Re: sort "b" option in pos2 has strange effect

[Pádraig Brady]

Jim Meyering wrote:
> For each fix, I usually try to determine when the bug was introduced
> and mention that in NEWS.
> 
> Both of these date back to the very beginning, since sort from
> textutils-1.13 (yes, I actually built it ;-) exhibits the same incorrect
> behavior, and the code in that function barely changed between my initial
> import in 1992 and that 1.13.
> 
> How many more +16-year-old bugs are lurking?

Interesting :)
I also noticed that freeBSD/Mac OS X use coreutils sort
so they have the same issue.
Also the i18n patch in fedora 8 at least seems
to be varying one of the problems somewhat:

upstream buggy coreutils:
  $ printf "a  y\na z\n" | sort -k1,1b #buggy
  a z
  a  y
  $ printf "a  y\na z\n" | sort -k1b,1 #ok
  a  y
  a z

fedora 8:
  $ printf "a  y\na z\n" | sort -k1,1b #ok
  a  y
  a z
  $ printf "a  y\na z\n" | sort -k1b,1 #buggy
  a z
  a  y

So I'll add the attached test I think to check for that.

cheers,
Pádraig

>From b857242886538be7c1ae387c85b86e9f96fecb80 Mon Sep 17 00:00:00 2001
From: =?utf-8?q?P=C3=A1draig=20Brady?= <address@hidden>
Date: Fri, 27 Feb 2009 08:40:42 +0000
Subject: [PATCH] tests: sort: Check skipping blanks in multibyte locales

* tests/misc/sort: On Fedora 8 at least, sort -k1b,1
mishandles blanks in multibyte locales, so add appropriate test.
---
 tests/misc/sort |    9 +++++++++
 1 files changed, 9 insertions(+), 0 deletions(-)

diff --git a/tests/misc/sort b/tests/misc/sort
index 3af2388..b6ee905 100755
--- a/tests/misc/sort
+++ b/tests/misc/sort
@@ -24,6 +24,10 @@ my $prog = 'sort';
 # Turn off localization of executable's output.
 @ENV{qw(LANGUAGE LANG LC_ALL)} = ('C') x 3;
 
+my $locale = $ENV{LOCALE_FR_UTF8};
+! defined $locale || $locale eq 'none'
+  and $locale = 'C';
+
 # Since each test is run with a file name and with redirected stdin,
 # the name in the diagnostic is either the file name or "-".
 # Normalize each diagnostic to use '-'.
@@ -216,6 +220,11 @@ my @Tests =
 # next field are not included in the sort. I.E. order should not change here.
 ["18f", '-k1,1b', {IN=>"a  y\na z\n"}, {OUT=>"a  y\na z\n"}],
 
+# When ignoring leading blanks for start position, ensure blanks from
+# next field are not included in the sort. I.E. order should not change here.
+# This was noticed as an issue on fedora 8 (only in multibyte locales).
+["18g", '-k1b,1', {IN=>"a  y\na z\n"}, {OUT=>"a  y\na z\n"}, {ENV => 
"LC_ALL=$locale"}],
+
 # This looks odd, but works properly -- 2nd keyspec is never
 # used because all lines are different.
 ["19a", '+0 +1nr', {IN=>"b 2\nb 1\nb 3\n"}, {OUT=>"b 1\nb 2\nb 3\n"}],
-- 
1.5.3.6