Re: sort -u -n error : sort v 4.5.3

[Eric Blake]

> > address@hidden data]# sort -n < /tmp/new_spammers
> > 64.202.165.131
> > 64.202.165.132
> > 64.202.165.133
> > 
> > 
> > address@hidden data]# sort -u -n < /tmp/new_spammers
> > 64.202.165.132
> 
> Not a bug.  This behavior is required by POSIX,
> http://www.opengroup.org/onlinepubs/009695399/utilities/sort.html,
> which states that -n restricts the key to the first numeric
> field, and that -u uniquifies based on the sort key.
> 
> In your sample file, the unique key is 64, and all
> three lines have the same key.  When -n or -u is
> used alone, a tiebreaker is used that sorts based
> on the rest of the line.  But when used together,
> the behavior correctly stops after one line.

However, I think this does point out some bugs in POSIX
compliance - GNU sort currently parses the decimal separator
in a numeric sort key, whereas POSIX does not permit that; and
treats signed zeros as significant, where POSIX does not
permit that:

   sort 5.94:
$ export LANG=C LC_ALL=C POSIXLY_CORRECT=1
$ cat blah
64.202
64.201
64.203
0
-0
0
$ sort -n blah
-0
0
0
64.201
64.202
64.203
$ sort -nu blah
0
64.201
64.202
64.203

  POSIX:
$ sort -n blah  # lines with equal numeric keys should stay stable during sort
0
-0
0
64.202
64.201
64.203
$ sort -nu blah # numeric keys do not include decimal points in LANG=C
0
64.202

-- 
Eric Blake