Telethon/telethon/client/downloads.py

import datetime
import io
import os
import pathlib
import typing
import inspect
import asyncio

from ..crypto import AES

from .. import utils, helpers, errors, hints
from ..requestiter import RequestIter
from ..tl import TLObject, types, functions

try:
    import aiohttp
except ImportError:
    aiohttp = None

if typing.TYPE_CHECKING:
    from .telegramclient import TelegramClient

# Chunk sizes for upload.getFile must be multiples of the smallest size
MIN_CHUNK_SIZE = 4096
MAX_CHUNK_SIZE = 512 * 1024

# 2021-01-15, users reported that `errors.TimeoutError` can occur while downloading files.
TIMED_OUT_SLEEP = 1

class _DirectDownloadIter(RequestIter):
    async def _init(
            self, file, dc_id, offset, stride, chunk_size, request_size, file_size, msg_data
    ):
        self.request = functions.upload.GetFileRequest(
            file, offset=offset, limit=request_size)

        self.total = file_size
        self._stride = stride
        self._chunk_size = chunk_size
        self._last_part = None
        self._msg_data = msg_data
        self._timed_out = False

        self._exported = dc_id and self.client.session.dc_id != dc_id
        if not self._exported:
            # The used sender will also change if ``FileMigrateError`` occurs
            self._sender = self.client._sender
        else:
            try:
                self._sender = await self.client._borrow_exported_sender(dc_id)
            except errors.DcIdInvalidError:
                # Can't export a sender for the ID we are currently in
                config = await self.client(functions.help.GetConfigRequest())
                for option in config.dc_options:
                    if option.ip_address == self.client.session.server_address:
                        self.client.session.set_dc(
                            option.id, option.ip_address, option.port)
                        self.client.session.save()
                        break

                # TODO Figure out why the session may have the wrong DC ID
                self._sender = self.client._sender
                self._exported = False

    async def _load_next_chunk(self):
        cur = await self._request()
        self.buffer.append(cur)
        if len(cur) < self.request.limit:
            self.left = len(self.buffer)
            await self.close()
        else:
            self.request.offset += self._stride

    async def _request(self):
        try:
            result = await self.client._call(self._sender, self.request)
            self._timed_out = False
            if isinstance(result, types.upload.FileCdnRedirect):
                raise NotImplementedError  # TODO Implement
            else:
                return result.bytes

        except errors.TimeoutError as e:
            if self._timed_out:
                self.client._log[__name__].warning('Got two timeouts in a row while downloading file')
                raise

            self._timed_out = True
            self.client._log[__name__].info('Got timeout while downloading file, retrying once')
            await asyncio.sleep(TIMED_OUT_SLEEP)
            return await self._request()

        except errors.FileMigrateError as e:
            self.client._log[__name__].info('File lives in another DC')
            self._sender = await self.client._borrow_exported_sender(e.new_dc)
            self._exported = True
            return await self._request()

        except errors.FilerefUpgradeNeededError as e:
            # Only implemented for documents which are the ones that may take that long to download
            if not self._msg_data \
                    or not isinstance(self.request.location, types.InputDocumentFileLocation) \
                    or self.request.location.thumb_size != '':
                raise

            self.client._log[__name__].info('File ref expired during download; refetching message')
            chat, msg_id = self._msg_data
            msg = await self.client.get_messages(chat, ids=msg_id)

            if not isinstance(msg.media, types.MessageMediaDocument):
                raise

            document = msg.media.document

            # Message media may have been edited for something else
            if document.id != self.request.location.id:
                raise

            self.request.location.file_reference = document.file_reference
            return await self._request()

    async def close(self):
        if not self._sender:
            return

        try:
            if self._exported:
                await self.client._return_exported_sender(self._sender)
            elif self._sender != self.client._sender:
                await self._sender.disconnect()
        finally:
            self._sender = None

    async def __aenter__(self):
        return self

    async def __aexit__(self, *args):
        await self.close()

