voice_client.py

# SPDX-License-Identifier: MIT

"""
Some documentation to refer to:

- Our main web socket (mWS) sends opcode 4 with a guild ID and channel ID.
- The mWS receives VOICE_STATE_UPDATE and VOICE_SERVER_UPDATE.
- We pull the session_id from VOICE_STATE_UPDATE.
- We pull the token, endpoint and server_id from VOICE_SERVER_UPDATE.
- Then we initiate the voice web socket (vWS) pointing to the endpoint.
- We send opcode 0 with the user_id, server_id, session_id and token using the vWS.
- The vWS sends back opcode 2 with an ssrc, port, modes(array) and hearbeat_interval.
- We send a UDP discovery packet to endpoint:port and receive our IP and our port in LE.
- Then we send our IP and port via vWS with opcode 1.
- When that's all done, we receive opcode 4 from the vWS.
- Finally we can transmit data to endpoint:port.
"""
from __future__ import annotations

import asyncio
import logging
import socket
import struct
import threading
from typing import TYPE_CHECKING, Any, Callable, List, Optional, Tuple

from . import opus, utils
from .backoff import ExponentialBackoff
from .errors import ClientException, ConnectionClosed
from .gateway import DiscordVoiceWebSocket
from .player import AudioPlayer, AudioSource
from .utils import MISSING

if TYPE_CHECKING:
    from . import abc
    from .client import Client
    from .guild import Guild
    from .opus import Encoder
    from .state import ConnectionState
    from .types.gateway import VoiceServerUpdateEvent
    from .types.voice import GuildVoiceState as GuildVoiceStatePayload, SupportedModes
    from .user import ClientUser


has_nacl: bool

try:
    import nacl.secret
    import nacl.utils

    has_nacl = True
except ImportError:
    has_nacl = False

__all__ = (
    "VoiceProtocol",
    "VoiceClient",
)


_log = logging.getLogger(__name__)


class VoiceProtocol:
    """A class that represents the Discord voice protocol.

    This is an abstract class. The library provides a concrete implementation
    under :class:`VoiceClient`.

    This class allows you to implement a protocol to allow for an external
    method of sending voice, such as Lavalink_ or a native library implementation.

    These classes are passed to :meth:`abc.Connectable.connect <VoiceChannel.connect>`.

    .. _Lavalink: https://github.com/freyacodes/Lavalink

    Parameters
    ----------
    client: :class:`Client`
        The client (or its subclasses) that started the connection request.
    channel: :class:`abc.Connectable`
        The voice channel that is being connected to.
    """

    def __init__(self, client: Client, channel: abc.Connectable) -> None:
        self.client: Client = client
        self.channel: abc.Connectable = channel

    async def on_voice_state_update(self, data: GuildVoiceStatePayload) -> None:
        """|coro|

        An abstract method that is called when the client's voice state
        has changed. This corresponds to ``VOICE_STATE_UPDATE``.

        Parameters
        ----------
        data: :class:`dict`
            The raw `voice state payload`__.

            .. _voice_state_update_payload: https://discord.com/developers/docs/resources/voice#voice-state-object

            __ voice_state_update_payload_
        """
        raise NotImplementedError

    async def on_voice_server_update(self, data: VoiceServerUpdateEvent) -> None:
        """|coro|

        An abstract method that is called when initially connecting to voice.
        This corresponds to ``VOICE_SERVER_UPDATE``.

        Parameters
        ----------
        data: :class:`dict`
            The raw `voice server update payload`__.

            .. _voice_server_update_payload: https://discord.com/developers/docs/topics/gateway#voice-server-update-voice-server-update-event-fields

            __ voice_server_update_payload_
        """
        raise NotImplementedError

    async def connect(self, *, timeout: float, reconnect: bool) -> None:
        """|coro|

        An abstract method called when the client initiates the connection request.

        When a connection is requested initially, the library calls the constructor
        under ``__init__`` and then calls :meth:`connect`. If :meth:`connect` fails at
        some point then :meth:`disconnect` is called.

