Merge mysql.com:/nfsdisk1/lars/bkroot/mysql-5.1-new-rpl

into  mysql.com:/nfsdisk1/lars/MERGE/mysql-5.1-merge


sql/field.cc:
  Auto merged
sql/log_event.cc:
  Auto merged
sql/rpl_utility.cc:
  Auto merged
sql/rpl_utility.h:
  Auto merged
sql/slave.cc:
  Auto merged
sql/sql_insert.cc:
  Auto merged
sql/sql_update.cc:
  Auto merged

8 files changed, 99 insertions, 78 deletions

diff --git a/sql/field.cc b/sql/field.cc
index fa93454c757..052f1cdf6d2 100644
--- a/sql/field.cc
+++ b/sql/field.cc
@@ -6732,6 +6732,7 @@ const uint Field_varstring::MAX_SIZE= UINT_MAX16;
 int Field_varstring::do_save_field_metadata(uchar *metadata_ptr)
 {
   char *ptr= (char *)metadata_ptr;
+  DBUG_ASSERT(field_length <= 65535);
   int2store(ptr, field_length);
   return 2;
 }
diff --git a/sql/log_event.cc b/sql/log_event.cc
index 44898ca8639..f03d4a25c38 100644
--- a/sql/log_event.cc
+++ b/sql/log_event.cc
@@ -6469,6 +6469,16 @@ void Rows_log_event::print_helper(FILE *file,
   data) in the table map are initialized as zero (0). The array size is the 
   same as the columns for the table on the slave.
 
+  Additionally, values saved for field metadata on the master are saved as a 
+  string of bytes (uchar) in the binlog. A field may require 1 or more bytes
+  to store the information. In cases where values require multiple bytes 
+  (e.g. values > 255), the endian-safe methods are used to properly encode 
+  the values on the master and decode them on the slave. When the field
+  metadata values are captured on the slave, they are stored in an array of
+  type uint16. This allows the least number of casts to prevent casting bugs
+  when the field metadata is used in comparisons of field attributes. When
+  the field metadata is used for calculating addresses in pointer math, the
+  type used is uint32. 
 */
 
