Patch 19a161b5  removed Ubuntu 20.10 from CI/CD for amd64 only.
Let's do it for arm64 as well.

Ubuntu 20.10 reached end-of-life few months ago,
remove it from CI/CD.

The uri is passed to the accept callback. Currently, it's unused,
but in future we will use uri parameters to configure connections
(e.g. setup encryption).

While we are at it, let's also add a test case for the move constructor.

We are planning to pass some extra connection parameters in uri
(e.g. encryption options).

Before this patch Tarantool http client did not support HTTP/2.
The reasons to enable HTTP/2 support are efficiency offered by
the protocol and potential ability to interact with a GRPC server.
The CMake version requirement is updated from 3.2 to 3.3, because
we need generator expressions in cmake for setting multiple paths
in CMAKE_FIND_ROOT_PATH for nghttp2 support.

Closes #5771

@TarantoolBot document
Title: Now we require CMake 3.3 to build tarantool

In tarantool/doc#2065 we requested to update
the CMake minimum version to 3.2. Now it is time for 3.3.
See details in the linked commit.

Macros SIGSTKSZ used to be an integral constant but
in glibc-2.34 it turned into a runtime function so it
cannot be used as constant known size for arrays anymore.

Beyond this, SIGSTKSZ is not enough for alt. signal stack size
when you use ASAN, so the size was increased.

Closes #6686

Closes #3349

@TarantoolBot document
Title: Document that takes_raw_args function option

A new function option was added - `takes_raw_args`:

```lua
box.schema.func.create('my_func', {takes_raw_args = true})
```

If set for a Lua function, the functions arguments will be passed
wrapped in a msgpack object (see `msgpack.object()`) when the function
is called over IPROTO or by `box.func.NAME:call()`:

```lua
local msgpack = require('msgpack')
local my_func = function(mp)
    assert(msgpack.is_object(mp))
    local args = mp:decode() -- array of arguments
end
```

Using this option might improve performance in case a function forwards
most of its arguments to another instance or writes them to a database,
because it eliminates msgpack decoding in Lua.

Currently, `is_multikey` function option must be wrapped in `opts`
sub-table, which looks ugly:

```lua
box.schema.func.create('my_func', {
    is_deterministic = true,
    opts = {is_multikey = true}
})
```

Let's allow to pass the option directly.

Note, the `is_multikey` option is reported by `box.func` at the same
level as `is_deterministic` so this would only be consistent:

```lua
box.func.my_func.is_deterministic
box.func.my_func.is_multikey
```

@TarantoolBot document
Title: Document that is_multikey func opt may be passed without sub-table

It's not necessary anymore to wrap `is_multikey` in `opts` table when
creating a function:

```lua
box.schema.func.create('my_func', {is_multikey = true})
```

is equivalent to

```lua
box.schema.func.create('my_func', {opts = {is_multikey = true}})
```

which still works, but deprecated.

Pease update [this][func-index] documentation entry.

[func-index]: https://www.tarantool.io/en/doc/latest/reference/reference_lua/box_space/create_index/#creating-a-functional-index

Closes #4861

@TarantoolBot document
Title: Document return_raw net.box option

If the `return_raw` flag is set, net.box will return response data
wrapped in a msgpack object (see `msgpack.object()`) instead of decoding
it to Lua. This can be useful when a response is supposed to be passed
through without decoding or with partial decoding - creating a msgpack
object can reduce pressure on the Lua garbage collector in this case.

Example:

```lua
local c = require('net.box').connect(uri)
local mp = c.eval('eval ...', {1, 2, 3}, {return_raw = true})
mp:decode() -- {1, 2, 3}
```

The option is ignored for methods that return nil (`begin`, `commit`,
`rollback`, `upsert`, `prepare` will return nil even if `return_raw` is
set) and for `index.count` (it returns a number). For `execute`, the
option is applied only to data (`rows`) while metadata is decoded
irrespective of the value of `return_raw`.

