Bump version of the `upload-artifact` and `download-artifact` actions
to v4 for fixing an annoying warning that appears in GitHub WebUI:

    Node.js 16 actions are deprecated.
    Please update the following actions to use Node.js 20

NO_DOC=ci
NO_TEST=ci
NO_CHANGELOG=ci

Let's add an `__autocomplete` method to enable error autocompletion and
make the error object easier to use. It will suggest error object fields
(including own and inherited payload fields) and methods.

Closes #9107

NO_DOC=<not a documentable feature>

Cdata objects cannot have an `__autocomplete` metamethod (it's our custom
metamethod). To work this around, we should also lookup the
`__autocomplete` method using an object's `__index` metamethod.

Part of #9107

NO_CHANGELOG=<not a reported bug>
NO_DOC=<bugfix>

Now on_access_denied and on_schema_init triggers do not throw error,
so it just sinks into oblivion, which is not good because a thrown error
indicates that something went wrong and database administrator should be
definitely notified about it.

The commit makes Tarantool log errors thrown from these triggers with
error level. Also, the commit adds a test checking if all triggers are
either raise an error or log it.

Closes #9309

@TarantoolBot document
Title: Mention that triggers are guaranteed to throw an error or log it

Now all triggers are either have a direct impact on the execution flow
(e.g. close a connection or throw an error) or print a message to error
log when they throw an error.

We have a separate entry for triggers in Tarantool documentation, which
lists the properties of the triggers - I think it's a perfect place for
this new property.

If we start to encode client error arguments as payload in C it is good
to have same behavior for Lua too. So that for example
`box.error.new(box.error.TUPLE_NOT_FOUND, 'space A', 'index B') will
create error with payload fields 'space' and 'index' with corresponding
values.

Follow up #9109

NO_CHANGELOG=internal
NO_DOC=internal

Follow up #9109

NO_TEST=refactoring
NO_CHANGELOG=refactoring
NO_DOC=refactoring

With new harness we can specify in 'errcode.h' for example

```c
	_(ER_TUPLE_NOT_FOUND, 4, "Tuple doesn't exist", "space", STRING, "index", STRING) \
```

and then on

```c
	diag_set(ClientError, ER_TUPLE_NOT_FOUND, "space A", "index B")
```

the error set will have payload fields "space" and "index" with
corresponding values.

Closes #9109

NO_CHANGELOG=internal
NO_DOC=internal

Part of #9109

NO_CHANGELOG=INTERNAL
NO_DOC=internal

Part of #9109

NO_CHANGELOG=internal
NO_DOC=internal

This makes backport patches with new error codes more clean. Also we can
use error codes with high numbers for testing in the next patch.

ER_SQL_CANT_RESOLVE_FIELD message is changed as it has non ascii symbol '`'
which is prohibited by checkpatch.

Part of #9109

NO_CHANGELOG=refactoring
NO_DOC=refactoring

Currently, it loads only instances defined in replicaset-001 of
group-001. Let's load all instances so that we can start a cluster
consisting of multiple replicasets.

NO_DOC=testing helper update
NO_CHANGELOG=see NO_DOC

Co-authored-by: Vladimir Davydov <vdavydov@tarantool.org>

A replicaset isn't enough for out goal. We need a whole cluster.
Let's start with renaming replicaset to cluster.

NO_DOC=testing helper update
NO_CHANGELOG=see NO_DOC

Co-authored-by: Vladimir Davydov <vdavydov@tarantool.org>

It's now possible to select a custom replicaset for following calls,
such as :add_instance(), in the config builder with the new methods
:use_group() and :use_replicaset(). This is necessary for creating
clusters spanning multiple replicasets.

NO_DOC=testing helper update
NO_CHANGELOG=see NO_DOC

Co-authored-by: Vladimir Davydov <vdavydov@tarantool.org>

Bump version of actions/checkout to v4.
Bump fixes an annoying warning that appears in Github WebUI:

Node.js 16 actions are deprecated. Please update the following actions
to use Node.js 20: actions/checkout@v3. For more information see:
https://github.blog/changelog/2023-09-22-github-actions-transitioning-from-node-16-to-node-20/.

