Re: [PATCH] df: Adding a --no-headers option, by request of Bruce Dubbs

[Assaf Gordon]

Hello,

On 2019-03-21 4:13 a.m., Erik Auerswald wrote:
> More general, table headers are useful for interactive use, but often need
> to be ignored when the tool output is used programmatically.

That is a very good point.

Note that Bruce's original request[1] was to have a way
to get the number of used blocks programmatically
(I assume for scripting).

[1] https://lists.gnu.org/r/coreutils/2019-03/msg00048.html

While 'df' was always geared towards interactive use
(e.g. the POSIX specification[2] explicitly set the header line,
and the output columns, with "-P" and "-t" option to control the
output).

[2] http://pubs.opengroup.org/onlinepubs/9699919799/utilities/df.html

For example, if you specify multiple fields, 'df' takes the extra effort
to right-align and space-pad the output - that's for interactive use / 
human consumption, not scripting.


Since the OP's goal was for scripting, perhaps it's good to take a step
back and consider whether other tools could provide better solutions:

stat(1) have this filesystem information, and allows explicit control
of the output.

For example, to get the number of used kilobytes on the system:

  USED=$(stat -f . -c "%s*(%b-%f)/1024" $PWD | bc)

While it's not shorter (because stat(1) doesn't have 'used blocks'
operand and uses blocks instead of kilobytes), it allows might tighter
control of the output.
(perhaps that's a reason to add 'used' operand to stat(1) and/or output
in bytes instead of blocks, but that's a different topic).

also:

> On Wed, Mar 20, 2019 at 02:22:42PM -0700, Kaz Kylheku (Coreutils)
> wrote:
[...]
> "df --no-h" would suffice, use half the processes of the above,
> avoid generating data that is read, ignored, and deleted, and avoid
> the additional data copies needed to process the data (i.e. do
> nothing) in sed. ;-)
These are valid upsides, and Eric nicely explains the downsides [3].
[3] https://lists.gnu.org/r/coreutils/2019-03/msg00049.html

A user who is familiar with "|sed 1d" or "|tail +2" will be know
how to remove headers from any program,  without the need
to learn how to disable headers specifically for each program (compare
'ps' vs 'lsblk' vs '--no-header' vs '--terse').


There's additional cost to adding such options: development and maintenance.

For example, several programs in the util-linux package [4]
(like lsblk, findmnt, lsns, lsmem) have dual role as both interactive
and scriptable programs.
The output formatting code is complicated enough that they created a
dedicated library just for that: libsmartcol [5].

[4] https://github.com/karelzak/util-linux
[5] 
https://mirrors.edge.kernel.org/pub/linux/utils/util-linux/v2.33/libsmartcols-docs/

In fact, "findmnt" is so versatile in its output options,
you might want to use it:

   UESD=$(findmnt -T $PWD --noheadings --output=USED --bytes)


regards,
 - assaf