import json import textwrap import urllib.parse import uuid from .common import InfoExtractor from ..utils import ( ExtractorError, determine_ext, filter_dict, get_first, int_or_none, parse_iso8601, update_url, url_or_none, variadic, ) from ..utils.traversal import traverse_obj class LoomIE(InfoExtractor): IE_NAME = 'loom' _VALID_URL = r'https?://(?:www\.)?loom\.com/(?:share|embed)/(?P[\da-f]{32})' _EMBED_REGEX = [rf']+\bsrc=["\'](?P{_VALID_URL})'] _TESTS = [{ # m3u8 raw-url, mp4 transcoded-url, cdn url == raw-url, json subs only 'url': 'https://www.loom.com/share/43d05f362f734614a2e81b4694a3a523', 'md5': 'bfc2d7e9c2e0eb4813212230794b6f42', 'info_dict': { 'id': '43d05f362f734614a2e81b4694a3a523', 'ext': 'mp4', 'title': 'A Ruler for Windows - 28 March 2022', 'uploader': 'wILLIAM PIP', 'upload_date': '20220328', 'timestamp': 1648454238, 'duration': 27, }, }, { # webm raw-url, mp4 transcoded-url, cdn url == transcoded-url, no subs 'url': 'https://www.loom.com/share/c43a642f815f4378b6f80a889bb73d8d', 'md5': '70f529317be8cf880fcc2c649a531900', 'info_dict': { 'id': 'c43a642f815f4378b6f80a889bb73d8d', 'ext': 'webm', 'title': 'Lilah Nielsen Intro Video', 'uploader': 'Lilah Nielsen', 'upload_date': '20200826', 'timestamp': 1598480716, 'duration': 20, }, }, { # m3u8 raw-url, mp4 transcoded-url, cdn url == raw-url, vtt sub and json subs 'url': 'https://www.loom.com/share/9458bcbf79784162aa62ffb8dd66201b', 'md5': '51737ec002969dd28344db4d60b9cbbb', 'info_dict': { 'id': '9458bcbf79784162aa62ffb8dd66201b', 'ext': 'mp4', 'title': 'Sharing screen with gpt-4', 'description': 'Sharing screen with GPT 4 vision model and asking questions to guide through blender.', 'uploader': 'Suneel Matham', 'chapters': 'count:3', 'upload_date': '20231109', 'timestamp': 1699518978, 'duration': 93, }, }, { # mpd raw-url, mp4 transcoded-url, cdn url == raw-url, no subs 'url': 'https://www.loom.com/share/24351eb8b317420289b158e4b7e96ff2', 'info_dict': { 'id': '24351eb8b317420289b158e4b7e96ff2', 'ext': 'webm', 'title': 'OMFG clown', 'description': 'md5:285c5ee9d62aa087b7e3271b08796815', 'uploader': 'MrPumkin B', 'upload_date': '20210924', 'timestamp': 1632519618, 'duration': 210, }, 'params': {'skip_download': 'dash'}, }, { # password-protected 'url': 'https://www.loom.com/share/50e26e8aeb7940189dff5630f95ce1f4', 'md5': '5cc7655e7d55d281d203f8ffd14771f7', 'info_dict': { 'id': '50e26e8aeb7940189dff5630f95ce1f4', 'ext': 'mp4', 'title': 'iOS Mobile Upload', 'uploader': 'Simon Curran', 'upload_date': '20200520', 'timestamp': 1590000123, 'duration': 35, }, 'params': {'videopassword': 'seniorinfants2'}, }, { # embed, transcoded-url endpoint sends empty JSON response 'url': 'https://www.loom.com/embed/ddcf1c1ad21f451ea7468b1e33917e4e', 'md5': '8488817242a0db1cb2ad0ea522553cf6', 'info_dict': { 'id': 'ddcf1c1ad21f451ea7468b1e33917e4e', 'ext': 'mp4', 'title': 'CF Reset User\'s Password', 'uploader': 'Aimee Heintz', 'upload_date': '20220707', 'timestamp': 1657216459, 'duration': 181, }, 'expected_warnings': ['Failed to parse JSON'], }] _WEBPAGE_TESTS = [{ 'url': 'https://www.loom.com/community/e1229802a8694a09909e8ba0fbb6d073-pg', 'md5': 'ec838cd01b576cf0386f32e1ae424609', 'info_dict': { 'id': 'e1229802a8694a09909e8ba0fbb6d073', 'ext': 'mp4', 'title': 'Rexie Jane Cimafranca - Founder\'s Presentation', 'uploader': 'Rexie Cimafranca', 'upload_date': '20230213', 'duration': 247, 'timestamp': 1676274030, }, }] _GRAPHQL_VARIABLES = { 'GetVideoSource': { 'acceptableMimes': ['DASH', 'M3U8', 'MP4'], }, } _GRAPHQL_QUERIES = { 'GetVideoSSR': textwrap.dedent('''\ query GetVideoSSR($videoId: ID!, $password: String) { getVideo(id: $videoId, password: $password) { __typename ... on PrivateVideo { id status message __typename } ... on VideoPasswordMissingOrIncorrect { id message __typename } ... on RegularUserVideo { id __typename createdAt description download_enabled folder_id is_protected needs_password owner { display_name __typename } privacy s3_id name video_properties { avgBitRate client camera_enabled client_version duration durationMs format height microphone_enabled os os_version recordingClient recording_type recording_version screen_type tab_audio trim_duration width __typename } playable_duration source_duration visibility } } }\n'''), 