that implement add_index
The problem was that ALTER TABLE blocked reads on an InnoDB table
while adding a secondary index, even if this was not needed. It is
only needed for the final step where the .frm file is updated.
The reason queries were blocked, was that ALTER TABLE upgraded the
metadata lock from MDL_SHARED_NO_WRITE (which blocks writes) to
MDL_EXCLUSIVE (which blocks all accesses) before index creation.
The way the server handles index creation, is that storage engines
publish their capabilities to the server and the server determines
which of the following three ways this can be handled: 1) build a
new version of the table; 2) change the existing table but with
exclusive metadata lock; 3) change the existing table but without
metadata lock upgrade.
For InnoDB and secondary index creation, option 3) should have been
selected. However this failed for two reasons. First, InnoDB did
not publish this capability properly.
Second, the ALTER TABLE code failed to made proper use of the
information supplied by the storage engine. A variable
need_lock_for_indexes was set accordingly, but was not later used.
This patch fixes this problem by only doing metadata lock upgrade
before index creation/deletion if this variable has been set.
This patch also changes some of the related terminology used
in the code. Specifically the use of "fast" and "online" with
respect to ALTER TABLE. "Fast" was used to indicate that an
ALTER TABLE operation could be done without involving a
temporary table. "Fast" has been renamed "in-place" to more
accurately describe the behavior.
"Online" meant that the operation could be done without taking
a table lock. However, in the current implementation writes
are always prohibited during ALTER TABLE and an exclusive
metadata lock is held while updating the .frm, so ALTER TABLE
is not completely online. This patch replaces "online" with
"in-place", with additional comments indicating if concurrent
reads are allowed during index creation/deletion or not.
An important part of this update of terminology is renaming
of the handler flags used by handlers to indicate if index
creation/deletion can be done in-place and if concurrent reads
are allowed. For example, the HA_ONLINE_ADD_INDEX_NO_WRITES
flag has been renamed to HA_INPLACE_ADD_INDEX_NO_READ_WRITE,
while HA_ONLINE_ADD_INDEX is now HA_INPLACE_ADD_INDEX_NO_WRITE.
Note that this is a rename to clarify current behavior, the
flag values have not changed and no flags have been removed or
added.
Test case added to innodb_mysql_sync.test.
ZERO
When dates are represented internally as strings, i.e. when a string constant
is compared to a date value, both values are converted to long integers,
ostensibly for fast comparisons. DATE typed integer values are converted to
DATETIME by multiplying by 1,000,000 (each digit pair representing hour,
minute and second, respectively). But the mechanism did not distuinguish
cached INTEGER values, already in correct format, from newly converted
strings.
Fixed by marking the INTEGER cache as being of DATETIME format.
Fix for bug#45740 introduced test case using SHOW TABLE STATUS against a Memory table using latin1 character in table name.
The test failed on Windows and FreeBSD due to a difference in the value for Avg_row_length.
The average row length normally depends on the values for data length and row count. According to the 5.5 manual data length is approximate with Memory tables.
With MyISAM and InnoDB the Avg_row_length is the same on Windows and Solaris.
The solution implemented by this patch is to mask out the value for Avg_row_length, as it may vary when using Memory tables.
Problem: the scanner function tested for strings "<![CDATA[" and
"-->" without checking input string boundaries, which led to valgrind's
"Conditional jump or move depends on uninitialised value(s)" error.
Fix: Adding boundary checking.
@ mysql-test/r/xml.result
@ mysql-test/t/xml.test
Adding test
@ strings/xml.c
Adding a helper function my_xml_parser_prefix_cmp(),
with input string boundary check.
Introduced by the fix for bug#44766.
Problem: it's not correct to use args[0]->str_value as a buffer,
because args[0] may need this buffer for its own purposes.
Fix: adding a new class member tmp_value to use as return value.
@ mysql-test/r/ctype_many.result
@ mysql-test/t/ctype_many.test
Adding tests
@ sql/item_strfunc.cc
Changing code into traditional style:
use "str" as a buffer for the argument and tmp_value for the result value.
@ sql/item_strfunc.h
Adding tmp_value
Problem: when processing a query like:
SELECT '' LIKE '1' ESCAPE COUNT(1);
escape_item->val_str() was never executed and the "escape" class member
stayed initialized, which led to valgrind uninitialized memory error.
Note, a query with some tables in "FROM" clause
returns ER_WRONG_ARGUMENTS in the same situation:
SELECT '' LIKE '1' ESCAPE COUNT(1) FROM t1;
ERROR 1210 (HY000): Incorrect arguments to ESCAPE
Fix: disallowing using aggregate functions in ESCAPE clause,
even if there are no tables used. There is no much use of that anyway.
When mysqldadmin is run with sleep and count options,
it goes into an infinite loop and keeps executing the
specified command.
This happened because the statement, responsible for
decrementing the count value, was missing.
Fixed by adding a statement which will decrement the
count value for each iteration.
IA64 and some other arcitectures use different float rounding mode and
i find no decent way to make it consistent.
So the test changed to be insensitive to this.
per-file messages:
mysql-test/t/gis.test
Bug#52208 gis fails on some platforms (Solaris, HP-UX, Linux)
--replace_result added
When mysqldump tries to dump information in xml format,
the result does not contain field level comments.
In order to retrieve various informations for a field/column,
mysqldump currently uses 'show fields from <tab>' statement.
The attributes returned by the statement lacks the information
regarding field comments.
Fixed by changing the query to one that probes I_S to retrieve
required field informations, including the field comment.
other crashes
Some string manipulating SQL functions use a shared string object intended to
contain an immutable empty string. This object was used by the SQL function
SUBSTRING_INDEX() to return an empty string when one argument was of the wrong
datatype. If the string object was then modified by the sql function INSERT(),
undefined behavior ensued.
Fixed by instead modifying the string object representing the function's
result value whenever string manipulating SQL functions return an empty
string.
Relevant code has also been documented.
The test case fails with out of memory while updating a table
with several multi-megabytes sized rows. This can probably be too
exhausting for PB2 env.
The quick fix here is to reduce the size of the biggest
row (256MB) so that it becomes a little smaller (64MB).
INVOKER-security view access check wrong".
When privilege checks were done for tables used from an
INVOKER-security view which in its turn was used from
a DEFINER-security view connection's active security
context was incorrectly used instead of security context
with privileges of the second view's creator.
This meant that users which had enough rights to access
the DEFINER-security view and as result were supposed to
be able successfully access it were unable to do so in
cases when they didn't have privileges on underlying tables
of the INVOKER-security view.
This problem was caused by the fact that for INVOKER-security
views TABLE_LIST::security_ctx member for underlying tables
were set to 0 even in cases when particular view was used from
another DEFINER-security view. This meant that when checks of
privileges on these underlying tables was done in
setup_tables_and_check_access() active connection security
context was used instead of context corresponding to the
creator of caller view.
This fix addresses the problem by ensuring that underlying
tables of an INVOKER-security view inherit security context
from the view and thus correct security context is used for
privilege checks on underlying tables in cases when such view
is used from another view with DEFINER-security.
Item_func_spatial_collection::fix_length_and_dec didn't call parent's method, so
the maybe_null was set to '0' after it. But in this case the result was
just NULL, that caused wrong behaviour.
per-file comments:
mysql-test/r/gis.result
Bug #57321 crashes and valgrind errors from spatial types
test result updated.
mysql-test/t/gis.test
Bug #57321 crashes and valgrind errors from spatial types
test case added.
sql/item_geofunc.h
Bug #57321 crashes and valgrind errors from spatial types
Item_func_geometry::fix_length_and_dec() called in
Item_func_spatial_collection::fix_length_and_dec().
tmptable needed
The function DEFAULT() works by modifying the the data buffer pointers (often
referred to as 'record' or 'table record') of its argument. This modification
is done during name resolution (fix_fields().) Unfortunately, the same
modification is done when creating a temporary table, because default values
need to propagate to the new table.
Fixed by skipping the pointer modification for fields that are arguments to
the DEFAULT function.
This assert could be triggered if -1 was inserted into
an auto increment column by a statement writing more than
one row.
Unless explicitly given, an interval of auto increment values
is generated when a statement first needs an auto increment
value. The triggered assert checks that the auto increment
counter is equal to or higher than the lower bound of this
interval.
Generally, the auto increment counter starts at 1 and is
incremented by 1 each time it is used. However, inserting an
explicit value into the auto increment column, sets the auto
increment counter to this value + 1 if this value is higher
than the current value of the auto increment counter.
This bug was triggered if the explicit value was -1. Since the
value was converted to unsigned before any comparisons were made,
it was found to be higher than the current vale of the auto
increment counter and the counter was set to -1 + 1. This value
was below the reserved interval and caused the assert to be
triggered the next time the statement tried to write a row.
With the patch for Bug#39828, this bug is no longer repeatable.
