If a configuration option value passed to `box.cfg` is the same as the
old one, the option handler isn't called. However, sometimes we might
need to do some extra work in the handler in this case. For example,
even if listen/replication URI is the same, we still need to recreate
the derived IO stream context, because a URI parameter may contain a
path to a file (for example, an SSL certificate), which could change.

So we remove the corresponding check from load_cfg.lua and instead move
it to option handlers. Most of the existing dynamic option handlers
already do nothing if the old and the new values are the same. There are
only three exceptions:

 1. box.cfg.memtx_memory. The problem is that the value is rounded up
    to a multiple of the quota unit size by the handler so even if we
    pass exactly the same value, it may mistakenly assume that the new
    value is less and fail. To avoid that, we make the handler round up
    the new value before checking it.

 2. box.cfg.replication. Even though commit 5994892c ("replication:
    fix replica disconnect upon reconfiguration") made the handler avoid
    disconnect, it still tries to connect to the new URIs and sync,
    which may pose a problem to our users. For example, resetting
    box.cfg.replication to the same value on the leader after right
    after a new cluster bootstrap (before granting access to replicas)
    would lead to a replication sync, which would obviously fail on
    timeout. To avoid that, we make the handler compare the new and old
    URIs internally and do nothing if they are the same.

 3. box.cfg.listen. Handled similarly to box.cfg.replication: to avoid
    futile rebinding of IPROTO ports, we store the last configured
    listen URIs in a C variable and do nothing on box.cfg if the
    configuration is the same.

Needed for tarantool/tarantool-ee#432

NO_DOC=refactoring
NO_TEST=refactoring
NO_CHANGELOG=refactoring

 - Call listen() once per socket in the TX thread instead of doing it in
   each IPROTO thread because calling listen() more than once has no
   effect.
 - Merge evio_service_bind and evio_service_listen into one method
   named evio_service_start to match evio_service_stop.
 - Rename IPROTO_CFG_LISTEN message to IPROTO_CFG_START to match
   IPROTO_CFG_STOP and drop the binary argument because it always
   attaches to tx_binary.
 - Drop the dead evio_service_is_active check from IPROTO_CFG_START
   because the service must be inactive at this point and we have an
   assertion for that in evio_service_attach.
 - Move ERRINJ_IPROTO_CFG_LISTEN error injection to iproto_listen and
   drop the error code of iproto_send_start_msg because it should never
   fail by design.

Needed for tarantool/tarantool-ee#432

NO_DOC=refactoring
NO_TEST=refactoring
NO_CHANGELOG=refactoring

To handle the case when box.cfg.replication isn't changed (required for
SSL certificate update), we need the uri set in box_set_replication.
We could obtain it with cfg_get_uri_set, but calling this function right
after box_check_replication, which already creates the uri set looks
ugly. Let's make box_check_uri_set, box_check_replication, and
box_check_listen return the uri_set, like box_check_uri does.

While we are at it, fix a memory leak in box_check_config, where we
forgot to destroy the uri returned by box_check_bootstrap_leader.
The leak is insignificant and may only happen once so this doesn't
deserve to be mentioned in the changelog.

Needed for tarantool/tarantool-ee#432

NO_DOC=refactoring
NO_TEST=refactoring
NO_CHANGELOG=refactoring

Historically, we create appliers in box.cc and then pass them to
replicaset_connect. Now, that we have struct uri_set, it's better
to pass URIs instead of appliers and let replicaset_connect create
appliers. It looks better because replicaset_connect may actually
delete some of the passed appliers on success, which makes the
function protocol difficult for understanding. Another reason to
pass URIs instead of appliers is that we need to store the configured
URIs in replication.cc so that we can detect the case when box.cfg
doesn't change box.cfg.replication (required for SSL certificate
update).

While we are at it,
 - Make replica_set_applier and replica_clear_applier static.
 - Mark the URI argument of applier_new as const and copy it instead
   of moving. This isn't a hot path and with copying it's easier to
   make assumptions about object lifetime.
 - Remove the error label from replicaset_connect. Use a guard instead.

Needed for tarantool/tarantool-ee#432

NO_DOC=refactoring
NO_TEST=refactoring
NO_CHANGELOG=refactoring

Appliers array and count are captured by value before initialization.
As a results, appliers won't be freed if replication configuration
fails.

