Fixed unique prefix key bug for multibyte character sets (BUG #4521) for

InnoDB. This fixes also a second part of the same problem with prefix keys
on a multibyte string column for InnoDB.


innobase/rem/rem0cmp.c:
  Fixed unique prefix key or prefix key using multibyte character set bugs for
  InnoDB (BUG #4521).
  
  The unique key with a prefix of N appears index 3*N bytes of the column not N
  characters. If you had two records which have the same first N characters 
  but differ in the first 3*N bytes, then you couldn't select the records using 
  an equality test.

1 files changed, 31 insertions, 0 deletions

diff --git a/innobase/rem/rem0cmp.c b/innobase/rem/rem0cmp.c
index 6e8f3d82ef3..f6c82102839 100644
--- a/innobase/rem/rem0cmp.c
+++ b/innobase/rem/rem0cmp.c
@@ -14,6 +14,9 @@ Created 7/1/1994 Heikki Tuuri
 
 #include "srv0srv.h"
 
+#include <m_ctype.h>
+#include <my_sys.h>
+
 /*		ALPHABETICAL ORDER
 		==================
 		
@@ -453,6 +456,8 @@ cmp_dtuple_rec_with_match(
 					in current field */
 	int		ret = 3333;	/* return value */
 	
+	CHARSET_INFO*   charset;        /* charset used in the field */
+
 	ut_ad(dtuple && rec && matched_fields && matched_bytes);
 	ut_ad(dtuple_check_typed(dtuple));
 
@@ -541,6 +546,32 @@ cmp_dtuple_rec_with_match(
 			&& dtype_get_charset_coll(cur_type->prtype) !=
 				data_mysql_latin1_swedish_charset_coll)) {
 
+		  	/* If character set is not latin1_swedish
+			we have to devide character length by the
+			maximum bytes needed for that character
+			set. For example if we have unique prefix
+			index for 1 utf8 character then we have
+			actually 3 bytes allocated in the index.
+			Therefore, we have to divide that with
+			maximum bytes needed for utf8 character i.e.
+			3 byges.*/
+
+                        if ( dtuple_f_len > 0) {
+			  charset = get_charset(
+				dtype_get_charset_coll(cur_type->prtype), 
+				MYF(MY_WME));
+
+			  ut_ad(charset);
+			  ut_ad(charset->mbmaxlen);
+
+			  dtuple_f_len = dtuple_f_len / charset->mbmaxlen;
+
+                          if ( dtuple_f_len == 0)
+			    dtuple_f_len = 1;
+
+			  rec_f_len = dtuple_f_len;
+			} 
+
 			ret = cmp_whole_field(cur_type,
 				dfield_get_data(dtuple_field), dtuple_f_len,
 				rec_b_ptr, rec_f_len);