import base64 import functools import re from .common import InfoExtractor from ..utils import ( clean_html, determine_ext, float_or_none, int_or_none, make_archive_id, mimetype2ext, orderedSet, parse_age_limit, parse_iso8601, remove_end, str_or_none, strip_jsonp, try_call, unified_strdate, url_or_none, ) from ..utils.traversal import traverse_obj class ORFRadioIE(InfoExtractor): IE_NAME = 'orf:radio' STATION_INFO = { 'fm4': ('fm4', 'fm4', 'orffm4'), 'noe': ('noe', 'oe2n', 'orfnoe'), 'wien': ('wie', 'oe2w', 'orfwie'), 'burgenland': ('bgl', 'oe2b', 'orfbgl'), 'ooe': ('ooe', 'oe2o', 'orfooe'), 'steiermark': ('stm', 'oe2st', 'orfstm'), 'kaernten': ('ktn', 'oe2k', 'orfktn'), 'salzburg': ('sbg', 'oe2s', 'orfsbg'), 'tirol': ('tir', 'oe2t', 'orftir'), 'vorarlberg': ('vbg', 'oe2v', 'orfvbg'), 'oe3': ('oe3', 'oe3', 'orfoe3'), 'oe1': ('oe1', 'oe1', 'orfoe1'), } _STATION_RE = '|'.join(map(re.escape, STATION_INFO.keys())) _VALID_URL = rf'''(?x) https?://(?: (?P{_STATION_RE})\.orf\.at/player| radiothek\.orf\.at/(?P{_STATION_RE}) )/(?P[0-9]+)/(?P\w+)''' _TESTS = [{ 'url': 'https://radiothek.orf.at/ooe/20220801/OGMO', 'info_dict': { 'id': 'OGMO', 'title': 'Guten Morgen OÖ', 'description': 'md5:a3f6083399ef92b8cbe2d421b180835a', }, 'playlist': [{ 'md5': 'f33147d954a326e338ea52572c2810e8', 'info_dict': { 'id': '2022-08-01_0459_tl_66_7DaysMon1_319062', 'ext': 'mp3', 'title': 'Guten Morgen OÖ', 'upload_date': '20220801', 'duration': 18000, 'timestamp': 1659322789, 'description': 'md5:a3f6083399ef92b8cbe2d421b180835a', }, }], }, { 'url': 'https://ooe.orf.at/player/20220801/OGMO', 'info_dict': { 'id': 'OGMO', 'title': 'Guten Morgen OÖ', 'description': 'md5:a3f6083399ef92b8cbe2d421b180835a', }, 'playlist': [{ 'md5': 'f33147d954a326e338ea52572c2810e8', 'info_dict': { 'id': '2022-08-01_0459_tl_66_7DaysMon1_319062', 'ext': 'mp3', 'title': 'Guten Morgen OÖ', 'upload_date': '20220801', 'duration': 18000, 'timestamp': 1659322789, 'description': 'md5:a3f6083399ef92b8cbe2d421b180835a', }, }], }, { 'url': 'http://fm4.orf.at/player/20170107/4CC', 'only_matching': True, }, { 'url': 'https://noe.orf.at/player/20200423/NGM', 'only_matching': True, }, { 'url': 'https://wien.orf.at/player/20200423/WGUM', 'only_matching': True, }, { 'url': 'https://burgenland.orf.at/player/20200423/BGM', 'only_matching': True, }, { 'url': 'https://steiermark.orf.at/player/20200423/STGMS', 'only_matching': True, }, { 'url': 'https://kaernten.orf.at/player/20200423/KGUMO', 'only_matching': True, }, { 'url': 'https://salzburg.orf.at/player/20200423/SGUM', 'only_matching': True, }, { 'url': 'https://tirol.orf.at/player/20200423/TGUMO', 'only_matching': True, }, { 'url': 'https://vorarlberg.orf.at/player/20200423/VGUM', 'only_matching': True, }, { 'url': 'https://oe3.orf.at/player/20200424/3WEK', 'only_matching': True, }, { 'url': 'http://oe1.orf.at/player/20170108/456544', 'md5': '34d8a6e67ea888293741c86a099b745b', 'info_dict': { 'id': '2017-01-08_0759_tl_51_7DaysSun6_256141', 'ext': 'mp3', 'title': 'Morgenjournal', 'duration': 609, 'timestamp': 1483858796, 'upload_date': '20170108', }, 'skip': 'Shows from ORF radios are only available for 7 days.', }] def _entries(self, data, station): _, loop_station, old_ie = self.STATION_INFO[station] for info in data['streams']: item_id = info.get('loopStreamId') if not item_id: continue video_id = item_id.replace('.mp3', '') yield { 'id': video_id, 'ext': 'mp3', 'url': f'https://loopstream01.apa.at/?channel={loop_station}&id={item_id}', '_old_archive_ids': [make_archive_id(old_ie, video_id)], 'title': data.get('title'), 'description': clean_html(data.get('subtitle')), 'duration': try_call(lambda: (info['end'] - info['start']) / 1000), 'timestamp': int_or_none(info.get('start'), scale=1000), 'series': data.get('programTitle'), } def _real_extract(self, url): station, station2, show_date, show_id = self._match_valid_url(url).group('station', 'station2', 'date', 'show') api_station, _, _ = self.STATION_INFO[station or station2] data = self._download_json( f'http://audioapi.orf.at/{api_station}/api/json/current/broadcast/{show_id}/{show_date}', show_id) return self.playlist_result( self._entries(data, station or station2), show_id, data.get('title'), clean_html(data.get('subtitle'))) class ORFPodcastIE(InfoExtractor): IE_NAME = 'orf:podcast' _STATION_RE = '|'.join(map(re.escape, ( 'bgl', 'fm4', 'ktn', 'noe', 'oe1', 'oe3', 'ooe', 'sbg', 'stm', 'tir', 'tv', 'vbg', 'wie'))) _VALID_URL = rf'https?://sound\.orf\.at/podcast/(?P{_STATION_RE})/(?P[\w-]+)/(?P[\w-]+)' _TESTS = [{ 'url': 'https://sound.orf.at/podcast/oe3/fruehstueck-bei-mir/nicolas-stockhammer-15102023', 'md5': '526a5700e03d271a1505386a8721ab9b', 'info_dict': { 'id': 'nicolas-stockhammer-15102023', 'ext': 'mp3', 'title': 'Nicolas Stockhammer (15.10.2023)', 'duration': 3396.0, 'series': 'Frühstück bei mir', }, 'skip': 'ORF podcasts are only available for a limited time', }] def _real_extract(self, url): station, show, show_id = self._match_valid_url(url).group('station', 'show', 'id') data = self._download_json( f'https://audioapi.orf.at/radiothek/api/2.0/podcast/{station}/{show}/{show_id}', show_id) return { 'id': show_id, 'ext': 'mp3', 'vcodec': 'none', **traverse_obj(data, ('payload', { 'url': ('enclosures', 0, 'url'), 'ext': ('enclosures', 0, 'type', {mimetype2ext}), 'title': 'title', 'description': ('description', {clean_html}), 'duration': ('duration', {functools.partial(float_or_none, scale=1000)}), 'series': ('podcast', 'title'), })), } class ORFIPTVIE(InfoExtractor): IE_NAME = 'orf:iptv' IE_DESC = 'iptv.ORF.at' _VALID_URL = r'https?://iptv\.orf\.at/(?:#/)?stories/(?P\d+)' _TEST = { 'url': 'http://iptv.orf.at/stories/2275236/', 'md5': 'c8b22af4718a4b4af58342529453e3e5', 'info_dict': { 'id': '350612', 'ext': 'flv', 'title': 'Weitere Evakuierungen um Vulkan Calbuco', 'description': 'md5:d689c959bdbcf04efeddedbf2299d633', 'duration': 68.197, 'thumbnail': r're:^https?://.*\.jpg$', 'upload_date': '20150425', }, } def _real_extract(self, url): story_id = self._match_id(url) webpage = self._download_webpage( f'http://iptv.orf.at/stories/{story_id}', story_id) video_id = self._search_regex( r'data-video(?:id)?="(\d+)"', webpage, 'video id') data = self._download_json( f'http://bits.orf.at/filehandler/static-api/json/current/data.json?file={video_id}', video_id)[0] duration = float_or_none(data['duration'], 1000) video = data['sources']['default'] load_balancer_url = video['loadBalancerUrl'] abr = int_or_none(video.get('audioBitrate')) vbr = int_or_none(video.get('bitrate')) fps = int_or_none(video.get('videoFps')) width = int_or_none(video.get('videoWidth')) height = int_or_none(video.get('videoHeight')) thumbnail = video.get('preview') rendition = self._download_json( load_balancer_url, video_id, transform_source=strip_jsonp) f = { 'abr': abr, 'vbr': vbr, 'fps': fps, 'width': width, 'height': height, } formats = [] for format_id, format_url in rendition['redirect'].items(): if format_id == 'rtmp': ff = f.copy() ff.update({ 'url': format_url, 'format_id': format_id, }) formats.append(ff) elif determine_ext(format_url) == 'f4m': formats.extend(self._extract_f4m_formats( format_url, video_id, f4m_id=format_id)) elif determine_ext(format_url) == 'm3u8': formats.extend(self._extract_m3u8_formats( format_url, video_id, 'mp4', m3u8_id=format_id)) else: continue title = remove_end(self._og_search_title(webpage), ' - iptv.ORF.at') description = self._og_search_description(webpage) upload_date = unified_strdate(self._html_search_meta( 'dc.date', webpage, 'upload date')) return { 'id': video_id, 'title': title, 'description': description, 'duration': duration, 'thumbnail': thumbnail, 'upload_date': upload_date, 'formats': formats, } class ORFFM4StoryIE(InfoExtractor): IE_NAME = 'orf:fm4:story' IE_DESC = 'fm4.orf.at stories' _VALID_URL = r'https?://fm4\.orf\.at/stories/(?P\d+)' _TEST = { 'url': 'http://fm4.orf.at/stories/2865738/', 'playlist': [{ 'md5': 'e1c2c706c45c7b34cf478bbf409907ca', 'info_dict': { 'id': '547792', 'ext': 'flv', 'title': 'Manu Delago und Inner Tongue live', 'description': 'Manu Delago und Inner Tongue haben bei der FM4 Soundpark Session live alles gegeben. Hier gibt es Fotos und die gesamte Session als Video.', 'duration': 1748.52, 'thumbnail': r're:^https?://.*\.jpg$', 'upload_date': '20170913', }, }, { 'md5': 'c6dd2179731f86f4f55a7b49899d515f', 'info_dict': { 'id': '547798', 'ext': 'flv', 'title': 'Manu Delago und Inner Tongue live (2)', 'duration': 1504.08, 'thumbnail': r're:^https?://.*\.jpg$', 'upload_date': '20170913', 'description': 'Manu Delago und Inner Tongue haben bei der FM4 Soundpark Session live alles gegeben. Hier gibt es Fotos und die gesamte Session als Video.', }, }], } def _real_extract(self, url): story_id = self._match_id(url) webpage = self._download_webpage(url, story_id) entries = [] all_ids = orderedSet(re.findall(r'data-video(?:id)?="(\d+)"', webpage)) for idx, video_id in enumerate(all_ids): data = self._download_json( f'http://bits.orf.at/filehandler/static-api/json/current/data.json?file={video_id}', video_id)[0] duration = float_or_none(data['duration'], 1000) video = data['sources']['q8c'] load_balancer_url = video['loadBalancerUrl'] abr = int_or_none(video.get('audioBitrate')) vbr = int_or_none(video.get('bitrate')) fps = int_or_none(video.get('videoFps')) width = int_or_none(video.get('videoWidth')) height = int_or_none(video.get('videoHeight')) thumbnail = video.get('preview') rendition = self._download_json( load_balancer_url, video_id, transform_source=strip_jsonp) f = { 'abr': abr, 'vbr': vbr, 'fps': fps, 'width': width, 'height': height, } formats = [] for format_id, format_url in rendition['redirect'].items(): if format_id == 'rtmp': ff = f.copy() ff.update({ 'url': format_url, 'format_id': format_id, }) formats.append(ff) elif determine_ext(format_url) == 'f4m': formats.extend(self._extract_f4m_formats( format_url, video_id, f4m_id=format_id)) elif determine_ext(format_url) == 'm3u8': formats.extend(self._extract_m3u8_formats( format_url, video_id, 'mp4', m3u8_id=format_id)) else: continue title = remove_end(self._og_search_title(webpage), ' - fm4.ORF.at') if idx >= 1: # Titles are duplicates, make them unique title += ' (' + str(idx + 1) + ')' description = self._og_search_description(webpage) upload_date = unified_strdate(self._html_search_meta( 'dc.date', webpage, 'upload date')) entries.append({ 'id': video_id, 'title': title, 'description': description, 'duration': duration, 'thumbnail': thumbnail, 'upload_date': upload_date, 'formats': formats, }) return self.playlist_result(entries) class ORFONIE(InfoExtractor): IE_NAME = 'orf:on' _VALID_URL = r'https?://on\.orf\.at/video/(?P\d+)(?:/(?P\d+))?' _TESTS = [{ 'url': 'https://on.orf.at/video/14210000/school-of-champions-48', 'info_dict': { 'id': '14210000', 'ext': 'mp4', 'duration': 2651.08, 'thumbnail': 'https://api-tvthek.orf.at/assets/segments/0167/98/thumb_16697671_segments_highlight_teaser.jpeg', 'title': 'School of Champions (4/8)', 'description': 'md5:d09ad279fc2e8502611e7648484b6afd', 'media_type': 'episode', 'timestamp': 1706558922, 'upload_date': '20240129', 'release_timestamp': 1706472362, 'release_date': '20240128', 'modified_timestamp': 1712756663, 'modified_date': '20240410', '_old_archive_ids': ['orftvthek 14210000'], }, }, { 'url': 'https://on.orf.at/video/3220355', 'md5': 'f94d98e667cf9a3851317efb4e136662', 'info_dict': { 'id': '3220355', 'ext': 'mp4', 'duration': 445.04, 'thumbnail': 'https://api-tvthek.orf.at/assets/segments/0002/60/thumb_159573_segments_highlight_teaser.png', 'title': '50 Jahre Burgenland: Der Festumzug', 'description': 'md5:1560bf855119544ee8c4fa5376a2a6b0', 'media_type': 'episode', 'timestamp': 52916400, 'upload_date': '19710905', 'release_timestamp': 52916400, 'release_date': '19710905', 'modified_timestamp': 1498536049, 'modified_date': '20170627', '_old_archive_ids': ['orftvthek 3220355'], }, }, { # Video with multiple segments selecting the second segment 'url': 'https://on.orf.at/video/14226549/15639808/jugendbande-einbrueche-aus-langeweile', 'md5': '90f4ebff86b4580837b8a361d0232a9e', 'info_dict': { 'id': '15639808', 'ext': 'mp4', 'duration': 97.707, 'thumbnail': 'https://api-tvthek.orf.at/assets/segments/0175/43/thumb_17442704_segments_highlight_teaser.jpg', 'title': 'Jugendbande: Einbrüche aus Langeweile', 'description': 'md5:193df0bf0d91cf16830c211078097120', 'media_type': 'segment', 'timestamp': 1715792400, 'upload_date': '20240515', 'modified_timestamp': 1715794394, 'modified_date': '20240515', '_old_archive_ids': ['orftvthek 15639808'], }, 'params': {'noplaylist': True}, }, { # Video with multiple segments and no combined version 'url': 'https://on.orf.at/video/14227864/formel-1-grosser-preis-von-monaco-2024', 'info_dict': { '_type': 'multi_video', 'id': '14227864', 'duration': 18410.52, 'thumbnail': 'https://api-tvthek.orf.at/assets/segments/0176/04/thumb_17503881_segments_highlight_teaser.jpg', 'title': 'Formel 1: Großer Preis von Monaco 2024', 'description': 'md5:aeeb010710ccf70ce28ccb4482243d4f', 'media_type': 'episode', 'timestamp': 1716721200, 'upload_date': '20240526', 'release_timestamp': 1716721802, 'release_date': '20240526', 'modified_timestamp': 1716967501, 'modified_date': '20240529', }, 'playlist_count': 42, }, { # Video with multiple segments, but with combined version 'url': 'https://on.orf.at/video/14228172', 'info_dict': { 'id': '14228172', 'ext': 'mp4', 'duration': 3294.878, 'thumbnail': 