Accept patches from Jörg Richter which fix a bug and implement
a todo item in the small allocator, and change slab_alloc_minimal to
16, to work well out of the box with small tuples.

Add missing comma to iproto_constants.c.

Required for #194

A fix and a test case for gh-791: call() return value can't contain
a Lua map.

Treat a map as a scalar, i.e. convert it to a tuple field on return
from call.

Use a simple heuristics to distinguish a map from an array:
if the first key of the map is integer 1, it's an array.
This approach works pretty well for JSON maps, which can't have
non-string keys anyway.

If failed to set locale to C.UTF-8 (the standard one which supports UTF-8),
try other options. Keep trying ad naeuseam to silence the warning,
since this raises support questions.

CentOS 6.6 has en_US.utf8.
Debian, FreeBSD and Darwin have en_US.utf-8.

Provide a good memory footprint out of the box.

Fix a crash due to memory corruption in fio.dirname().
A test case would test nothing since the corrupted Lua heap
doesn't crash on you immediately..

fio.dirname() is tested in fio.test.lua

Setting password for user guest has no effect, add a check
that prohibits setting a password for a user just like it does
laready for a role.

Add a test case.

The bug was caused by incorrect check for universal access in SetuidGuard
and missing user id change thanks to it.

Fix a coding bug and add a test case.
Closes gh-508.

A fix and a test case for gh-664 (bad syntax is accepted in
schema.role.*, schema.user.*, etc).

Check the second argument of role.create()/role.drop(),
user.create()/user.drop(), func.create()/func.drop().

Implement if_exists option for user.drop(), role.drop(),
func.drop().

Add a test case.

Don't start event loop in interactive mode after it ends.

Implement box.schema.user.create(name, { if_not_exists = true })
box.schema.role.create(name, { if_not_exists = true })
box.schema.func.create(name, { if_not_exists = true })

Add tests.

Fixes gh-637.

Use a correct server id to relay snapshot records to a replica
(server id 0 is server id of the snapshot).

Move recover_snap() implementation to memtx_engine.

Fixes gh-778: fiber.id() throws an error for a dead fiber.

Always return fiber id, even for a dead fiber.
Add a test case.

* transaction is yet active on CALL return
* transaction conflicts on commit

Don't run event loop if socket is ready to write. *Always* write
optimistically. With fiber.channel(), channel push  wakes up the waiter
without in the same event loop iteration. This means that whenever we have
data on an incoming connection, we can push messages to outgoing connections
in the same iteration, without epoll_ctl syscalls or unnecessary
epoll_wait.
With 1 incoming and 5 outgoing connections, for example, this saves
~10 syscalls per each incoming request.

If a fiber becomes ready on a user-defined event, schedule it
in the same event loop, not in the next loop, to save one epoll_wait()
invocation in cases when something is added to a fiber channel,
and there is a fiber ready for execution waiting on the channel.

Add a micro-bench to the unit test suite.

Rollback d29469f2, it makes build bug on FreeBSD build and doesn't affect on OSX