Internally, net.box can do some transformations of a key passed to
select/update/delete, see luamp_covert_key. We need to skip those if
a msgpack object is passed.

Part of #3349

Msgpack object is already handled by the msgpack encoder.
All we need to do is to remove the args check from call/eval.

Part of #3349

Closes #1629
Closes #3909
Closes #5316
Needed for #3349
Needed for #4861

@TarantoolBot document
Title: Document msgpack.object

The following new methods were added to the Lua `msgpack` module:
 - `msgpack.object()`: Encodes an arbitrary Lua object given in its only
   argument in MsgPack and returns the encoded MsgPack data encapsulated
   in a MsgPack object.
 - `msgpack.object_from_raw()`: Creates a MsgPack object from raw data
   given either as a string or as a pointer and size.
 - `msgpack.is_object()`: Checks is the given argument is a MsgPack
   object.

Example:

```lua
local buffer = require('buffer')
local msgpack = require('msgpack')
-- Create an object from a Lua object.
local mp = msgpack.object(123)
local mp = msgpack.object("foobar")
local mp = msgpack.object({1, 2, 3})
local mp = msgpack.object({foo = 1, bar = 2})
local mp = msgpack.object(box.tuple.new(1, 2, 3))
-- Create an object from raw data given as a string.
local data = msgpack.encode({1, 2, 3})
local mp = msgpack.object_from_raw(data)
-- Create an object from raw data given as ptr and size.
local buf = buffer.ibuf()
msgpack.encode({1, 2, 3}, buf)
local mp = msgpack.object_from_raw(buf.buf, buf:size())
-- Check object.
msgpack.is_object(mp) -- true
msgpack.is_object({}) -- false
```

A MsgPack object has the following methods:

 - `mp:decode()`: Decodes MsgPack and returns a Lua object.
 - `mp:iterator()`: Returns an iterator over MsgPack data.

A MsgPack iterator has the following methods:

 - `it:decode_array_header()`: Decodes a MsgPack array header under the
   cursor and returns the number of elements in the array. After calling
   this function the iterator points to the first element of the array
   or to the value following the array if the array is empty. Raises an
   error if the type of the value under the iterator cursor is not
   `MP_ARRAY`.
 - `it:decode_map_header()`: Decodes a MsgPack map header under the
   cursor and returns the number of key value pairs in the map. After
   calling this function the iterator points to the first key stored in
   the map or to the value following the map if the map is empty. Raises
   an error if the type of the value under the iterator cursor is not
   `MP_MAP`.
 - `it:decode()`: Decodes a MsgPack value under the iterator cursor and
   advances the cursor. Returns a Lua object corresponding to the
   MsgPack value. Raises a Lua error if there's no data to decode.
 - `it:take()`: Returns a MsgPack value under the iterator cursor as a
   MsgPack object (without decoding). Raises a Lua error if there's no
   data to decode. This function doesn't copy MsgPack data - instead it
   takes a reference to the original object.
 - `it:skip()`: Advances the iterator cursor by skipping one MsgPack
   value under the cursor. Returns nothing. Raises a Lua error if
   there's not data to skip.

Usage example:

```lua
local msgpack = require('msgpack')
local mp = msgpack.object({foo = 123, bar = {1, 2, 3}})
local it = mp:iterator()
it:decode_map_header()  -- returns 2
it:decode()             -- returns 'foo'
it:decode()             -- returns 123
it:skip()               -- returns none, skips 'bar'
local mp2 = it:take()
mp2:decode()            -- returns {1, 2, 3}
```

For more examples, see `test/app-luatest/msgpack_test.lua`.

A MsgPack object can be passed to MsgPack encoder with the effect being
the same as passing the original Lua object:

```lua
local msgpack = require('msgpack')
local mp = msgpack.object(123)
msgpack.object({mp, mp}):decode()         -- returns {123, 123}
msgpack.decode(msgpack.encode({mp, mp}))  -- returns {123, 123}
```

In particular, this means that if a MsgPack object stores an array,
it can be inserted into a database space:

```lua
box.space.my_space:insert(msgpack.object({1, 2, 3}))
```

* storage hints implemented for `struct datetime` values;
* proper comparison for indices of those datetime values;

* introduced `engine-tap/` for engine tests, and added `datetime.test.lua`
  there for testing of storing and sorting of datetime values.

