bzr push into mysql-trunk branch (gopal.shankar:4114 to 4115) Bug#14364558

[Gopal Shankar]

 4115 Gopal Shankar	2012-07-25 [merge]
      BUG#14364558 - ASSERT `TABLE_LIST->PRELOCKING_PLACEHOLDER == FALSE'
                     FAILED IN CHECK_LOCK_AND_START_STMT
      
      Problem:
      rqg_signal_resignal test hits the assert 
      DBUG_ASSERT(table_list->prelocking_placeholder == false), in
      check_lock_and_start_stmt();
      
      Analysis:
      The above assert was added as part of fix for Bug#13036505,
      to capture the fact that prelocking_placeholder should always
      be false in check_lock_and_start_stmt() in theory. But in practice,
      this assertion is hit. Even without the fix Bug#13036505 this
      problem exists. So, this problem is not an effect to Bug#13036505,
      and hence can be treated as a new issue.
      
      For now we shall remove this assert which will keep the
      old behavior. And we keep this bug open, so as to
      dig into the issue more and provide proper fix.
      
      Current observation is that, the problem is not reproducible with
      --threads=[<=3] and could be due to concurrent operations.
      
      perl runall.pl --threads=10 --queries=1M --duration=300 
      --grammar=conf/runtime/signal_resignal.yy 
      --mysqld=--max-sp-recursion-depth=10 
      --reporters=Deadlock,ErrorLog,Backtrace,Shutdown 
      --basedir=$BASEDIR --vardir=vardir  --mysqld=--log-output=file 
      --mysqld=--loose-lock-wait-timeout=1 
      --mysqld=--loose-innodb-lock-wait-timeout=1 
      --testname=rqg_signal_resignal 

    modified:
      sql/sql_base.cc
 4114 Thayumanavar	2012-07-25 [merge]
      Merging patch for bug#13699303 from mysql-5.6

    modified:
      include/mysql/thread_pool_priv.h
      sql/sql_class.cc
=== modified file 'sql/sql_base.cc'
--- a/sql/sql_base.cc	2012-07-24 07:09:42 +0000
+++ b/sql/sql_base.cc	2012-07-25 10:35:49 +0000
@@ -5352,7 +5352,6 @@ static bool check_lock_and_start_stmt(TH
     Last argument routine_modifies_data for read_lock_type_for_table()
     is ignored, as prelocking placeholder will never be set here.
   */
-  DBUG_ASSERT(table_list->prelocking_placeholder == false);
   if (table_list->lock_type == TL_WRITE_DEFAULT)
     lock_type= thd->update_lock_default;
   else if (table_list->lock_type == TL_WRITE_CONCURRENT_DEFAULT)

No bundle (reason: useless for push emails).