    __enter__ = helpers._sync_enter
    __exit__ = helpers._sync_exit


class _GenericDownloadIter(_DirectDownloadIter):
    async def _load_next_chunk(self):
        # 1. Fetch enough for one chunk
        data = b''

        # 1.1. ``bad`` is how much into the data we have we need to offset
        bad = self.request.offset % self.request.limit
        before = self.request.offset

        # 1.2. We have to fetch from a valid offset, so remove that bad part
        self.request.offset -= bad

        done = False
        while not done and len(data) - bad < self._chunk_size:
            cur = await self._request()
            self.request.offset += self.request.limit

            data += cur
            done = len(cur) < self.request.limit

        # 1.3 Restore our last desired offset
        self.request.offset = before

        # 2. Fill the buffer with the data we have
        # 2.1. Slicing `bytes` is expensive, yield `memoryview` instead
        mem = memoryview(data)

        # 2.2. The current chunk starts at ``bad`` offset into the data,
        #      and each new chunk is ``stride`` bytes apart of the other
        for i in range(bad, len(data), self._stride):
            self.buffer.append(mem[i:i + self._chunk_size])

            # 2.3. We will yield this offset, so move to the next one
            self.request.offset += self._stride

        # 2.4. If we are in the last chunk, we will return the last partial data
        if done:
            self.left = len(self.buffer)
            await self.close()
            return

        # 2.5. If we are not done, we can't return incomplete chunks.
        if len(self.buffer[-1]) != self._chunk_size:
            self._last_part = self.buffer.pop().tobytes()

            # 3. Be careful with the offsets. Re-fetching a bit of data
            #    is fine, since it greatly simplifies things.
            # TODO Try to not re-fetch data
            self.request.offset -= self._stride


class DownloadMethods:

    # region Public methods

    async def download_profile_photo(
            self: 'TelegramClient',
            entity: 'hints.EntityLike',
            file: 'hints.FileLike' = None,
            *,
            download_big: bool = True) -> typing.Optional[str]:
        """
        Downloads the profile photo from the given user, chat or channel.

        Arguments
            entity (`entity`):
                From who the photo will be downloaded.

                .. note::

                    This method expects the full entity (which has the data
                    to download the photo), not an input variant.

                    It's possible that sometimes you can't fetch the entity
                    from its input (since you can get errors like
                    ``ChannelPrivateError``) but you already have it through
                    another call, like getting a forwarded message from it.

            file (`str` | `file`, optional):
                The output file path, directory, or stream-like object.
                If the path exists and is a file, it will be overwritten.
                If file is the type `bytes`, it will be downloaded in-memory
                as a bytestring (e.g. ``file=bytes``).

            download_big (`bool`, optional):
                Whether to use the big version of the available photos.

        Returns
            `None` if no photo was provided, or if it was Empty. On success
            the file path is returned since it may differ from the one given.

        Example
            .. code-block:: python

                # Download your own profile photo
                path = await client.download_profile_photo('me')
                print(path)
        """
        # hex(crc32(x.encode('ascii'))) for x in
        # ('User', 'Chat', 'UserFull', 'ChatFull')
        ENTITIES = (0x2da17977, 0xc5af5d94, 0x1f4661b9, 0xd49a2697)
        # ('InputPeer', 'InputUser', 'InputChannel')
        INPUTS = (0xc91c90b6, 0xe669bf46, 0x40f202fd)
        if not isinstance(entity, TLObject) or entity.SUBCLASS_OF_ID in INPUTS:
            entity = await self.get_entity(entity)

        thumb = -1 if download_big else 0

        possible_names = []
        if entity.SUBCLASS_OF_ID not in ENTITIES:
            photo = entity
        else:
            if not hasattr(entity, 'photo'):
                # Special case: may be a ChatFull with photo:Photo
                # This is different from a normal UserProfilePhoto and Chat
                if not hasattr(entity, 'chat_photo'):
                    return None

                return await self._download_photo(
                    entity.chat_photo, file, date=None,
                    thumb=thumb, progress_callback=None
                )

