Telethon/telethon/network/mtprotosender.py

import asyncio
import collections
import struct

from . import authenticator
from ..extensions.messagepacker import MessagePacker
from .mtprotoplainsender import MTProtoPlainSender
from .requeststate import RequestState
from .mtprotostate import MTProtoState
from ..tl.tlobject import TLRequest
from .. import helpers, utils
from ..errors import (
    BadMessageError, InvalidBufferError, AuthKeyNotFound, SecurityError,
    TypeNotFoundError, rpc_message_to_error
)
from ..extensions import BinaryReader
from ..tl.core import RpcResult, MessageContainer, GzipPacked
from ..tl.functions.auth import LogOutRequest
from ..tl.functions import PingRequest, DestroySessionRequest, DestroyAuthKeyRequest
from ..tl.types import (
    MsgsAck, Pong, BadServerSalt, BadMsgNotification, FutureSalts,
    MsgNewDetailedInfo, NewSessionCreated, MsgDetailedInfo, MsgsStateReq,
    MsgsStateInfo, MsgsAllInfo, MsgResendReq, upload, DestroySessionOk, DestroySessionNone,
    DestroyAuthKeyOk, DestroyAuthKeyNone, DestroyAuthKeyFail
)
from ..tl import types as _tl
from ..crypto import AuthKey
from ..helpers import retry_range


class MTProtoSender:
    """
    MTProto Mobile Protocol sender
    (https://core.telegram.org/mtproto/description).

    This class is responsible for wrapping requests into `TLMessage`'s,
    sending them over the network and receiving them in a safe manner.

    Automatic reconnection due to temporary network issues is a concern
    for this class as well, including retry of messages that could not
    be sent successfully.

    A new authorization key will be generated on connection if no other
    key exists yet.
    """
    def __init__(self, auth_key, *, loggers,
                 retries=5, delay=1, auto_reconnect=True, connect_timeout=None,
                 auth_key_callback=None,
                 updates_queue=None, auto_reconnect_callback=None):
        self._connection = None
        self._loggers = loggers
        self._log = loggers[__name__]
        self._retries = retries
        self._delay = delay
        self._auto_reconnect = auto_reconnect
        self._connect_timeout = connect_timeout
        self._auth_key_callback = auth_key_callback
        self._updates_queue = updates_queue
        self._auto_reconnect_callback = auto_reconnect_callback
        self._connect_lock = asyncio.Lock()
        self._ping = None

        # Whether the user has explicitly connected or disconnected.
        #
        # If a disconnection happens for any other reason and it
        # was *not* user action then the pending messages won't
        # be cleared but on explicit user disconnection all the
        # pending futures should be cancelled.
        self._user_connected = False
        self._reconnecting = False
        self._disconnected = helpers.get_running_loop().create_future()
        self._disconnected.set_result(None)

        # We need to join the loops upon disconnection
        self._send_loop_handle = None
        self._recv_loop_handle = None

        # Preserving the references of the AuthKey and state is important
        self.auth_key = auth_key or AuthKey(None)
        self._state = MTProtoState(self.auth_key, loggers=self._loggers)

        # Outgoing messages are put in a queue and sent in a batch.
        # Note that here we're also storing their ``_RequestState``.
        self._send_queue = MessagePacker(self._state, loggers=self._loggers)

        # Sent states are remembered until a response is received.
        self._pending_state = {}

        # Responses must be acknowledged, and we can also batch these.
        self._pending_ack = set()

        # Similar to pending_messages but only for the last acknowledges.
        # These can't go in pending_messages because no acknowledge for them
        # is received, but we may still need to resend their state on bad salts.
        self._last_acks = collections.deque(maxlen=10)

