[Pkg-shadow-devel] Bug#343595: shadow: [a-z] must be used in maintainer scripts only under C locale

[Denis Barbier]

Package: shadow
Severity: normal
Tags: patch

Hi,

[a-z] may not represent all 26 ASCII lowercase letters in regular
expressions, eg. Estonian collation sorts z between s and t:
  $ echo rstuvwxyz | LC_ALL=en_US.UTF-8 sed -e 's/[a-z]//g'

  $ echo rstuvwxyz | LC_ALL=et_EE.UTF-8 sed -e 's/[a-z]//g'
  tuvwxy
As shown above, this applies to sed, but also awk, grep, expr
  $ echo tuv | LC_ALL=et_EE.UTF-8 awk '/[a-z]/ {print}'
  $ echo tuv | LC_ALL=et_EE.UTF-8 grep [a-z]
  $ LC_ALL=et_EE.UTF-8 expr tuv : [a-z]
  $
and many more.
On the other hand, some commands always work with C collation rules,
most notably tr:
  $ echo 1rstuvwxyz2 | LC_ALL=et_EE.UTF-8 tr -d a-z
  12
or perl and python if they are not told to switch to user's locale.

One must then switch to C locale to avoid those collation issues,
see attached patch.  I also added switches to 'tr' even if it is not
needed for now because documentation tells that 'tr' may get fixed.

Thanks

Denis
-------------- next part --------------
diff -ur debian.orig/passwd.config debian/passwd.config
--- debian.orig/passwd.config	2005-12-16 13:10:49.000000000 +0100
+++ debian/passwd.config	2005-12-16 13:19:56.000000000 +0100
@@ -225,11 +225,11 @@
 					userdefault="tbm"
 				    ;;
 				    *)
-					userdefault=`echo $RET | sed 's/ .*//' | tr A-Z a-z`
+					userdefault=`echo $RET | sed 's/ .*//' | LC_ALL=C tr A-Z a-z`
 				    ;;
 				esac
 				if test -n "$userdefault" && \
-				   expr "$userdefault" : '[a-z][-a-z0-9]*$' >/dev/null; then
+				   LC_ALL=C expr "$userdefault" : '[a-z][-a-z0-9]*$' >/dev/null; then
 					db_set passwd/username "$userdefault"
 				fi
 			fi
@@ -243,7 +243,7 @@
 			# Verify the user name, loop with message if bad.
 			db_get passwd/username
 			USER="$RET"
-			if ! expr "$USER" : '[a-z][-a-z0-9]*$' >/dev/null; then
+			if ! LC_ALL=C expr "$USER" : '[a-z][-a-z0-9]*$' >/dev/null; then
 				db_fset passwd/username seen false
 				db_fset passwd/username-bad seen false
 				db_input critical passwd/username-bad