'https://api-tvthek.orf.at/assets/segments/0176/17/thumb_17516455_segments_highlight_teaser.jpg', 'title': 'Willkommen Österreich mit Stermann & Grissemann', 'description': 'md5:5de034d033a9c27f989343be3bbd4839', 'media_type': 'episode', 'timestamp': 1716926584, 'upload_date': '20240528', 'release_timestamp': 1716919202, 'release_date': '20240528', 'modified_timestamp': 1716968045, 'modified_date': '20240529', '_old_archive_ids': ['orftvthek 14228172'], }, }] @staticmethod def _parse_metadata(api_json): return traverse_obj(api_json, { 'id': ('id', {int}, {str_or_none}), 'age_limit': ('age_classification', {parse_age_limit}), 'duration': ('exact_duration', {functools.partial(float_or_none, scale=1000)}), 'title': (('title', 'headline'), {str}), 'description': (('description', 'teaser_text'), {str}), 'media_type': ('video_type', {str}), 'thumbnail': ('_embedded', 'image', 'public_urls', 'highlight_teaser', 'url', {url_or_none}), 'timestamp': (('date', 'episode_date'), {parse_iso8601}), 'release_timestamp': ('release_date', {parse_iso8601}), 'modified_timestamp': ('updated_at', {parse_iso8601}), }, get_all=False) def _extract_video_info(self, video_id, api_json): formats, subtitles = [], {} for manifest_type in traverse_obj(api_json, ('sources', {dict.keys}, ...)): for manifest_url in traverse_obj(api_json, ('sources', manifest_type, ..., 'src', {url_or_none})): if manifest_type == 'hls': fmts, subs = self._extract_m3u8_formats_and_subtitles( manifest_url, video_id, fatal=False, m3u8_id='hls') elif manifest_type == 'dash': fmts, subs = self._extract_mpd_formats_and_subtitles( manifest_url, video_id, fatal=False, mpd_id='dash') else: continue formats.extend(fmts) self._merge_subtitles(subs, target=subtitles) for sub_url in traverse_obj(api_json, ( '_embedded', 'subtitle', ('xml_url', 'sami_url', 'stl_url', 'ttml_url', 'srt_url', 'vtt_url'), {url_or_none})): self._merge_subtitles({'de': [{'url': sub_url}]}, target=subtitles) return { 'id': video_id, 'formats': formats, 'subtitles': subtitles, '_old_archive_ids': [make_archive_id('ORFTVthek', video_id)], **self._parse_metadata(api_json), } def _real_extract(self, url): video_id, segment_id = self._match_valid_url(url).group('id', 'segment') encrypted_id = base64.b64encode(f'3dSlfek03nsLKdj4Jsd{video_id}'.encode()).decode() api_json = self._download_json( f'https://api-tvthek.orf.at/api/v4.3/public/episode/encrypted/{encrypted_id}', video_id) if traverse_obj(api_json, 'is_drm_protected'): self.report_drm(video_id) segments = traverse_obj(api_json, ('_embedded', 'segments', lambda _, v: v['id'])) selected_segment = traverse_obj(segments, (lambda _, v: str(v['id']) == segment_id, any)) # selected_segment will be falsy if input URL did not include a valid segment_id if selected_segment and not self._yes_playlist(video_id, segment_id, playlist_label='episode', video_label='segment'): return self._extract_video_info(segment_id, selected_segment) # Even some segmented videos have an unsegmented version available in API response root if (self._configuration_arg('prefer_segments_playlist') or not traverse_obj(api_json, ('sources', ..., ..., 'src', {url_or_none}))): return self.playlist_result( (self._extract_video_info(str(segment['id']), segment) for segment in segments), video_id, **self._parse_metadata(api_json), multi_video=True) return self._extract_video_info(video_id, api_json)