        # Jump table from response ID to method that handles it
        self._handlers = {
            RpcResult.CONSTRUCTOR_ID: self._handle_rpc_result,
            MessageContainer.CONSTRUCTOR_ID: self._handle_container,
            GzipPacked.CONSTRUCTOR_ID: self._handle_gzip_packed,
            Pong.CONSTRUCTOR_ID: self._handle_pong,
            BadServerSalt.CONSTRUCTOR_ID: self._handle_bad_server_salt,
            BadMsgNotification.CONSTRUCTOR_ID: self._handle_bad_notification,
            MsgDetailedInfo.CONSTRUCTOR_ID: self._handle_detailed_info,
            MsgNewDetailedInfo.CONSTRUCTOR_ID: self._handle_new_detailed_info,
            NewSessionCreated.CONSTRUCTOR_ID: self._handle_new_session_created,
            MsgsAck.CONSTRUCTOR_ID: self._handle_ack,
            FutureSalts.CONSTRUCTOR_ID: self._handle_future_salts,
            MsgsStateReq.CONSTRUCTOR_ID: self._handle_state_forgotten,
            MsgResendReq.CONSTRUCTOR_ID: self._handle_state_forgotten,
            MsgsAllInfo.CONSTRUCTOR_ID: self._handle_msg_all,
            DestroySessionOk.CONSTRUCTOR_ID: self._handle_destroy_session,
            DestroySessionNone.CONSTRUCTOR_ID: self._handle_destroy_session,
            DestroyAuthKeyOk.CONSTRUCTOR_ID: self._handle_destroy_auth_key,
            DestroyAuthKeyNone.CONSTRUCTOR_ID: self._handle_destroy_auth_key,
            DestroyAuthKeyFail.CONSTRUCTOR_ID: self._handle_destroy_auth_key,
        }

    # Public API

    async def connect(self, connection):
        """
        Connects to the specified given connection using the given auth key.
        """
        async with self._connect_lock:
            if self._user_connected:
                self._log.info('User is already connected!')
                return False

            self._connection = connection
            await self._connect()
            self._user_connected = True
            return True

    def is_connected(self):
        return self._user_connected

    def _transport_connected(self):
        return (
            not self._reconnecting
            and self._connection is not None
            and self._connection._connected
        )

    async def disconnect(self):
        """
        Cleanly disconnects the instance from the network, cancels
        all pending requests, and closes the send and receive loops.
        """
        await self._disconnect()

    def send(self, request, ordered=False):
        """
        This method enqueues the given request to be sent. Its send
        state will be saved until a response arrives, and a ``Future``
        that will be resolved when the response arrives will be returned:

        .. code-block:: python

            async def method():
                # Sending (enqueued for the send loop)
                future = sender.send(request)
                # Receiving (waits for the receive loop to read the result)
                result = await future

        Designed like this because Telegram may send the response at
        any point, and it can send other items while one waits for it.
        Once the response for this future arrives, it is set with the
        received result, quite similar to how a ``receive()`` call
        would otherwise work.

        Since the receiving part is "built in" the future, it's
        impossible to await receive a result that was never sent.
        """
        if not self._user_connected:
            raise ConnectionError('Cannot send requests while disconnected')

        if not utils.is_list_like(request):
            try:
                state = RequestState(request)
            except struct.error as e:
                # "struct.error: required argument is not an integer" is not
                # very helpful; log the request to find out what wasn't int.
                self._log.error('Request caused struct.error: %s: %s', e, request)
                raise

            self._send_queue.append(state)
            return state.future
        else:
            states = []
            futures = []
            state = None
            for req in request:
                try:
                    state = RequestState(req, after=ordered and state)
                except struct.error as e:
                    self._log.error('Request caused struct.error: %s: %s', e, request)
                    raise

                states.append(state)
                futures.append(state.future)

            self._send_queue.extend(states)
            return futures

    @property
    def disconnected(self):
        """
        Future that resolves when the connection to Telegram
        ends, either by user action or in the background.