NO_DOC=bug fix
NO_CHANGELOG=minor
NO_TEST=should be checked by ASAN, see #7326

There's absolutely no reason to create a new iostream context for a
ballot watcher, because it'll be exactly the same as the one used by
the applier because it's created from the same URI.

NO_DOC=refactoring
NO_TEST=refactoring
NO_CHANGELOG=refactoring

Needed for tarantool/tarantool-ee#432

NO_DOC=internal
NO_CHANGELOG=internal

Bump test-run to new version with the following improvements:

- Bump luatest to 0.5.7-38-ge6a2093 [1]

[1] tarantool/test-run@affa13f

NO_DOC=testing stuff
NO_TEST=testing stuff
NO_CHANGELOG=testing stuff

This patch fixes a possible 'use-after-free' of field pOrderBy in
generateWithRecursiveQuery().

Closes tarantool/security#120

NO_DOC=bugfix

This patch fixes SQL memory leaks found by static analyzers and SQL
fuzzer.

Part of tarantool/security#120

NO_DOC=fix for memleak
NO_TEST=fix for memleak
NO_CHANGELOG=fix for memleak

The sql_xfree() function can take NULL as an argument, and this is not
an error because db->lookaside.pStart > NULL. However, this causes
warnings from static analyzers. Let's add an explicit check to avoid
these warnings.

Part of tarantool/security#120

NO_DOC=refactoring
NO_TEST=refactoring
NO_CHANGELOG=refactoring

This patch removes some unnecessary checks that should have been removed
in the 'sql: refactor memory allocation system' patch.

Also, this patch removes the unnecessary check that the return value of
space_by_id() is NULL in sql_check_create(). Return values is not NULL
because sql_check_create() can be called in two cases:
1) space was just created, and space_id is its ID;
1) space_id was obtained directly from the space that existed at the
time the VDBE was created.

Part of tarantool/security#120

NO_DOC=refactoring
NO_TEST=refactoring
NO_CHANGELOG=refactoring

This patch removes some dead code in SQL. This is done to silence static
analyzers. In addition, this patch removes the mmapSizeActual field from
the struct unixFile, since this field is identical to the mmapSize field
in the same structure.

Part of tarantool/security#120

NO_DOC=refactoring
NO_TEST=refactoring
NO_CHANGELOG=refactoring

This patch adds 4 methods of the key_def module instance to the
`index_object.parts`, see the docbot request for details. The rest
methods (new() and totable()) are not applicable here, because the
instance is already created and pushed to the stack as a table.

Closes #7356

@TarantoolBot document
Title: `index_object.parts` methods
Product: Tarantool
Since: 3.0
Root document: https://www.tarantool.io/en/doc/latest/reference/reference_lua/box_index/parts/

`index_object.parts` has the following methods: `extract_key()`,
`compare()`, `compare_with_key()`, `merge()`.
For their description and usage examples, refer to
[Module key_def](https://www.tarantool.io/en/doc/latest/reference/reference_lua/key_def/#lua-object.key_def.key_def_object).

---

Root document: https://www.tarantool.io/en/doc/latest/reference/reference_lua/key_def/

`index_object.parts` can be used like a key_def module instance for calling
the corresponding methods. Example:

```lua
box.schema.space.create('T')
i = box.space.T:create_index('I', {parts={3, 'string', 1, 'unsigned'}})
box.space.T:insert{1, 99.5, 'X', nil, 99.5}
i.parts:extract_key(box.space.T:get({'X', 1}))
```

The code above is equivalent to:

```lua
key_def = require('key_def')
box.schema.space.create('T')
i = box.space.T:create_index('I', {parts={3, 'string', 1, 'unsigned'}})
box.space.T:insert{1, 99.5, 'X', nil, 99.5}
k = key_def.new(i.parts)
k:extract_key(box.space.T:get({'X', 1}))
```

Introduced command 'tt-list-walk'

Changes in 'tt-list' options:
- dropped options -pre/post (previously used to implement walking
  functionality; no longer needed)
- option -filter was renamed to -predicate (so it can be specified with
  just -p while previously at least -fil need to be used to avoid
  ambiguity with -fields option)

Closes #8524

NO_DOC=gdb extension
NO_TEST=gdb extension
NO_CHANGELOG=gdb extension

There is a system include file TargetConditionals.h on macOS, which
defines TARGET_OS_LINUX (and others) to 0 or 1. On the other side,
TARGET_OS_LINUX is also defined by trivia/config.h.cmake, but there it
has another possible values: undefined or 1. This inconsistency causes
issues like #8445, when TARGET_OS_LINUX is defined (to 0) in one file
and undefined in another. Let's always define it to 0 or 1.

