Path: utzoo!utgpu!water!watmath!clyde!att!osu-cis!tut.cis.ohio-state.edu!mailrus!ames!ucsd!ucsdhub!hp-sdd!hplabs!hpda!hpcupt1!hpsal2!morrell
From: morrell@hpsal2.HP.COM (Michael Morrell)
Newsgroups: comp.unix.questions
Subject: Re: stable sort in unix tools?
Message-ID: <2450010@hpsal2.HP.COM>
Date: 19 May 88 18:49:36 GMT
References: <5456@umn-cs.cs.umn.edu>
Organization: HP System Architecture Lab, Cupertino
Lines: 32

/ hpsal2:comp.unix.questions / ries@trwrb.UUCP (Marc Ries) /  9:30 am  May 18, 1988 /
>I'm stuck with a little sorting problem... not big enough to write my own
[...]
>(sorted by the number in the first column) that I would like to get
>sorted by the classification in the second column (A1, A2, A3...), but
>keeping the numbers in order within each classification.  The above list,
>for instance, would end up:
>
>1644    A1      blah blah blah
>1718    A1      still more
>1237    A2      text comments
>1357    A3      more text
>
>I need a stable sort, in other words, and the standard (4.3 bsd) UNIX sort
>doesn't seem to provide that.  Is there a way around this using UNIX tools?
  
   In the past, "sort +1 -2 +0 -1 data_file" seemed to work ok 8-).
---

 Sort does not have a way to perform a stable sort.  I have felt for a long
time that there should be an option to do this (-s?).  With your example, doing
what Marc Ries suggests will work because it re-sorts the 1st colummn (which
was already sorted in the input).  The problem occurs if the input order cannot
be recreated by sorting.  For now, the following kludge is the only way I know
of to get a stable sort:

  awk '{print NR "\t" $0}' data_file | sort +2 -3 +0n -1 | cut -f2-

This prefixes each line by its line number, sorts using the line number
as the final sort key, then strips off the line number.

  Michael Morrell