        Note that it may resolve in either a ``ConnectionError``
        or any other unexpected error that could not be handled.
        """
        return asyncio.shield(self._disconnected)

    # Private methods

    async def _connect(self):
        """
        Performs the actual connection, retrying, generating the
        authorization key if necessary, and starting the send and
        receive loops.
        """
        self._log.info('Connecting to %s...', self._connection)

        connected = False

        for attempt in retry_range(self._retries):
            if not connected:
                connected = await self._try_connect(attempt)
                if not connected:
                    continue  # skip auth key generation until we're connected

            if not self.auth_key:
                try:
                    if not await self._try_gen_auth_key(attempt):
                        continue  # keep retrying until we have the auth key
                except (IOError, asyncio.TimeoutError) as e:
                    # Sometimes, specially during user-DC migrations,
                    # Telegram may close the connection during auth_key
                    # generation. If that's the case, we will need to
                    # connect again.
                    self._log.warning('Connection error %d during auth_key gen: %s: %s',
                                      attempt, type(e).__name__, e)

                    # Whatever the IOError was, make sure to disconnect so we can
                    # reconnect cleanly after.
                    await self._connection.disconnect()
                    connected = False
                    await asyncio.sleep(self._delay)
                    continue  # next iteration we will try to reconnect

            break  # all steps done, break retry loop
        else:
            if not connected:
                raise ConnectionError('Connection to Telegram failed {} time(s)'.format(self._retries))

            e = ConnectionError('auth_key generation failed {} time(s)'.format(self._retries))
            await self._disconnect(error=e)
            raise e

        loop = helpers.get_running_loop()
        self._log.debug('Starting send loop')
        self._send_loop_handle = loop.create_task(self._send_loop())

        self._log.debug('Starting receive loop')
        self._recv_loop_handle = loop.create_task(self._recv_loop())

        # _disconnected only completes after manual disconnection
        # or errors after which the sender cannot continue such
        # as failing to reconnect or any unexpected error.
        if self._disconnected.done():
            self._disconnected = loop.create_future()

        self._log.info('Connection to %s complete!', self._connection)

    async def _try_connect(self, attempt):
        try:
            self._log.debug('Connection attempt %d...', attempt)
            await self._connection.connect(timeout=self._connect_timeout)
            self._log.debug('Connection success!')
            return True
        except (IOError, asyncio.TimeoutError) as e:
            self._log.warning('Attempt %d at connecting failed: %s: %s',
                              attempt, type(e).__name__, e)
            await asyncio.sleep(self._delay)
            return False

    async def _try_gen_auth_key(self, attempt):
        plain = MTProtoPlainSender(self._connection, loggers=self._loggers)
        try:
            self._log.debug('New auth_key attempt %d...', attempt)
            self.auth_key.key, self._state.time_offset = \
                await authenticator.do_authentication(plain)

            # This is *EXTREMELY* important since we don't control
            # external references to the authorization key, we must
            # notify whenever we change it. This is crucial when we
            # switch to different data centers.
            if self._auth_key_callback:
                self._auth_key_callback(self.auth_key)

            self._log.debug('auth_key generation success!')
            return True
        except (SecurityError, AssertionError) as e:
            self._log.warning('Attempt %d at new auth_key failed: %s', attempt, e)
            await asyncio.sleep(self._delay)
            return False

    async def _disconnect(self, error=None):
        if self._connection is None:
            self._log.info('Not disconnecting (already have no connection)')
            return

        self._log.info('Disconnecting from %s...', self._connection)
        self._user_connected = False
        try:
            self._log.debug('Closing current connection...')
            await self._connection.disconnect()
        finally:
            self._log.debug('Cancelling %d pending message(s)...', len(self._pending_state))
            for state in self._pending_state.values():
                if error and not state.future.done():
                    state.future.set_exception(error)
                else:
                    state.future.cancel()

            self._pending_state.clear()
            await helpers._cancel(
                self._log,
                send_loop_handle=self._send_loop_handle,
                recv_loop_handle=self._recv_loop_handle
            )

            self._log.info('Disconnection from %s complete!', self._connection)
            self._connection = None

        if self._disconnected and not self._disconnected.done():
            if error:
                self._disconnected.set_exception(error)
            else:
                self._disconnected.set_result(None)

    async def _reconnect(self, last_error):
        """
        Cleanly disconnects and then reconnects.
        """
        self._log.info('Closing current connection to begin reconnect...')
        await self._connection.disconnect()

        await helpers._cancel(
            self._log,
            send_loop_handle=self._send_loop_handle,
            recv_loop_handle=self._recv_loop_handle
        )