        Within this method, to start the voice connection flow it is recommended to
        use :meth:`Guild.change_voice_state` to start the flow. After which,
        :meth:`on_voice_server_update` and :meth:`on_voice_state_update` will be called.
        The order that these two are called is unspecified.

        Parameters
        ----------
        timeout: :class:`float`
            The timeout for the connection.
        reconnect: :class:`bool`
            Whether reconnection is expected.
        """
        raise NotImplementedError

    async def disconnect(self, *, force: bool) -> None:
        """|coro|

        An abstract method called when the client terminates the connection.

        See :meth:`cleanup`.

        Parameters
        ----------
        force: :class:`bool`
            Whether the disconnection was forced.
        """
        raise NotImplementedError

    def cleanup(self) -> None:
        """This method *must* be called to ensure proper clean-up during a disconnect.

        It is advisable to call this from within :meth:`disconnect` when you are
        completely done with the voice protocol instance.

        This method removes it from the internal state cache that keeps track of
        currently alive voice clients. Failure to clean-up will cause subsequent
        connections to report that it's still connected.
        """
        key_id, _ = self.channel._get_voice_client_key()
        self.client._connection._remove_voice_client(key_id)


class VoiceClient(VoiceProtocol):
    """Represents a Discord voice connection.

    You do not create these, you typically get them from
    e.g. :meth:`VoiceChannel.connect`.

    Warning
    --------
    In order to use PCM based AudioSources, you must have the opus library
    installed on your system and loaded through :func:`opus.load_opus`.
    Otherwise, your AudioSources must be opus encoded (e.g. using :class:`FFmpegOpusAudio`)
    or the library will not be able to transmit audio.

    Attributes
    ----------
    session_id: :class:`str`
        The voice connection session ID.
    token: :class:`str`
        The voice connection token.
    endpoint: :class:`str`
        The endpoint we are connecting to.
    channel: :class:`abc.Connectable`
        The voice channel connected to.
    loop: :class:`asyncio.AbstractEventLoop`
        The event loop that the voice client is running on.
    """

    endpoint_ip: str
    voice_port: int
    secret_key: List[int]
    ssrc: int
    ip: str
    port: int

    def __init__(self, client: Client, channel: abc.Connectable):
        if not has_nacl:
            raise RuntimeError("PyNaCl library needed in order to use voice")

        super().__init__(client, channel)
        state = client._connection
        self.token: str = MISSING
        self.socket: socket.socket = MISSING
        self.loop: asyncio.AbstractEventLoop = state.loop
        self._state: ConnectionState = state
        # this will be used in the AudioPlayer thread
        self._connected: threading.Event = threading.Event()

        self._handshaking: bool = False
        self._potentially_reconnecting: bool = False
        self._voice_state_complete: asyncio.Event = asyncio.Event()
        self._voice_server_complete: asyncio.Event = asyncio.Event()

        self.mode: str = MISSING
        self._connections: int = 0
        self.sequence: int = 0
        self.timestamp: int = 0
        self.timeout: float = 0
        self._runner: asyncio.Task = MISSING
        self._player: Optional[AudioPlayer] = None
        self.encoder: Encoder = MISSING
        self._lite_nonce: int = 0
        self.ws: DiscordVoiceWebSocket = MISSING

    warn_nacl = not has_nacl
    supported_modes: Tuple[SupportedModes, ...] = (
        "xsalsa20_poly1305_lite",
        "xsalsa20_poly1305_suffix",
        "xsalsa20_poly1305",
    )

    @property
    def guild(self) -> Guild:
        """:class:`Guild`: The guild we're connected to."""
        return self.channel.guild

    @property
    def user(self) -> ClientUser:
        """:class:`ClientUser`: The user connected to voice (i.e. ourselves)."""
        return self._state.user

    def checked_add(self, attr, value, limit):
        val = getattr(self, attr)
        if val + value > limit:
            setattr(self, attr, 0)
        else:
            setattr(self, attr, val + value)

