GNU bug report logs - #7157
df should default to -P if output is not a tty

Previous Next

To add a comment to this bug, you must first unarchive it, by sending
a message to control AT debbugs.gnu.org, with unarchive 7157 in the body.
You can then email your comments to 7157 AT debbugs.gnu.org in the normal way.

Toggle the display of automated, internal messages from the tracker.

Message #5 received at submit <at> debbugs.gnu.org (full text, mbox):

From: Alain Knaff <alain <at> knaff.lu>
To: bug-coreutils <at> gnu.org
Subject: df should default to -P if output is not a tty
Date: Mon, 04 Oct 2010 11:44:15 +0200

Hello,

We just had a case where an overfull disk went unnoticed by logwatch.

The reason turned out to be its long device name (/dev/mapper
/VolGroup00-LogVol00), which caused df to break the line, messing up
the column count. Indeed, logwatch looks for the use% in the fifth
field, but due to this line breaking, the use% for this disk ended up
in the _fourth_ field of the next line instead.

There is an option to prevent this behavior (-P), but apparently the
logwatch authors were not aware of it. Understandably, I might say,
because if their test cases happen to only have disks with short names,
they'll never stumble upon this.

So, in the name of the "principle of least surprise" wouldn't it be
possible to make -P the default if the output is not a tty (and add an
additional flag for those rare cases where this line-breaking behavior
is actually wanted on non-tty's)

Other utilities, such as Debian's dpkg -l, also default to this
behavior (trimming lines to window width if output is a tty, and not
trimming it if output is not a tty). Same thing for ls (column output
on a tty, one file-name per line on non-tty)

The problem was observed on version coreutils-8.4-8.fc13.i686:

# df | cat
Filesystem           1K-blocks      Used Available Use% Mounted on
/dev/mapper/VolGroup00-LogVol00
                      66688656  47463424  15782956  76% /
/dev/cciss/c0d0p1       197546     19124    168223  11% /boot
tmpfs                   517468        12    517456   1% /dev/shm

Thanks,

Alain

Message #8 received at 7157 <at> debbugs.gnu.org (full text, mbox):

From: Eric Blake <eblake <at> redhat.com>
To: Alain Knaff <alain <at> knaff.lu>
Cc: 7157 <at> debbugs.gnu.org
Subject: Re: bug#7157: df should default to -P if output is not a tty
Date: Mon, 04 Oct 2010 08:30:16 -0600

On 10/04/2010 03:44 AM, Alain Knaff wrote:
> There is an option to prevent this behavior (-P), but apparently the
> logwatch authors were not aware of it. Understandably, I might say,
> because if their test cases happen to only have disks with short names,
> they'll never stumble upon this.

Thanks for the report.  However, I'm reluctant to make the change, on 
the grounds that df -P is documented by POSIX:

http://www.opengroup.org/onlinepubs/9699919799/utilities/df.html

Anyone that wants portable output MUST use -P, because POSIX explicitly 
implies that not using -P leads to implementation-defined formatting, 
which is therefore not machine-parseable.

Furthermore, GNU Coding Standards strongly discourage making behavior 
dependent on whether stdout is a tty.  Yes, some actions, like ls 
behavior, are mandated by POSIX, and some actions, like --color=auto 
only make sense with a tty check, but we should not be adding new 
exceptions without good cause.

One thing that might be possible, however, is to improve df output to 
adjust column alignment if $COLUMNS is not set (or is set to something 
large), so as to avoid breaking lines in the case where the output is 
usable in a long-line context.  But this still doesn't help the case 
that logwatch should have been using -P in the first place.

-- 
Eric Blake   eblake <at> redhat.com    +1-801-349-2682
Libvirt virtualization library http://libvirt.org

Message #11 received at 7157 <at> debbugs.gnu.org (full text, mbox):

From: Eric Blake <eblake <at> redhat.com>
To: Alain Knaff <alain <at> knaff.lu>, 7157 <at> debbugs.gnu.org
Subject: Re: bug#7157: df should default to -P if output is not a tty
Date: Mon, 04 Oct 2010 09:06:39 -0600

[please keep the list in the loop]

On 10/04/2010 08:54 AM, Alain Knaff wrote:
>> One thing that might be possible, however, is to improve df output to
>> adjust column alignment if $COLUMNS
>
> Rather than using $COLUMNS (which is not set by all shells...), it might
> be preferable to do an ioctl(1, TIOCGWINSZ, ...) instead. Indeed, the
> ioctl returns the window size as it is now (even if it has been resized).
> That is what ls does, in any case.

Yep, for consistency among coreutils, the ioctl approach is definitely best.

>
>> is not set (or is set to something
>> large), so as to avoid breaking lines in the case where the output is
>> usable in a long-line context.  But this still doesn't help the case
>> that logwatch should have been using -P in the first place.
>>
>
> Agreed. I now also reported the problem to Kirk Bauer from logwatch.

Glad to hear it.

-- 
Eric Blake   eblake <at> redhat.com    +1-801-349-2682
Libvirt virtualization library http://libvirt.org

Message #14 received at 7157 <at> debbugs.gnu.org (full text, mbox):

From: Alain Knaff <alain <at> knaff.lu>
To: Eric Blake <eblake <at> redhat.com>
Cc: 7157 <at> debbugs.gnu.org
Subject: Re: bug#7157: df should default to -P if output is not a tty
Date: Mon, 04 Oct 2010 17:10:47 +0200

On 04/10/10 17:06, Eric Blake wrote:
> [please keep the list in the loop]
> 
> On 10/04/2010 08:54 AM, Alain Knaff wrote:
>>> One thing that might be possible, however, is to improve df output to
>>> adjust column alignment if $COLUMNS
>>
>> Rather than using $COLUMNS (which is not set by all shells...), it might
>> be preferable to do an ioctl(1, TIOCGWINSZ, ...) instead. Indeed, the
>> ioctl returns the window size as it is now (even if it has been resized).
>> That is what ls does, in any case.
> 
> Yep, for consistency among coreutils, the ioctl approach is definitely best.

... and interestingly enough, the ioctl has the "advantage" of telling
us whether the output is indeed a tty or not (it fails with EINVAL if it
is not). AFAIU, this is how ls does the trick of finding out whether
stdout is a tty or not. In a way, a non-tty could be considered
equivalent to an infinetly wide window.

> 
>>
>>> is not set (or is set to something
>>> large), so as to avoid breaking lines in the case where the output is
>>> usable in a long-line context.  But this still doesn't help the case
>>> that logwatch should have been using -P in the first place.
>>>
>>
>> Agreed. I now also reported the problem to Kirk Bauer from logwatch.
> 
> Glad to hear it.
> 

Regards,

Alain

Message #19 received at 7157-done <at> debbugs.gnu.org (full text, mbox):

From: Pádraig Brady <P <at> draigBrady.com>
To: 7157-done <at> debbugs.gnu.org, 6511-done <at> debbugs.gnu.org
Subject: df no longer wraps lines
Date: Thu, 24 Mar 2011 00:11:40 +0000

Since I've just changed df to no longer split
entries with long device names over two lines
(since it now aligns columns consistently),
I'm closing these.

cheers,
Pádraig

Previous Next

GNU bug tracking system
Copyright (C) 1999 Darren O. Benham, 1997,2003 nCipher Corporation Ltd, 1994-97 Ian Jackson.