 /**
diff --git a/sql/rpl_utility.cc b/sql/rpl_utility.cc
index d1ce5bf3b7b..b3ca26d4c2c 100644
--- a/sql/rpl_utility.cc
+++ b/sql/rpl_utility.cc
@@ -31,31 +31,34 @@ uint32 table_def::calc_field_size(uint col, uchar *master_data) const
   switch (type(col)) {
   case MYSQL_TYPE_NEWDECIMAL:
     length= my_decimal_get_binary_size(m_field_metadata[col] >> 8, 
-             m_field_metadata[col] - ((m_field_metadata[col] >> 8) << 8));
+                                       m_field_metadata[col] & 0xff);
     break;
   case MYSQL_TYPE_DECIMAL:
   case MYSQL_TYPE_FLOAT:
   case MYSQL_TYPE_DOUBLE:
     length= m_field_metadata[col];
     break;
+  /*
+    The cases for SET and ENUM are include for completeness, however
+    both are mapped to type MYSQL_TYPE_STRING and their real types
+    are encoded in the field metadata.
+  */
   case MYSQL_TYPE_SET:
   case MYSQL_TYPE_ENUM:
   case MYSQL_TYPE_STRING:
   {
-    if (((m_field_metadata[col] & 0xff00) == (MYSQL_TYPE_SET << 8)) ||
-        ((m_field_metadata[col] & 0xff00) == (MYSQL_TYPE_ENUM << 8)))
+    uchar type= m_field_metadata[col] >> 8U;
+    if ((type == MYSQL_TYPE_SET) || (type == MYSQL_TYPE_ENUM))
       length= m_field_metadata[col] & 0x00ff;
     else
     {
-      length= m_field_metadata[col] & 0x00ff;
-      DBUG_ASSERT(length > 0);
-      if (length > 255)
-      {
-        DBUG_ASSERT(uint2korr(master_data) > 0);
-        length= uint2korr(master_data) + 2;
-      }
-      else
-        length= (uint) *master_data + 1;
+      /*
+        We are reading the actual size from the master_data record
+        because this field has the actual lengh stored in the first
+        byte.
+      */
+      length= (uint) *master_data + 1;
+      DBUG_ASSERT(length != 0);
     }
     break;
   }
@@ -95,6 +98,13 @@ uint32 table_def::calc_field_size(uint col, uchar *master_data) const
     break;
   case MYSQL_TYPE_BIT:
   {
+    /*
+      Decode the size of the bit field from the master.
+        from_len is the length in bytes from the master
+        from_bit_len is the number of extra bits stored in the master record
+      If from_bit_len is not 0, add 1 to the length to account for accurate
+      number of bytes needed.
+    */
     uint from_len= (m_field_metadata[col] >> 8U) & 0x00ff;
     uint from_bit_len= m_field_metadata[col] & 0x00ff;
     DBUG_ASSERT(from_bit_len <= 7);
@@ -136,7 +146,7 @@ uint32 table_def::calc_field_size(uint col, uchar *master_data) const
       length= *master_data;
       break;
     case 2:
-      length= sint2korr(master_data);
+      length= uint2korr(master_data);
       break;
     case 3:
       length= uint3korr(master_data);
diff --git a/sql/rpl_utility.h b/sql/rpl_utility.h
index 26edbdd1405..375715c7858 100644
--- a/sql/rpl_utility.h
+++ b/sql/rpl_utility.h
@@ -99,7 +99,7 @@ public:
           /*
             These types store a single byte.
           */
-          m_field_metadata[i]= (uchar)field_metadata[index];
+          m_field_metadata[i]= field_metadata[index];
           index++;
           break;
         }
@@ -107,14 +107,14 @@ public:
         case MYSQL_TYPE_ENUM:
         case MYSQL_TYPE_STRING:
         {
-          short int x= field_metadata[index++] << 8U; // real_type
-          x = x + field_metadata[index++];            // pack or field length
+          uint16 x= field_metadata[index++] << 8U; // real_type
+          x+= field_metadata[index++];            // pack or field length
           m_field_metadata[i]= x;
           break;
         }
         case MYSQL_TYPE_BIT:
         {
-          short int x= field_metadata[index++]; 
+          uint16 x= field_metadata[index++]; 
           x = x + (field_metadata[index++] << 8U);
           m_field_metadata[i]= x;
           break;
@@ -125,14 +125,14 @@ public:
             These types store two bytes.
           */
           char *ptr= (char *)&field_metadata[index];
-          m_field_metadata[i]= sint2korr(ptr);
+          m_field_metadata[i]= uint2korr(ptr);
           index= index + 2;
           break;
         }
         case MYSQL_TYPE_NEWDECIMAL:
         {
-          short int x= field_metadata[index++] << 8U; // precision
-          x = x + field_metadata[index++];            // decimals
+          uint16 x= field_metadata[index++] << 8U; // precision
+          x+= field_metadata[index++];            // decimals
           m_field_metadata[i]= x;
           break;
         }
diff --git a/sql/sql_delete.cc b/sql/sql_delete.cc
index 303918f42a2..d5e88ab8761 100644
--- a/sql/sql_delete.cc
+++ b/sql/sql_delete.cc
@@ -345,7 +345,7 @@ cleanup:
     thd->transaction.stmt.modified_non_trans_table= TRUE;
   