    # connection related

    async def on_voice_state_update(self, data: GuildVoiceStatePayload) -> None:
        self.session_id = data["session_id"]
        channel_id = data["channel_id"]

        if not self._handshaking or self._potentially_reconnecting:
            # If we're done handshaking then we just need to update ourselves
            # If we're potentially reconnecting due to a 4014, then we need to differentiate
            # a channel move and an actual force disconnect
            if channel_id is None:
                # We're being disconnected so cleanup
                await self.disconnect()
            else:
                guild = self.guild
                self.channel = channel_id and guild and guild.get_channel(int(channel_id))  # type: ignore
        else:
            self._voice_state_complete.set()

    async def on_voice_server_update(self, data: VoiceServerUpdateEvent) -> None:
        if self._voice_server_complete.is_set():
            _log.info("Ignoring extraneous voice server update.")
            return

        self.token = data["token"]
        self.server_id = int(data["guild_id"])
        endpoint = data.get("endpoint")

        if endpoint is None or self.token is None:
            _log.warning(
                "Awaiting endpoint... This requires waiting. "
                "If timeout occurred considering raising the timeout and reconnecting."
            )
            return

        self.endpoint, _, _ = endpoint.rpartition(":")
        if self.endpoint.startswith("wss://"):
            # Just in case, strip it off since we're going to add it later
            self.endpoint = self.endpoint[6:]

        # This gets set later
        self.endpoint_ip = MISSING

        if self.socket:
            self.socket.close()
        self.socket = socket.socket(socket.AF_INET, socket.SOCK_DGRAM)
        self.socket.setblocking(False)

        if not self._handshaking:
            # If we're not handshaking then we need to terminate our previous connection in the websocket
            if self.ws:
                await self.ws.close(4000)
            return

        self._voice_server_complete.set()

    async def voice_connect(self) -> None:
        await self.channel.guild.change_voice_state(channel=self.channel)

    async def voice_disconnect(self) -> None:
        _log.info(
            "The voice handshake is being terminated for Channel ID %s (Guild ID %s)",
            self.channel.id,
            self.guild.id,
        )
        await self.channel.guild.change_voice_state(channel=None)

    def prepare_handshake(self) -> None:
        self._voice_state_complete.clear()
        self._voice_server_complete.clear()
        self._handshaking = True
        _log.info("Starting voice handshake... (connection attempt %d)", self._connections + 1)
        self._connections += 1

    def finish_handshake(self) -> None:
        _log.info("Voice handshake complete. Endpoint found %s", self.endpoint)
        self._handshaking = False
        self._voice_server_complete.clear()
        self._voice_state_complete.clear()

    async def connect_websocket(self) -> DiscordVoiceWebSocket:
        ws = await DiscordVoiceWebSocket.from_client(self)
        self._connected.clear()
        while ws.secret_key is None:
            await ws.poll_event()
        self._connected.set()
        return ws

    async def connect(self, *, reconnect: bool, timeout: float) -> None:
        _log.info("Connecting to voice...")
        self.timeout = timeout

        for i in range(5):
            self.prepare_handshake()

            # This has to be created before we start the flow.
            futures = [
                self._voice_state_complete.wait(),
                self._voice_server_complete.wait(),
            ]

            # Start the connection flow
            await self.voice_connect()

            try:
                await utils.sane_wait_for(futures, timeout=timeout)
            except asyncio.TimeoutError:
                await self.disconnect(force=True)
                raise

            self.finish_handshake()

            try:
                self.ws = await self.connect_websocket()
                break
            except (ConnectionClosed, asyncio.TimeoutError):
                if reconnect:
                    _log.exception("Failed to connect to voice... Retrying...")
                    await asyncio.sleep(1 + i * 2.0)
                    await self.voice_disconnect()
                    continue
                else:
                    raise

        if self._runner is MISSING:
            self._runner = self.loop.create_task(self.poll_voice_ws(reconnect))

    async def potential_reconnect(self) -> bool:
        # Attempt to stop the player thread from playing early
        self._connected.clear()
        self.prepare_handshake()
        self._potentially_reconnecting = True
        try:
            # We only care about VOICE_SERVER_UPDATE since VOICE_STATE_UPDATE can come before we get disconnected
            await asyncio.wait_for(self._voice_server_complete.wait(), timeout=self.timeout)
        except asyncio.TimeoutError:
            self._potentially_reconnecting = False
            await self.disconnect(force=True)
            return False

        self.finish_handshake()
        self._potentially_reconnecting = False
        try:
            self.ws = await self.connect_websocket()
        except (ConnectionClosed, asyncio.TimeoutError):
            return False
        else:
            return True

    @property
    def latency(self) -> float:
        """:class:`float`: Latency between a HEARTBEAT and a HEARTBEAT_ACK in seconds.

        This could be referred to as the Discord Voice WebSocket latency and is
        an analogue of user's voice latencies as seen in the Discord client.

        .. versionadded:: 1.4
        """
        ws = self.ws
        return float("inf") if not ws else ws.latency

    @property
    def average_latency(self) -> float:
        """:class:`float`: Average of most recent 20 HEARTBEAT latencies in seconds.

        .. versionadded:: 1.4
        """
        ws = self.ws
        return float("inf") if not ws else ws.average_latency

    async def poll_voice_ws(self, reconnect: bool) -> None:
        backoff = ExponentialBackoff()
        while True:
            try:
                await self.ws.poll_event()
            except (ConnectionClosed, asyncio.TimeoutError) as exc:
                if isinstance(exc, ConnectionClosed):
                    # The following close codes are undocumented so I will document them here.
                    # 1000 - normal closure (obviously)
                    # 4014 - voice channel has been deleted.
                    # 4015 - voice server has crashed
                    if exc.code in (1000, 4015):
                        _log.info("Disconnecting from voice normally, close code %d.", exc.code)
                        await self.disconnect()
                        break
                    if exc.code == 4014:
                        _log.info("Disconnected from voice by force... potentially reconnecting.")
                        successful = await self.potential_reconnect()
                        if not successful:
                            _log.info(
                                "Reconnect was unsuccessful, disconnecting from voice normally..."
                            )
                            await self.disconnect()
                            break
                        else:
                            continue

                if not reconnect:
                    await self.disconnect()
                    raise

                retry = backoff.delay()
                _log.exception("Disconnected from voice... Reconnecting in %.2fs.", retry)
                self._connected.clear()
                await asyncio.sleep(retry)
                await self.voice_disconnect()
                try:
                    await self.connect(reconnect=True, timeout=self.timeout)
                except asyncio.TimeoutError:
                    # at this point we've retried 5 times... let's continue the loop.
                    _log.warning("Could not connect to voice... Retrying...")
                    continue

    async def disconnect(self, *, force: bool = False) -> None:
        """|coro|

        Disconnects this voice client from voice.
        """
        if not force and not self.is_connected():
            return

        self.stop()
        self._connected.clear()

        try:
            if self.ws:
                await self.ws.close()

            await self.voice_disconnect()
        finally:
            self.cleanup()
            if self.socket:
                self.socket.close()

    async def move_to(self, channel: abc.Snowflake) -> None:
        """|coro|

        Moves you to a different voice channel.

        Parameters
        ----------
        channel: :class:`abc.Snowflake`
            The channel to move to. Must be a voice channel.
        """
        await self.channel.guild.change_voice_state(channel=channel)

    def is_connected(self) -> bool:
        """Indicates if the voice client is connected to voice."""
        return self._connected.is_set()

    # audio related

    def _get_voice_packet(self, data):
        header = bytearray(12)

        # Formulate rtp header
        header[0] = 0x80
        header[1] = 0x78
        struct.pack_into(">H", header, 2, self.sequence)
        struct.pack_into(">I", header, 4, self.timestamp)
        struct.pack_into(">I", header, 8, self.ssrc)

        encrypt_packet = getattr(self, f"_encrypt_{self.mode}")
        return encrypt_packet(header, data)

    def _encrypt_xsalsa20_poly1305(self, header: bytes, data) -> bytes:
        box = nacl.secret.SecretBox(bytes(self.secret_key))
        nonce = bytearray(24)
        nonce[:12] = header

        return header + box.encrypt(bytes(data), bytes(nonce)).ciphertext

    def _encrypt_xsalsa20_poly1305_suffix(self, header: bytes, data) -> bytes:
        box = nacl.secret.SecretBox(bytes(self.secret_key))
        nonce = nacl.utils.random(nacl.secret.SecretBox.NONCE_SIZE)