        # TODO See comment in `_start_reconnect`
        # Perhaps this should be the last thing to do?
        # But _connect() creates tasks which may run and,
        # if they see that reconnecting is True, they will end.
        # Perhaps that task creation should not belong in connect?
        self._reconnecting = False

        # Start with a clean state (and thus session ID) to avoid old msgs
        self._state.reset()

        retries = self._retries if self._auto_reconnect else 0

        attempt = 0
        ok = True
        # We're already "retrying" to connect, so we don't want to force retries
        for attempt in retry_range(retries, force_retry=False):
            try:
                await self._connect()
            except (IOError, asyncio.TimeoutError) as e:
                last_error = e
                self._log.info('Failed reconnection attempt %d with %s',
                               attempt, e.__class__.__name__)
                await asyncio.sleep(self._delay)
            except BufferError as e:
                # TODO there should probably only be one place to except all these errors
                if isinstance(e, InvalidBufferError) and e.code == 404:
                    self._log.info('Server does not know about the current auth key; the session may need to be recreated')
                    last_error = AuthKeyNotFound()
                    ok = False
                    break
                else:
                    self._log.warning('Invalid buffer %s', e)

            except Exception as e:
                last_error = e
                self._log.exception('Unexpected exception reconnecting on '
                                    'attempt %d', attempt)

                await asyncio.sleep(self._delay)
            else:
                self._send_queue.extend(self._pending_state.values())
                self._pending_state.clear()

                if self._auto_reconnect_callback:
                    helpers.get_running_loop().create_task(self._auto_reconnect_callback())

                break
        else:
            ok = False

        if not ok:
            self._log.error('Automatic reconnection failed %d time(s)', attempt)
            # There may be no error (e.g. automatic reconnection was turned off).
            error = last_error.with_traceback(None) if last_error else None
            await self._disconnect(error=error)

    def _start_reconnect(self, error):
        """Starts a reconnection in the background."""
        if self._user_connected and not self._reconnecting:
            # We set reconnecting to True here and not inside the new task
            # because it may happen that send/recv loop calls this again
            # while the new task hasn't had a chance to run yet. This race
            # condition puts `self.connection` in a bad state with two calls
            # to its `connect` without disconnecting, so it creates a second
            # receive loop. There can't be two tasks receiving data from
            # the reader, since that causes an error, and the library just
            # gets stuck.
            # TODO It still gets stuck? Investigate where and why.
            self._reconnecting = True
            helpers.get_running_loop().create_task(self._reconnect(error))

    def _keepalive_ping(self, rnd_id):
        """
        Send a keep-alive ping. If a pong for the last ping was not received
        yet, this means we're probably not connected.
        """
        # TODO this is ugly, update loop shouldn't worry about this, sender should
        if self._ping is None:
            self._ping = rnd_id
            self.send(PingRequest(rnd_id))
        else:
            self._start_reconnect(None)

    # Loops

    async def _send_loop(self):
        """
        This loop is responsible for popping items off the send
        queue, encrypting them, and sending them over the network.

