#!/usr/bin/env python # # A library that provides a Python interface to the Telegram Bot API # Copyright (C) 2015-2023 # Leandro Toledo de Souza # # This program is free software: you can redistribute it and/or modify # it under the terms of the GNU General Public License as published by # the Free Software Foundation, either version 3 of the License, or # (at your option) any later version. # # This program is distributed in the hope that it will be useful, # but WITHOUT ANY WARRANTY; without even the implied warranty of # MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the # GNU General Public License for more details. # # You should have received a copy of the GNU General Public License # along with this program. If not, see [http://www.gnu.org/licenses/]. import asyncio import copy from collections.abc import Sequence from typing import Optional import pytest from telegram import ( InputFile, InputMediaAnimation, InputMediaAudio, InputMediaDocument, InputMediaPhoto, InputMediaVideo, Message, MessageEntity, ) from telegram.constants import ParseMode # noinspection PyUnresolvedReferences from telegram.error import BadRequest from telegram.request import RequestData from tests._files.test_animation import animation, animation_file # noqa: F401 from tests.auxil.files import data_file from tests.auxil.networking import expect_bad_request from tests.auxil.slots import mro_slots # noinspection PyUnresolvedReferences from tests.test_forum import emoji_id, real_topic # noqa: F401 # noinspection PyUnresolvedReferences from .test_audio import audio, audio_file # noqa: F401 # noinspection PyUnresolvedReferences from .test_document import document, document_file # noqa: F401 # noinspection PyUnresolvedReferences from .test_photo import photo, photo_file, photolist, thumb # noqa: F401 # noinspection PyUnresolvedReferences from .test_video import video, video_file # noqa: F401 @pytest.fixture(scope="module") def input_media_video(class_thumb_file): return InputMediaVideo( media=TestInputMediaVideoBase.media, caption=TestInputMediaVideoBase.caption, width=TestInputMediaVideoBase.width, height=TestInputMediaVideoBase.height, duration=TestInputMediaVideoBase.duration, parse_mode=TestInputMediaVideoBase.parse_mode, caption_entities=TestInputMediaVideoBase.caption_entities, thumbnail=class_thumb_file, supports_streaming=TestInputMediaVideoBase.supports_streaming, has_spoiler=TestInputMediaVideoBase.has_spoiler, ) @pytest.fixture(scope="module") def input_media_photo(): return InputMediaPhoto( media=TestInputMediaPhotoBase.media, caption=TestInputMediaPhotoBase.caption, parse_mode=TestInputMediaPhotoBase.parse_mode, caption_entities=TestInputMediaPhotoBase.caption_entities, has_spoiler=TestInputMediaPhotoBase.has_spoiler, ) @pytest.fixture(scope="module") def input_media_animation(class_thumb_file): return InputMediaAnimation( media=TestInputMediaAnimationBase.media, caption=TestInputMediaAnimationBase.caption, parse_mode=TestInputMediaAnimationBase.parse_mode, caption_entities=TestInputMediaAnimationBase.caption_entities, width=TestInputMediaAnimationBase.width, height=TestInputMediaAnimationBase.height, thumbnail=class_thumb_file, duration=TestInputMediaAnimationBase.duration, has_spoiler=TestInputMediaAnimationBase.has_spoiler, ) @pytest.fixture(scope="module") def input_media_audio(class_thumb_file): return InputMediaAudio( media=TestInputMediaAudioBase.media, caption=TestInputMediaAudioBase.caption, duration=TestInputMediaAudioBase.duration, performer=TestInputMediaAudioBase.performer, title=TestInputMediaAudioBase.title, thumbnail=class_thumb_file, parse_mode=TestInputMediaAudioBase.parse_mode, caption_entities=TestInputMediaAudioBase.caption_entities, ) @pytest.fixture(scope="module") def input_media_document(class_thumb_file): return InputMediaDocument( media=TestInputMediaDocumentBase.media, caption=TestInputMediaDocumentBase.caption, thumbnail=class_thumb_file, parse_mode=TestInputMediaDocumentBase.parse_mode, caption_entities=TestInputMediaDocumentBase.caption_entities, disable_content_type_detection=TestInputMediaDocumentBase.disable_content_type_detection, ) class TestInputMediaVideoBase: type_ = "video" media = "NOTAREALFILEID" caption = "My Caption" width = 3 height = 4 duration = 5 parse_mode = "HTML" supports_streaming = True caption_entities = [MessageEntity(MessageEntity.BOLD, 0, 2)] has_spoiler = True class TestInputMediaVideoWithoutRequest(TestInputMediaVideoBase): def test_slot_behaviour(self, input_media_video): inst = input_media_video for attr in inst.__slots__: assert getattr(inst, attr, "err") != "err", f"got extra slot '{attr}'" assert len(mro_slots(inst)) == len(set(mro_slots(inst))), "duplicate slot" def test_expected_values(self, input_media_video): assert input_media_video.type == self.type_ assert input_media_video.media == self.media assert input_media_video.caption == self.caption assert input_media_video.width == self.width assert input_media_video.height == self.height assert input_media_video.duration == self.duration assert input_media_video.parse_mode == self.parse_mode assert input_media_video.caption_entities == tuple(self.caption_entities) assert input_media_video.supports_streaming == self.supports_streaming assert isinstance(input_media_video.thumbnail, InputFile) assert input_media_video.has_spoiler == self.has_spoiler def test_caption_entities_always_tuple(self): input_media_video = InputMediaVideo(self.media) assert input_media_video.caption_entities == () def test_to_dict(self, input_media_video): input_media_video_dict = input_media_video.to_dict() assert input_media_video_dict["type"] == input_media_video.type assert input_media_video_dict["media"] == input_media_video.media assert input_media_video_dict["caption"] == input_media_video.caption assert input_media_video_dict["width"] == input_media_video.width assert input_media_video_dict["height"] == input_media_video.height assert input_media_video_dict["duration"] == input_media_video.duration assert input_media_video_dict["parse_mode"] == input_media_video.parse_mode assert input_media_video_dict["caption_entities"] == [ ce.to_dict() for ce in input_media_video.caption_entities ] assert input_media_video_dict["supports_streaming"] == input_media_video.supports_streaming assert input_media_video_dict["has_spoiler"] == input_media_video.has_spoiler def test_with_video(self, video): # noqa: F811 # fixture found in test_video input_media_video = InputMediaVideo(video, caption="test 3") assert input_media_video.type == self.type_ assert input_media_video.media == video.file_id assert input_media_video.width == video.width assert input_media_video.height == video.height assert input_media_video.duration == video.duration assert input_media_video.caption == "test 3" def test_with_video_file(self, video_file): # noqa: F811 # fixture found in test_video input_media_video = InputMediaVideo(video_file, caption="test 3") assert input_media_video.type == self.type_ assert isinstance(input_media_video.media, InputFile) assert input_media_video.caption == "test 3" def test_with_local_files(self): input_media_video = InputMediaVideo( data_file("telegram.mp4"), thumbnail=data_file("telegram.jpg") ) assert input_media_video.media == data_file("telegram.mp4").as_uri() assert input_media_video.thumbnail == data_file("telegram.jpg").as_uri() class TestInputMediaPhotoBase: type_ = "photo" media = "NOTAREALFILEID" caption = "My Caption" parse_mode = "Markdown" caption_entities = [MessageEntity(MessageEntity.BOLD, 0, 2)] has_spoiler = True class TestInputMediaPhotoWithoutRequest(TestInputMediaPhotoBase): def test_slot_behaviour(self, input_media_photo): inst = input_media_photo for attr in inst.__slots__: assert getattr(inst, attr, "err") != "err", f"got extra slot '{attr}'" assert len(mro_slots(inst)) == len(set(mro_slots(inst))), "duplicate slot" def test_expected_values(self, input_media_photo): assert input_media_photo.type == self.type_ assert input_media_photo.media == self.media assert input_media_photo.caption == self.caption assert input_media_photo.parse_mode == self.parse_mode assert input_media_photo.caption_entities == tuple(self.caption_entities) assert input_media_photo.has_spoiler == self.has_spoiler def test_caption_entities_always_tuple(self): input_media_photo = InputMediaPhoto(self.media) assert input_media_photo.caption_entities == () def test_to_dict(self, input_media_photo): input_media_photo_dict = input_media_photo.to_dict() assert input_media_photo_dict["type"] == input_media_photo.type assert input_media_photo_dict["media"] == input_media_photo.media assert input_media_photo_dict["caption"] == input_media_photo.caption assert input_media_photo_dict["parse_mode"] == input_media_photo.parse_mode assert input_media_photo_dict["caption_entities"] == [ ce.to_dict() for ce in input_media_photo.caption_entities ] assert input_media_photo_dict["has_spoiler"] == input_media_photo.has_spoiler def test_with_photo(self, photo): # noqa: F811 # fixture found in test_photo input_media_photo = InputMediaPhoto(photo, caption="test 2") assert input_media_photo.type == self.type_ assert input_media_photo.media == photo.file_id assert input_media_photo.caption == "test 2" def test_with_photo_file(self, photo_file): # noqa: F811 # fixture found in test_photo input_media_photo = InputMediaPhoto(photo_file, caption="test 2") assert input_media_photo.type == self.type_ assert isinstance(input_media_photo.media, InputFile) assert input_media_photo.caption == "test 2" def test_with_local_files(self): input_media_photo = InputMediaPhoto(data_file("telegram.mp4")) assert input_media_photo.media == data_file("telegram.mp4").as_uri() class TestInputMediaAnimationBase: type_ = "animation" media = "NOTAREALFILEID" caption = "My Caption" parse_mode = "Markdown" caption_entities = [MessageEntity(MessageEntity.BOLD, 0, 2)] width = 30 height = 30 duration = 1 has_spoiler = True class TestInputMediaAnimationWithoutRequest(TestInputMediaAnimationBase): def test_slot_behaviour(self, input_media_animation): inst = input_media_animation for attr in inst.__slots__: assert getattr(inst, attr, "err") != "err", f"got extra slot '{attr}'" assert len(mro_slots(inst)) == len(set(mro_slots(inst))), "duplicate slot" def test_expected_values(self, input_media_animation): assert input_media_animation.type == self.type_ assert input_media_animation.media == self.media assert input_media_animation.caption == self.caption assert input_media_animation.parse_mode == self.parse_mode assert input_media_animation.caption_entities == tuple(self.caption_entities) assert isinstance(input_media_animation.thumbnail, InputFile) assert input_media_animation.has_spoiler == self.has_spoiler def test_caption_entities_always_tuple(self): input_media_animation = InputMediaAnimation(self.media) assert input_media_animation.caption_entities == () def test_to_dict(self, input_media_animation): input_media_animation_dict = input_media_animation.to_dict() assert input_media_animation_dict["type"] == input_media_animation.type assert input_media_animation_dict["media"] == input_media_animation.media assert input_media_animation_dict["caption"] == input_media_animation.caption assert input_media_animation_dict["parse_mode"] == input_media_animation.parse_mode assert input_media_animation_dict["caption_entities"] == [ ce.to_dict() for ce in input_media_animation.caption_entities ] assert input_media_animation_dict["width"] == input_media_animation.width assert input_media_animation_dict["height"] == input_media_animation.height assert input_media_animation_dict["duration"] == input_media_animation.duration assert input_media_animation_dict["has_spoiler"] == input_media_animation.has_spoiler def test_with_animation(self, animation): # noqa: F811 # fixture found in test_animation input_media_animation = InputMediaAnimation(animation, caption="test 2") assert input_media_animation.type == self.type_ assert input_media_animation.media == animation.file_id assert input_media_animation.caption == "test 2" def test_with_animation_file(self, animation_file): # noqa: F811 # fixture found in test_animation input_media_animation = InputMediaAnimation(animation_file, caption="test 2") assert input_media_animation.type == self.type_ assert isinstance(input_media_animation.media, InputFile) assert input_media_animation.caption == "test 2" def test_with_local_files(self): input_media_animation = InputMediaAnimation( data_file("telegram.mp4"), thumbnail=data_file("telegram.jpg") ) assert input_media_animation.media == data_file("telegram.mp4").as_uri() assert input_media_animation.thumbnail == data_file("telegram.jpg").as_uri() class TestInputMediaAudioBase: type_ = "audio" media = "NOTAREALFILEID" caption = "My Caption" duration = 3 performer = "performer" title = "title" parse_mode = "HTML" caption_entities = [MessageEntity(MessageEntity.BOLD, 0, 2)] class TestInputMediaAudioWithoutRequest(TestInputMediaAudioBase): def test_slot_behaviour(self, input_media_audio): inst = input_media_audio for attr in inst.__slots__: assert getattr(inst, attr, "err") != "err", f"got extra slot '{attr}'" assert len(mro_slots(inst)) == len(set(mro_slots(inst))), "duplicate slot" def test_expected_values(self, input_media_audio): assert input_media_audio.type == self.type_ assert input_media_audio.media == self.media assert input_media_audio.caption == self.caption assert input_media_audio.duration == self.duration assert input_media_audio.performer == self.performer assert input_media_audio.title == self.title assert input_media_audio.parse_mode == self.parse_mode assert input_media_audio.caption_entities == tuple(self.caption_entities) assert isinstance(input_media_audio.thumbnail, InputFile) def test_caption_entities_always_tuple(self): input_media_audio = InputMediaAudio(self.media) assert input_media_audio.caption_entities == () def test_to_dict(self, input_media_audio): input_media_audio_dict = input_media_audio.to_dict() assert input_media_audio_dict["type"] == input_media_audio.type assert input_media_audio_dict["media"] == input_media_audio.media assert input_media_audio_dict["caption"] == input_media_audio.caption assert input_media_audio_dict["duration"] == input_media_audio.duration assert input_media_audio_dict["performer"] == input_media_audio.performer assert input_media_audio_dict["title"] == input_media_audio.title assert input_media_audio_dict["parse_mode"] == input_media_audio.parse_mode assert input_media_audio_dict["caption_entities"] == [ ce.to_dict() for ce in input_media_audio.caption_entities ] def test_with_audio(self, audio): # noqa: F811 # fixture found in test_audio input_media_audio = InputMediaAudio(audio, caption="test 3") assert input_media_audio.type == self.type_ assert input_media_audio.media == audio.file_id assert input_media_audio.duration == audio.duration assert input_media_audio.performer == audio.performer assert input_media_audio.title == audio.title assert input_media_audio.caption == "test 3" def test_with_audio_file(self, audio_file): # noqa: F811 # fixture found in test_audio input_media_audio = InputMediaAudio(audio_file, caption="test 3") assert input_media_audio.type == self.type_ assert isinstance(input_media_audio.media, InputFile) assert input_media_audio.caption == "test 3" def test_with_local_files(self): input_media_audio = InputMediaAudio( data_file("telegram.mp4"), thumbnail=data_file("telegram.jpg") ) assert input_media_audio.media == data_file("telegram.mp4").as_uri() assert input_media_audio.thumbnail == data_file("telegram.jpg").as_uri() class TestInputMediaDocumentBase: type_ = "document" media = "NOTAREALFILEID" caption = "My Caption" parse_mode = "HTML" caption_entities = [MessageEntity(MessageEntity.BOLD, 0, 2)] disable_content_type_detection = True class TestInputMediaDocumentWithoutRequest(TestInputMediaDocumentBase): def test_slot_behaviour(self, input_media_document): inst = input_media_document for attr in inst.__slots__: assert getattr(inst, attr, "err") != "err", f"got extra slot '{attr}'" assert len(mro_slots(inst)) == len(set(mro_slots(inst))), "duplicate slot" def test_expected_values(self, input_media_document): assert input_media_document.type == self.type_ assert input_media_document.media == self.media assert input_media_document.caption == self.caption assert input_media_document.parse_mode == self.parse_mode assert input_media_document.caption_entities == tuple(self.caption_entities) assert ( input_media_document.disable_content_type_detection == self.disable_content_type_detection ) assert isinstance(input_media_document.thumbnail, InputFile) def test_caption_entities_always_tuple(self): input_media_document = InputMediaDocument(self.media) assert input_media_document.caption_entities == () def test_to_dict(self, input_media_document): input_media_document_dict = input_media_document.to_dict() assert input_media_document_dict["type"] == input_media_document.type assert input_media_document_dict["media"] == input_media_document.media assert input_media_document_dict["caption"] == input_media_document.caption assert input_media_document_dict["parse_mode"] == input_media_document.parse_mode assert input_media_document_dict["caption_entities"] == [ ce.to_dict() for ce in input_media_document.caption_entities ] assert ( input_media_document["disable_content_type_detection"] == input_media_document.disable_content_type_detection ) def test_with_document(self, document): # noqa: F811 # fixture found in test_document input_media_document = InputMediaDocument(document, caption="test 3") assert input_media_document.type == self.type_ assert input_media_document.media == document.file_id assert input_media_document.caption == "test 3" def test_with_document_file(self, document_file): # noqa: F811 # fixture found in test_document input_media_document = InputMediaDocument(document_file, caption="test 3") assert input_media_document.type == self.type_ assert isinstance(input_media_document.media, InputFile) assert input_media_document.caption == "test 3" def test_with_local_files(self): input_media_document = InputMediaDocument( data_file("telegram.mp4"), thumbnail=data_file("telegram.jpg") ) assert input_media_document.media == data_file("telegram.mp4").as_uri() assert input_media_document.thumbnail == data_file("telegram.jpg").as_uri() @pytest.fixture(scope="module") def media_group(photo, thumb): # noqa: F811 return [ InputMediaPhoto(photo, caption="*photo* 1", parse_mode="Markdown"), InputMediaPhoto(thumb, caption="photo 2", parse_mode="HTML"), InputMediaPhoto( photo, caption="photo 3", caption_entities=[MessageEntity(MessageEntity.BOLD, 0, 5)] ), ] @pytest.fixture(scope="module") def media_group_no_caption_args(photo, thumb): # noqa: F811 return [InputMediaPhoto(photo), InputMediaPhoto(thumb), InputMediaPhoto(photo)] @pytest.fixture(scope="module") def media_group_no_caption_only_caption_entities(photo, thumb): # noqa: F811 return [ InputMediaPhoto(photo, caption_entities=[MessageEntity(MessageEntity.BOLD, 0, 5)]), InputMediaPhoto(photo, caption_entities=[MessageEntity(MessageEntity.BOLD, 0, 5)]), ] @pytest.fixture(scope="module") def media_group_no_caption_only_parse_mode(photo, thumb): # noqa: F811 return [ InputMediaPhoto(photo, parse_mode="Markdown"), InputMediaPhoto(thumb, parse_mode="HTML"), ] class TestSendMediaGroupWithoutRequest: async def test_send_media_group_throws_error_with_group_caption_and_individual_captions( self, bot, chat_id, media_group, media_group_no_caption_only_caption_entities, media_group_no_caption_only_parse_mode, ): for group in ( media_group, media_group_no_caption_only_caption_entities, media_group_no_caption_only_parse_mode, ): with pytest.raises( ValueError, match="You can only supply either group caption or media with captions.", ): await bot.send_media_group(chat_id, group, caption="foo") async def test_send_media_group_custom_filename( self, bot, chat_id, photo_file, # noqa: F811 animation_file, # noqa: F811 audio_file, # noqa: F811 video_file, # noqa: F811 monkeypatch, ): async def make_assertion(url, request_data: RequestData, *args, **kwargs): result = all( field_tuple[0] == "custom_filename" for field_tuple in request_data.multipart_data.values() ) if result is True: raise Exception("Test was successful") monkeypatch.setattr(bot.request, "post", make_assertion) media = [ InputMediaAnimation(animation_file, filename="custom_filename"), InputMediaAudio(audio_file, filename="custom_filename"), InputMediaPhoto(photo_file, filename="custom_filename"), InputMediaVideo(video_file, filename="custom_filename"), ] with pytest.raises(Exception, match="Test was successful"): await bot.send_media_group(chat_id, media) async def test_send_media_group_with_thumbs( self, bot, chat_id, video_file, photo_file, monkeypatch # noqa: F811 ): async def make_assertion(method, url, request_data: RequestData, *args, **kwargs): nonlocal input_video files = request_data.multipart_data video_check = files[input_video.media.attach_name] == input_video.media.field_tuple thumb_check = ( files[input_video.thumbnail.attach_name] == input_video.thumbnail.field_tuple ) result = video_check and thumb_check raise Exception(f"Test was {'successful' if result else 'failing'}") monkeypatch.setattr(bot.request, "_request_wrapper", make_assertion) input_video = InputMediaVideo(video_file, thumbnail=photo_file) with pytest.raises(Exception, match="Test was successful"): await bot.send_media_group(chat_id, [input_video, input_video]) async def test_edit_message_media_with_thumb( self, bot, chat_id, video_file, photo_file, monkeypatch # noqa: F811 ): async def make_assertion( method: str, url: str, request_data: Optional[RequestData] = None, *args, **kwargs ): files = request_data.multipart_data video_check = files[input_video.media.attach_name] == input_video.media.field_tuple thumb_check = ( files[input_video.thumbnail.attach_name] == input_video.thumbnail.field_tuple ) result = video_check and thumb_check raise Exception(f"Test was {'successful' if result else 'failing'}") monkeypatch.setattr(bot.request, "_request_wrapper", make_assertion) input_video = InputMediaVideo(video_file, thumbnail=photo_file) with pytest.raises(Exception, match="Test was successful"): await bot.edit_message_media(chat_id=chat_id, message_id=123, media=input_video) class CustomSequence(Sequence): def __init__(self, items): self.items = items def __getitem__(self, index): return self.items[index] def __len__(self): return len(self.items) class TestSendMediaGroupWithRequest: async def test_send_media_group_photo(self, bot, chat_id, media_group): messages = await bot.send_media_group(chat_id, media_group) assert isinstance(messages, tuple) assert len(messages) == 3 assert all(isinstance(mes, Message) for mes in messages) assert all(mes.media_group_id == messages[0].media_group_id for mes in messages) assert all(mes.caption == f"photo {idx+1}" for idx, mes in enumerate(messages)) assert all( mes.caption_entities == (MessageEntity(MessageEntity.BOLD, 0, 5),) for mes in messages ) async def test_send_media_group_new_files( self, bot, chat_id, video_file, photo_file # noqa: F811 ): async def func(): return await bot.send_media_group( chat_id, [ InputMediaVideo(video_file), InputMediaPhoto(photo_file), InputMediaPhoto(data_file("telegram.jpg").read_bytes()), ], ) messages = await expect_bad_request( func, "Type of file mismatch", "Telegram did not accept the file." ) assert isinstance(messages, tuple) assert len(messages) == 3 assert all(isinstance(mes, Message) for mes in messages) assert all(mes.media_group_id == messages[0].media_group_id for mes in messages) @pytest.mark.parametrize("sequence_type", [list, tuple, CustomSequence]) @pytest.mark.parametrize("bot_class", ["raw_bot", "ext_bot"]) async def test_send_media_group_different_sequences( self, bot, chat_id, media_group, sequence_type, bot_class, raw_bot ): """Test that send_media_group accepts different sequence types. This test ensures that Bot._insert_defaults works for arbitrary sequence types.""" bot = bot if bot_class == "ext_bot" else raw_bot messages = await bot.send_media_group(chat_id, sequence_type(media_group)) assert isinstance(messages, tuple) assert len(messages) == 3 assert all(isinstance(mes, Message) for mes in messages) assert all(mes.media_group_id == messages[0].media_group_id for mes in messages) async def test_send_media_group_with_message_thread_id( self, bot, real_topic, forum_group_id, media_group # noqa: F811 ): messages = await bot.send_media_group( forum_group_id, media_group, message_thread_id=real_topic.message_thread_id, ) assert isinstance(messages, tuple) assert len(messages) == 3 assert all(isinstance(mes, Message) for mes in messages) assert all(i.message_thread_id == real_topic.message_thread_id for i in messages) @pytest.mark.parametrize( ("caption", "parse_mode", "caption_entities"), [ # same combinations of caption options as in media_group fixture ("*photo* 1", "Markdown", None), ("photo 1", "HTML", None), ("photo 1", None, [MessageEntity(MessageEntity.BOLD, 0, 5)]), ], ) async def test_send_media_group_with_group_caption( self, bot, chat_id, media_group_no_caption_args, caption, parse_mode, caption_entities, ): # prepare a copy to check later on if calling the method has caused side effects copied_media_group = media_group_no_caption_args.copy() messages = await bot.send_media_group( chat_id, media_group_no_caption_args, caption=caption, parse_mode=parse_mode, caption_entities=caption_entities, ) # Check that the method had no side effects: # original group was not changed and 1st item still points to the same object # (1st item must be copied within the method before adding the caption) assert media_group_no_caption_args == copied_media_group assert media_group_no_caption_args[0] is copied_media_group[0] assert not any(item.parse_mode for item in media_group_no_caption_args) assert isinstance(messages, tuple) assert len(messages) == 3 assert all(isinstance(mes, Message) for mes in messages) first_message, other_messages = messages[0], messages[1:] assert all(mes.media_group_id == first_message.media_group_id for mes in messages) # Make sure first message got the caption, which will lead # to Telegram displaying its caption as group caption assert first_message.caption assert first_message.caption_entities == (MessageEntity(MessageEntity.BOLD, 0, 5),) # Check that other messages have no captions assert all(mes.caption is None for mes in other_messages) assert not any(mes.caption_entities for mes in other_messages) async def test_send_media_group_all_args(self, bot, raw_bot, chat_id, media_group): ext_bot = bot # We need to test 1) below both the bot and raw_bot and setting this up with # pytest.parametrize appears to be difficult ... aws = {b.send_message(chat_id, text="test") for b in (ext_bot, raw_bot)} for msg_task in asyncio.as_completed(aws): m1 = await msg_task copied_media_group = copy.copy(media_group) messages = await m1.get_bot().send_media_group( chat_id, media_group, disable_notification=True, reply_to_message_id=m1.message_id, protect_content=True, ) # 1) # make sure that the media_group was not modified assert media_group == copied_media_group assert all( a.parse_mode == b.parse_mode for a, b in zip(media_group, copied_media_group) ) assert isinstance(messages, tuple) assert len(messages) == 3 assert all(isinstance(mes, Message) for mes in messages) assert all(mes.media_group_id == messages[0].media_group_id for mes in messages) assert all(mes.caption == f"photo {idx+1}" for idx, mes in enumerate(messages)) assert all( mes.caption_entities == (MessageEntity(MessageEntity.BOLD, 0, 5),) for mes in messages ) assert all(mes.has_protected_content for mes in messages) async def test_send_media_group_with_spoiler( self, bot, chat_id, photo_file, video_file # noqa: F811 ): # Media groups can't contain Animations, so that is tested in test_animation.py media = [ InputMediaPhoto(photo_file, has_spoiler=True), InputMediaVideo(video_file, has_spoiler=True), ] messages = await bot.send_media_group(chat_id, media) assert isinstance(messages, tuple) assert len(messages) == 2 assert all(isinstance(mes, Message) for mes in messages) assert all(mes.media_group_id == messages[0].media_group_id for mes in messages) assert all(mes.has_media_spoiler for mes in messages) async def test_edit_message_media(self, bot, raw_bot, chat_id, media_group): ext_bot = bot # We need to test 1) below both the bot and raw_bot and setting this up with # pytest.parametrize appears to be difficult ... aws = {b.send_media_group(chat_id, media_group) for b in (ext_bot, raw_bot)} for msg_task in asyncio.as_completed(aws): messages = await msg_task cid = messages[-1].chat.id mid = messages[-1].message_id copied_media = copy.copy(media_group[0]) new_message = ( await messages[-1] .get_bot() .edit_message_media(chat_id=cid, message_id=mid, media=media_group[0]) ) assert isinstance(new_message, Message) # 1) # make sure that the media was not modified assert media_group[0].parse_mode == copied_media.parse_mode async def test_edit_message_media_new_file(self, bot, chat_id, media_group, thumb_file): messages = await bot.send_media_group(chat_id, media_group) cid = messages[-1].chat.id mid = messages[-1].message_id new_message = await bot.edit_message_media( chat_id=cid, message_id=mid, media=InputMediaPhoto(thumb_file) ) assert isinstance(new_message, Message) @pytest.mark.parametrize( ("default_bot", "custom"), [ ({"allow_sending_without_reply": True}, None), ({"allow_sending_without_reply": False}, None), ({"allow_sending_without_reply": False}, True), ], indirect=["default_bot"], ) async def test_send_media_group_default_allow_sending_without_reply( self, default_bot, chat_id, media_group, custom ): reply_to_message = await default_bot.send_message(chat_id, "test") await reply_to_message.delete() if custom is not None: messages = await default_bot.send_media_group( chat_id, media_group, allow_sending_without_reply=custom, reply_to_message_id=reply_to_message.message_id, ) assert [m.reply_to_message is None for m in messages] elif default_bot.defaults.allow_sending_without_reply: messages = await default_bot.send_media_group( chat_id, media_group, reply_to_message_id=reply_to_message.message_id ) assert [m.reply_to_message is None for m in messages] else: with pytest.raises(BadRequest, match="message not found"): await default_bot.send_media_group( chat_id, media_group, reply_to_message_id=reply_to_message.message_id ) @pytest.mark.parametrize("default_bot", [{"protect_content": True}], indirect=True) async def test_send_media_group_default_protect_content( self, chat_id, media_group, default_bot ): tasks = asyncio.gather( default_bot.send_media_group(chat_id, media_group), default_bot.send_media_group(chat_id, media_group, protect_content=False), ) protected, unprotected = await tasks