From: Chaithra Gopalareddy
Date: April 18 2012 9:49am
Subject: bzr push into mysql-trunk branch (chaithra.gopalareddy:3894 to 3895)
 Bug#12713907 WL#5558
List-Archive: http://lists.mysql.com/commits/143542
X-Bug: 12713907
Message-Id: <201204180949.q3I9nKt9019543@acsmt358.oracle.com>
MIME-Version: 1.0
Content-Type: text/plain; charset="us-ascii"
Content-Transfer-Encoding: 7bit

 3895 Chaithra Gopalareddy	2012-04-18
      PROBLEM:
      This is a follow up patch on 5.6 for the bug#12713907.
      This bug is not reproducible on 5.6 as changes for WL#5558
      hides the problem.
      
      In 5.6, since keyread is set to true in test_if_skip_sot_order
      for both 'QUICK_RANGE_SELECT' and 'JT_INDEX_SCAN', we need
      to move the call for set_keyread to join_init_read_record.
      Access function for JT_INDEX_SCAN already has the call to
      'set_keyread'.
      
      Note that a Work log is being created to clean up the
      set_keyread calls from places other than access functions.
      For now, we are only removing the call from
      test_if_skip_sort_order and moving it to
      join_init_read_record.
     @ sql/sql_executor.cc
        Call set_keyread from access function join_init_read_record
     @ sql/sql_select.cc
        Remove the call to set_keyread from test_if_skip_sort_order

    modified:
      sql/sql_executor.cc
      sql/sql_select.cc
 3894 Nuno Carvalho	2012-04-18 [merge]
      WL#6236: Allow SHOW MASTER LOGS and SHOW BINARY LOGS with REPLICATION CLIENT
      
      Merge from 5.5 into 5.6.

    modified:
      mysql-test/suite/binlog/r/binlog_grant.result
      mysql-test/suite/binlog/t/binlog_grant.test
      sql/sql_parse.cc
=== modified file 'sql/sql_executor.cc'
--- a/sql/sql_executor.cc	2012-04-12 15:31:01 +0000
+++ b/sql/sql_executor.cc	2012-04-18 09:46:51 +0000
@@ -3145,6 +3145,18 @@ int join_init_read_record(JOIN_TAB *tab)
   if (init_read_record(&tab->read_record, tab->join->thd, tab->table,
                        tab->select, 1, 1, FALSE))
     return 1;
+  /*
+    set keyread to TRUE if quick index is covering.
+    @todo: Call set_keyread only from access functions. Currently this is
+    also done in make_join_readinfo.
+  */
+  if (tab->select && tab->select->quick)
+  {
+    TABLE *table= tab->table;
+    if(!table->no_keyread && tab->select->quick->index != MAX_KEY //not index merge
+       && table->covering_keys.is_set(tab->select->quick->index))
+      table->set_keyread(true);
+  }
   return (*tab->read_record.read_record)(&tab->read_record);
 }
 

=== modified file 'sql/sql_select.cc'
--- a/sql/sql_select.cc	2012-04-18 06:07:33 +0000
+++ b/sql/sql_select.cc	2012-04-18 09:46:51 +0000
@@ -3974,17 +3974,12 @@ check_reverse_order:
       bool quick_created= 
         (select && select->quick && select->quick!=save_quick);
 
-      /* 
-         If 'best_key' has changed from  prev. 'ref_key':
-         Update strategy for using index tree reading only
-         ('Using index' in EXPLAIN)
+      /*
+        If ref_key used index tree reading only ('Using index' in EXPLAIN),
+        and best_key doesn't, then revert the decision.
       */
-      if (best_key != ref_key)
-      {
-        const bool using_index= 
-          (table->covering_keys.is_set(best_key) && !table->no_keyread);
-        table->set_keyread(using_index);
-      }
+      if(!table->covering_keys.is_set(best_key))
+        table->set_keyread(false);
       if (!quick_created)
       {
         if (select)                  // Throw any existing quick select

No bundle (reason: useless for push emails).