        Besides `connect`, only this method ever sends data.
        """
        while self._user_connected and not self._reconnecting:
            if self._pending_ack:
                ack = RequestState(MsgsAck(list(self._pending_ack)))
                self._send_queue.append(ack)
                self._last_acks.append(ack)
                self._pending_ack.clear()

            self._log.debug('Waiting for messages to send...')
            # TODO Wait for the connection send queue to be empty?
            # This means that while it's not empty we can wait for
            # more messages to be added to the send queue.
            batch, data = await self._send_queue.get()

            if not data:
                continue

            self._log.debug('Encrypting %d message(s) in %d bytes for sending',
                            len(batch), len(data))

            data = self._state.encrypt_message_data(data)

            # Whether sending succeeds or not, the popped requests are now
            # pending because they're removed from the queue. If a reconnect
            # occurs, they will be removed from pending state and re-enqueued
            # so even if the network fails they won't be lost. If they were
            # never re-enqueued, the future waiting for a response "locks".
            for state in batch:
                if not isinstance(state, list):
                    if isinstance(state.request, TLRequest):
                        self._pending_state[state.msg_id] = state
                else:
                    for s in state:
                        if isinstance(s.request, TLRequest):
                            self._pending_state[s.msg_id] = s

            try:
                await self._connection.send(data)
            except IOError as e:
                self._log.info('Connection closed while sending data')
                self._start_reconnect(e)
                return

            self._log.debug('Encrypted messages put in a queue to be sent')

    async def _recv_loop(self):
        """
        This loop is responsible for reading all incoming responses
        from the network, decrypting and handling or dispatching them.

        Besides `connect`, only this method ever receives data.
        """
        while self._user_connected and not self._reconnecting:
            self._log.debug('Receiving items from the network...')
            try:
                body = await self._connection.recv()
            except IOError as e:
                self._log.info('Connection closed while receiving data')
                self._start_reconnect(e)
                return
            except InvalidBufferError as e:
                if e.code == 429:
                    self._log.warning('Server indicated flood error at transport level: %s', e)
                    await self._disconnect(error=e)
                else:
                    self._log.exception('Server sent invalid buffer')
                    self._start_reconnect(e)
                return
            except Exception as e:
                self._log.exception('Unhandled error while receiving data')
                self._start_reconnect(e)
                return

            try:
                message = self._state.decrypt_message_data(body)
                if message is None:
                    continue  # this message is to be ignored
            except TypeNotFoundError as e:
                # Received object which we don't know how to deserialize
                self._log.info('Type %08x not found, remaining data %r',
                               e.invalid_constructor_id, e.remaining)
                continue
            except SecurityError as e:
                # A step while decoding had the incorrect data. This message
                # should not be considered safe and it should be ignored.
                self._log.warning('Security error while unpacking a '
                                  'received message: %s', e)
                continue
            except BufferError as e:
                if isinstance(e, InvalidBufferError) and e.code == 404:
                    self._log.info('Server does not know about the current auth key; the session may need to be recreated')
                    await self._disconnect(error=AuthKeyNotFound())
                else:
                    self._log.warning('Invalid buffer %s', e)
                    self._start_reconnect(e)
                return
            except Exception as e:
                self._log.exception('Unhandled error while decrypting data')
                self._start_reconnect(e)
                return

            try:
                await self._process_message(message)
            except Exception:
                self._log.exception('Unhandled error while processing msgs')

    # Response Handlers

    async def _process_message(self, message):
        """
        Adds the given message to the list of messages that must be
        acknowledged and dispatches control to different ``_handle_*``
        method based on its type.
        """
        self._pending_ack.add(message.msg_id)
        handler = self._handlers.get(message.obj.CONSTRUCTOR_ID,
                                     self._handle_update)
        await handler(message)

    def _pop_states(self, msg_id):
        """
        Pops the states known to match the given ID from pending messages.