            for attr in ('username', 'first_name', 'title'):
                possible_names.append(getattr(entity, attr, None))

            photo = entity.photo

        if isinstance(photo, (types.UserProfilePhoto, types.ChatPhoto)):
            dc_id = photo.dc_id
            loc = types.InputPeerPhotoFileLocation(
                # min users can be used to download profile photos
                # self.get_input_entity would otherwise not accept those
                peer=utils.get_input_peer(entity, check_hash=False),
                photo_id=photo.photo_id,
                big=download_big
            )
        else:
            # It doesn't make any sense to check if `photo` can be used
            # as input location, because then this method would be able
            # to "download the profile photo of a message", i.e. its
            # media which should be done with `download_media` instead.
            return None

        file = self._get_proper_filename(
            file, 'profile_photo', '.jpg',
            possible_names=possible_names
        )

        try:
            result = await self.download_file(loc, file, dc_id=dc_id)
            return result if file is bytes else file
        except errors.LocationInvalidError:
            # See issue #500, Android app fails as of v4.6.0 (1155).
            # The fix seems to be using the full channel chat photo.
            ie = await self.get_input_entity(entity)
            ty = helpers._entity_type(ie)
            if ty == helpers._EntityType.CHANNEL:
                full = await self(functions.channels.GetFullChannelRequest(ie))
                return await self._download_photo(
                    full.full_chat.chat_photo, file,
                    date=None, progress_callback=None,
                    thumb=thumb
                )
            else:
                # Until there's a report for chats, no need to.
                return None

    async def download_media(
            self: 'TelegramClient',
            message: 'hints.MessageLike',
            file: 'hints.FileLike' = None,
            *,
            thumb: 'typing.Union[int, types.TypePhotoSize]' = None,
            progress_callback: 'hints.ProgressCallback' = None) -> typing.Optional[typing.Union[str, bytes]]:
        """
        Downloads the given media from a message object.

        Note that if the download is too slow, you should consider installing
        ``cryptg`` (through ``pip install cryptg``) so that decrypting the
        received data is done in C instead of Python (much faster).

        See also `Message.download_media() <telethon.tl.custom.message.Message.download_media>`.

        Arguments
            message (`Message <telethon.tl.custom.message.Message>` | :tl:`Media`):
                The media or message containing the media that will be downloaded.

            file (`str` | `file`, optional):
                The output file path, directory, or stream-like object.
                If the path exists and is a file, it will be overwritten.
                If file is the type `bytes`, it will be downloaded in-memory
                as a bytestring (e.g. ``file=bytes``).

            progress_callback (`callable`, optional):
                A callback function accepting two parameters:
                ``(received bytes, total)``.

            thumb (`int` | :tl:`PhotoSize`, optional):
                Which thumbnail size from the document or photo to download,
                instead of downloading the document or photo itself.

                If it's specified but the file does not have a thumbnail,
                this method will return `None`.

                The parameter should be an integer index between ``0`` and
                ``len(sizes)``. ``0`` will download the smallest thumbnail,
                and ``len(sizes) - 1`` will download the largest thumbnail.
                You can also use negative indices, which work the same as
                they do in Python's `list`.

                You can also pass the :tl:`PhotoSize` instance to use.
                Alternatively, the thumb size type `str` may be used.

                In short, use ``thumb=0`` if you want the smallest thumbnail
                and ``thumb=-1`` if you want the largest thumbnail.

                .. note::
                    The largest thumbnail may be a video instead of a photo,
                    as they are available since layer 116 and are bigger than
                    any of the photos.

        Returns
            `None` if no media was provided, or if it was Empty. On success
            the file path is returned since it may differ from the one given.