Closes #8445

NO_DOC=bugfix

This commit adds two new Lua functions:
 - `box.malloc.info()` is a public function that returns the total
   memory usage of internal data structures allocated with `malloc()`.
   This information is extracted from the `malloc_info()` output.
   See the doc bot request for details.
 - `box.malloc.internal.info()` is an internal function that returns
   a parsed XML document returned by `malloc_info()` as is. It isn't
   documented. We may need it to get the detailed information about
   the system memory allocator.

Both functions are useful only on Linux and only if ASAN is disabled
because otherwise `malloc_info()` is unavailable. On other platforms
`box.malloc.info()` reports zero memory usage while `internal.info()`
returns an empty table.

Closes #7311

@TarantoolBot document
Title: Document `box.malloc.info()`

`box.malloc.info()` is a function that returns a table with two fields:

```
tarantool> box.malloc.info()
---
- size: 2498560
  used: 1835726
...
```

The `used` value is the total amount of memory in bytes allocated by
Tarantool for internal data structures with [`malloc()`][1].

The `size` value is the total amount of memory in bytes allocated from
the system by the `malloc()` allocator. It may not be less than `used`.

The function may be used before `box.cfg()` is called.

Currently, the function is available only on Linux. On other systems it
reports zero memory usage (`{size = 0, used = 0}`).

[1]: https://man7.org/linux/man-pages/man3/malloc.3.html

We need an XML parser to parse the string returned by malloc_info.
This commit implements a simple parser in Lua C that transforms
an XML string into a Lua table. The parser is available both from
C (luaT_xml_decode) and Lua (internal.xml.decode).

Needed for #7311

NO_DOC=internal
NO_CHANGELOG=internal

This patch limits the amount of allocated memory in the built-in
REPLACE() function.

Follow-up tarantool/security#119

NO_DOC=bugfix
NO_CHANGELOG=already exists

Function tuple_extract_key_raw for nullable partially sequential key_def
can extract wrong key from tuples without trailing nils. Fortunately,
it does not affect tarantool - tuple_extract_key_raw is used only with
cmp_def, which ends with non-nullable primary key definition, hence
contains no trailing nils.

Closes #8504

NO_CHANGELOG=no visible changes
NO_DOC=no visible changes

For an unknown reason default fio.open() mode used to be set as
0777 & umask that basically equals 18, meaning no read/write/execute
rights for anyone.

Now the default is 0666 before umask, meaning read and write access.

Closes #7981

NO_DOC=bugfix

Provide the `report-job-status` action with the `job-name` param to get
the direct link to the failed job in the failure message.

NO_DOC=ci
NO_TEST=ci
NO_CHANGELOG=ci

Call the `checkout` action with the `path` param to avoid the following
error:

  Error: File was unable to be removed Error: EACCES: permission denied,
  unlink '/home/runner/work/tarantool/tarantool/build-out/csv_fuzzer'

NO_DOC=ci
NO_TEST=ci
NO_CHANGELOG=ci

NO_DOC=ci
NO_TEST=ci
NO_CHANGELOG=ci

Closes #8523

NO_DOC=gdb extension
NO_TEST=gdb extension
NO_CHANGELOG=gdb extension

This patch introduces the SHOW CREATE TABLE statement. This statement
can be used to obtain a description of a space in the form of a
corresponding CREATE TABLE and CREATE INDEX statements.

Closes #8098

@TarantoolBot document
Title: SHOW CREATE TABLE statement

Statement can be used to obtain a description of a space in the form of
a corresponding `CREATE TABLE` and `CREATE INDEX` statements. Result
will be in form of set of statements and set of found errors. If errors
were not detected, set of the statements should be enough to completely
serialize space definition.

