5.0 and 4.1 (where it fails).

One test in subselect.test still wrong.
Had to reconstruct all errmsg.txt files completely, since auto-merge made
a mess of the lot. :-(

into myvenu.com:/home/venu/work/sql/dev-4.1

using a hash table instead (and made it work with lex pointers).
Some additional code cleanup too.

into sanja.is.com.ua:/home/bell/mysql/bk/work-alloc_group-4.1

(now it return correct results. Test was commited before, but it is not pushed,
BUG# is not registered, because bug was introduced after release)

fixed test func_gconcat to be repeatable independent of presend tables and avoid removing user tables

initialization of item_thd moved to constructor (in any case we need thd in constructor)
initialization of group_concat_max_len to constructor to avoid incorrect length reporting (BUG#757)
removed Item_func_group_concat::fix_length_and_dec() because item have its own fix_fields and will not have inherited items

into sanja.is.com.ua:/home/bell/mysql/bk/work-alloc_group-4.1

into myvenu.com:/home/venu/work/sql/dev-4.1

into bar.mysql.r18.ru:/usr/home/bar/mysql-4.1

  Item_func::agg_arg_collations()
  Item_func::arr_arg_collations_for_comparison()
 to aggregate argument collations.
It helps to reuse a lot of code.

into serg.mylan:/usr/home/serg/Abk/mysql-4.1

into mysql.r18.ru:/usr/home/ram/work/4.1

into serg.mylan:/usr/home/serg/Abk/mysql-4.1

into narttu.mysql.fi:/my/mysql-4.1

into bar.mysql.r18.ru:/usr/home/bar/mysql-4.1

into sanja.is.com.ua:/home/bell/mysql/bk/work-4.1

  ROW('a','b','c) = ROW('A' collate latin1_bin,'b','c') returns 0
When a number is compared to a string, character sets and 
collations are not aggregated. e.g. this returned error in 4.1.0:

SELECT 1=_latin2'1';
because character sets was aggregated, and 1 was considered as
a string of latin1 charset during this aggregation.

into narttu.mysql.fi:/my/mysql-4.1

into sanja.is.com.ua:/home/bell/mysql/bk/work-4.1

into sanja.is.com.ua:/home/bell/mysql/bk/work-alloc_group-4.1