WL#3228 (NDB) : RBR using different table defs on slave/master

This patch corrects a problem found during testing on Solaris. The code
changes how length values are retrieved on big endian machines. The
patch allows the rpl_extraColmaster tests to run on these machines.


mysql-test/suite/rpl/r/rpl_row_create_table.result:
  WL#3228 (NDB) : RBR using different table defs on slave/master
  
  New result file with changes from merge of 5.1 main.
mysql-test/suite/rpl/t/disabled.def:
  WL#3228 (NDB) : RBR using different table defs on slave/master
  
  Disable the rpl_rwo_extraColmaster_ndb test (WL#3915) because the code
  fails on Big Endian machines. See BUG#29549 for more details.
sql/field.cc:
  WL#3228 (NDB) : RBR using different table defs on slave/master
  
  This patch corrects a problem found during testing on Solaris. The code
  changes how the store_length method processes requests for values on
  big endian machines.
sql/field.h:
  WL#3228 (NDB) : RBR using different table defs on slave/master
  
  This patch corrects a problem found during testing on Solaris. The code
  changes how the store_length method processes requests for values on
  big endian machines. It also changes the get_packed_length() method to
  use the endian-ness of the host in getting the length + packlength.
sql/rpl_record.cc:
  WL#3228 (NDB) : RBR using different table defs on slave/master
  
  This patch turns on the little endian switch (db_low_byte_first) in 
  order to ensure the values are unpack correctly from binlog as they
  are stored in little endian format in binlog.
sql/rpl_utility.cc:
  WL#3228 (NDB) : RBR using different table defs on slave/master
  
  This patch corrects a problem found during testing on Solaris. The code
  changes how the calculated field size method processes requests for 
  values on big endian machines.

1 files changed, 36 insertions, 1 deletions

diff --git a/sql/rpl_utility.cc b/sql/rpl_utility.cc
index 40937e98b27..9c0eb9891a0 100644
--- a/sql/rpl_utility.cc
+++ b/sql/rpl_utility.cc
@@ -114,8 +114,43 @@ uint32 table_def::calc_field_size(uint col, uchar *master_data)
   case MYSQL_TYPE_BLOB:
   case MYSQL_TYPE_GEOMETRY:
   {
+#if 1
+    /*
+      BUG#29549: 
+      This is currently broken for NDB, which is using big-endian
+      order when packing length of BLOB. Once they have decided how to
+      fix the issue, we can enable the code below to make sure to
+      always read the length in little-endian order.
+    */
     Field_blob fb(m_field_metadata[col]);
-    length= fb.get_packed_size(master_data);
+    length= fb.get_packed_size(master_data, TRUE);
+#else
+    /*
+      Compute the length of the data. We cannot use get_length() here
+      since it is dependent on the specific table (and also checks the
+      packlength using the internal 'table' pointer) and replication
+      is using a fixed format for storing data in the binlog.
+    */
+    switch (m_field_metadata[col]) {
+    case 1:
+      length= *master_data;
+      break;
+    case 2:
+      length= sint2korr(master_data);
+      break;
+    case 3:
+      length= uint3korr(master_data);
+      break;
+    case 4:
+      length= uint4korr(master_data);
+      break;
+    default:
+      DBUG_ASSERT(0);		// Should not come here
+      break;
+    }
+
+    length+= m_field_metadata[col];
+#endif
     break;
   }
   default: