MDEV-22337 Assertion `Alloced_length >= (str_length + length + net_le… …ngth_size(length))' failed in Binary_string::q_net_store_data on long MULTIPOLYGON query with session_track_user_variables=1 (optimized builds).

We have to reserve enough space in String to use q_something().
Also pointer calculations fixed.

…while using json_arrayagg() as a window function.

We don't support JSON_ARRAYAGG and JSON_OBJECTAGG in window functions
yet.

Problem:
=======
SET @@GLOBAL.replicate_wild_ignore_table='';
SET @@GLOBAL.replicate_wild_do_table='';

Reports following valgrind error.

Conditional jump or move depends on uninitialised value(s)
Rpl_filter::set_wild_ignore_table(char const*) (rpl_filter.cc:439)

Conditional jump or move depends on uninitialised value(s)
at 0xF60390: delete_dynamic (array.c:304)
by 0x74F3F2: Rpl_filter::set_wild_do_table(char const*) (rpl_filter.cc:421)

Analysis:
========
List of values provided for options "wild_do_table" and "wild_ignore_table" are
stored in DYNAMIC_ARRAYS.  When an empty list is provided these dynamic arrays
are not initialized.  Existing code treats empty element list as an error and
tries to clean the uninitialized list. This results in above valgrind issue.

Fix:
===
The clean up should be initiated only when there is an error while parsing the
'wild_do_table' or 'wild_ignore_table' list and the dynamic_array is in
initialized state. Otherwise for empty list it should simply return success.

- buf_page_get_gen() fails to set error value when it initializes the
page with redo logs.

We need to make sure that hton is not same as view_pseudo_hton
that is not NULL but not usable.

The json_arrayagg::val_str should handle NULL result.

Replace all fsync() with fdatasync() when possible (e.g. On Linux)

InnoDB doesn't care about file timestamps. So, to achieve a better
performance it makes sense to use fdatasync() everywhere.

file_io::flush(): renamed from flush_data_only()

os_file_flush_data(): removed

os_file_sync_posix(): renamed from os_file_fsync_posix(). Now it uses
fdatasync() when it's available.

If any of default names clashes with existing names find next large
enough name gap for the requested number of partitions.

This is continuation of MDEV-22153 bug when contiguity of history
partitions is broken. ha_partition::open_read_partitions() can not
handle non-contiguous list of default partitions.

Fix: when default partition is dropped convert list of partitions to
non-default.

The functional changes of commit 5836191c
(MDEV-21168) are omitted due to MDEV-742 having addressed the issue.

The reason why we have wsrep_on() at all is that the macro WSREP(thd)
depends on the definition of THD, and that is intentionally an opaque
data type for InnoDB. So, we cannot avoid invoking wsrep_on(), but
we can evaluate the less expensive conditions thd && WSREP_ON before
calling the function.

Global_read_lock: Use WSREP_NNULL(thd) instead of wsrep_on(thd)
because we not only know the definition of THD but also that
the pointer is not null.

wsrep_open(): Use WSREP(thd) instead of wsrep_on(thd).

InnoDB: Replace thd && wsrep_on(thd) with wsrep_on(thd), now that
the condition has been merged to the definition of the macro
wsrep_on().

If the server is compiled WITH_WSREP=OFF, we should avoid evaluating
conditions on a global variable that is constant.

WSREP_ON_: Renamed from WSREP_ON. Defined only WITH_WSREP=ON.

WSREP_ON: Defined as unlikely(WSREP_ON_).

wsrep_on(): Defined as WSREP_ON && wsrep_service->wsrep_on_func().

The reason why we have wsrep_on() at all is that the macro WSREP(thd)
depends on the definition of THD, and that is intentionally an opaque
data type for InnoDB. So, we cannot avoid invoking wsrep_on(), but
we can evaluate the less expensive condition WSREP_ON before calling
the function.

Replaced WSREP_ON macro by single global variable WSREP_ON
that is then updated at server statup and on wsrep_on and
wsrep_provider update functions.

https://gcc.gnu.org/bugzilla/show_bug.cgi?id=89357

In short, alignas of big alignments doesn't work for gcc on ARM.
The bug is fixed in gcc-9.3 and won't be fixed in older branches.
F.ex gcc-7 is not supported already.

In rare cases, both listener and worker can do poll at the same time
(one of them will return events, another one no events)

Fix test query to account for it.

This is not crtitical as if Aria doesn't start, the server will exit at once.

The bug was that the script tried to start mysqld while there was already
a running mysqld server. Fixed by killing the running one.

Stop making new symlinks and use the existing ones directly as they have
been available since d2f5e82f.

Originally added in cc08c134 in 2015
for backwards compatibility with old packages.

It no longer makes much sense in providing mysql-* packages since the
virtual-mysql-* packages paradign has been in effect since 2016 and
all current MySQL installs should already follow the virtual-mysql-*
paradigm controlled by https://tracker.debian.org/pkg/mysql-defaults.

- Include binary wsrep_sst_rsync_wan previously omitted.
- Don't build the embedded server. This saves a lot of time and
  around 500 MB of disk space.
- Clean away unused override_dh_installcron. The package does not
  have any *.cron files, so this section is not needed.
- Make mariadb-server run time depend on procps so pgrep works.
  The command pgrep is used in the maintainer scripts of mariadb-server,
  thus we should ensure it is found on the system.
- Include build plugin func_test.so in test package. Don't include yet
  type_test.so since it fails (MDEV-22243).
- Clean away obsolete Lintian overrides.

This was partially already done in 9e1b3af4
as the symlink mysqlreport pointed to mariadb-report. There just was no
script with that name, so this rename complements it and now both
mariadb-report and mysqlreport work again.

Likewise to TokuDB and Cassandra, if this plugin is maintained it should
be either fully enabled or not present at all. Packaging can be recovered
from MariaDB 10.2 if needed, or if build outside of this packaging, then
also the packaging scripts and snippets should be injected from the
outside.

Cassandra has deprecated their Thrift interface years ago (a few months
after Cassandra SE was released). The new way to access Cassandra is to
use the CQL Connector, which does not use Thrift.

There have been talks to make a CassandraSE V2 but no usable code have
been produced. The motivation to keep CassandraSE in the MariaDB source
code is to have it as an example for such future attempts.

It can stay in the source code but it should be removed from packaging
to avoid Debian packaging maintenance work around trying to compile it
and have tests pass on new platform.

If somebody wants Cassandra, they can use the 10.4 branch.

It was already disabled in 1589cf1c but
turns out that as long as the packaging is there, is requires maintenance,
ensuring it still builds, debugging failing tests etc and that takes its
toll and is not worth all the extra work. Having it "only disabled" does
not help, since either we have it and keep it tested and working, or we
don't have it and don't spend time on keeping it working and tested.

In MDEV-19780 there was already a decision to drop TokuDB from MariaDB.

If somebody wants TokuDB, they can use the 10.4 branch.

As initially most tests fail, they have allow_failures defined so that
testing anyway proceeds all the way to the final 'upgrade extras' stage.

All of these tests work for downstream Debian packaging of MariaDB 10.4
and should eventually pass on upstream MariaDB 10.5 as well.

Also upstream the Debian autopkgtests from MariaDB 10.4 in Debian so that
pipeline includes running mtr.

This are filed as issues so I trust these will not be forgotten:
- MDEV-20372
- MDEV-21965
- MDEV-21968
- MDEV-22119

Having these tests fail the MTR permanently decreases the overall quality
of the software as:
- developers start neglecting the results from CI systems and do not spot
  the new regressions their changes introduce
- reviews take many hours as a diligent reviewer needs to open all build
  logs that are failing, see what is failing, and check if those are new
  failures or not – alternatively reviewers stop looking at CI altogether

It is better to disable these test now and prioritize the MDEVs for the
test failure bugs than keep the MTR red and allow more test failures to
creep in.

Fix a test result.

Test do we have galera debug library with debug_sync functionality
needs to be earlier.

Also add a check for tmp file being empty and bail out with a clear
error message in such a case, as mysqld_safe prevents normal stderr
from being displayed anywhere and would fail silently on this.

MDEV-22181 : galera.galera_sst_mysqldump_with_key MTR failed: INSERT failed: 1146: Table 'test.t1' doesn't exist

Add wait condition to make sure table test.t1 is replicated
to node_2 before we insert to it.

Commit 5defdc38
(which aimed to reduce sizeof(mtr_t) for non-debug builds)
replaced the ternary mtr_t::status with two debug-only bool
data members m_start, m_commit and inadvertently made the
(now debug-only) predicate mtr_t::is_active() wrongly hold
after mtr_t::commit().

mtr_t::is_active(): Evaluate both m_start and m_commit,
to be compatible with the old definition.

row_merge_read_clustered_index(): Correct a debug assertion.