Re: Du feature request - group reporting

[Assaf Gordon]

Hello Dan,

Expanding on Eric's comments:

On Thu, Jan 25, 2018 at 02:42:32PM -0600, Eric Blake wrote:
> On 01/25/2018 12:11 PM, Daniel Gall wrote:
> > coreutils-8.26> !diff
> 
> We prefer 'git diff' output against the latest coreutils.git,
> but any program which can produce unified diffs (diff -u) is better than
> an ed script diff.

Good starting points are here:
 https://git.savannah.gnu.org/cgit/coreutils.git/tree/README-hacking
 https://git.savannah.gnu.org/cgit/coreutils.git/tree/HACKING
 
https://git.savannah.gnu.org/cgit/coreutils.git/tree/.github/PULL_REQUEST_TEMPLATE.txt

> A feature addition requires documentation, NEWS update, and preferably
> testsuite additions to be complete 

A typical example of these required changes is here:
https://git.savannah.gnu.org/cgit/coreutils.git/commit/?id=57dea5ed07471b2192cc5edf08993e663a3f6802



Additionally, a work-around would be to combine several existing programs
to get approximately similar information:

First, use `find` to print the size (%s) and group (%g) of each file/directory:

  $ find /home -printf "%g %s\n"
  root    4096  /home
  gordon  4096  /home/gordon
  gordon  59    /home/gordon/.Xauthority
  gordon  4096  /home/gordon/.cache
  gordon  4096  /home/gordon/.cache/RStudio
  ...

Then, use `awk` to sum up the sizes per group:

  $ find /home -printf "%g %s\n" \
       | awk '{a[$1] += $2} END {for(i in a) { print a[i],i }}'
  1044086087 gordon
  542342 mike
  4123 root

And optionally, use `numfmt` to print human sizes:

  $ find /home -printf "%g %s\n" \
       | awk '{a[$1] += $2} END {for(i in a) { print a[i],i }}' \
       | numfmt --to=iec
  997M gordon
  530K mike
  4.1K root


The above commands are rather naive, counting hard-links as many times
as they appear (similar to 'du -l'), and showing the apparent size
instead of allocated blocks (similar to 'du --apparent-size').

To show allocated blocks, replace '%s' with '%k'.

To count hardlinked files just once, print the device(%D) and inode number (%i) 
of
each file, then use 'sort -u' to keep only one of each:

  find /home -printf "%g %s %D %i\n" \
    | sort -k3n,3 -k4n,4 -u \
    | awk '{a[$1] += $2} END {for(i in a) { print a[i],i }}' \
    | numfmt --to=iec

This isn't as efficient as 'du', but could be used with existing programs
without code modifications (and using find's many predicates allows fine-tuning
of the summaries, e.g. per-user, per-user-and-group, etc.).

regards,
 - assaf