This was a bug in the code of MDEV-12387 "Push conditions into materialized
subqueries". The bug manifested itself in rather rare situations. An
affected query must contain IN subquery predicate whose left operand
was an outer field of a mergeable derived table or view and right operand
was a materialized subquery.
The erroneous code in fact stripped off the Item_direct_ref wrapper from
the left operand of the IN subquery predicate when building equalities
produced by the conversion of the predicate into a semi-join. As a result
the left operand was not considered as an outer reference anymore and
used_tables() was calculated incorrectly. This caused a crash in the
function optimize_keyuse().
- remove function prototype for shared memory (no more used), and VIO
members that are unused
- Do not call DisconnectNamedPipe on pipe handle. CloseHandle() is enough.
Remove threads that are doing nothing but wait
- main thread now handles the connections
(if threadpool is used, also threadpool threads would wait for connections)
- thread for socket and pipe connections are removed
- shutdown thread is now removed, we wait for shutdown
notification in main thread as well
- kill_server() is also called inside the main thread, after connection
loop finished.
btr_cur_instant_root_init(): Check the "infimum" and "supremum"
record strings already here, and not later in btr_cur_instant_root_init().
In this way, we can properly reject files from later versions where
instant ALTER TABLE could support further operations that change the
format of InnoDB clustered indexes.
Because changes of the FIL_PAGE_TYPE or PAGE_INSTANT in the root
page are not undo-logged, it is possible that the fields suggest
that instant ADD COLUMN is in effect, even though no metadata
record exists. If the fields are set, proceed to fetch the
metadata record. If the metadata record does not exist, return
success if !index->is_instant().
Also, check that the "infimum" and "supremum" records carry the
strings in the root page. In a later format that supports instant
DROP COLUMN, we will have to store more information in the root
page, so that index->n_core_null_bytes can be determined accurately.
btr_cur_instant_init_low(): If columns were instantly added and dropped,
then index->is_instant() might not hold even though the root page type
was FIL_PAGE_TYPE_INSTANT. MariaDB 10.3 must refuse to open such files,
because instant DROP COLUMN is not supported.
Also, refuse to open the table if the metadata record has
wrong info OR status bits. Previously, we only refused to open
if both bits were wrong.
truncating a temporary table
TRUNCATE expects only one TABLE instance (which is used by TRUNCATE
itself) to be open. However this requirement wasn't enforced after
"MDEV-5535: Cannot reopen temporary table".
Fixed by closing unused table instances before performing TRUNCATE.
Patch fixes two bugs:
1) BEGIN_TIMESTAMP of MYSQL.TRANSACTION_REGISTY is '0-0-0' on replication record insertion
2) BEGIN_TIMESTAMP equals COMMMIT_TIMESTAMP in MYSQL.TRANSACTION_REGISTRY
Fixed by calling THD::set_time() at appropriate places
Fixing C++ function check_date() to get the "fuzzydate" as
date_mode_t rather than ulonglong, so conversion from
date_time_t to ulonglong is now done inside C++ check_date(),
and no conversion is needed in the callers' code.
As an additional safety, modified the code not to pass
TIME_FUZZY_DATE to the low level C functions:
- check_date()
- str_to_datetime()
- str_to_time()
- number_to_datetime()
because TIME_FUZZY_DATE is known only on the C++ level,
C functions do not know it.
Soon we'll be adding more flags into the C++ level (i.e. to date_time_t),
e.g. for rounding. It's a good idea to prevent passing C++ specific
flags into pure C routines before this change.
Asserts were added into the affected C functions to verify
that the caller passed only known C level flags.
Also fixes:
MDEV-17330 Wrong result for 0 + LEAST(TIME'-10:00:00',TIME'10:00:00')
Problems:
1. These methods did not take into account the current session date flags
and passed date_mode_t(0) to func->get_date():
Type_handler_temporal_result::Item_func_min_max_val_real
Type_handler_temporal_result::Item_func_min_max_val_int
Type_handler_temporal_result::Item_func_min_max_val_decimal
Fixing to pass sql_mode_for_dates(thd) instead of date_mode_t(0).
Note, sql_mode_for_dates(thd) is only needed for DATE/DATETIME
data types. It is not needed for TIME.
So splitting value methods Type_handler_temporal_result::Item_func_min_max_xxx
into individual implementations for
Type_handler_{time|date|datetime|timestamp}_common
and, instead of calling get_date(), reusing inside classes
Time(), Date(), Datetime() and their methods to_longlong().
sql_mode_for_dates(thd) is automatically passed to get_date()
inside Date() and Datetime() constructors.
The switch to classes also fixed the problem reported in MDEV-17330.
Type_handler_temporal_result::Item_func_min_max_val_int() used to
call TIME_to_ulonglong(), which was not correct for TIME.
Changing the code to use Time().to_longlong() solved this.
2. Type_handler_temporal_result::Item_func_min_max_get_date
also did not take into account the current session
date flags in case of conversion from DATE/DATETIME to time
and passed date_mode_t(0) to get_date_native().
Fixing to pass sql_mode_for_dates(thd) in case of conversion
from DATE/DATETIME to TIME.
storage/rocksdb/rdb_datadic.cc: In member function 'int myrocks::Rdb_key_def::unpack_integer(myrocks::Rdb_field_packing*, Field*, uchar*, myrocks::Rdb_string_reader*, myrocks::Rdb_string_reader*) const'
storage/rocksdb/rdb_datadic.cc:1781:1: internal compiler error: Segmentation fault
}
on ppc64le, ubuntu bionic gcc 7.3.0 and debian stretch gcc 6.3.0
The error happens with -ftree-loop-vectorize when trying to vectorize
a particular loop (see Rdb_key_def::unpack_integer())
Compiler gets confused by __attribute__((optimize("O0")) that comes from
ha_rocksdb_proto.h. The intention of this __attribute__ was to prevent
function from being inlined (see ha_rocksdb.cc). Let's use a more
specific attribute that prevents inlining but does not confuse
loop vectorizer.
An error in "group commit with MariaDB's binlog" code: we would flush
the WAL even when the transaction did not do any writes (and so the logic
in myrocks::Rdb_transaction::commit caused it to rollback).
Unary minus operation for the smallest possible signed long long value
(LONLONG_MIN) is undefined in C++. Because of this, func_time.test
failed on ppc64 buildbot machines.
Fixing the code to avod using undefined operations.
This is fix is similar to "MDEV-7973 bigint fail with gcc 5.0"
