Re: [Coreutils] Sort enhancement request: 1 < 1K < 1M < 1G etc.

[Philip Ganchev]

2010/10/8 Pádraig Brady <address@hidden>:
> On 08/10/10 17:15, Philip Ganchev wrote:
>> On a related note, couldn't all GNU programs (such as "du" and "ls")
>> be made to print (and read?) numbers and dates formatted according to
>> the locale setting, including thousands-separators, etc? That would
>> avoid having a special option for every program.
>
> BLOCK_SIZE can be used to control a few output formats:
>
> export BLOCK_SIZE=\'1; export LC_ALL=de_DE.utf8
>
> $ \ls -ld .
> drwxrwxr-x. 4 padraig padraig 4.096 15. Sep 23:04 .
> $ du -s .
> 15.368.192
> $ df .
> Dateisystem          1B-Blöcke   Benutzt Verfügbar Ben% Eingehängt auf
> /dev/sda5            5.155.188.736 4.930.887.680 171.970.560  97% /

OK, great -- thanks!  Maybe BLOCK_SIZE should be mentioned in the
manual page or the info page of the programs that respect it? It does
not seem to be widely known -- I did not know about it. (Maybe it is
already in the latest versions.)

(It's too bad that non-GNU programs like "ps" don't use those
variables. And I think there should be a more general number format
variable to allow other programs outputting numbers, such as "nl",
should be able to output numbers with thousands separators.)

> I do notice that stat doesn't pass on a single
> quote to the underlying printf to get locale specific output.
> Maybe we want to support that?
>
> $ stat -c%\'s .
> ?s
>
> $ stat -c%08s .
> 00004096

It seems to my naive eyes that it ought to be supported. It should
also respect the BLOCK_SIZE (or BLOCKSIZE?) variable.

> Also sort can read locale specific separators:
>
> $ printf '1 234\n1 123\n' | LC_ALL=fr_FR.utf8 sort --debug -sn
> sort: using `fr_FR.utf8' sorting rules
> 1 123
> _____
> 1 234
> _____

Great!

> As for dates, it's hard enough to read in a single
> locale, never mind supporting multiple locales.

Can't "sort" rely on the "date" program (library) to recognize and
convert dates to an unambiguous time stamp? Maybe it would be slow
check every string if it's a date, but just do it when an option is
specified?


Philip