safe thing to
> do.

Ahh, I see the problem. Actually Eric Biggers has fixed a very similar
problem in commit 9bcb266cd778 ("generic/397: be compatible with ignored
SIGPIPE"), but at that time he only saw generic/397 failure, so he
decided to fix that test only.

How about print out warnings and exit if check finds SIGPIPE is ignored?
Because seems we can't un-ignore it in a sub-shell, if it's already
ignored on start. e.g.

if trap -p | grep -q "'' SIGPIPE"; then
	echo "SIGPIPE is ignored by current shell, this will cause unexpected"
	echo "broken pipe error in tests. Please unignore SIGPIPE and rerun."
	echo "For bash, just run 'trap - SIGPIPE'"
	exit
fi

Because I think ignoring SIGPIPE indicates bugs/issues somewhere, as the
python bug mentioned in comment 9bcb266cd778.

> 
> Regarding your question about failed tests: Tests generic/081, generic/347,
> generic/361, shared/298, xfs/074 failed due to this and they all just had
> this additional line in the output.

Thanks,
Eryu

> 
> -- 
> Jan Kara <jack@suse.com>
> SUSE Labs, CR
> --
> To unsubscribe from this list: send the line "unsubscribe fstests" in
> the body of a message to majordomo@vger.kernel.org
> More majordomo info at  http://vger.kernel.org/majordomo-info.html