        Example
            .. code-block:: python

                path = await client.download_media(message)
                await client.download_media(message, filename)
                # or
                path = await message.download_media()
                await message.download_media(filename)

                # Printing download progress
                def callback(current, total):
                    print('Downloaded', current, 'out of', total,
                          'bytes: {:.2%}'.format(current / total))

                await client.download_media(message, progress_callback=callback)
        """
        # Downloading large documents may be slow enough to require a new file reference
        # to be obtained mid-download. Store (input chat, message id) so that the message
        # can be re-fetched.
        msg_data = None

        # TODO This won't work for messageService
        if isinstance(message, types.Message):
            date = message.date
            media = message.media
            msg_data = (message.input_chat, message.id) if message.input_chat else None
        else:
            date = datetime.datetime.now()
            media = message

        if isinstance(media, str):
            media = utils.resolve_bot_file_id(media)

        if isinstance(media, types.MessageService):
            if isinstance(message.action,
                          types.MessageActionChatEditPhoto):
                media = media.photo

        if isinstance(media, types.MessageMediaWebPage):
            if isinstance(media.webpage, types.WebPage):
                media = media.webpage.document or media.webpage.photo

        if isinstance(media, (types.MessageMediaPhoto, types.Photo)):
            return await self._download_photo(
                media, file, date, thumb, progress_callback
            )
        elif isinstance(media, (types.MessageMediaDocument, types.Document)):
            return await self._download_document(
                media, file, date, thumb, progress_callback, msg_data
            )
        elif isinstance(media, types.MessageMediaContact) and thumb is None:
            return self._download_contact(
                media, file
            )
        elif isinstance(media, (types.WebDocument, types.WebDocumentNoProxy)) and thumb is None:
            return await self._download_web_document(
                media, file, progress_callback
            )

    async def download_file(
            self: 'TelegramClient',
            input_location: 'hints.FileLike',
            file: 'hints.OutFileLike' = None,
            *,
            part_size_kb: float = None,
            file_size: int = None,
            progress_callback: 'hints.ProgressCallback' = None,
            dc_id: int = None,
            key: bytes = None,
            iv: bytes = None) -> typing.Optional[bytes]:
        """
        Low-level method to download files from their input location.

        .. note::

            Generally, you should instead use `download_media`.
            This method is intended to be a bit more low-level.

        Arguments
            input_location (:tl:`InputFileLocation`):
                The file location from which the file will be downloaded.
                See `telethon.utils.get_input_location` source for a complete
                list of supported types.

            file (`str` | `file`, optional):
                The output file path, directory, or stream-like object.
                If the path exists and is a file, it will be overwritten.

                If the file path is `None` or `bytes`, then the result
                will be saved in memory and returned as `bytes`.

            part_size_kb (`int`, optional):
                Chunk size when downloading files. The larger, the less
                requests will be made (up to 512KB maximum).

            file_size (`int`, optional):
                The file size that is about to be downloaded, if known.
                Only used if ``progress_callback`` is specified.

            progress_callback (`callable`, optional):
                A callback function accepting two parameters:
                ``(downloaded bytes, total)``. Note that the
                ``total`` is the provided ``file_size``.

            dc_id (`int`, optional):
                The data center the library should connect to in order
                to download the file. You shouldn't worry about this.

            key ('bytes', optional):
                In case of an encrypted upload (secret chats) a key is supplied

            iv ('bytes', optional):
                In case of an encrypted upload (secret chats) an iv is supplied


        Example
            .. code-block:: python

                # Download a file and print its header
                data = await client.download_file(input_file, bytes)
                print(data[:16])
        """
        return await self._download_file(
            input_location,
            file,
            part_size_kb=part_size_kb,
            file_size=file_size,
            progress_callback=progress_callback,
            dc_id=dc_id,
            key=key,
            iv=iv,
        )

    async def _download_file(
            self: 'TelegramClient',
            input_location: 'hints.FileLike',
            file: 'hints.OutFileLike' = None,
            *,
            part_size_kb: float = None,
            file_size: int = None,
            progress_callback: 'hints.ProgressCallback' = None,
            dc_id: int = None,
            key: bytes = None,
            iv: bytes = None,
            msg_data: tuple = None) -> typing.Optional[bytes]:
        if not part_size_kb:
            if not file_size:
                part_size_kb = 64  # Reasonable default
            else:
                part_size_kb = utils.get_appropriated_part_size(file_size)

        part_size = int(part_size_kb * 1024)
        if part_size % MIN_CHUNK_SIZE != 0:
            raise ValueError(
                'The part size must be evenly divisible by 4096.')