'GetVideoSource': textwrap.dedent('''\ query GetVideoSource($videoId: ID!, $password: String, $acceptableMimes: [CloudfrontVideoAcceptableMime]) { getVideo(id: $videoId, password: $password) { ... on RegularUserVideo { id nullableRawCdnUrl(acceptableMimes: $acceptableMimes, password: $password) { url __typename } __typename } __typename } }\n'''), 'FetchVideoTranscript': textwrap.dedent('''\ query FetchVideoTranscript($videoId: ID!, $password: String) { fetchVideoTranscript(videoId: $videoId, password: $password) { ... on VideoTranscriptDetails { id video_id source_url captions_source_url __typename } ... on GenericError { message __typename } __typename } }\n'''), 'FetchChapters': textwrap.dedent('''\ query FetchChapters($videoId: ID!, $password: String) { fetchVideoChapters(videoId: $videoId, password: $password) { ... on VideoChapters { video_id content __typename } ... on EmptyChaptersPayload { content __typename } ... on InvalidRequestWarning { message __typename } ... on Error { message __typename } __typename } }\n'''), } _APOLLO_GRAPHQL_VERSION = '0a1856c' def _call_graphql_api(self, operations, video_id, note=None, errnote=None): password = self.get_param('videopassword') return self._download_json( 'https://www.loom.com/graphql', video_id, note or 'Downloading GraphQL JSON', errnote or 'Failed to download GraphQL JSON', headers={ 'Accept': 'application/json', 'Content-Type': 'application/json', 'x-loom-request-source': f'loom_web_{self._APOLLO_GRAPHQL_VERSION}', 'apollographql-client-name': 'web', 'apollographql-client-version': self._APOLLO_GRAPHQL_VERSION, }, data=json.dumps([{ 'operationName': operation_name, 'variables': { 'videoId': video_id, 'password': password, **self._GRAPHQL_VARIABLES.get(operation_name, {}), }, 'query': self._GRAPHQL_QUERIES[operation_name], } for operation_name in variadic(operations)], separators=(',', ':')).encode()) def _call_url_api(self, endpoint, video_id): response = self._download_json( f'https://www.loom.com/api/campaigns/sessions/{video_id}/{endpoint}', video_id, f'Downloading {endpoint} JSON', f'Failed to download {endpoint} JSON', fatal=False, headers={'Accept': 'application/json', 'Content-Type': 'application/json'}, data=json.dumps({ 'anonID': str(uuid.uuid4()), 'deviceID': None, 'force_original': False, # HTTP error 401 if True 'password': self.get_param('videopassword'), }, separators=(',', ':')).encode()) return traverse_obj(response, ('url', {url_or_none})) def _extract_formats(self, video_id, metadata, gql_data): formats = [] video_properties = traverse_obj(metadata, ('video_properties', { 'width': ('width', {int_or_none}), 'height': ('height', {int_or_none}), 'acodec': ('microphone_enabled', {lambda x: 'none' if x is False else None}), })) def get_formats(format_url, format_id, quality): if not format_url: return ext = determine_ext(format_url) query = urllib.parse.urlparse(format_url).query if ext == 'm3u8': # Extract pre-merged HLS formats to avoid buggy parsing of metadata in split playlists format_url = format_url.replace('-split.m3u8', '.m3u8') m3u8_formats = self._extract_m3u8_formats( format_url, video_id, 'mp4', m3u8_id=f'hls-{format_id}', fatal=False, quality=quality) for fmt in m3u8_formats: yield { **fmt, 'url': update_url(fmt['url'], query=query), 'extra_param_to_segment_url': query, } elif ext == 'mpd': dash_formats = self._extract_mpd_formats( format_url, video_id, mpd_id=f'dash-{format_id}', fatal=False) for fmt in dash_formats: yield { **fmt, 'extra_param_to_segment_url': query, 'quality': quality, } else: yield { 'url': format_url, 'ext': ext, 'format_id': f'http-{format_id}', 'quality': quality, **video_properties, } raw_url = self._call_url_api('raw-url', video_id) formats.extend(get_formats(raw_url, 'raw', quality=1)) # original quality transcoded_url = self._call_url_api('transcoded-url', video_id) formats.extend(get_formats(transcoded_url, 'transcoded', quality=-1)) # transcoded quality cdn_url = get_first(gql_data, ('data', 'getVideo', 'nullableRawCdnUrl', 'url', {url_or_none})) # cdn_url is usually a dupe, but the raw-url/transcoded-url endpoints could return errors valid_urls = [update_url(url, query=None) for url in (raw_url, transcoded_url) if url] if cdn_url and update_url(cdn_url, query=None) not in valid_urls: formats.extend(get_formats(cdn_url, 'cdn', quality=0)) # could be original or transcoded return formats def _real_extract(self, url): video_id = self._match_id(url) metadata = get_first( self._call_graphql_api('GetVideoSSR', video_id, 'Downloading GraphQL metadata JSON'), ('data', 'getVideo', {dict})) or {} if metadata.get('__typename') == 'VideoPasswordMissingOrIncorrect': if not self.get_param('videopassword'): raise ExtractorError( 'This video is password-protected, use the --video-password option', expected=True) raise ExtractorError('Invalid video password', expected=True) gql_data = self._call_graphql_api(['FetchChapters', 'FetchVideoTranscript', 'GetVideoSource'], video_id) duration = traverse_obj(metadata, ('video_properties', 'duration', {int_or_none})) return { 'id': video_id, 'duration': duration, 'chapters': self._extract_chapters_from_description( get_first(gql_data, ('data', 'fetchVideoChapters', 'content', {str})), duration) or None, 'formats': self._extract_formats(video_id, metadata, gql_data), 'subtitles': filter_dict({ 'en': traverse_obj(gql_data, ( ..., 'data', 'fetchVideoTranscript', ('source_url', 'captions_source_url'), { 'url': {url_or_none}, })) or None, }), **traverse_obj(metadata, { 'title': ('name', {str}), 'description': ('description', {str}), 'uploader': ('owner', 'display_name', {str}), 'timestamp': ('createdAt', {parse_iso8601}), }), } class LoomFolderIE(InfoExtractor): IE_NAME = 'loom:folder' _VALID_URL = r'https?://(?:www\.)?loom\.com/share/folder/(?P[\da-f]{32})' _TESTS = [{ # 2 subfolders, no videos in root 'url': 'https://www.loom.com/share/folder/997db4db046f43e5912f10dc5f817b5c', 'playlist_mincount': 16, 'info_dict': { 'id': '997db4db046f43e5912f10dc5f817b5c', 'title': 'Blending Lessons', }, }, { # only videos, no subfolders 'url': 'https://www.loom.com/share/folder/9a8a87f6b6f546d9a400c8e7575ff7f2', 'playlist_mincount': 12, 'info_dict': { 'id': '9a8a87f6b6f546d9a400c8e7575ff7f2', 'title': 'List A- a, i, o', }, }, { # videos in root and empty subfolder 'url': 'https://www.loom.com/share/folder/886e534218c24fd292e97e9563078cc4', 'playlist_mincount': 21, 'info_dict': { 'id': '886e534218c24fd292e97e9563078cc4', 'title': 'Medicare Agent Training videos', }, }, { # videos in root and videos in subfolders 'url': 'https://www.loom.com/share/folder/b72c4ecdf04745da9403926d80a40c38', 'playlist_mincount': 21, 'info_dict': { 'id': 'b72c4ecdf04745da9403926d80a40c38', 'title': 'Quick Altos Q & A Tutorials', }, }, { # recursive folder extraction 'url': 'https://www.loom.com/share/folder/8b458a94e0e4449b8df9ea7a68fafc4e', 'playlist_count': 23, 'info_dict': { 'id': '8b458a94e0e4449b8df9ea7a68fafc4e', 'title': 'Sezer Texting Guide', }, }, { # more than 50 videos in 1 folder 'url': 'https://www.loom.com/share/folder/e056a91d290d47ca9b00c9d1df56c463', 'playlist_mincount': 61, 'info_dict': { 'id': 'e056a91d290d47ca9b00c9d1df56c463', 'title': 'User Videos', }, }, { # many subfolders 'url': 'https://www.loom.com/share/folder/c2dde8cc67454f0e99031677279d8954', 'playlist_mincount': 75, 'info_dict': { 'id': 'c2dde8cc67454f0e99031677279d8954', 'title': 'Honors 1', }, }, { 'url': 'https://www.loom.com/share/folder/bae17109a68146c7803454f2893c8cf8/Edpuzzle', 'only_matching': True, }] def _extract_folder_data(self, folder_id): return self._download_json( f'https://www.loom.com/v1/folders/{folder_id}', folder_id, 'Downloading folder info JSON', query={'limit': '10000'}) def _extract_folder_entries(self, folder_id, initial_folder_data=None): folder_data = initial_folder_data or self._extract_folder_data(folder_id) for video in traverse_obj(folder_data, ('videos', lambda _, v: v['id'])): video_id = video['id'] yield self.url_result( f'https://www.loom.com/share/{video_id}', LoomIE, video_id, video.get('name')) # Recurse into subfolders for subfolder_id in traverse_obj(folder_data, ( 'folders', lambda _, v: v['id'] != folder_id, 'id', {str})): yield from self._extract_folder_entries(subfolder_id) def _real_extract(self, url): playlist_id = self._match_id(url) playlist_data = self._extract_folder_data(playlist_id) return self.playlist_result( self._extract_folder_entries(playlist_id, playlist_data), playlist_id, traverse_obj(playlist_data, ('folder', 'name', {str.strip})))