Version of the action in workflow static_build_pack_test_deploy.yml
remains the same because Node.js 20 is failed on CentOS 7 with error:

/__e/node20/bin/node: /lib64/libm.so.6: version `GLIBC_2.27' not found (required by /__e/node20/bin/node)
/__e/node20/bin/node: /lib64/libstdc++.so.6: version `GLIBCXX_3.4.20' not found (required by /__e/node20/bin/node)
/__e/node20/bin/node: /lib64/libstdc++.so.6: version `CXXABI_1.3.9' not found (required by /__e/node20/bin/node)
/__e/node20/bin/node: /lib64/libstdc++.so.6: version `GLIBCXX_3.4.21' not found (required by /__e/node20/bin/node)
/__e/node20/bin/node: /lib64/libc.so.6: version `GLIBC_2.28' not found (required by /__e/node20/bin/node)
/__e/node20/bin/node: /lib64/libc.so.6: version `GLIBC_2.25' not found (required by /__e/node20/bin/node)

NO_CHANGELOG=ci
NO_DOC=ci
NO_TEST=ci

Commit 71de4b2c ("box: fix schema downgrade replication") introduces
check that decides when DDL is disabled if current schema is not equal
to required schema. This patch adds a condition for DDL operations to be
enabled during recovery, so new bootstrap snapshot can be properly
generated.

Follow-up #9049

NO_DOC=bugfix
NO_TEST=internal
NO_CHANGELOG=internal

Co-authored-by: Vladimir Davydov <vdavydov@tarantool.org>

Closes #9775

@TarantoolBot document
Title: The `sharding.weight` option

The `sharding.weight` option can be set to facilitate rebalancing. The
larger the weight of a replicaset, the more data it can store. This is a
relative value, so a replicaset with a value N times larger than a value
of the another replicaset can store N times more data. The default value
is 1.

Previously, the config status was set after the "apply" phase, and then
set again after the "post_apply" phase. This can lead to a situation
where the status becomes "ready" after the "apply" phase and changes to
"check_errors" or "check_warnings" after the "post_apply" phase.
Because of this, it was possible that config:reload() may be called from
a role or an app even if though the config loading is not finished.

Note that due to this change, the "startup_in_progress" and
"reload_in_progress" statuses are now expanded to include the
"post_apply" phase.

Also note that the status still changes before the "extras.post_apply"
stage.

Needed for https://github.com/tarantool/tarantool-ee/issues/643

NO_DOC=internal

Currently, the error object's `__serialize` metamethod and the Lua
serializer only display the `message` field omitting all other potentially
useful fields.

Let's call the error object's `__serialize` metamethod from the Lua
serializer and call `error:unpack` from the `__serialize` metamethod to
display all other error object fields. This will transparently allow to
display the whole error stack (i.e., the cause chain).

Let's make the error object's  `__tostring` metamethod return the `message`
field followed by a whitespace and the rest of the fields encoded to JSON.
Let's print the error stack (i.e., the cause chain) on separate lines.

Since this change may potentially break existing users, let's add a new
`box_error_serialize_verbose` option to `compat` to retain old behavior,
and disable the new behaviour by default. Let's also retain the old
behaviour for the MsgPack serializers unconditionally.

Closes #9105

@TarantoolBot document
Title: Document increased error serialization verbosity
Product: Tarantool
Since: 3.1
Root documents: https://www.tarantool.io/en/doc/latest/reference/reference_lua/box_error/
and https://www.tarantool.io/en/doc/latest/reference/reference_lua/box_error/error
and https://tarantool.io/compat/box_error_serialize_verbose

[Link to the design document](https://www.notion.so/tarantool/Error-subsystem-improvements-90faa0a4714b4143abaf8bed2c10b2fc?pvs=4#072087684e094ea28cba88002236178a)

Please add a new https://tarantool.io/compat/box_error_serialize_verbose
page for the `box_error_serialize_verbose` compatibility option.

Currently, we delegate the work of serializing extensions to serializers in
case of known extensions. In case of unknown extensions we try to convert
them from Lua using the serialization and string conversion metamethods.

However, there are cases where we would prefer to do the serialization in
Lua instead of delegating to serializers. Let's always try to serialize
extensions first using Lua via `luaL_convertfield`, and, if it fails, or
the conversion returns an extension again, fallback to the serializers.

Needed for #9105

NO_CHANGELOG=<refactoring>
NO_DOC=<refactoring>
NO_TEST=<refactoring>

The #8350 was introduced by the commit b42302f5 ("lua-yaml: enable
aliasing for objects returned by __serialize") so the patch is
effectively reversed.

The idea is to call all object __serialize methods recursively before
finding references. The new serialization pass stores the mapping from
the original object to the serialized representation.