Follow-up to #5941
Part of #5946

@TarantoolBot document

Title: Storage support for datetime values

It's now possible to store datetime values to spaces and to create
indexed datetime fields.

Serialize `struct datetime` as newly introduced MP_EXT type.
It saves 1 required integer field and 3 optional fields:
- epoch is required field;
- but nsec, tzoffset and tzindex are optional;

* supported json, yaml serialization formats, lua output mode;

Please refer to the
https://hackmd.io/@Mons/S1Vfc_axK#%D0%A3%D0%BF%D0%B0%D0%BA%D0%BE%D0%B2%D0%BA%D0%B0-%D0%B2-msgpack
for a description of a messagepack serialization schema for datetime
values.

Follow-up to #5941
Part of #5946

Add interval arithmetics as date:add{} and date:sub{}, or
overloaded '+' and '-' operators (__add/_sub metamethods).
There is no special processing for timezone specific
daylight saving time changes, which impacts result of arithmetics.
This will be completed after Olsen database support added.

If one needs to control a way arithmetic proceeded for days
of months then there is optional argument to `:add()` or `:sub()`,
which defines how we round day in month after arithmetic
operation:

- 'last' - mode when day snaps to the end of month, if happens:

  -- 29.02.2004 -> 31.03.2004
  dt:add{month = 1, adjust = "last" }

- 'none' only truncation toward the end of month performed
  (**default mode**):

  -- 29.02.* -> 29.03.*
  dt:add{month = 1, adjust = "none" }

- 'excess' - overflow mode, without any snap or truncation to
  the end of month, straight addition of days in month, stopping
  over month boundaries if there is less number of days.

  dt = new{year = 2004, month = 1, day = 31}
  dt:add{month = 1, adjust = 'excess'} -> 02.03.2004

  -- i.e. there were 29 days in February 2004, 2 excessive days
     from original date 31 make result 2nd March 2004

Intervals objects, created via `datetime.interval.new()` call
and objects of the same attributes are equivalent for interval
arithmetics and exchangeable. i.e.

   dt + date.interval.new{year = 1, month = 2}

is the same as
   dt + {year = 1, month = 2}

Follow-up to #5941

Negative Julian dates used to be incorrectly formatted, if there were
non-zero hour/minutes/second.

```
tarantool> T = date.new{year = -1, month = 6, day = 10, hour = 12, min = 10, sec = 10}

tarantool> T
- -001-06-11T-11:-49:-50Z
```

The problem is due to operator % in C/C++ which is returning negative
modulo toward 0. Use Euclidian (floored) modulo instead.

Correct both `tostring(T)`/`T:format()` and `strftime`-like
'T:format(fmt)'

```
tarantool> T:format()
- -001-06-10T12:10:10Z

tarantool> T:format('%FT%T.%f%z')
- -001-06-10T12:10:10.000+0000
```

Follow-up to #5941

uri.parse always creates params table even if there's no query
parameters in the given uri:

tarantool> require('uri').parse('localhost:3301')
---
- host: localhost
  service: '3301'
  params: []
...

This is confusing and inconsistent with other uri properties, which are
created only if they actually exist. Let's create the params table only
if the given uri actually has any params.

Needed for Tarantool EE. Off by default.

After updating the uri library, the "params" table was added
to the URI structure. It breaks backward compatibility in
case when URI passed to "uri.format" function in old way,
without params table.

This commit defines two cmake variables:

 - EXTRA_REEXPORT_LIBRARIES: list of libraries appended to
   reexport_libraries.
 - EXTRA_EXPORTS_FILE_SOURCES: list of exports file sources appended to
   export_file_sources.