   /* See similar binlogging code in sql_update.cc, for comments */
-  if ((error < 0) || (deleted && !transactional_table))
+  if ((error < 0) || thd->transaction.stmt.modified_non_trans_table)
   {
     if (mysql_bin_log.is_open())
     {
@@ -862,7 +862,8 @@ bool multi_delete::send_eof()
   {
     query_cache_invalidate3(thd, delete_tables, 1);
   }
-  if ((local_error == 0) || (deleted && normal_tables))
+  DBUG_ASSERT(!normal_tables || !deleted || thd->transaction.stmt.modified_non_trans_table);
+  if ((local_error == 0) || thd->transaction.stmt.modified_non_trans_table)
   {
     if (mysql_bin_log.is_open())
     {
@@ -879,7 +880,6 @@ bool multi_delete::send_eof()
     if (thd->transaction.stmt.modified_non_trans_table)
       thd->transaction.all.modified_non_trans_table= TRUE;
   }
-  DBUG_ASSERT(!normal_tables || !deleted || thd->transaction.stmt.modified_non_trans_table);
 
   /* Commit or rollback the current SQL statement */
   if (transactional_tables)
diff --git a/sql/sql_insert.cc b/sql/sql_insert.cc
index ebbf4cafb19..41e4fe4957e 100644
--- a/sql/sql_insert.cc
+++ b/sql/sql_insert.cc
@@ -839,59 +839,58 @@ bool mysql_insert(THD *thd,TABLE_LIST *table_list,
     }
     transactional_table= table->file->has_transactions();
 
-    if ((changed= (info.copied || info.deleted || info.updated)) ||
-        was_insert_delayed)
+    if ((changed= (info.copied || info.deleted || info.updated)))
     {
       /*
         Invalidate the table in the query cache if something changed.
         For the transactional algorithm to work the invalidation must be
         before binlog writing and ha_autocommit_or_rollback
       */
-      if (changed)
-        query_cache_invalidate3(thd, table_list, 1);
-      if (error <= 0 || !transactional_table)
+      query_cache_invalidate3(thd, table_list, 1);
+    }
+    if (changed && error <= 0 || thd->transaction.stmt.modified_non_trans_table
+	|| was_insert_delayed)
+    {
+      if (mysql_bin_log.is_open())
       {
-        if (mysql_bin_log.is_open())
+	if (error <= 0)
         {
-          if (error <= 0)
-          {
-            /*
-              [Guilhem wrote] Temporary errors may have filled
-              thd->net.last_error/errno.  For example if there has
-              been a disk full error when writing the row, and it was
-              MyISAM, then thd->net.last_error/errno will be set to
-              "disk full"... and the my_pwrite() will wait until free
-              space appears, and so when it finishes then the
-              write_row() was entirely successful
-            */
-            /* todo: consider removing */
-            thd->clear_error();
-          }
-          /* bug#22725:
-
-          A query which per-row-loop can not be interrupted with
-          KILLED, like INSERT, and that does not invoke stored
-          routines can be binlogged with neglecting the KILLED error.
-          
-          If there was no error (error == zero) until after the end of
-          inserting loop the KILLED flag that appeared later can be
-          disregarded since previously possible invocation of stored
-          routines did not result in any error due to the KILLED.  In
-          such case the flag is ignored for constructing binlog event.
-          */
-          DBUG_ASSERT(thd->killed != THD::KILL_BAD_DATA || error > 0);
-          if (thd->binlog_query(THD::ROW_QUERY_TYPE,
-                                thd->query, thd->query_length,
-                                transactional_table, FALSE,
-                                (error>0) ? thd->killed : THD::NOT_KILLED) &&
-              transactional_table)
-          {
-            error=1;
-          }
-        }
-        if (thd->transaction.stmt.modified_non_trans_table)
-          thd->transaction.all.modified_non_trans_table= TRUE;
+	  /*
+	    [Guilhem wrote] Temporary errors may have filled
+	    thd->net.last_error/errno.  For example if there has
+	    been a disk full error when writing the row, and it was
+	    MyISAM, then thd->net.last_error/errno will be set to
+	    "disk full"... and the my_pwrite() will wait until free
+	    space appears, and so when it finishes then the
+	    write_row() was entirely successful
+	  */
+	  /* todo: consider removing */
+	  thd->clear_error();
+	}
+	/* bug#22725:
+
+	A query which per-row-loop can not be interrupted with
+	KILLED, like INSERT, and that does not invoke stored
+	routines can be binlogged with neglecting the KILLED error.
+        
+	If there was no error (error == zero) until after the end of
+	inserting loop the KILLED flag that appeared later can be
+	disregarded since previously possible invocation of stored
+	routines did not result in any error due to the KILLED.  In
+	such case the flag is ignored for constructing binlog event.
+	*/
+	DBUG_ASSERT(thd->killed != THD::KILL_BAD_DATA || error > 0);
+	if (thd->binlog_query(THD::ROW_QUERY_TYPE,
+			      thd->query, thd->query_length,
+			      transactional_table, FALSE,
+			      (error>0) ? thd->killed : THD::NOT_KILLED) &&
+	    transactional_table)
+        {
+	  error=1;
+	}
       }
+      if (thd->transaction.stmt.modified_non_trans_table)
+	thd->transaction.all.modified_non_trans_table= TRUE;
     }
     DBUG_ASSERT(transactional_table || !changed || 
                 thd->transaction.stmt.modified_non_trans_table);
@@ -3164,6 +3163,7 @@ void select_insert::abort() {
    */
   if (table)
   {
+    bool changed, transactional_table;
     /*
       If we are not in prelocked mode, we end the bulk insert started
       before.
@@ -3185,20 +3185,20 @@ void select_insert::abort() {
       If table creation failed, the number of rows modified will also be
       zero, so no check for that is made.
     */
-    if (info.copied || info.deleted || info.updated)
+    changed= (info.copied || info.deleted || info.updated);
+    transactional_table= table->file->has_transactions();
+    if (thd->transaction.stmt.modified_non_trans_table)
     {
-      DBUG_ASSERT(table != NULL);
-      if (!table->file->has_transactions())
-      {
         if (mysql_bin_log.is_open())
           thd->binlog_query(THD::ROW_QUERY_TYPE, thd->query, thd->query_length,
-                            table->file->has_transactions(), FALSE);
-        if (!thd->current_stmt_binlog_row_based && !table->s->tmp_table &&
-            !can_rollback_data())
+                            transactional_table, FALSE);
+        if (!thd->current_stmt_binlog_row_based && !can_rollback_data())
           thd->transaction.all.modified_non_trans_table= TRUE;
-        query_cache_invalidate3(thd, table, 1);
-      }
+	if (changed)
+	  query_cache_invalidate3(thd, table, 1);
     }
+    DBUG_ASSERT(transactional_table || !changed ||
+		thd->transaction.stmt.modified_non_trans_table);
     table->file->ha_release_auto_increment();
   }
 
