Use git from CMAKE_SOURCE_DIR instead of PROJECT_SOURCE_DIR so that if
Tarantool is built as a sub-project, we will use the version of the main
project.

To build audit log as a part of the box library, set the following cmake
variables:
 - ENABLE_AUDIT_LOG: ON
 - AUDIT_LOG_SOURCES: audit log source files
 - EXTRA_BOX_INCLUDE_DIRS: header files needed for compilation

and add "audit_impl.h" to EXTRA_BOX_INCLUDE_DIRS - then it will be
included by "audit.h".

If ENABLE_AUDIT_LOG is unset, then a stub implementation is built, which
spits a warning to the log on an attempt to configure audit log.

Needed for the tests to pass with enabled audit log.

Now there's a lot of load_*/store_* functions that are designed
for unaligned access to values in data stream. Unfortunately
they are written in a way that makes new compilers to warn about
outside bounds.

Rewrite it in the most safe, cross-platform and efficient way -
using memcpy. Note that memcpy with compile-time-defined size is
recognized by the magority of compilers and they make the most
efficient operation.

Closes #6703

If the tarantool repository is used as a submodule named <foobar> in
another repository, then the statically built binary will be placed in

  <binary_dir>/<foobar>/src/tarantool

not in

  <binary_dir>/src/tarantool

where static-build/CMakeLists.txt currently tries to look it up in order
to run `ctest` and so we can't use static-build/CMakeLists.txt as is.

Let's instead use

  <install_dir>/bin/tarantool

This way `ctest` will work for static-build in both open-source and EE
repository without requiring any modifications.

We anyway run all tests on graviton machines in centos_{7,8}_aarch64 and
fedora_{33,34}_aarch64 jobs.

Regarding debug_aarch64: we had old git on odroid machines (see
PR #6180), but now we use graviton ones and, it seems, it becomes
irrelevant.

Regarding memtx_allocator_based_on_malloc: it seems, checkout@v1 here is
due to usage of Debian Stretch (with git 2.11) in docker, as in older
release workflow. Updated it in the spirit of PR #5949.

Left actions/checkout@v1 in containers jobs (coverity and jepsen*): it
should be revisited separately.

Left actions/checkout@v1 in perf_* jobs: they run on special machines
and I don't want to dig inside this part of the infrastructure ATM.

AFAIK we anyway run self-hosted runners from root due to all those
problems like [1]. No reason to include the hack into every workflow
file. If we'll going to start runners from a non-root user, it is better
to pass `--user $PROPER_UID` to all docker jobs instead, see [2]: at
least it does not require to place a boilerplate into all workflow
files.

Didn't touch perf_* jobs: don't want to dig inside this part of the
infrastructure ATM.

It reverts PR #5953.

[1]: https://github.com/actions/checkout/issues/211
[2]: https://github.com/actions/runner/issues/691

AFAIU it only has meaning for jobs constructed with matrix expansion.
See the documentation: [1].

It is relevant to [2] as well, but already fixed on nektos/act side.

[1]: https://docs.github.com/en/actions/learn-github-actions/workflow-syntax-for-github-actions#jobsjob_idstrategyfail-fast
[2]: https://github.com/tarantool/tarantool-qa/issues/118

Workaround for #6599 where we may randomly fail on AWS Graviton machines,
while it's working properly when jit disabled.

Follow-up to #5941
Relates to #6599

There was a build problem with CMAKE_BUILD_TYPE=Debug build on ARM64. It
 was fixed in 224cb68c (build: fix
 build on Linux ARM64 with CMAKE_BUILD_TYPE=Debug), before this fix
 '-DCMAKE_C_FLAGS="-Wno-type-limits "' was used to avoid the issue,
 implemented by 72c77166 (github-ci:
 add GitHub Actions workflow for Odroid).

Follow-up #6143

tarantool debian package MUST NOT depends on binutils package. This is
due to the fact that binutils include linker and assembler, what in most
cases forbidden on production servers.