Now, -1 + 1 is detected as an "overflow" which causes the auto
increment counter to be set to ULONGLONG_MAX. This avoids hitting
the assert for the next insert and causes a new interval of
auto increment values to be generated. This resolves the issue.
This patch therefore only contains a regression test and no code
changes. Test case added to auto_increment.test.
Enabled test snippet for bug 4374, tested on Mac OS X 10.6 as well as Solaris.
Moved test snippet to a different place in the file, in order to avoid having
to save and restore "SET NAMES" setting. New surroundings expect latin1, as is
used in the testsnippet.
An extra copy of the commented test snippet is removed, a comment is added,
SQL keywords are converted to uppercase, and engine name "heap" is updated to
"Memory".
Also added Copyright statement and a notice about the file's encoding(s).
Fixed incorrect checks in join_read_const_table() for when to
accept a non-existing, or empty const-row as a part of the const'ified
set of tables.
Intention of this test is to only accept NULL-rows if this table is outer joined
into the resultset. (In case of an inner-join we can conclude at this point that
resultset will be empty, end we want to return 'error' to signal this.)
Initially 'maybe_null' is set to the same value as 'outer_join' in
setup_table_map(), mysql_priv.h ~line 2424. Later simplify_joins() will
attemp to replace outer joins by inner join whenever possible. This
will cause 'outer_join' to be updated. However, 'maybe_null' is *not* updated
to reflect this rewrite as this field is used to currectly set the 'nullability'
property for the columns in the resultset.
We should therefore change join_read_const_table() to check the 'outer_join'
property instead of 'maybe_null', as this correctly reflect the nullability of
the *execution plan* (not *resultset*).
Item_equal::val_int() checked for NULL-values by checking Item::null_value
*before* the respective ::store_value() and ::cmp(Item*) metods where called.
As Item::null_value is set by these metods, the value of 'null_value'
is not valid until *after* ::store_value() or ::cmp() has
been called for the Item object.
Fix is to swap order of ::store_value()/::cmp() and checking of Item::null_value.
This pattern is widely used other places inside item_cmpfunc.cc .
TIMESTAMP.
Item_cache::get_cache wasn't treating TIMESTAMP as a DATETIME value thus
returning string cache for items with TIMESTAMP type. This led to incorrect
TIMESTAMP -> INT conversion and to a wrong query result.
Fixed by using Item::is_datetime function to check for DATETIME type group.
If the ::single_value_transformer() find an existing HAVING condition it used
to do the transformation:
1) HAVING cond -> (HAVING Cond) AND (cond_guard (Item_ref_null_helper(...))
As the AND condition in 1) is Mc'Carty evaluated, the
right side of the AND cond should be executed only if the
original 'HAVING evaluated' to true.
However, as we failed to set 'top_level' for the tranformed HAVING condition,
'abort_on_null' was FALSE after transformation. An
UNKNOWN having condition will then not terminate evaluation of the
transformed having condition, and we incorrectly continued
into the Item_ref_null_helper() part.
Put descriptions of plugins into a separate file read by MTR
MTR itself has generalised code to read this and set env. variables
Removed the *SO variables, updated some tests accordingly
New commit: added optional list of plugin names for _LOAD variable
Also made changes for the new AUTH_* plugins
OPTIMIZE TABLE
OPTIMIZE TABLE for InnoDB tables is handled as recreate + analyze.
The triggered assert checked that an error had been reported if either
recreate or analyze failed. However the assert failed to take into
account that they could have failed because OPTIMIZE TABLE had been
victim of KILL QUERY, KILL CONNECTION or server shutdown.
This patch adjusts the assert to take this possibility into account.
The problem was only noticeable on debug versions of the server.
Test case added to innodb_mysql_sync.test.
and Order By
When having a UNION statement in a subquery, with no
referenced tables (or only a reference to the virtual
table 'dual'), the UNION did not allow an ORDER BY clause.
i.e:
SELECT(SELECT 1 AS a UNION
SELECT 0 AS a
ORDER BY a) AS b or
SELECT(SELECT 1 AS a FROM dual UNION
SELECT 0 as a
ORDER BY a) AS b
In addition, an ORDER BY / LIMIT clause was not accepted
in subqueries even for single SELECT statements with no
referenced tables (or with 'dual' as table reference)
i.e:
SELECT(SELECT 1 AS a ORDER BY a) AS b or
SELECT(SELECT 1 AS a FROM dual ORDER BY a) AS b
The fix was to allow an optional ORDER BY/LIMIT clause to
the grammar for these cases.
See also: Bug#57986