        if isinstance(file, pathlib.Path):
            file = str(file.absolute())

        in_memory = file is None or file is bytes
        if in_memory:
            f = io.BytesIO()
        elif isinstance(file, str):
            # Ensure that we'll be able to download the media
            helpers.ensure_parent_dir_exists(file)
            f = open(file, 'wb')
        else:
            f = file

        try:
            async for chunk in self._iter_download(
                    input_location, request_size=part_size, dc_id=dc_id, msg_data=msg_data):
                if iv and key:
                    chunk = AES.decrypt_ige(chunk, key, iv)
                r = f.write(chunk)
                if inspect.isawaitable(r):
                    await r

                if progress_callback:
                    r = progress_callback(f.tell(), file_size)
                    if inspect.isawaitable(r):
                        await r

            # Not all IO objects have flush (see #1227)
            if callable(getattr(f, 'flush', None)):
                f.flush()

            if in_memory:
                return f.getvalue()
        finally:
            if isinstance(file, str) or in_memory:
                f.close()

    def iter_download(
            self: 'TelegramClient',
            file: 'hints.FileLike',
            *,
            offset: int = 0,
            stride: int = None,
            limit: int = None,
            chunk_size: int = None,
            request_size: int = MAX_CHUNK_SIZE,
            file_size: int = None,
            dc_id: int = None
    ):
        """
        Iterates over a file download, yielding chunks of the file.

        This method can be used to stream files in a more convenient
        way, since it offers more control (pausing, resuming, etc.)

        .. note::

            Using a value for `offset` or `stride` which is not a multiple
            of the minimum allowed `request_size`, or if `chunk_size` is
            different from `request_size`, the library will need to do a
            bit more work to fetch the data in the way you intend it to.

            You normally shouldn't worry about this.

        Arguments
            file (`hints.FileLike`):
                The file of which contents you want to iterate over.

            offset (`int`, optional):
                The offset in bytes into the file from where the
                download should start. For example, if a file is
                1024KB long and you just want the last 512KB, you
                would use ``offset=512 * 1024``.

            stride (`int`, optional):
                The stride of each chunk (how much the offset should
                advance between reading each chunk). This parameter
                should only be used for more advanced use cases.

                It must be bigger than or equal to the `chunk_size`.

            limit (`int`, optional):
                The limit for how many *chunks* will be yielded at most.

            chunk_size (`int`, optional):
                The maximum size of the chunks that will be yielded.
                Note that the last chunk may be less than this value.
                By default, it equals to `request_size`.

            request_size (`int`, optional):
                How many bytes will be requested to Telegram when more
                data is required. By default, as many bytes as possible
                are requested. If you would like to request data in
                smaller sizes, adjust this parameter.

                Note that values outside the valid range will be clamped,
                and the final value will also be a multiple of the minimum
                allowed size.

            file_size (`int`, optional):
                If the file size is known beforehand, you should set
                this parameter to said value. Depending on the type of
                the input file passed, this may be set automatically.

            dc_id (`int`, optional):
                The data center the library should connect to in order
                to download the file. You shouldn't worry about this.

        Yields

            `bytes` objects representing the chunks of the file if the
            right conditions are met, or `memoryview` objects instead.

        Example
            .. code-block:: python

                # Streaming `media` to an output file
                # After the iteration ends, the sender is cleaned up
                with open('photo.jpg', 'wb') as fd:
                    async for chunk in client.iter_download(media):
                        fd.write(chunk)

                # Fetching only the header of a file (32 bytes)
                # You should manually close the iterator in this case.
                #
                # "stream" is a common name for asynchronous generators,
                # and iter_download will yield `bytes` (chunks of the file).
                stream = client.iter_download(media, request_size=32)
                header = await stream.__anext__()  # "manual" version of `async for`
                await stream.close()
                assert len(header) == 32
        """
        return self._iter_download(
            file,
            offset=offset,
            stride=stride,
            limit=limit,
            chunk_size=chunk_size,
            request_size=request_size,
            file_size=file_size,
            dc_id=dc_id,
        )

    def _iter_download(
            self: 'TelegramClient',
            file: 'hints.FileLike',
            *,
            offset: int = 0,
            stride: int = None,
            limit: int = None,
            chunk_size: int = None,
            request_size: int = MAX_CHUNK_SIZE,
            file_size: int = None,
            dc_id: int = None,
            msg_data: tuple = None
    ):
        info = utils._get_file_info(file)
        if info.dc_id is not None:
            dc_id = info.dc_id

        if file_size is None:
            file_size = info.size

        file = info.location

        if chunk_size is None:
            chunk_size = request_size

        if limit is None and file_size is not None:
            limit = (file_size + chunk_size - 1) // chunk_size

        if stride is None:
            stride = chunk_size
        elif stride < chunk_size:
            raise ValueError('stride must be >= chunk_size')

        request_size -= request_size % MIN_CHUNK_SIZE
        if request_size < MIN_CHUNK_SIZE:
            request_size = MIN_CHUNK_SIZE
        elif request_size > MAX_CHUNK_SIZE:
            request_size = MAX_CHUNK_SIZE

        if chunk_size == request_size \
                and offset % MIN_CHUNK_SIZE == 0 \
                and stride % MIN_CHUNK_SIZE == 0 \
                and (limit is None or offset % limit == 0):
            cls = _DirectDownloadIter
            self._log[__name__].info('Starting direct file download in chunks of '
                                     '%d at %d, stride %d', request_size, offset, stride)
        else:
            cls = _GenericDownloadIter
            self._log[__name__].info('Starting indirect file download in chunks of '
                                     '%d at %d, stride %d', request_size, offset, stride)

        return cls(
            self,
            limit,
            file=file,
            dc_id=dc_id,
            offset=offset,
            stride=stride,
            chunk_size=chunk_size,
            request_size=request_size,
            file_size=file_size,
            msg_data=msg_data,
        )

    # endregion

    # region Private methods

    @staticmethod
    def _get_thumb(thumbs, thumb):
        # Seems Telegram has changed the order and put `PhotoStrippedSize`
        # last while this is the smallest (layer 116). Ensure we have the
        # sizes sorted correctly with a custom function.
        def sort_thumbs(thumb):
            if isinstance(thumb, types.PhotoStrippedSize):
                return 1, len(thumb.bytes)
            if isinstance(thumb, types.PhotoCachedSize):
                return 1, len(thumb.bytes)
            if isinstance(thumb, types.PhotoSize):
                return 1, thumb.size
            if isinstance(thumb, types.PhotoSizeProgressive):
                return 1, max(thumb.sizes)
            if isinstance(thumb, types.VideoSize):
                return 2, thumb.size

            # Empty size or invalid should go last
            return 0, 0

        thumbs = list(sorted(thumbs, key=sort_thumbs))

        for i in reversed(range(len(thumbs))):
            # :tl:`PhotoPathSize` is used for animated stickers preview, and the thumb is actually
            # a SVG path of the outline. Users expect thumbnails to be JPEG files, so pretend this
            # thumb size doesn't actually exist (#1655).
            if isinstance(thumbs[i], types.PhotoPathSize):
                thumbs.pop(i)

        if thumb is None:
            return thumbs[-1]
        elif isinstance(thumb, int):
            return thumbs[thumb]
        elif isinstance(thumb, str):
            return next((t for t in thumbs if t.type == thumb), None)
        elif isinstance(thumb, (types.PhotoSize, types.PhotoCachedSize,
                                types.PhotoStrippedSize, types.VideoSize)):
            return thumb
        else:
            return None

