bug#20120: wc output padding differs when "-" is in the file list

[Bernhard Voelker]

On 03/16/2015 06:42 AM, Eric Mrak wrote:
> It seems that whenever STDIN is involved the results padding
> reverts to the BSD-style 7/8 padding. When files are given
> as input (excluding STDIN) the padding reflects the width of
> the largest count.
> When files are given as input and one of these is "-", the
> padding reverts again to the BSD 7/8 padding.

Thanks for the report.
This effect is there at least since the last bigger change in
this area, the introduction of the function compute_number_width(),
back in 2003.

Furthermore, strange formatting also happend in other cases,
e.g. for other non-regular files ...

  $ wc /etc/hosts /dev/null
       41     124    1355 /etc/hosts
        0       0       0 /dev/null
       41     124    1355 total

... or where stat() returns a wrong value like for /proc files ...

  $ wc /proc/cpuinfo x
  52 256 1276 /proc/cpuinfo
  1 0 1 x
  53 256 1277 total

... or with the --files0-from=FILE option:

  $ printf '%s\0' x /etc/hosts | wc --files0=-
  1 0 1 x
  41 124 1355 /etc/hosts
  42 124 1356 total

The number width is determined before reading the actual files.
I'm asking myself if it would hurt to save the values for all files
until all of them are read, and then do the calculation of the
number width and the printing of all values.
OTOH this would delay output until all files are read (besides
the memory footprint).
Any opinions if a proper output format warrants this disadvantages?

Regarding the number width fallback of %7d: this is mentioned in
the POSIX specification (in 'Rationale'), but I'm unsure if it's
mandated/ recommended/deprecated behavior.
http://pubs.opengroup.org/onlinepubs/9699919799/utilities/wc.html

Have a nice day,
Berny