diff --git a/admin/charsets/mapconv b/admin/charsets/mapconv
index 6fd13c61c4a..245dce114a0 100755
--- a/admin/charsets/mapconv
+++ b/admin/charsets/mapconv
@@ -35,6 +35,11 @@
 #	GLIBC-1 GLIBC-2 GLIBC-2-7 CZYBORRA IANA UNICODE UNICODE2 YASUOKA
 #   $4: awk script
 
+## So that eg [A-F] as used by KANJI-DATABASE branch below works as expected.
+## Otherwise with LANG=en_US.utf8, CNS-6.map was generated with a
+## bogus entry.  By experiment, LC_COLLATE=C was not enough.
+export LC_ALL=C
+
 BASE=`expr "$1" : '.*/\(.*\)' '|' "$1"` # basename
 FILE="admin/charsets/mapfiles/$BASE"
 BASE=`expr "$BASE" : '\(.*\)\.gz$' '|' "$BASE"` # remove any .gz suffix