After this, the reference analysis pass and the encoding pass use this
mapping to replace original objects with the serialized representation.

As result, the reference analysis has a complete information about
objects and no references are missed.

Closes #8350
Closes #8310
Closes #8321

NO_DOC=bugfix

Co-authored-by: Nikolay Shirokovskiy <nshirokovskiy@tarantool.org>

Add workflow job for integration testing of the etcd-client module.

Closes #9093

NO_DOC=ci
NO_TEST=ci
NO_CHANGELOG=ci

The commit introduces new `unit/port.cc` test that covers almost all
port implementations and almost all port methods. This test was
intended as a testing point for future port changes.

Implementations `port_sql` and `port_vdbemem` and method  `dump_vdbemem`
are not tested because I don't know much about the sql subsystem. For
the same reason, I didn't tested `port_dump_msgpack_with_context`.
Method `dump_msgpack_16` is not actually supported, but it is
still used for compatibility with very old connectors. The problem is I
found a bug in `port_c_dump_msgpack_16` while writing the test, and we
don't want to fix it - probably, we will get rid of this method in
future. Method `dump_plain` is not tested because it requires functions
from `console.lua` file - it is easier to test `console.push` from Lua
then loading the Lua file in unit test.

The file has .cc extension because the test requires user_cache
susbystem, which has init and free methods available only from C++
source files.

NO_CHANGELOG=test
NO_DOC=test

Lua headers don't have C++ guards which means all Lua symbols are
mangled as C++ ones when the files included in a C++ source file,
but then the linker fails because all Lua functions are exported
with C mangling. Let's include lua headers in extern "C" scope to
allow to use "lua_test_utils.h" in unit tests written in C++.

NO_CHANGELOG=test
NO_DOC=test

The method actually supports all modes of dump to Lua, but the assertion
checks that mp_object mode is never used - the commit drops the
assertion.

This mode is tested with a tuple constraint taking raw args. Along the
way, some test cases are enabled back because the bug that caused them
to be disabled has been fixed.

NO_CHANGELOG=internal
NO_DOC=internal

Closes #9101

@TarantoolBot document
Title: hide redundant fields of box.error.unpack()

Do not show redundant fields of box.error.unpack(). These are `base_type`,
`custom_type` and `code` if the latter is 0.

New behaviour is available if `box_error_unpack_type_and_code` compat
option is 'new'. Default value is 'old` currently.

(The https://tarantool.io/compat/box_error_unpack_type_and_code is
to be added.)

Leads nowhere after commit 216b6243 ("tools: remove tarantoolctl
utility").

NO_DOC=cleanup
NO_TEST=cleanup
NO_CHANGELOG=cleanup

Now the error message is allocated by `malloc' if it doesn't fit into the
static buffer.

Closes #4975

NO_DOC=bugfix

The benchmark tests the tree build, key search, insert and delete
operations performance. The latter are tested both including and
excluding the tree reballancing overhead.

A very simple allocator had been introduced to mitigate the memory
management overhead and noise.

The benchmark functions are templated. This allows to test multiple
tree configurations using the same benchmarking routines.

The simplest tree configuration is used, though it's possible to add
new configurations to the benchmark. Example on how to create a tree
similar to the one used by the memtx index is shown in the appendix A.

