From 45293fc346e0294a33f35e09575da8ca14f062b9 Mon Sep 17 00:00:00 2001
From: "bar@mysql.com" <>
Date: Thu, 13 Apr 2006 10:55:48 +0500
Subject: [PATCH] Bug#18691: Converting number to UNICODE string returns
 invalid result.

Conversion from int and real numbers to UCS2 didn't work fine:
CONVERT(100, CHAR(50) UNICODE)
CONVERT(103.9, CHAR(50) UNICODE)

The problem appeared because numbers have binary charset, so,
simple charset recast binary->ucs2 was performed
instead of real conversion.

Fixed to make numbers pretend to be non-binary.
---
 mysql-test/r/ctype_ucs.result | 12 ++++++++++++
 mysql-test/t/ctype_ucs.test   |  8 ++++++++
 sql/item_timefunc.cc          | 28 +++++++++++++++++++++++-----
 sql/item_timefunc.h           |  2 +-
 4 files changed, 44 insertions(+), 6 deletions(-)

diff --git a/mysql-test/r/ctype_ucs.result b/mysql-test/r/ctype_ucs.result
index 0e12ec88662..8869604b128 100644
--- a/mysql-test/r/ctype_ucs.result
+++ b/mysql-test/r/ctype_ucs.result
@@ -666,6 +666,18 @@ Warnings:
 Warning	1265	Data truncated for column 'Field1' at row 1
 DROP TABLE t1;
 SET NAMES latin1;
+SELECT CONVERT(103, CHAR(50) UNICODE);
+CONVERT(103, CHAR(50) UNICODE)
+103
+SELECT CONVERT(103.0, CHAR(50) UNICODE);
+CONVERT(103.0, CHAR(50) UNICODE)
+103.0
+SELECT CONVERT(-103, CHAR(50) UNICODE);
+CONVERT(-103, CHAR(50) UNICODE)
+-103
+SELECT CONVERT(-103.0, CHAR(50) UNICODE);
+CONVERT(-103.0, CHAR(50) UNICODE)
+-103.0
 CREATE TABLE t1 (
 a varchar(255) NOT NULL default '',
 KEY a (a)
diff --git a/mysql-test/t/ctype_ucs.test b/mysql-test/t/ctype_ucs.test
index bd37d008173..2cbabf88ee0 100644
--- a/mysql-test/t/ctype_ucs.test
+++ b/mysql-test/t/ctype_ucs.test
@@ -407,6 +407,14 @@ INSERT INTO t1 VALUES ('-1');
 DROP TABLE t1;
 SET NAMES latin1;
 
+#
+# Bug#18691 Converting number to UNICODE string returns invalid result
+#
+SELECT CONVERT(103, CHAR(50) UNICODE);
+SELECT CONVERT(103.0, CHAR(50) UNICODE);
+SELECT CONVERT(-103, CHAR(50) UNICODE);
+SELECT CONVERT(-103.0, CHAR(50) UNICODE);
+
 #
 # Bug#9557 MyISAM utf8 table crash
 #
diff --git a/sql/item_timefunc.cc b/sql/item_timefunc.cc
index d6b57464d59..f3d6858755c 100644
--- a/sql/item_timefunc.cc
+++ b/sql/item_timefunc.cc
@@ -2221,8 +2221,8 @@ String *Item_char_typecast::val_str(String *str)
     // Convert character set if differ
     uint dummy_errors;
     if (!(res= args[0]->val_str(&tmp_value)) ||
-	str->copy(res->ptr(), res->length(), res->charset(),
-                  cast_cs, &dummy_errors))
+        str->copy(res->ptr(), res->length(), from_cs,
+        cast_cs, &dummy_errors))
     {
       null_value= 1;
       return 0;
@@ -2261,14 +2261,32 @@ void Item_char_typecast::fix_length_and_dec()
      For single-byte character sets we allow just to copy
      from the argument. A single-byte character sets string
      is always well-formed. 
+     
+     There is a special trick to convert form a number to ucs2.
+     As numbers have my_charset_bin as their character set,
+     it wouldn't do conversion to ucs2 without an additional action.
+     To force conversion, we should pretend to be non-binary.
+     Let's choose from_cs this way:
+     - If the argument in a number and cast_cs is ucs2 (i.e. mbminlen > 1),
+       then from_cs is set to latin1, to perform latin1 -> ucs2 conversion.
+     - If the argument is a number and cast_cs is ASCII-compatible
+       (i.e. mbminlen == 1), then from_cs is set to cast_cs,
+       which allows just to take over the args[0]->val_str() result
+       and thus avoid unnecessary character set conversion.
+     - If the argument is not a number, then from_cs is set to
+       the argument's charset.
   */
+  from_cs= (args[0]->result_type() == INT_RESULT || 
+            args[0]->result_type() == REAL_RESULT) ?
+           (cast_cs->mbminlen == 1 ? cast_cs : &my_charset_latin1) :
+           args[0]->collation.collation;
   charset_conversion= (cast_cs->mbmaxlen > 1) ||
-                      !my_charset_same(args[0]->collation.collation, cast_cs) &&
-                      args[0]->collation.collation != &my_charset_bin &&
+                      !my_charset_same(from_cs, cast_cs) &&
+                      from_cs != &my_charset_bin &&
                       cast_cs != &my_charset_bin;
   collation.set(cast_cs, DERIVATION_IMPLICIT);
   char_length= (cast_length >= 0) ? cast_length : 
-	       args[0]->max_length/args[0]->collation.collation->mbmaxlen;
+	       args[0]->max_length/from_cs->mbmaxlen;
   max_length= char_length * cast_cs->mbmaxlen;
 }
 
diff --git a/sql/item_timefunc.h b/sql/item_timefunc.h
index ce9d6b0a7aa..163b1591e52 100644
--- a/sql/item_timefunc.h
+++ b/sql/item_timefunc.h
@@ -681,7 +681,7 @@ public:
 class Item_char_typecast :public Item_typecast
 {
   int cast_length;
-  CHARSET_INFO *cast_cs;
+  CHARSET_INFO *cast_cs, *from_cs;
   bool charset_conversion;
   String tmp_value;
 public: