in projection list of the queries with GROUP BY and ORDER BY.
Removed support for Item_equal b/c it is an internal format MDB uses
and shouldn't be used outside of MDB's optimizer.
any function in projection looking for either aggregation function or
GROUP BY key column.
MDB doesn't convert Item_equal into corresponding Item_func_eq w/o
OPTIMIZER_SWITCH_COND_PUSHDOWN_FROM_HAVING optimizer flag.
Activation of the flag fixes queries from having test scenario w
equal operator in HAVING, e.g. HAVING c1 = 50.
This happens in external_lock() whilst locking the table.
Fixes LIMIT=1 optimization for EXISTS_SUBS subqueries.
external_lock() contains if condition that gives false positive
for SH + pushed conditions.
external_lock() now resets in_subquery_conversion_threshold
variable that governs IN_INTO_SUBQUERY optimization for
queries run in table mode.
external_lock() now purges dynamicall allocated condInfo for
SH and DH execution path.
Commented out UNION check b/c if condition gives false positives
and silently enables table mode execution for queries w/o
UNION.
This patch:
* Moves ColumnStore my.cnf to /etc/my.cnf.d/columnstore.cnf
* Removes unneeded entries from columnstore.cnf
* Removes some things that used my.cnf and are now dead
* Also removes utils/scenarios
* Modifies things that use extra defaults file to use the standard one
* Makes sure that C++11 standard is used for older CMake versions
With this patch we no longer need to set -DINSTALL_SYSCONF2DIR and
-DINSTALL_SYSCONFDIR when building MariaDB server.
2. Fallback to using sql_select_limit system variable value if an explicit limit is not supplied.
3. Remove checks that did not allow (2^64 - 1) as a valid limit value.
I returned if predicate that guards FE-BE initial connection
block. This predicated had been removed by accident in cb36041
Since we switched to internal ORDER BY FE now reduces the
default limit value to uint64-2 this broke a predicate check
for correlated subqueries. I replaced a predicate with more
reasonable to avoid false positives.
call.
CS doesn't use SH for SELECT..INTO OUTFILE queries.
Clean up gwi::physTableList when processing Storage API request.
SH now explicitly set an execution error in THD::stmt_da.
SH now set queryState in select_next() to mark a begging of the execution.
SH query execution migrated from SH::init() into create_SH().
There is a session variable columnstore_processing_handlers_fallback
that allows to fallback to DH, GBH if SH fails. DH now uses semantic
tree check for unsupported features to allow to fallback to GBH or
storage API.
Fixes GBH related bug when create_GBH() returns a handler for
queries with impossible WHERE/HAVING.
Fixed bug in FromSubquery::transform() where isUnion is set to true.
Enabled RTTI b/c server team enabled it for MDB.
Removed unused code supposed to be used with vtable.
2. CASE function should return false when it evaluates to NULL (e.g. due to absense of ELSE clause)
3. Set the operation type of IN function to varchar if all parameters are char/varchar/text
Intro* INSERT statements could face a non-existant block when MCOL-498 feature
is enabled. writeRow() guard blocks was supposed to proactively create empty
blocks. The pre-patch logic failed when first value in the block has been
removed by DELETE and this overwrites the whole valid block with empty magics.
This patch moves proactive creation logic into allocRowId().
MariaDB 10.4 only allows the system root user to log in as the root DB
user. This means the SQL commands to install the plugins won't run on a
non-root install. This patch skips the grant tables for plugin install
and does some workarounds to get the plugins in.
The MariaDB handler API changed for write_row(). The 'buf' parameter is
now a 'const'. This meant that our implementation didn't match the
virtual call so ours was no longer called. This implemented the 'const'.
We preserve union related condition blocks until UNION-capable smart handler.
Removed a number of commented code blocks.
Add projection list REAL and TIME constants, e.g
SELECT TIME'10:20:30', TIMESTAMP'2001-01-01 10:20:30', DATE'2001-01-01' FROM cs1;
Marked potentially useless block in ORDER BY processing for future removal.
Removed string variables used for reconstructed query in the pre-1.4 fork.
Removed cp_get_plan().
GBH doesn't step in if SH processes the query.
All smart handlers now correctly processes impossible WHERE and HAVING.