    def _download_cached_photo_size(self: 'TelegramClient', size, file):
        # No need to download anything, simply write the bytes
        if isinstance(size, types.PhotoStrippedSize):
            data = utils.stripped_photo_to_jpg(size.bytes)
        else:
            data = size.bytes

        if file is bytes:
            return data
        elif isinstance(file, str):
            helpers.ensure_parent_dir_exists(file)
            f = open(file, 'wb')
        else:
            f = file

        try:
            f.write(data)
        finally:
            if isinstance(file, str):
                f.close()
        return file

    async def _download_photo(self: 'TelegramClient', photo, file, date, thumb, progress_callback):
        """Specialized version of .download_media() for photos"""
        # Determine the photo and its largest size
        if isinstance(photo, types.MessageMediaPhoto):
            photo = photo.photo
        if not isinstance(photo, types.Photo):
            return

        # Include video sizes here (but they may be None so provide an empty list)
        size = self._get_thumb(photo.sizes + (photo.video_sizes or []), thumb)
        if not size or isinstance(size, types.PhotoSizeEmpty):
            return

        if isinstance(size, types.VideoSize):
            file = self._get_proper_filename(file, 'video', '.mp4', date=date)
        else:
            file = self._get_proper_filename(file, 'photo', '.jpg', date=date)

        if isinstance(size, (types.PhotoCachedSize, types.PhotoStrippedSize)):
            return self._download_cached_photo_size(size, file)

        if isinstance(size, types.PhotoSizeProgressive):
            file_size = max(size.sizes)
        else:
            file_size = size.size

        result = await self.download_file(
            types.InputPhotoFileLocation(
                id=photo.id,
                access_hash=photo.access_hash,
                file_reference=photo.file_reference,
                thumb_size=size.type
            ),
            file,
            file_size=file_size,
            progress_callback=progress_callback
        )
        return result if file is bytes else file

    @staticmethod
    def _get_kind_and_names(attributes):
        """Gets kind and possible names for :tl:`DocumentAttribute`."""
        kind = 'document'
        possible_names = []
        for attr in attributes:
            if isinstance(attr, types.DocumentAttributeFilename):
                possible_names.insert(0, attr.file_name)

            elif isinstance(attr, types.DocumentAttributeAudio):
                kind = 'audio'
                if attr.performer and attr.title:
                    possible_names.append('{} - {}'.format(
                        attr.performer, attr.title
                    ))
                elif attr.performer:
                    possible_names.append(attr.performer)
                elif attr.title:
                    possible_names.append(attr.title)
                elif attr.voice:
                    kind = 'voice'

        return kind, possible_names

    async def _download_document(
            self, document, file, date, thumb, progress_callback, msg_data):
        """Specialized version of .download_media() for documents."""
        if isinstance(document, types.MessageMediaDocument):
            document = document.document
        if not isinstance(document, types.Document):
            return

        if thumb is None:
            kind, possible_names = self._get_kind_and_names(document.attributes)
            file = self._get_proper_filename(
                file, kind, utils.get_extension(document),
                date=date, possible_names=possible_names
            )
            size = None
        else:
            file = self._get_proper_filename(file, 'photo', '.jpg', date=date)
            size = self._get_thumb(document.thumbs, thumb)
            if isinstance(size, (types.PhotoCachedSize, types.PhotoStrippedSize)):
                return self._download_cached_photo_size(size, file)

        result = await self._download_file(
            types.InputDocumentFileLocation(
                id=document.id,
                access_hash=document.access_hash,
                file_reference=document.file_reference,
                thumb_size=size.type if size else ''
            ),
            file,
            file_size=size.size if size else document.size,
            progress_callback=progress_callback,
            msg_data=msg_data,
        )

        return result if file is bytes else file

    @classmethod
    def _download_contact(cls, mm_contact, file):
        """
        Specialized version of .download_media() for contacts.
        Will make use of the vCard 4.0 format.
        """
        first_name = mm_contact.first_name
        last_name = mm_contact.last_name
        phone_number = mm_contact.phone_number