There is two types of `SHOW CREATE TABLE` statement:
1) Get a description of a single space:
```
SHOW CREATE TABLE table_name;
```

This statement can be used to obtain a description of a space in the
form of the corresponding `CREATE TABLE` and `CREATE INDEX` statements.
The result will be in the form of a set of statements and a set of found
errors. If no errors are found, the set of statements should be
sufficient to fully serialize the space definition. Otherwise, it will
certainly not be a complete space definition, and a `CREATE TABLE`
statement is generally not guaranteed to be syntactically correct.

2) Get descriptions of all available non-system spaces:
```
SHOW CREATE TABLE;
```

This statement returns descriptions for each available non-system space
in the form described above.

In some cases we need to convert a string to upper case (e.g., when
normalizing a lower-case field name for SQL): add helper functions
that do this in-place or by creating a copy of the original string.

Needed for #8098

NO_DOC=internal
NO_CHANGELOG=internal

We have the limited hardware resources with macOS 12, and full testing
is time-consuming. So let's check only the release build on the x86_64
platform.

NO_DOC=ci
NO_TEST=ci
NO_CHANGELOG=ci

All tests were updated according to tarantool/luatest@930b63b. So this
should resolve the problem with Unix socket collisions that we tried to
fix, for example, in tarantool/tarantool@7ac2685.

NO_DOC=testing stuff
NO_TEST=testing stuff
NO_CHANGELOG=testing stuff

Bump test-run to new version with the following improvements:

- Properly ignore Unix sockets on artifact saving [1]
- Bump luatest to 0.5.7-34-g930b63b [2]

[1] tarantool/test-run@3afaccc
[2] tarantool/test-run@02c4828

NO_DOC=testing stuff
NO_TEST=testing stuff
NO_CHANGELOG=testing stuff

This reverts commit 7ac2685b.

NO_DOC=testing stuff
NO_TEST=testing stuff
NO_CHANGELOG=testing stuff

NO_CHANGELOG=changelog
NO_DOC=changelog
NO_TEST=changelog

Refactor the way how we run various debugger scenarios, getting rid
of external static file, and producing debuggee scripts on the fly.
The idea is to have everything in the single source: both source
script to be used for debugging, and corresponding debugger commands.

NO_DOC=internal
NO_CHANGELOG=internal

It's frequently needed to see interaction of a luatest script
and its chield launched via popen.
Now we display extra debugging information received from
children pipes directly to the parent screen. It's hidden
by default by luatest (if everything goes well), but could
be displayed with `luatest -c` option.

```
luatest -c -v app-luatest/console_debugger_session_test.lua
```

NO_DOC=internal
NO_CHANGELOG=internal

NO_DOC=Test update
NO_CHANGELOG=Test update

Rework "running tests with built-in package" assert since now checks is
a callable table package with subpackages instead of a single function.

NO_DOC=No tagged version since checks initial embedding
NO_CHANGELOG=No tagged version since checks initial embedding

This patch replaces the type for some int and uint32_t values with
size_t to avoid problems with integer overflow.

Closes tarantool/security#119

NO_DOC=bugfix

Skip unstable tests of `replication-luatest/linearizable_test.lua` due
to tarantool-qa#277.

NO_DOC=testing stuff
NO_TEST=testing stuff
NO_CHANGELOG=testing stuff

The fixed tests often failed with

  fail | 2023-03-01 15:54:30.550 [3724975] main/103/server_instance.lua
  F> can't initialize storage: unlink, called on fd 63,
  aka unix/:(socket), peer of unix/:(socket): Address already in use

We fixed a similar issue in commit 3d3e9dea ("test: fix flaky
box-luatest/gh_7917_log_row_on_recovery_error_test") by using unique
instance names. Let's do the same here.

NO_DOC=testing stuff
NO_TEST=testing stuff
NO_CHANGELOG=testing stuff

Server API from the luatest has been changed: `server:clean()` method
has been removed. Use `server:drop()` instead.

Follows up tarantool/luatest#296

NO_DOC=test fix
NO_TEST=test fix
NO_CHANGELOG=test fix