Needed to export extra symbols in Tarantool EE build.

Currently, we use the same luarocks hardcoded config for static build as
for regular build. As a result, LUA_BINDIR, LUA_INCDIR, SYSCONFDIR, and
PREFIX point to locations inside the build directory, which doesn't make
any sense.

Let's introduce a special hardcoded config which will set those
directories to locations relative to the binary location:

 - LUA_BINDIR: directory where the binary is located (bindir)
 - PREFIX: <dir> if bindir is <dir>/bin, otherwise bindir.
 - LUA_INCDIR: PREFIX/include/tarantool
 - SYSCONFDIR: PREFIX/etc/tarantool/rocks

Also, let's add the PREFIX/rocks to the rocks path if present.

This is needed for Tarantool SDK bundle to work as expected.
In particular the check for <dir>/bin is needed, because SDK
installs binaries in bundle root, not in /bin.

The options are not used anymore since update to 3.1.1,
see third_party/luarocks/src/luarocks/core/cfg.lua.

They should have been removed by commit
4222c1f6
("rocks: update luarocks to 3.1.1")

It contains Lua code. Format it appropriately. No functional changes.

On heavy loaded hosts either slow machines like VMware found the
following issues:

  [010] --- replication/long_row_timeout.result   Fri May  8 08:56:08 2020
  [010] +++ var/rejects/replication/long_row_timeout.reject       Mon Jun 21 04:39:08 2021
  [010] @@ -23,7 +23,7 @@
  [010]  ...
  [010]  box.info.replication[2].downstream.status
  [010]  ---
  [010] -- follow
  [010] +- stopped
  [010]  ...
  [010]  -- make applier incapable of reading rows in one go, so that it
  [010]  -- yields a couple of times.
  [010]

It happened because replication downstream status check occurred too
early, when it was only in 'stopped' state. This situation happens
when replica done its initial join, but not reached subscription yet.
To give the replication status check routine ability to reach the
needed 'follow' state, it need for it using test_run:wait_downstream()
routine.

We don't see this issue anymore, but let's fix it still in case we ever
encounter it again.

Also remove the test from fragile list.

Closes #4351

Co-developed-by: Serge Petrenko <sergepetrenko@tarantool.org>

After this patch, MAP values can be used as bind variables. However, due
to the current syntax for binding in Lua, the only possible way is to
use MAP values as the named bind variable.

Part of #4763

After this patch, ARRAY values can be used as bind variables. However,
due to the current syntax for binding in Lua, the only possible way is
to use ARRAY values as the named bind variable.

Part of #4762

What is bad: a considerable amount of boilerplate code should be added
to just run a simple script. I hope we'll do something with this
in #6604.

After this change the release notes are able to be built.

Changed '---' to '----', so comments for a release manager are parsed
correctly.

The idea is to allow to add more checks here: I'm going to add
changelogs check in a next commit.

This commit allows to override Tarantool version detection by setting
CMake TARANTOOL_VERSION variable. This is needed to build Tarantool as a
part of tarantool/sdk, which sets its own, extended version.

Also, this commit updates box-py/args.test.py to allow arbitrary tags
appended to the version (tarantool/sdk appends -rNNN).

Submodules are auto-updated in two cases:
1. When cmake is called for the first time (and there's no
   src/lib/small/CMakeLists.txt).
2. If git-describe version doesn't match the version stored in the
   VERSION file.

The first case looks okay, but the second one is confusing. It means
that the following sequence of commands does NOT update submodules:

```sh
cmake . # initial configuration, fetches small
cd src/lib/small && git checkout HEAD^ && cd - # checkout small
cmake . # does NOT update small
```

while the following sequence of commands does:

```sh
git commit --allow-empty -m dummy
cmake . # updates small!
```

This looks confusing, inconsistent, and dangerous. Let's drop this code
and let the developers update submodules manually when they need to.

This effectively reverts commit c83eea60
("CMake: automatically update submodules").

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.