From mboxrd@z Thu Jan  1 00:00:00 1970
Received: from mx1.redhat.com ([209.132.183.28]:58488 "EHLO mx1.redhat.com"
        rhost-flags-OK-OK-OK-OK) by vger.kernel.org with ESMTP
        id S1755407AbdCHGrZ (ORCPT <rfc822;fstests@vger.kernel.org>);
        Wed, 8 Mar 2017 01:47:25 -0500
Date: Wed, 8 Mar 2017 14:09:59 +0800
From: Eryu Guan <eguan@redhat.com>
Subject: Re: generic/411 clash with TEST_DIR=/mnt
Message-ID: <20170308060959.GB14226@eguan.usersys.redhat.com>
References: <20170307165027.GA15724@birch.djwong.org>
 <20170308035045.GN14075@dhcp12-143.nay.redhat.com>
MIME-Version: 1.0
Content-Type: text/plain; charset=us-ascii
Content-Disposition: inline
In-Reply-To: <20170308035045.GN14075@dhcp12-143.nay.redhat.com>
Sender: fstests-owner@vger.kernel.org
To: Zorro Lang <zlang@redhat.com>
Cc: "Darrick J. Wong" <darrick.wong@oracle.com>, fstests@vger.kernel.org
List-ID: <fstests@vger.kernel.org>

On Wed, Mar 08, 2017 at 11:50:45AM +0800, Zorro Lang wrote:
> On Tue, Mar 07, 2017 at 08:50:27AM -0800, Darrick J. Wong wrote:
> > Hi Zorro,
> > 
> > I noticed that generic/411 creates some sort of directory structure
> > involving $TEST_DIR/$seq/$$_mpA/mnt1/mnt2, then pipes the findmnt output
> > through _filter_test_dir.
> > 
> > Unfortunately, on my test system I have TEST_DIR=/mnt, so the filtering

Ah, that's a rare setup, I'm curious what you use for SCRATCH_MNT :)

> > produces this output:
> > 
> > QA output created by 411
> > ------
> > TEST_DIR/411 SCRATCH_DEV
> > mpA SCRATCH_DEV
> > mpATEST_DIR1 SCRATCH_DEV
> > mpB SCRATCH_DEV
> > mpBTEST_DIR1 SCRATCH_DEV
> > mpBTEST_DIR1/TEST_DIR2 SCRATCH_DEV
> > mpC SCRATCH_DEV
> > mpCTESTDIR1 SCRATCH_DEV
> > ======
> 
> Hmm, that's really a problem, thanks for finding it:) If someone
> use "/mnt" as TEST_DIR, _filter_test_dir will change all "/mnt"
> things to "TEST_DIR".
> 
> Except we recommend all cases shouldn't contain "/mnt"
> in its test path name. Or we'd better change _filter_test_dir,
> make it to be more picky.
> 
> I prefer the second way, except someone has a third way :)
> 
> > crash test passed
> > 
> > Which means that the golden output comparison fails. :(
> > 
> > I'm not sure what's a proper fix here: changing _filter_test_dir to be
> > more picky about what gets sed'ed?
> > 
> > e.g. sed -e "s,\([[:space:]]\)$TEST_DIR,\1TEST_DIR,g/"
> 
> Except [[:space:]]$TEST_DIR, maybe we should think about "^$TEST_DIR" too,
> e.g:
> 
> echo "/mnt/mnt1/mnt2     /mnt/mnt3/mnt4" | \
>      sed -e "s,\([[:space:]]\)$TEST_DIR,\1TEST_DIR,g" | \
>      sed -e "s,^$TEST_DIR,TEST_DIR,g"
> 
> (Is there a simpler way to do above things ^^ ? :)
> 
> But this change will affect lots of cases, we need to run full around
> xfstests to make sure it won't change some cases' golden output. What
> do you think?
> 
> Thanks,
> Zorro
> 
> > 
> > Or just to change the test to use directory names that are less likely to be
> > mistaken for TEST_* and SCRATCH_*?

I'd prefer this easier way :) IMO, getting the filters right is tricky
(see Amir's recent patch for example, "4e965d8 fstests: fix test and
scratch filters for overlapping DEV/MNT paths"), and, as you said, will
affect more tests. Changing the directory names used in generic/411 is
much easier, prefixing with $seq should do the work, e.g. ${seq}_mnt1

Thanks,
Eryu