This dependency is a residual from times, when tarantool did use libbfd
for stack unwinding. Now it simply does not required at all.

Fixes #6699

Previously, for "listen" and "replication" options URI can be passed
as a number, as a string value or as a table, which contains numbers
and strings, no special properties for URIs was available. Now user
can pass URIs in different ways, same as in tarantool "uri" library:
as before, as a table which contains URI and it's parameters in "param"
table, as a table which contains URI strings and URI tables. Same
as in "uri" library, which internally used to parse new "listen" and
"replication" there are different ways to specify properties for URI:
in a string which contains URI, after '?' delimiter, in a table which
contains URI in "params" table, in "default_params" table if it is
default parameters for all URIs.

Closes #5928

@TarantoolBot document
Title: implement ability to pass several URIs in different ways
Now there are several ways to pass multiple URIs for box.cfg.listen
and box.cfg.replication:
```lua
-- As a string with one or several URIs separated by commas
-- (provides backward compatibility):
box.cfg { listen = "127.0.0.1:3301, /unix.sock, 3302" }
-- As an array which contains string URIs (unambiguous short form):
box.cfg { listen = {"127.0.0.1:3301", "/unix.sock", "3302"} }
-- As an array of tables with 'uri' field
-- (may be extended with more fields in future):
box.cfg {  listen = {
        {uri = "127.0.0.1:3301"},
        {uri = "/unix.sock"},
        {uri = 3302}
    }
}
```

Previously, URI can be passed as a string, which contains one URI
or several URIs separated by commas. Now URIs can be passed in
different ways: as before, as a table which contains URI and it's
parameters in "param" table, as a table which contains URI strings
and URI tables. Also there are different ways to specify properties
for URI: in a string which contains URI, after '?' delimiter, in a
table which contains URI in "params" table, in "default_params" table
if it is default parameters for all URIs.
For this purposes new method `parse_many` was implemented in tarantool
`uri` library. Also `parse` method was updated to make possible the
same as new `parse_many` method but only for single URI.
```lua
uri = require('uri')
-- Single URI, passed as before
uri.parse_many("/tmp/unix.sock")
-- Single URI, with query paramters
uri.parse_many("/tmp/unix.sock?q1=v1&q2=v2")
-- Several URIs with parameters in one string, separated by commas
uri.parse_many("/tmp/unix.sock_1?q=v, /tmp/unix.sock_2?q=v")
-- Single URI passed in table, with additional parameters, passed
-- in "params" table. This parameters overwrite parameters from
-- URI string (q1 = "v2" in example below).
uri.parse_many({"/tmp/unix.sock?q1=v1", params = {q1 = "v2"}})
-- For parse it's also works now
uri.parse({"/tmp/unix.sock?q1=v1", params = {q1 = "v2"}})
-- Several URIs passed in table with default parameters, passed
-- in "default_params" table, which are used for parameters, which
-- not specified for URI (q3 parameter with "v3" value corresponds
-- to all URIs, and used if there is no such parameter in URI).
uri.parse_many({
    "/tmp/unix.sock_1?q1=v1",
    { uri = "/tmp/unix.sock_2", params = { q2 = "v2" } },
    default_params = { q3 = "v3" }
})
```

Implement ability to parse a string, which contains several URIs
separated by commas. Each URI can contain different query parameters
after '?'. All created URIs saved in new implemented struct `uri_set`.

Part of #5928

Previously, the part of the URI string that comes after the first '?' was
saved as a single unit in the field `query` as a part of `struct uri`. Now
an array of URI paramters has been added to the uri structure, and query is
subjected to the additional parsing and URI query parameters are extracted
from it. The separator symbol between URI query parameters is '&', the
separator symbol between parameter and it's value is '='. For example
URI = "/tmp/unix.sock?q1=v11&q1=v12&q2=v2" has two parameters: parameter
'q1' with two values 'v11' and 'v12' and parameter 'q2' with value 'v2'.

Part of #5928
Closes #1175

In the future, it is planned to extend the URI structure to allow its
passing with different options and in different formats (see next commit
`uri: implement ability to parse URI query paramters` for example). For
these purposes, it is planned to use functions that modify the source
string, for example `strtok_r`, so we need to rework the URI structure
to create copies of the string for each of the URI components.

Part of #5928

In the next commit we implement new `struct uri` and
rename current `struct uri` to `struct uri_raw` which
will be located in uri_parser.* files. Since git does
not understand that we renamed file, if in the same
commit we create file with the same name that file had
before renaming, we will perform the renaming in a
separate commit.

Part of #5928

Previously in case when unix socket URI with query didn't
contain "unix/:" prefix, parsing of this URI was incorrect.
For example "/unix.sock?q=v" was parsed into "path: /unix.sock"
and "query: x=y" instead of "host: unix/", "service: /unix.sock",
"unix: /unix.sock" and "query: x=y". Now parsing of unix socket
URI is independent from "unix/:" prefix and correct.
Pay attention, all parsing logic is realized in uri.rl file which
compiled in uri.c using 'ragel'.

Part of #5928

The third parameter passed to the 'test:is' function is a message
and usually should described what test checked.

Currently, select/pairs/get look at an :array part of passed table,
that is why different problems occurs when passed to select() table
is not a regular array.

Now we will check if passed table is not "clear dict"
(not empty table without array part). Invalid queries
like s:select{key = 'value'} will fail with an appropriate
error, but the problems with dicts containing array part
still remain (for example, s:select{1, key = 'value'} is
the same as s:select{1}).

Closes #6167

Without it `tarantoolctl rocks` would require luarocks to be installed
on the system, which isn't always possible.

To avoid possible name clashes: e.g. we might want to add a module
'dump' to the built-in lib, but the symbol name dump_lua is already
taken for 'jit.dump' - so better rename dump_lua to jit_dump_lua.

Currently, the symbol name is deducted from the filename, but it may
lead to name clashes if different packages have modules with same names.
Let's pass the symbol name explicitly to avoid that.

This is needed to embed luarocks in static build.

This is needed to embed luarocks in static build.

Zlib is used in OpenSSL and in libCURL. Besides, we will need it for
lua-zlib. So better define FindZLIB.

Note, `add_library(ZLIB::ZLIB UNKNOWN IMPORTED)` and setting the
corresponding properties in FindZLIB.cmake is needed to build libcurl,
which depends on zlib.

Use of PROJECT_ prefix gives ability to build the project as a
submodule of other projects.

This patch extends the 'integration.yml' workflow and adds a new
workflow call for running tests to verify integration between tarantool
and go-tarantool connector.

Part of #5265
Part of #6056
Closes #6607

Replication of data in the test should happened before the data presence
test on the replica. Wait for lsn helper is used.

Closes #6663

Test for #6198 handles an edge case (box.schema.FIELD_MAX tuple
field count), hence consuming a huge amount of memory (see #6684): mark
it as long run.

Closes #6684

This is a second attempt to stabilize #5568 test since
5105c2d7 ("test: fix flaky
read_only_reason test").

It had several failures with various frequency.

* test_read_only_reason_synchro() could see ro_reason as nil even
  after box.error.READONLY was raised (and some yields passed);

* test_read_only_reason_orphan() could do the same;

* `box.ctl.demote()` could raise error "box.ctl.demote does not
  support simultaneous invocations".

The orphan failure couldn't be reproduced. It was caught only
locally, so maybe it was just some unnoticed diff breaking the
test.

Failure of test_read_only_reason_synchro() could happen when
demote() was called in a previous test case, then the current test
case called promote(), then got box.error.READONLY on the replica,
then that old demote() was delivered to replica, and the attempt
to get ro_reason returned nil. It is attempted to be fixed with
replication_synchro_quorum = 2, so master promote()/demote() will
implicitly push the previous operation to the replica. Via term
bump and quorum wait.