Closes #9630

NO_DOC=new benchmark
NO_TEST=new benchmark
NO_CHANGELOG=new benchmark

APPENDIX A: Adding a new tree configuration to the benchmark.

```C
/* Instantiate the tree. */

#define tree_s128_EXTENT_SIZE 16 * 1024
#define tree_s128_elem_t struct tree_s128_elem
#define tree_s128_key_t struct tree_s128_key

struct tree_s128_elem {
	void *tuple; /* Unused. */
	int64_t hint;

	tree_s128_elem() = default;
	tree_s128_elem(int64_t hint) : hint(hint) {}
};

struct tree_s128_key {
	void *key; /* Unused. */
	uint32_t part_count; /* Unused. */
	int64_t hint;

	tree_s128_key(int64_t hint) : hint(hint) {}
};

#define BPS_TREE_NAME tree_s128_t
#define BPS_TREE_BLOCK_SIZE 512
#define BPS_TREE_EXTENT_SIZE tree_s128_EXTENT_SIZE
#define BPS_TREE_IS_IDENTICAL(a, b) ((a).hint == (b).hint)
#define BPS_TREE_COMPARE(a, b, arg) ((a).hint - (b).hint)
#define BPS_TREE_COMPARE_KEY(a, b, arg) ((a).hint - (b).hint)
#define bps_tree_elem_t tree_s128_elem_t
#define bps_tree_key_t tree_s128_key_t
#define bps_tree_arg_t int
#include "salad/bps_tree.h"
#undef BPS_TREE_NAME
#undef BPS_TREE_BLOCK_SIZE
#undef BPS_TREE_EXTENT_SIZE
#undef BPS_TREE_IS_IDENTICAL
#undef BPS_TREE_COMPARE
#undef BPS_TREE_COMPARE_KEY
#undef bps_tree_elem_t
#undef bps_tree_key_t
#undef bps_tree_arg_t

/** Add the new tree to the `generate_benchmarks` macro. */

#define generate_benchmarks(generator, func, arg) \
	generator(tree_i64, func, arg); \
	generator(tree_s128, func, arg) /* < The line to be added. */

/** Create the new tree class. */

CREATE_TREE_CLASS(tree_i64);
CREATE_TREE_CLASS(tree_s128); /* < The line to be added. */
```

In development of the configuration module we met several cases, when an
order of appearance of some items is as important as a constant time
access using a key.

This commit adds a collection that serves such a need. It is to be used
internally in the src/box/lua/config code.

The module is inspired by Python's collections.OrderedDict. See the
description in the module code for details.

NO_DOC=the module is for internal use, at least for now
NO_CHANGELOG=see NO_DOC

It brings newer codespell version: 2.1.0. Ubuntu Focal offers 1.16.0.

Fixes tarantool/checkpatch#70

NO_DOC=CI adjustment
NO_CHANGELOG=see NO_DOC
NO_TEST=see NO_DOC

If we run a static build with ASAN enabled via Clang 16, the build will
fail unless `libresolv` is in the white list of static dependencies.
It looks like it is a peculiarity of Clang 16 and higher.

Fixes #9740

NO_DOC=build
NO_TEST=build
NO_CHANGELOG=build

There are two reasons for this changeset:

* The positive one: Tarantool supports -b and -j options to use LuaJIT
  modules since the commit bf8b76a4
  ("lua: proxy -j and -b flags"), so the related tests from lua-Harness
  suite can be partially (since -O option is still not implemented in
  Tarantool) enabled.

* The negative one: Tarantool diff-based tests for CLI interfaces are
  hard to maintain, if any change occurs in LuaJIT modules, since the
  aforementioned tests implement dumb comparison of the output, produced
  by the current CLI version against the expected one, managed by the
  .result file. Hence, to rule the tests related to the LuaJIT CLI
  interface in a more convenient way, the corresponding tests should be
  moved from the tests in the Tarantool repository to the tests in the
  LuaJIT repository.

The recent LuaJIT bump landed to the master in the scope of commit
0dcf6759 ("luajit: bump new version")
enables the nice checks implemented in the lua-Harness suite; this patch
removes the barely maintainable diff-based tests from this repository.

Follows up #5541

NO_DOC=test
NO_CHANGELOG=test

* cmake: introduce AddTestLib macro
* test: prepare lauxilarily libs for LuaJIT-tests
* test: separate LuaJIT helpers from ffi_util.inc
* test: enable <ffi_arith_ptr.lua> in LuaJIT-tests
* test: enable <ffi_bitfield.lua> in LuaJIT-tests
* test: enable <ffi_call.lua> in LuaJIT-tests
* test: enable <ffi_callback.lua> in LuaJIT-tests
* test: enable <ffi_const.lua> in LuaJIT-tests
* test: enable <ffi_convert.lua> in LuaJIT-tests
* test: enable <ffi_enum.lua> in LuaJIT-tests
* test: enable <ffi_gcstep_recursive.lua>
* test: enable <ffi_jit_arith.lua> in LuaJIT-tests
* test: enable <ffi_jit_call.lua> in LuaJIT-tests
* test: enable <ffi_jit_conv.lua> in LuaJIT-tests
* test: enable <ffi_lex_number.lua> in LuaJIT-tests
* test: enable <ffi_metatype.lua> in LuaJIT-tests
* test: enable <ffi_new.lua> in LuaJIT-tests
* test: enable <ffi_parse_array.lua> in LuaJIT-tests
* test: enable <ffi_parse_basic.lua> in LuaJIT-tests
* test: enable <ffi_parse_cdef.lua> in LuaJIT-tests
* test: enable <ffi_parse_struct.lua> LuaJIT test
* test: enable <ffi_tabov.lua> LuaJIT test
* test: enable <lightud.lua> LuaJIT test
* test: enable <api_call.lua> LuaJIT test
* test: enable <catch_wrap.lua> LuaJIT test
* test: enable <catch_cpp.lua> LuaJIT test
* test: introduce routine to build error message
* test: enable CLI-related lua-Harness tests back

