Auto-merge

mysql-test/r/ctype_utf8.result

@@ -1848,3 +1848,35 @@ select hex(_utf8 B'001111111111');
 ERROR HY000: Invalid utf8 character string: 'FF'
 select (_utf8 X'616263FF');
 ERROR HY000: Invalid utf8 character string: 'FF'
+CREATE TABLE t1 (a INT NOT NULL, b INT NOT NULL);
+INSERT INTO t1 VALUES (70000, 1092), (70001, 1085), (70002, 1065);
+SELECT CONVERT(a, CHAR), CONVERT(b, CHAR) FROM t1 GROUP BY b;
+CONVERT(a, CHAR)	CONVERT(b, CHAR)
+70002	1065
+70001	1085
+70000	1092
+SELECT CONVERT(a, CHAR), CONVERT(b, CHAR) FROM t1;
+CONVERT(a, CHAR)	CONVERT(b, CHAR)
+70000	1092
+70001	1085
+70002	1065
+ALTER TABLE t1 ADD UNIQUE (b);
+SELECT CONVERT(a, CHAR), CONVERT(b, CHAR) FROM t1 GROUP BY b;
+CONVERT(a, CHAR)	CONVERT(b, CHAR)
+70002	1065
+70001	1085
+70000	1092
+DROP INDEX b ON t1;
+SELECT CONVERT(a, CHAR), CONVERT(b, CHAR) FROM t1 GROUP BY b;
+CONVERT(a, CHAR)	CONVERT(b, CHAR)
+70002	1065
+70001	1085
+70000	1092
+ALTER TABLE t1 ADD INDEX (b);
+SELECT CONVERT(a, CHAR), CONVERT(b, CHAR) from t1 GROUP BY b;
+CONVERT(a, CHAR)	CONVERT(b, CHAR)
+70002	1065
+70001	1085
+70000	1092
+DROP TABLE t1;
+End of 5.0 tests

mysql-test/t/ctype_utf8.test

@@ -1439,3 +1439,20 @@ select hex(_utf8 X'616263FF');
 select hex(_utf8 B'001111111111');
 --error ER_INVALID_CHARACTER_STRING
 select (_utf8 X'616263FF');
+
+#
+# Bug #36772: When using UTF8, CONVERT with GROUP BY returns truncated results
+#
+CREATE TABLE t1 (a INT NOT NULL, b INT NOT NULL);
+INSERT INTO t1 VALUES (70000, 1092), (70001, 1085), (70002, 1065);
+SELECT CONVERT(a, CHAR), CONVERT(b, CHAR) FROM t1 GROUP BY b;
+SELECT CONVERT(a, CHAR), CONVERT(b, CHAR) FROM t1;
+ALTER TABLE t1 ADD UNIQUE (b);
+SELECT CONVERT(a, CHAR), CONVERT(b, CHAR) FROM t1 GROUP BY b;
+DROP INDEX b ON t1;
+SELECT CONVERT(a, CHAR), CONVERT(b, CHAR) FROM t1 GROUP BY b;
+ALTER TABLE t1 ADD INDEX (b);
+SELECT CONVERT(a, CHAR), CONVERT(b, CHAR) from t1 GROUP BY b;
+DROP TABLE t1;
+
+--echo End of 5.0 tests

sql/item_timefunc.cc

@@ -2504,6 +2504,8 @@ void Item_char_typecast::fix_length_and_dec()
        and thus avoid unnecessary character set conversion.
      - If the argument is not a number, then from_cs is set to
        the argument's charset.
+
+       Note (TODO): we could use repertoire technique here.
   */
   from_cs= (args[0]->result_type() == INT_RESULT || 
             args[0]->result_type() == DECIMAL_RESULT ||
@@ -2511,12 +2513,13 @@ void Item_char_typecast::fix_length_and_dec()
            (cast_cs->mbminlen == 1 ? cast_cs : &my_charset_latin1) :
            args[0]->collation.collation;
   charset_conversion= (cast_cs->mbmaxlen > 1) ||
-                      !my_charset_same(from_cs, cast_cs) &&
+                      (!my_charset_same(from_cs, cast_cs) &&
                        from_cs != &my_charset_bin &&
-                      cast_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/from_cs->mbmaxlen;
+  char_length= (cast_length >= 0) ?
+                 cast_length :
+                 args[0]->max_length / args[0]->collation.collation->mbmaxlen;
   max_length= char_length * cast_cs->mbmaxlen;
 }