        This method should be used when the response isn't specific.
        """
        state = self._pending_state.pop(msg_id, None)
        if state:
            return [state]

        to_pop = []
        for state in self._pending_state.values():
            if state.container_id == msg_id:
                to_pop.append(state.msg_id)

        if to_pop:
            return [self._pending_state.pop(x) for x in to_pop]

        for ack in self._last_acks:
            if ack.msg_id == msg_id:
                return [ack]

        return []

    async def _handle_rpc_result(self, message):
        """
        Handles the result for Remote Procedure Calls:

            rpc_result#f35c6d01 req_msg_id:long result:bytes = RpcResult;

        This is where the future results for sent requests are set.
        """
        rpc_result = message.obj
        state = self._pending_state.pop(rpc_result.req_msg_id, None)
        self._log.debug('Handling RPC result for message %d',
                        rpc_result.req_msg_id)

        if not state:
            # TODO We should not get responses to things we never sent
            # However receiving a File() with empty bytes is "common".
            # See #658, #759 and #958. They seem to happen in a container
            # which contain the real response right after.
            #
            # But, it might also happen that we get an *error* for no parent request.
            # If that's the case attempting to read from body which is None would fail with:
            # "BufferError: No more data left to read (need 4, got 0: b''); last read None".
            # This seems to be particularly common for "RpcError(error_code=-500, error_message='No workers running')".
            if rpc_result.error:
                self._log.info('Received error without parent request: %s', rpc_result.error)
            else:
                try:
                    with BinaryReader(rpc_result.body) as reader:
                        if not isinstance(reader.tgread_object(), upload.File):
                            raise ValueError('Not an upload.File')
                except (TypeNotFoundError, ValueError):
                    self._log.info('Received response without parent request: %s', rpc_result.body)
            return

        if rpc_result.error:
            error = rpc_message_to_error(rpc_result.error, state.request)
            self._send_queue.append(
                RequestState(MsgsAck([state.msg_id])))

            if not state.future.cancelled():
                state.future.set_exception(error)
        else:
            try:
                with BinaryReader(rpc_result.body) as reader:
                    result = state.request.read_result(reader)
            except Exception as e:
                # e.g. TypeNotFoundError, should be propagated to caller
                if not state.future.cancelled():
                    state.future.set_exception(e)
            else:
                self._store_own_updates(result)
                if not state.future.cancelled():
                    state.future.set_result(result)

    async def _handle_container(self, message):
        """
        Processes the inner messages of a container with many of them:

            msg_container#73f1f8dc messages:vector<%Message> = MessageContainer;
        """
        self._log.debug('Handling container')
        for inner_message in message.obj.messages:
            await self._process_message(inner_message)

    async def _handle_gzip_packed(self, message):
        """
        Unpacks the data from a gzipped object and processes it:

            gzip_packed#3072cfa1 packed_data:bytes = Object;
        """
        self._log.debug('Handling gzipped data')
        with BinaryReader(message.obj.data) as reader:
            message.obj = reader.tgread_object()
            await self._process_message(message)

    async def _handle_update(self, message):
        try:
            assert message.obj.SUBCLASS_OF_ID == 0x8af52aac  # crc32(b'Updates')
        except AssertionError:
            self._log.warning(
                'Note: %s is not an update, not dispatching it %s',
                message.obj.__class__.__name__,
                message.obj
            )
            return

        self._log.debug('Handling update %s', message.obj.__class__.__name__)
        self._updates_queue.put_nowait(message.obj)

    def _store_own_updates(self, obj, *, _update_ids=frozenset((
        _tl.UpdateShortMessage.CONSTRUCTOR_ID,
        _tl.UpdateShortChatMessage.CONSTRUCTOR_ID,
        _tl.UpdateShort.CONSTRUCTOR_ID,
        _tl.UpdatesCombined.CONSTRUCTOR_ID,
        _tl.Updates.CONSTRUCTOR_ID,
        _tl.UpdateShortSentMessage.CONSTRUCTOR_ID,
    )), _update_like_ids=frozenset((
        _tl.messages.AffectedHistory.CONSTRUCTOR_ID,
        _tl.messages.AffectedMessages.CONSTRUCTOR_ID,
        _tl.messages.AffectedFoundMessages.CONSTRUCTOR_ID,
    ))):
        try:
            if obj.CONSTRUCTOR_ID in _update_ids:
                obj._self_outgoing = True  # flag to only process, but not dispatch these
                self._updates_queue.put_nowait(obj)
            elif obj.CONSTRUCTOR_ID in _update_like_ids:
                # Ugly "hack" (?) - otherwise bots reliably detect gaps when deleting messages.
                #
                # Note: the `date` being `None` is used to check for `updatesTooLong`, so `0` is
                # used instead. It is still not read, because `updateShort` has no `seq`.
                #
                # Some requests, such as `readHistory`, also return these types. But the `pts_count`
                # seems to be zero, so while this will produce some bogus `updateDeleteMessages`,
                # it's still one of the "cleaner" approaches to handling the new `pts`.
                # `updateDeleteMessages` is probably the "least-invasive" update that can be used.
                upd = _tl.UpdateShort(_tl.UpdateDeleteMessages([], obj.pts, obj.pts_count), 0)
                upd._self_outgoing = True
                self._updates_queue.put_nowait(upd)
        except AttributeError:
            pass

