though unneeded". It's indeed unneeded, as slave is only interested in
permanent tables, and permanent tables don't depend on temporary tables
when in row-based binlogging mode. And other CREATE TEMPORARY TABLE
(referring no table or with LIKE) already don't write the CREATE to
binlog in row-based mode. 


mysql-test/r/rpl_row_create_table.result:
  result update (note that no lines appear in SHOW BINLOG EVENTS further below, which is how we
  see the bug is fixed)
mysql-test/t/rpl_row_create_table.test:
  testing if a CREATE TEMPORARY TABLE SELECT goes to binlog (it should not)
  when in row-based binlogging mode. A few lines after, there is a SHOW
  BINLOG EVENTS; before the bugfix it showed the CREATE TEMPORARY TABLE.
sql/sql_class.h:
  a method to access select_create::create_info from outside of select_create
  ("read-only" access). Making get_thd() "read-only" too.
sql/sql_insert.cc:
  The function (hook) which writes CREATE TABLE to binlog when in row-based
  binlogging mode, for CREATE TABLE SELECT, now does nothing if the table
  is temporary (as in row-based mode, temp tables are not replicated).
  This is consistent with CREATE TEMPORARY TABLE LIKE and
  CREATE TEMPORARY TABLE, which don't write any CREATE to binlog in row-based
  mode.