        return header + box.encrypt(bytes(data), nonce).ciphertext + nonce

    def _encrypt_xsalsa20_poly1305_lite(self, header: bytes, data) -> bytes:
        box = nacl.secret.SecretBox(bytes(self.secret_key))
        nonce = bytearray(24)

        nonce[:4] = struct.pack(">I", self._lite_nonce)
        self.checked_add("_lite_nonce", 1, 4294967295)

        return header + box.encrypt(bytes(data), bytes(nonce)).ciphertext + nonce[:4]

    def play(
        self, source: AudioSource, *, after: Optional[Callable[[Optional[Exception]], Any]] = None
    ) -> None:
        """Plays an :class:`AudioSource`.

        The finalizer, ``after`` is called after the source has been exhausted
        or an error occurred.

        If an error happens while the audio player is running, the exception is
        caught and the audio player is then stopped.  If no after callback is
        passed, any caught exception will be displayed as if it were raised.

        Parameters
        ----------
        source: :class:`AudioSource`
            The audio source we're reading from.
        after: Callable[[Optional[:class:`Exception`]], Any]
            The finalizer that is called after the stream is exhausted.
            This function must have a single parameter, ``error``, that
            denotes an optional exception that was raised during playing.

        Raises
        ------
        ClientException
            Already playing audio or not connected.
        TypeError
            Source is not a :class:`AudioSource` or after is not a callable.
        OpusNotLoaded
            Source is not opus encoded and opus is not loaded.
        """

        if not self.is_connected():
            raise ClientException("Not connected to voice.")

        if self.is_playing():
            raise ClientException("Already playing audio.")

        if not isinstance(source, AudioSource):
            raise TypeError(f"source must be an AudioSource not {source.__class__.__name__}")

        if not self.encoder and not source.is_opus():
            self.encoder = opus.Encoder()

        self._player = AudioPlayer(source, self, after=after)
        self._player.start()

    def is_playing(self) -> bool:
        """Indicates if we're currently playing audio."""
        return self._player is not None and self._player.is_playing()

    def is_paused(self) -> bool:
        """Indicates if we're playing audio, but if we're paused."""
        return self._player is not None and self._player.is_paused()

    def stop(self) -> None:
        """Stops playing audio."""
        if self._player:
            self._player.stop()
            self._player = None

    def pause(self) -> None:
        """Pauses the audio playing."""
        if self._player:
            self._player.pause()

    def resume(self) -> None:
        """Resumes the audio playing."""
        if self._player:
            self._player.resume()

    @property
    def source(self) -> Optional[AudioSource]:
        """Optional[:class:`AudioSource`]: The audio source being played, if playing.

        This property can also be used to change the audio source currently being played.
        """
        return self._player.source if self._player else None

    @source.setter
    def source(self, value: AudioSource) -> None:
        if not isinstance(value, AudioSource):
            raise TypeError(f"expected AudioSource not {value.__class__.__name__}.")

        if self._player is None:
            raise ValueError("Not playing anything.")

        self._player._set_source(value)

    def send_audio_packet(self, data: bytes, *, encode: bool = True) -> None:
        """Sends an audio packet composed of the data.

        You must be connected to play audio.

        Parameters
        ----------
        data: :class:`bytes`
            The :term:`py:bytes-like object` denoting PCM or Opus voice data.
        encode: :class:`bool`
            Indicates if ``data`` should be encoded into Opus.

        Raises
        ------
        ClientException
            You are not connected.
        opus.OpusError
            Encoding the data failed.
        """

        self.checked_add("sequence", 1, 65535)
        if encode:
            encoded_data = self.encoder.encode(data, self.encoder.SAMPLES_PER_FRAME)
        else:
            encoded_data = data
        packet = self._get_voice_packet(encoded_data)
        try:
            self.socket.sendto(packet, (self.endpoint_ip, self.voice_port))
        except BlockingIOError:
            _log.warning(
                "A packet has been dropped (seq: %s, timestamp: %s)", self.sequence, self.timestamp
            )

        self.checked_add("timestamp", opus.Encoder.SAMPLES_PER_FRAME, 4294967295)