        # Remove these pesky characters
        first_name = first_name.replace(';', '')
        last_name = (last_name or '').replace(';', '')
        result = (
            'BEGIN:VCARD\n'
            'VERSION:4.0\n'
            'N:{f};{l};;;\n'
            'FN:{f} {l}\n'
            'TEL;TYPE=cell;VALUE=uri:tel:+{p}\n'
            'END:VCARD\n'
        ).format(f=first_name, l=last_name, p=phone_number).encode('utf-8')

        if file is bytes:
            return result
        elif isinstance(file, str):
            file = cls._get_proper_filename(
                file, 'contact', '.vcard',
                possible_names=[first_name, phone_number, last_name]
            )
            f = open(file, 'wb')
        else:
            f = file

        try:
            f.write(result)
        finally:
            # Only close the stream if we opened it
            if isinstance(file, str):
                f.close()

        return file

    @classmethod
    async def _download_web_document(cls, web, file, progress_callback):
        """
        Specialized version of .download_media() for web documents.
        """
        if not aiohttp:
            raise ValueError(
                'Cannot download web documents without the aiohttp '
                'dependency install it (pip install aiohttp)'
            )

        # TODO Better way to get opened handles of files and auto-close
        in_memory = file is bytes
        if in_memory:
            f = io.BytesIO()
        elif isinstance(file, str):
            kind, possible_names = cls._get_kind_and_names(web.attributes)
            file = cls._get_proper_filename(
                file, kind, utils.get_extension(web),
                possible_names=possible_names
            )
            f = open(file, 'wb')
        else:
            f = file

        try:
            async with aiohttp.ClientSession() as session:
                # TODO Use progress_callback; get content length from response
                # https://github.com/telegramdesktop/tdesktop/blob/c7e773dd9aeba94e2be48c032edc9a78bb50234e/Telegram/SourceFiles/ui/images.cpp#L1318-L1319
                async with session.get(web.url) as response:
                    while True:
                        chunk = await response.content.read(128 * 1024)
                        if not chunk:
                            break
                        f.write(chunk)
        finally:
            if isinstance(file, str) or file is bytes:
                f.close()

        return f.getvalue() if in_memory else file

    @staticmethod
    def _get_proper_filename(file, kind, extension,
                             date=None, possible_names=None):
        """Gets a proper filename for 'file', if this is a path.

           'kind' should be the kind of the output file (photo, document...)
           'extension' should be the extension to be added to the file if
                       the filename doesn't have any yet
           'date' should be when this file was originally sent, if known
           'possible_names' should be an ordered list of possible names

           If no modification is made to the path, any existing file
           will be overwritten.
           If any modification is made to the path, this method will
           ensure that no existing file will be overwritten.
        """
        if isinstance(file, pathlib.Path):
            file = str(file.absolute())

        if file is not None and not isinstance(file, str):
            # Probably a stream-like object, we cannot set a filename here
            return file

        if file is None:
            file = ''
        elif os.path.isfile(file):
            # Make no modifications to valid existing paths
            return file

        if os.path.isdir(file) or not file:
            try:
                name = None if possible_names is None else next(
                    x for x in possible_names if x
                )
            except StopIteration:
                name = None

            if not name:
                if not date:
                    date = datetime.datetime.now()
                name = '{}_{}-{:02}-{:02}_{:02}-{:02}-{:02}'.format(
                    kind,
                    date.year, date.month, date.day,
                    date.hour, date.minute, date.second,
                )
            file = os.path.join(file, name)

        directory, name = os.path.split(file)
        name, ext = os.path.splitext(name)
        if not ext:
            ext = extension

        result = os.path.join(directory, name + ext)
        if not os.path.isfile(result):
            return result

        i = 1
        while True:
            result = os.path.join(directory, '{} ({}){}'.format(name, i, ext))
            if not os.path.isfile(result):
                return result
            i += 1

    # endregion