assert all(msg.has_protected_content for msg in protected) assert not all(msg.has_protected_content for msg in unprotected) @pytest.mark.parametrize("default_bot", [{"parse_mode": ParseMode.HTML}], indirect=True) async def test_send_media_group_default_parse_mode( self, chat_id, media_group_no_caption_args, default_bot ): default = await default_bot.send_media_group( chat_id, media_group_no_caption_args, caption="photo 1" ) # make sure no parse_mode was set as a side effect assert not any(item.parse_mode for item in media_group_no_caption_args) tasks = asyncio.gather( default_bot.send_media_group( chat_id, media_group_no_caption_args.copy(), caption="*photo* 1", parse_mode=ParseMode.MARKDOWN_V2, ), default_bot.send_media_group( chat_id, media_group_no_caption_args.copy(), caption="photo 1", parse_mode=None, ), ) overridden_markdown_v2, overridden_none = await tasks # Make sure first message got the caption, which will lead to Telegram # displaying its caption as group caption assert overridden_none[0].caption == "photo 1" assert not overridden_none[0].caption_entities # First messages in these two groups have to have caption "photo 1" # because of parse mode (default or explicit) for mes_group in (default, overridden_markdown_v2): first_message = mes_group[0] assert first_message.caption == "photo 1" assert first_message.caption_entities == (MessageEntity(MessageEntity.BOLD, 0, 5),) # This check is valid for all 3 groups of messages for mes_group in (default, overridden_markdown_v2, overridden_none): first_message, other_messages = mes_group[0], mes_group[1:] assert all(mes.media_group_id == first_message.media_group_id for mes in mes_group) # Check that messages from 2nd message onwards have no captions assert all(mes.caption is None for mes in other_messages) assert not any(mes.caption_entities for mes in other_messages) @pytest.mark.parametrize( "default_bot", [{"parse_mode": ParseMode.HTML}], indirect=True, ids=["HTML-Bot"] ) @pytest.mark.parametrize("media_type", ["animation", "document", "audio", "photo", "video"]) async def test_edit_message_media_default_parse_mode( self, chat_id, default_bot, media_type, animation, # noqa: F811 document, # noqa: F811 audio, # noqa: F811 photo, # noqa: F811 video, # noqa: F811 ): html_caption = "bold italic code" markdown_caption = "*bold* _italic_ `code`" test_caption = "bold italic code" test_entities = [ MessageEntity(MessageEntity.BOLD, 0, 4), MessageEntity(MessageEntity.ITALIC, 5, 6), MessageEntity(MessageEntity.CODE, 12, 4), ] def build_media(parse_mode, med_type): kwargs = {} if parse_mode != ParseMode.HTML: kwargs["parse_mode"] = parse_mode kwargs["caption"] = markdown_caption else: kwargs["caption"] = html_caption if med_type == "animation": return InputMediaAnimation(animation, **kwargs) if med_type == "document": return InputMediaDocument(document, **kwargs) if med_type == "audio": return InputMediaAudio(audio, **kwargs) if med_type == "photo": return InputMediaPhoto(photo, **kwargs) if med_type == "video": return InputMediaVideo(video, **kwargs) return None message = await default_bot.send_photo(chat_id, photo) media = build_media(parse_mode=ParseMode.HTML, med_type=media_type) copied_media = copy.copy(media) message = await default_bot.edit_message_media( media, message.chat_id, message.message_id, ) assert message.caption == test_caption assert message.caption_entities == tuple(test_entities) # make sure that the media was not modified assert media.parse_mode == copied_media.parse_mode # Remove caption to avoid "Message not changed" await message.edit_caption() media = build_media(parse_mode=ParseMode.MARKDOWN_V2, med_type=media_type) copied_media = copy.copy(media) message = await default_bot.edit_message_media( media, message.chat_id, message.message_id, ) assert message.caption == test_caption assert message.caption_entities == tuple(test_entities) # make sure that the media was not modified assert media.parse_mode == copied_media.parse_mode # Remove caption to avoid "Message not changed" await message.edit_caption() media = build_media(parse_mode=None, med_type=media_type) copied_media = copy.copy(media) message = await default_bot.edit_message_media( media, message.chat_id, message.message_id, ) assert message.caption == markdown_caption assert message.caption_entities == () # make sure that the media was not modified assert media.parse_mode == copied_media.parse_mode