Closes #7834
Part of #9398
Follows up #5541

NO_DOC=LuaJIT submodule bump
NO_TEST=LuaJIT submodule bump
NO_CHANGELOG=add new tests

We assume that if the new format can store tuples matching the old
format, we can update the space format without calling the engine
check_format callback. This is true for both memtx and vinyl but
not for memcs (EE), which doesn't support extending field types
(e.g. changing int16 to int32).

Let's call the engine check_format callback unconditionally and let
it decide whether tuple format checking can be skipped.

Needed for tarantool/tarantool-ee#694

NO_DOC=refactoring
NO_TEST=refactoring
NO_CHANGELOG=refactoring

In the commit 22d507d5 ("iproto: don't hang on uncancellable iproto
request") we used TEST_BUILD define which is absent, we only have
a CMake build option with such name. Let's add a define too.

While at it let's also show this flag in the `tarantool.build` table.

Follow-up #8423

@TarantoolBot document
Title: new tarantool.build.test_build flag

It is `true` if `TEST_BUILD` build option is set and `false` otherwise.

We need to handle case of dropping new connection. When
net_send_greeting() is executed the connection can be closed due to
iproto_drop_connections() call.

Note that in the test the Tarantool crashes for another reason. Due to
access after sleep to the connection that is destroyed so its memory is
poisoned. Yet we visit net_send_greeting() too in the test with patch so
original issue is verified too. We also need to test that such
a connection is closed. This will be done in EE version.

Closes #9717

NO_DOC=bugfix

Suppose an error has a cause with some payload fields, for example:
  local e1 = box.error.new{'e1', foo = 'bar'} -- cause
  local e2 = box.error.new{'e2', prev = e1}   -- effect

Now it is possible to access cause payload fields via e2 directly:
  e2.foo -- 'bar'

While looking for a payload field with a given name, we always stop at
the topmost (closest to the effect) field. If there's a field with the
same name deeper in the stack it is masked.

Closes #9106

@TarantoolBot document
Title: Document inheritance of error payload fields
Product: Tarantool
Since: 3.1
Root document: https://www.tarantool.io/en/doc/latest/reference/reference_lua/box_error/error_object/

[Link to the design document](https://www.notion.so/tarantool/Error-subsystem-improvements-90faa0a4714b4143abaf8bed2c10b2fc?pvs=4#c080fe2ac28b46c8b0eda7234a8852ce)

With commit eb2c6a4a ("box/lua: allow to set custom error payload
fields") it is possible to add a field named "raise", but it will
override the `error_object:raise()' method. Forbid it. The user can
still access masked payload fields through the `unpack()' method.

NO_DOC=unreleased
NO_CHANGELOG=unreleased

At last we can drop usage of pthread_cancel and associated functions.
And remove related leak suppressions.

Let's keep memory protection disabling under ASAN. Otherwise
leak sanitizer may misbehave on Tarantool panic as below.

```
  #   Tracer caught signal 11: addr=0x705236d1e000 pc=0x57b7605b10d0 sp=0x705232a00ca0\
  #   ==1022907==LeakSanitizer has encountered a fatal error.\
  #   ==1022907==HINT: For debugging, try setting environment variable LSAN_OPTIONS=verbosity=1:log_threads=1\
  #   ==1022907==HINT: LeakSanitizer does not work under ptrace (strace, gdb, etc)",
```

Let's also add missing pipe/endpoint destroy in wal while at it.

Close #8423

NO_CHANGELOG=internal
NO_DOC=internal

In the commit d40ce0fa ("core: disable fibers creation after
shutdown started") we disable creation of new fibers in the process of
shutdown. This may cause subsystem shutdown hanging.

The thing is we need fiber pool working during shutdown. For example
vinyl engine uses it through "tx" endpoint. Fibers in the pool are
finished after idle timeout. So we may have a situation when there is no
idle fibers in the pool and we cannot create a new one.

Part of #8423

NO_CHANGELOG=bugfix for unreleased version
NO_DOC=bugfix for unreleased version