Additionally, huge replication_synchro_timeout is added for manual
promotions. Automatic promotion is retried so here the timeout is
not so important.

`box.ctl.demote` failure due to `simultaneous invocations` seems
to be happening because the original auto-election win didn't
finish limbo transition yet. Hence the instance calling demote()
now would think it is called 'simultaneously' with another
promote()/demote().

It is attempted to be fixed from 2 sides:

- Add waiting for `not box.info.ro` on the leader node after
  auto-promotion. To ensure the limbo is taken by the leader;

- The first option didn't help much, so `box.ctl.demote()` is
  simply called in a loop until it succeeds.

Closes #6670

assert(cur < end) after mp_load_u8 in mp_check_binl was failing

 | [100%] Linking C static library libcurl-d.a
 | <...>
 | make[3]: *** No rule to make target `build/curl/dest/lib/libcurl.a',
 |     needed by `test/unit/luaL_iterator.test'. Stop.

The problem appears on Mac OS. It is a bit strange that we don't see it
on Linux. However I didn't dig into this, just fixed the observed
problem.

Related: https://github.com/curl/curl/issues/2121

Fixes #6656

The test fails quite often with one of the following results.

Either

[001] @@ -60,7 +60,7 @@
[001]  ...
[001]  test_timeout()
[001]  ---
[001] -- true
[001] +- false
[001]  ...
[001]  test_run:cmd("switch default")
[001]  ---
[001]

Or

[034]  test_timeout()
[034]  ---
[034] -- true
[034] +- error: 'replicas are not in the follow status'
[034]  ...

Both errors are caused by wait_cond checking saved
`box.info.replication` values instead of actual `box.info.replication`
output. Fix this.

Also, the test's quite long for no reason.

The wait_cond waiting for replication to break lasts a whole
`replication_timeout`, which is excess.

The test's idea, as stated in the commit that has introduced it
(195d4462: Send relay heartbeat if wal changes won't be send) is to
constantly relay some data to the remote peers and make
it so their relays never send heartbeats.

If we wait for a whole replication timeout between inserts, there's a high
chance of peer relays waking up naturally (after replication_timeout passes).

In this case the test tests nothing. Fix the issue by waiting for ~ 1/5
of replication_timeout in between the despatches. This also reduces test
run time from ~4.5 to ~1.5 seconds on my machine.

Remove the test from fragile list, since it shouldn't be flaky anymore.

Closes #4940

vfork got deprecated on macOS: for now, change the vfork deprecation
error to a warning until we find a suitable solution. We plan to review
popen's design and get rid of vfork completely or change vfork to
posix_spawn only on macOS in #6674.

Closes #6576

vfork got deprecated on macOS: change vfork to more portable
posix_spawn.

Needed for #6576

If slab order of first several pools was equal to zero,
all these pools added to same group with pools with slab
order is equal to one. It's leds to incorrect behavior
of `smfree` function, because this function relies on
same slab order size of all pools in mempool group.
Fix this and add assertion and test to check it.

Closes #6659

Tuple field count overflow is not handled: before inserting, explicitly
compare the current tuple size to box.schema.FIELD_MAX and set a newly
introduced diagnostic message if overflow occurs.

Closes #6198

Tests inserting fields into tuples, causing it to append lots of
nil's, work unacceptably slow: optimize mp_next.

Needed for #6198

On completion of compaction tasks we remove compacted run files created
after the last checkpoint immediately to save disk space. In order to
perform this optimization we compare the unused runs' dump LSN with the
last checkpoint's one.

But during replica's initial JOIN stage we set the LSN of all rows
received from remote master to 0 (see
box/box.cc/boostrap_journal_write). Considering that the LSN of an
initial checkpoint is also 0, our optimization stops working, and we
receive a huge disk space usage spike (as the unused run files will
only get removed when garbage collection occurs).

We should check the vinyl space engine's status and perform
our optimization unconditionally if we are in replica's initial JOIN
stage.

Closes #6568