Putting initialization into .cc will reduce compiler's abilities
to optimize this constant away.
Defines are not OK as they bloat global namespace.
Looking for a way to declare an efficient named constant in reduced
namespace (i. e. in  a class).
Let's try enums: normally they should be implicitly casted to int.
Let's see if we really have a compiler which won't do that.

into sanja.is.com.ua:/home/bell/mysql/bk/work-5.0

fixed other Item_ref methods where result_field involved

into mysql.com:/home/pem/work/mysql-5.0-merge

This finishes (almost) WL#2002: Implement stored procedure GOTO.
Only the syntax issue for free labels remains ("label L;" vs "L:").

into sanja.is.com.ua:/home/bell/mysql/bk/work-5.0

items for functions which is converted to constants internally to support its correct printing added (BUG#4663)

and BUG#336: Subselects with tables does not work as values for
local SP variables (which was closed before with a temp. fix, but not
actually fixed).

into mysql.com:/home/pem/work/mysql-5.0-merge

unions (crashed randomly), and made "label" not a reserved word.

into sanja.is.com.ua:/home/bell/mysql/bk/work-5.0

  Fix mysql-test failure in the 5.0 tree on Aug 23: label seems to be a reserved word in 5.0, and we to use it a s a column name we must but it inside backticks

into mysql.com:/home/pem/work/mysql-5.0-merge

Mostly done, it works, but the temporary LABEL syntax still to be fixed.

into sanja.is.com.ua:/home/bell/mysql/bk/mysql-5.0

resolve UPDATE fields belonged to VIEWs as Item_field (instead of Item_ref) in prepared statements, too (BUG#4999)

into mysql.com:/dbdata/psergey/mysql-5.0-root

into brandersnatch.localdomain:/home/dlenev/src/mysql-5.0

  Added a case for bug #4976 when one of the inner tables is empty.
select.result, join.result:
  Reversed the previous change of the erronious fix for bug #4976.
sql_select.cc:
  The previous fix for bug 4976 was reversed as it
  erroniously converted an outer join into an innner join
  when on_expression does not refer to outer tables.
  This is not valid if inner tables return an empty set.
  Setting dependency on outer tables was added for the above cases.
  To fix the crash in the test case of bug #4976  
  a guard was added that blocks running the crashing code for
  nested outer joins.

 The crash is eliminated but still it is weird/inefficent that ROR-intersection is used when performing updates in empty table.