diff --git a/sql/sql_load.cc b/sql/sql_load.cc
index 8bbe1e413b3..23547349a37 100644
--- a/sql/sql_load.cc
+++ b/sql/sql_load.cc
@@ -445,7 +445,7 @@ bool mysql_load(THD *thd,sql_exchange *ex,TABLE_LIST *table_list,
 	/* If the file was not empty, wrote_create_file is true */
 	if (lf_info.wrote_create_file)
 	{
-	  if ((info.copied || info.deleted) && !transactional_table)
+	  if (thd->transaction.stmt.modified_non_trans_table)
 	    write_execute_load_query_log_event(thd, handle_duplicates,
 					       ignore, transactional_table);
 	  else
diff --git a/sql/sql_update.cc b/sql/sql_update.cc
index d4df805d006..2a469220fa7 100644
--- a/sql/sql_update.cc
+++ b/sql/sql_update.cc
@@ -797,7 +797,7 @@ int mysql_update(THD *thd,
     Sometimes we want to binlog even if we updated no rows, in case user used
     it to be sure master and slave are in same state.
   */
-  if ((error < 0) || (updated && !transactional_table))
+  if ((error < 0) || thd->transaction.stmt.modified_non_trans_table)
   {
     if (mysql_bin_log.is_open())
     {