From: Eric Whitney <enwlinux@gmail.com>
Subject: Re: current e2fsprogs maint branch test results
Date: Sun, 22 Dec 2013 21:50:44 -0500
Message-ID: <20131223025044.GB2396@wallace>
References: <20131217234313.GA2183@wallace>
 <20131218213657.GA25946@birch.djwong.org>
Mime-Version: 1.0
Content-Type: text/plain; charset=us-ascii
Cc: Eric Whitney <enwlinux@gmail.com>, linux-ext4@vger.kernel.org,
	tytso@mit.edu
To: "Darrick J. Wong" <darrick.wong@oracle.com>
Content-Disposition: inline
In-Reply-To: <20131218213657.GA25946@birch.djwong.org>
Sender: linux-ext4-owner@vger.kernel.org

* Darrick J. Wong <darrick.wong@oracle.com>:
> On Tue, Dec 17, 2013 at 06:43:13PM -0500, Eric Whitney wrote:
> > I've built the contents of the current e2fsprogs maint branch (4727c67dc2)
> > and run make check on both a Pandaboard (ARM) and an x86-64 VM. In each
> > case I used the following arguments to configure: --disable-uuidd
> > --disable-libuuid --disable-libblkid.
> > 
> > Both the ARM and x86-64 runs produced warnings when compiling ea_refcount.c
> > during make check.  I posted a patch yesterday (e2fsck: fix printf
> > conversion specs in ea_refcount.c) that clears those up for me on both
> > platforms.
> > 
> > The ARM build and make check were otherwise clean.
> > 
> > The x86-64 build and make check contained one more problem - a warning while
> > compiling debugfs.c:
> > ../../debugfs/debugfs.c:2462:5: warning: too many arguments for format
> >  [-Wformat-extra-args]
> > 
> > This code (commit fe56188b07) is part of that used to check superblock block
> > numbers specified on the command line, and the error reporting has become
> > a little fuzzy relative to what we had previously.
> > 
> > Before:
> > 
> > root@debug1:~# debugfs -s 327b /dev/vdc
> > debugfs 1.42.8 (20-Jun-2013)
> > debugfs: Bad superblock number - 327b
> > 
> > After:
> > 
> > root@debug1:~# debugfs -s 327b /dev/vdc
> > debugfs 1.42.8 (20-Jun-2013)
> > debugfs: Bad block number - 327b
> > debugfs: Invalid block number: 327b
> > debugfs: Operation not permitted 
> > 
> > Both strtoblk() and parseulonglong() (which it calls) output error messages
> > for bad/invalid block numbers, which is redundant in this case.  The last
> > (erroneous) error message is output by the call to com_err() which also 
> > causes the warning noted above.
> > 
> > It seems to me that the call to com_err() ought to be deleted, and maybe
> > the immediately preceding call to strtoblk() ought to be converted to a call
> > to parseulonglong() to restore the original messaging.  I'd like to post a
> > patch, but there are a number of other calls to strtoblk() in debugfs now
> > that will produce two messages on errors, and the intent isn't clear to me.
> > Is this just an area that needed a little more polish?
> 
> Back in October, I wrote a patch[1] that did a straight conversion to
> parse_ulonglong for 64bit support, but Lukas suggested using strtoblk instead.
> So I changed it, and that's what's in there now... uglified.  Sorry about that.
> I should have argued more strongly against strtoblk. :(
> 
> I don't have a problem with changing it to parse_ulonglong.
> 
> [1] http://patchwork.ozlabs.org/patch/279295/
> 

Thanks for the background.  Since Ted accepted the strtoblk() changes, maybe
he's ok with the changed error reporting.  For now, since I think he's
interested in getting the next e2fsprogs maint release out the door, I posted
a simple patch just to address the compile-time warning and that last broken
error message.  There are enough strtoblk() call sites that it's probably
better to address them all and apply a common solution or not address any of
them rather than to address just one.

Thanks,
Eric