BUG#8585 ndb_cache2

 - When deleting from a cursor the m_rows_changed variable was not properly incremented to indicate that m_share->commit_count should be cleared at end of trans.
  

BitKeeper/etc/logging_ok

@@ -120,6 +120,7 @@ kostja@oak.local
 lars@mysql.com
 lenz@kallisto.mysql.com
 lenz@mysql.com
+magnus@msdesk.mysql.com
 magnus@neptunus.(none)
 magnus@shellback.(none)
 marko@hundin.mysql.fi

sql/ha_ndbcluster.cc

@@ -2171,6 +2171,7 @@ int ha_ndbcluster::delete_row(const byte *record)
   DBUG_ENTER("delete_row");
 
   statistic_increment(thd->status_var.ha_delete_count,&LOCK_status);
+  m_rows_changed++;
 
   if (cursor)
   {
@@ -2221,8 +2222,6 @@ int ha_ndbcluster::delete_row(const byte *record)
     }
   }
 
-  m_rows_changed++;
-
   // Execute delete operation
   if (execute_no_commit(this,trans) != 0) {
     no_uncommitted_rows_execute_failure();