    async def _handle_pong(self, message):
        """
        Handles pong results, which don't come inside a ``rpc_result``
        but are still sent through a request:

            pong#347773c5 msg_id:long ping_id:long = Pong;
        """
        pong = message.obj
        self._log.debug('Handling pong for message %d', pong.msg_id)
        if self._ping == pong.ping_id:
            self._ping = None

        state = self._pending_state.pop(pong.msg_id, None)
        if state:
            state.future.set_result(pong)

    async def _handle_bad_server_salt(self, message):
        """
        Corrects the currently used server salt to use the right value
        before enqueuing the rejected message to be re-sent:

            bad_server_salt#edab447b bad_msg_id:long bad_msg_seqno:int
            error_code:int new_server_salt:long = BadMsgNotification;
        """
        bad_salt = message.obj
        self._log.debug('Handling bad salt for message %d', bad_salt.bad_msg_id)
        self._state.salt = bad_salt.new_server_salt
        states = self._pop_states(bad_salt.bad_msg_id)
        self._send_queue.extend(states)

        self._log.debug('%d message(s) will be resent', len(states))

    async def _handle_bad_notification(self, message):
        """
        Adjusts the current state to be correct based on the
        received bad message notification whenever possible:

            bad_msg_notification#a7eff811 bad_msg_id:long bad_msg_seqno:int
            error_code:int = BadMsgNotification;
        """
        bad_msg = message.obj
        states = self._pop_states(bad_msg.bad_msg_id)

        self._log.debug('Handling bad msg %s', bad_msg)
        if bad_msg.error_code in (16, 17):
            # Sent msg_id too low or too high (respectively).
            # Use the current msg_id to determine the right time offset.
            to = self._state.update_time_offset(
                correct_msg_id=message.msg_id)
            self._log.info('System clock is wrong, set time offset to %ds', to)
        elif bad_msg.error_code == 32:
            # msg_seqno too low, so just pump it up by some "large" amount
            # TODO A better fix would be to start with a new fresh session ID
            self._state._sequence += 64
        elif bad_msg.error_code == 33:
            # msg_seqno too high never seems to happen but just in case
            self._state._sequence -= 16
        else:
            for state in states:
                state.future.set_exception(
                    BadMessageError(state.request, bad_msg.error_code))
            return

        # Messages are to be re-sent once we've corrected the issue
        self._send_queue.extend(states)
        self._log.debug('%d messages will be resent due to bad msg',
                        len(states))

    async def _handle_detailed_info(self, message):
        """
        Updates the current status with the received detailed information:

            msg_detailed_info#276d3ec6 msg_id:long answer_msg_id:long
            bytes:int status:int = MsgDetailedInfo;
        """
        # TODO https://goo.gl/VvpCC6
        msg_id = message.obj.answer_msg_id
        self._log.debug('Handling detailed info for message %d', msg_id)
        self._pending_ack.add(msg_id)

    async def _handle_new_detailed_info(self, message):
        """
        Updates the current status with the received detailed information:

            msg_new_detailed_info#809db6df answer_msg_id:long
            bytes:int status:int = MsgDetailedInfo;
        """
        # TODO https://goo.gl/G7DPsR
        msg_id = message.obj.answer_msg_id
        self._log.debug('Handling new detailed info for message %d', msg_id)
        self._pending_ack.add(msg_id)

    async def _handle_new_session_created(self, message):
        """
        Updates the current status with the received session information:

            new_session_created#9ec20908 first_msg_id:long unique_id:long
            server_salt:long = NewSession;
        """
        # TODO https://goo.gl/LMyN7A
        self._log.debug('Handling new session created')
        self._state.salt = message.obj.server_salt

    async def _handle_ack(self, message):
        """
        Handles a server acknowledge about our messages. Normally
        these can be ignored except in the case of ``auth.logOut``:

            auth.logOut#5717da40 = Bool;

        Telegram doesn't seem to send its result so we need to confirm
        it manually. No other request is known to have this behaviour.

        Since the ID of sent messages consisting of a container is
        never returned (unless on a bad notification), this method
        also removes containers messages when any of their inner
        messages are acknowledged.
        """
        ack = message.obj
        self._log.debug('Handling acknowledge for %s', str(ack.msg_ids))
        for msg_id in ack.msg_ids:
            state = self._pending_state.get(msg_id)
            if state and isinstance(state.request, LogOutRequest):
                del self._pending_state[msg_id]
                if not state.future.cancelled():
                    state.future.set_result(True)

    async def _handle_future_salts(self, message):
        """
        Handles future salt results, which don't come inside a
        ``rpc_result`` but are still sent through a request:

            future_salts#ae500895 req_msg_id:long now:int
            salts:vector<future_salt> = FutureSalts;
        """
        # TODO save these salts and automatically adjust to the
        # correct one whenever the salt in use expires.
        self._log.debug('Handling future salts for message %d', message.msg_id)
        state = self._pending_state.pop(message.msg_id, None)
        if state:
            state.future.set_result(message.obj)

    async def _handle_state_forgotten(self, message):
        """
        Handles both :tl:`MsgsStateReq` and :tl:`MsgResendReq` by
        enqueuing a :tl:`MsgsStateInfo` to be sent at a later point.
        """
        self._send_queue.append(RequestState(MsgsStateInfo(
            req_msg_id=message.msg_id, info=chr(1) * len(message.obj.msg_ids)
        )))

    async def _handle_msg_all(self, message):
        """
        Handles :tl:`MsgsAllInfo` by doing nothing (yet).
        """

    async def _handle_destroy_session(self, message):
        """
        Handles both :tl:`DestroySessionOk` and :tl:`DestroySessionNone`.
        It behaves pretty much like handling an RPC result.
        """
        for msg_id, state in self._pending_state.items():
            if isinstance(state.request, DestroySessionRequest)\
                    and state.request.session_id == message.obj.session_id:
                break
        else:
            return

        del self._pending_state[msg_id]
        if not state.future.cancelled():
            state.future.set_result(message.obj)

    async def _handle_destroy_auth_key(self, message):
        """
        Handles :tl:`DestroyAuthKeyFail`, :tl:`DestroyAuthKeyNone`, and :tl:`DestroyAuthKeyOk`.

        :tl:`DestroyAuthKey` is not intended for users to use, but they still
        might, and the response won't come in `rpc_result`, so thhat's worked
        around here.
        """
        self._log.debug('Handling destroy auth key %s', message.obj)
        for msg_id, state in list(self._pending_state.items()):
            if isinstance(state.request, DestroyAuthKeyRequest):
                del self._pending_state[msg_id]
                if not state.future.cancelled():
                    state.future.set_result(message.obj)

        # If the auth key has been destroyed, that pretty much means the
        # library can't continue as our auth key will no longer be found
        # on the server.
        # Even if the library didn't disconnect, the server would (and then
        # the library would reconnect and learn about auth key being invalid).
        if isinstance(message.obj, DestroyAuthKeyOk):
            await self._disconnect(error=AuthKeyNotFound())