Added some more info in a number of fail cases
(re-commit for administrative reasons)

main.mysqltest skipped on Windows because a perl intentionally does exit(1)
Use exit(2), as exit(1) on Windows is indistinguishable from failing to
execute perl.

Added sanity check, similar to the one preventing send without reap

Trying to run perl fails, just like it does when perl is started but fails
Trap the case that perl was not found/could not be started, and skip test
Also force a restart of servers since test may already have done something
mtr now also appends path of current perl to PATH to aid mysqltest

This happens when creating new tmpdir due to too long socket path
Don't delete it if --start-and-exit, but warn user to do it.

Added --enable-connect-log, somewhet similar to --enable-query-log
If query log is disabled, disable connect log too
Also some related cleanup in mysqltest.test: removing duplicate test loop

Follow-up: don't call pthread_join() on Windows
This change only valid for 5.1

When stepping backward to end of `` expression, check for illegal chars

Bug #55546 mysqltest fails to create a new thread on HPUX
Missing call to pthread_join(), in embedded mode
This independently solves both problems, see 55426 for details.
Addendum: cannot test against a pthread_t, adds boolean flag instead

Moved an init_dynamic_string() from before to after potential die()

Added --force-restart

Allow --testcase-timeout=<mins> to be set in .opt file for test

Added code resulted in strange linking problem for embedded on Windows
Avoided by not doing this for embedded mode
It's irrelevant for embedded server anyway, --protocol will be ignored

Allow escaped quotes also in statements not starting with --
But will not support single unescaped ' or `

Follow-up patch: added test cases for -0 and while

if() treated any non-numeric string as false
Fixed to treat those as true instead
Added some test cases
Fixed missing $ in variable name in include/mix2.inc

A CREATE...SELECT that fails is written to the binary log if a non-transactional
statement is updated. If the logging format is ROW, the CREATE statement and the
changes are written to the binary log as distinct events and by consequence the
created table is not rolled back in the slave.

In this patch, we opted to let the slave goes out of sync by not writting to the
binary log the CREATE statement. We do this by simply reseting the binary log's
cache.

The expect file can now include "restart:<server options>"
Also drop check-testcase if this has been done
Added comment explaining the restart: syntax

Queries may crash, if
  1) the GREATEST or the LEAST function has a mixed list of
     numeric and LONGBLOB arguments and
  2) the result of such a function goes through an intermediate
     temporary table.

An Item that references a LONGBLOB field has max_length of
UINT_MAX32 == (2^32 - 1).

The current implementation of GREATEST/LEAST returns REAL
result for a mixed list of numeric and string arguments (that
contradicts with the current documentation, this contradiction
was discussed and it was decided to update the documentation).

The max_length of such a function call was calculated as a
maximum of argument max_length values (i.e. UINT_MAX32).

That max_length value of UINT_MAX32 was used as a length for
the intermediate temporary table Field_double to hold
GREATEST/LEAST function result.

The Field_double::val_str() method call on that field
allocates a String value.

Since an allocation of String reserves an additional byte
for a zero-termination, the size of String buffer was
set to (UINT_MAX32 + 1), that caused an integer overflow:
actually, an empty buffer of size 0 was allocated.

An initialization of the "first" byte of that zero-size
buffer with '\0' caused a crash.

The Item_func_min_max::fix_length_and_dec() has been
modified to calculate max_length for the REAL result like
we do it for arithmetical operators.


******
Bug #54461: crash with longblob and union or update with subquery

Queries may crash, if
  1) the GREATEST or the LEAST function has a mixed list of
     numeric and LONGBLOB arguments and
  2) the result of such a function goes through an intermediate
     temporary table.

An Item that references a LONGBLOB field has max_length of
UINT_MAX32 == (2^32 - 1).

The current implementation of GREATEST/LEAST returns REAL
result for a mixed list of numeric and string arguments (that
contradicts with the current documentation, this contradiction
was discussed and it was decided to update the documentation).

The max_length of such a function call was calculated as a
maximum of argument max_length values (i.e. UINT_MAX32).

That max_length value of UINT_MAX32 was used as a length for
the intermediate temporary table Field_double to hold
GREATEST/LEAST function result.

The Field_double::val_str() method call on that field
allocates a String value.

Since an allocation of String reserves an additional byte
for a zero-termination, the size of String buffer was
set to (UINT_MAX32 + 1), that caused an integer overflow:
actually, an empty buffer of size 0 was allocated.

An initialization of the "first" byte of that zero-size
buffer with '\0' caused a crash.

The Item_func_min_max::fix_length_and_dec() has been
modified to calculate max_length for the REAL result like
we do it for arithmetical operators.

Fix compiler warnings.

mostly because existing test result files were not updated.

In order to be able to check if the set of the grouping fields in a 
GROUP BY has changed (and thus to start a new group) the optimizer
caches the current values of these fields in a set of Cached_item 
derived objects.
The Cached_item_str, used for caching varchar and TEXT columns,
is limited in length by the max_sort_length variable.
A String buffer to store the value with an alloced length of either
the max length of the string or the value of max_sort_length 
(whichever is smaller) in Cached_item_str's constructor.
Then, at compare time the value of the string to compare to was 
truncated to the alloced length of the string buffer inside 
Cached_item_str.
This is all fine and valid, but only if you're not assigning 
values near or equal to the alloced length of this buffer.
Because when assigning values like this the alloced length is 
rounded up and as a result the next set of data will not match the
group buffer, thus leading to wrong results because of the changed
alloced_length.
Fixed by preserving the original maximum length in the 
Cached_item_str's constructor and using this instead of the 
alloced_length to limit the string to compare to.
Test case added.

Fix a regression (due to a typo) which caused spurious incorrect
argument errors for long data stream parameters if all forms of
logging were disabled (binary, general and slow logs).

Fix a regression (due to a typo) which caused spurious incorrect
argument errors for long data stream parameters if all forms of
logging were disabled (binary, general and slow logs).