import datetime import io import os import pathlib from .users import UserMethods from .. import utils, helpers, errors from ..tl import TLObject, types, functions try: import aiohttp except ImportError: aiohttp = None class DownloadMethods(UserMethods): # region Public methods async def download_profile_photo( self, entity, file=None, *, download_big=True): """ Downloads the profile photo of the given entity (user/chat/channel). Args: 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. """ # 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) 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, 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)): loc = photo.photo_big if download_big else photo.photo_small else: try: loc = utils.get_input_location(photo) except TypeError: return None file = self._get_proper_filename( file, 'profile_photo', '.jpg', possible_names=possible_names ) try: result = await self.download_file(loc, file) 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) if isinstance(ie, types.InputPeerChannel): full = await self(functions.channels.GetFullChannelRequest(ie)) return await self._download_photo( full.full_chat.chat_photo, file, date=None, progress_callback=None ) else: # Until there's a report for chats, no need to. return None async def download_media(self, message, file=None, *, progress_callback=None): """ Downloads the given media, or the media from a specified Message. 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). 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)``. 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. """ # TODO This won't work for messageService if isinstance(message, types.Message): date = message.date media = message.media else: date = datetime.datetime.now() media = message if isinstance(media, str): media = utils.resolve_bot_file_id(media) 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, types.PhotoSize, types.PhotoCachedSize, types.PhotoStrippedSize)): return await self._download_photo( media, file, date, progress_callback ) elif isinstance(media, (types.MessageMediaDocument, types.Document)): return await self._download_document( media, file, date, progress_callback ) elif isinstance(media, types.MessageMediaContact): return self._download_contact( media, file ) elif isinstance(media, (types.WebDocument, types.WebDocumentNoProxy)): return await self._download_web_document( media, file, progress_callback ) async def download_file( self, input_location, file=None, *, part_size_kb=None, file_size=None, progress_callback=None): """ Downloads the given input location to a file. Args: input_location (:tl:`FileLocation` | :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``. """ 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) # https://core.telegram.org/api/files says: # > part_size % 1024 = 0 (divisible by 1KB) # # But https://core.telegram.org/cdn (more recent) says: # > limit must be divisible by 4096 bytes # So we just stick to the 4096 limit. if part_size % 4096 != 0: raise ValueError( 'The part size must be evenly divisible by 4096.') 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 dc_id, input_location = utils.get_input_location(input_location) exported = dc_id and self.session.dc_id != dc_id if exported: try: sender = await self._borrow_exported_sender(dc_id) except errors.DcIdInvalidError: # Can't export a sender for the ID we are currently in config = await self(functions.help.GetConfigRequest()) for option in config.dc_options: if option.ip_address == self.session.server_address: self.session.set_dc( option.id, option.ip_address, option.port) self.session.save() break # TODO Figure out why the session may have the wrong DC ID sender = self._sender exported = False else: # The used sender will also change if ``FileMigrateError`` occurs sender = self._sender self._log[__name__].info('Downloading file in chunks of %d bytes', part_size) try: offset = 0 while True: try: result = await sender.send(functions.upload.GetFileRequest( input_location, offset, part_size )) if isinstance(result, types.upload.FileCdnRedirect): # TODO Implement raise NotImplementedError except errors.FileMigrateError as e: self._log[__name__].info('File lives in another DC') sender = await self._borrow_exported_sender(e.new_dc) exported = True continue offset += part_size if not result.bytes: if in_memory: f.flush() return f.getvalue() else: return getattr(result, 'type', '') self._log[__name__].debug('Saving %d more bytes', len(result.bytes)) f.write(result.bytes) if progress_callback: progress_callback(f.tell(), file_size) finally: if exported: await self._return_exported_sender(sender) elif sender != self._sender: await sender.disconnect() if isinstance(file, str) or in_memory: f.close() # endregion # region Private methods async def _download_photo(self, photo, file, date, 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 isinstance(photo, types.Photo): for size in reversed(photo.sizes): if not isinstance(size, types.PhotoSizeEmpty): photo = size break else: return if not isinstance(photo, (types.PhotoSize, types.PhotoCachedSize, types.PhotoStrippedSize)): return file = self._get_proper_filename(file, 'photo', '.jpg', date=date) if isinstance(photo, (types.PhotoCachedSize, types.PhotoStrippedSize)): # No need to download anything, simply write the bytes if file is bytes: return photo.bytes elif isinstance(file, str): helpers.ensure_parent_dir_exists(file) f = open(file, 'wb') else: f = file try: f.write(photo.bytes) finally: if isinstance(file, str): f.close() return file result = await self.download_file( photo.location, file, file_size=photo.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, progress_callback): """Specialized version of .download_media() for documents.""" if isinstance(document, types.MessageMediaDocument): document = document.document if not isinstance(document, types.Document): return 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 ) result = await self.download_file( document, file, file_size=document.size, progress_callback=progress_callback) 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, 'w', encoding='utf-8') 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: 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