# coding: utf-8 from __future__ import unicode_literals import json import uuid from .common import InfoExtractor from ..compat import compat_HTTPError from ..utils import ( determine_ext, ExtractorError, float_or_none, int_or_none, strip_or_none, try_get, unified_timestamp, ) class DPlayBaseIE(InfoExtractor): _PATH_REGEX = r'/(?P[^/]+/[^/?#]+)' _auth_token_cache = {} def _get_auth(self, disco_base, display_id, realm, needs_device_id=True): key = (disco_base, realm) st = self._get_cookies(disco_base).get('st') token = (st and st.value) or self._auth_token_cache.get(key) if not token: query = {'realm': realm} if needs_device_id: query['deviceId'] = uuid.uuid4().hex token = self._download_json( disco_base + 'token', display_id, 'Downloading token', query=query)['data']['attributes']['token'] # Save cache only if cookies are not being set if not self._get_cookies(disco_base).get('st'): self._auth_token_cache[key] = token return f'Bearer {token}' def _process_errors(self, e, geo_countries): info = self._parse_json(e.cause.read().decode('utf-8'), None) error = info['errors'][0] error_code = error.get('code') if error_code == 'access.denied.geoblocked': self.raise_geo_restricted(countries=geo_countries) elif error_code in ('access.denied.missingpackage', 'invalid.token'): raise ExtractorError( 'This video is only available for registered users. You may want to use --cookies.', expected=True) raise ExtractorError(info['errors'][0]['detail'], expected=True) def _update_disco_api_headers(self, headers, disco_base, display_id, realm): headers['Authorization'] = self._get_auth(disco_base, display_id, realm, False) def _download_video_playback_info(self, disco_base, video_id, headers): streaming = self._download_json( disco_base + 'playback/videoPlaybackInfo/' + video_id, video_id, headers=headers)['data']['attributes']['streaming'] streaming_list = [] for format_id, format_dict in streaming.items(): streaming_list.append({ 'type': format_id, 'url': format_dict.get('url'), }) return streaming_list def _get_disco_api_info(self, url, display_id, disco_host, realm, country, domain=''): geo_countries = [country.upper()] self._initialize_geo_bypass({ 'countries': geo_countries, }) disco_base = 'https://%s/' % disco_host headers = { 'Referer': url, } self._update_disco_api_headers(headers, disco_base, display_id, realm) try: video = self._download_json( disco_base + 'content/videos/' + display_id, display_id, headers=headers, query={ 'fields[channel]': 'name', 'fields[image]': 'height,src,width', 'fields[show]': 'name', 'fields[tag]': 'name', 'fields[video]': 'description,episodeNumber,name,publishStart,seasonNumber,videoDuration', 'include': 'images,primaryChannel,show,tags' }) except ExtractorError as e: if isinstance(e.cause, compat_HTTPError) and e.cause.code == 400: self._process_errors(e, geo_countries) raise video_id = video['data']['id'] info = video['data']['attributes'] title = info['name'].strip() formats = [] subtitles = {} try: streaming = self._download_video_playback_info( disco_base, video_id, headers) except ExtractorError as e: if isinstance(e.cause, compat_HTTPError) and e.cause.code == 403: self._process_errors(e, geo_countries) raise for format_dict in streaming: if not isinstance(format_dict, dict): continue format_url = format_dict.get('url') if not format_url: continue format_id = format_dict.get('type') ext = determine_ext(format_url) if format_id == 'dash' or ext == 'mpd': dash_fmts, dash_subs = self._extract_mpd_formats_and_subtitles( format_url, display_id, mpd_id='dash', fatal=False) formats.extend(dash_fmts) subtitles = self._merge_subtitles(subtitles, dash_subs) elif format_id == 'hls' or ext == 'm3u8': m3u8_fmts, m3u8_subs = self._extract_m3u8_formats_and_subtitles( format_url, display_id, 'mp4', entry_protocol='m3u8_native', m3u8_id='hls', fatal=False) formats.extend(m3u8_fmts) subtitles = self._merge_subtitles(subtitles, m3u8_subs) else: formats.append({ 'url': format_url, 'format_id': format_id, }) self._sort_formats(formats) creator = series = None tags = [] thumbnails = [] included = video.get('included') or [] if isinstance(included, list): for e in included: attributes = e.get('attributes') if not attributes: continue e_type = e.get('type') if e_type == 'channel': creator = attributes.get('name') elif e_type == 'image': src = attributes.get('src') if src: thumbnails.append({ 'url': src, 'width': int_or_none(attributes.get('width')), 'height': int_or_none(attributes.get('height')), }) if e_type == 'show': series = attributes.get('name') elif e_type == 'tag': name = attributes.get('name') if name: tags.append(name) return { 'id': video_id, 'display_id': display_id, 'title': title, 'description': strip_or_none(info.get('description')), 'duration': float_or_none(info.get('videoDuration'), 1000), 'timestamp': unified_timestamp(info.get('publishStart')), 'series': series, 'season_number': int_or_none(info.get('seasonNumber')), 'episode_number': int_or_none(info.get('episodeNumber')), 'creator': creator, 'tags': tags, 'thumbnails': thumbnails, 'formats': formats, 'subtitles': subtitles, 'http_headers': { 'referer': domain, }, } class DPlayIE(DPlayBaseIE): _VALID_URL = r'''(?x)https?:// (?P (?:www\.)?(?Pd (?: play\.(?Pdk|fi|jp|se|no)| iscoveryplus\.(?Pdk|es|fi|it|se|no) ) )| (?Pes|it)\.dplay\.com )/[^/]+''' + DPlayBaseIE._PATH_REGEX _TESTS = [{ # non geo restricted, via secure api, unsigned download hls URL 'url': 'https://www.dplay.se/videos/nugammalt-77-handelser-som-format-sverige/nugammalt-77-handelser-som-format-sverige-101', 'info_dict': { 'id': '13628', 'display_id': 'nugammalt-77-handelser-som-format-sverige/nugammalt-77-handelser-som-format-sverige-101', 'ext': 'mp4', 'title': 'Svensken lär sig njuta av livet', 'description': 'md5:d3819c9bccffd0fe458ca42451dd50d8', 'duration': 2649.856, 'timestamp': 1365453720, 'upload_date': '20130408', 'creator': 'Kanal 5', 'series': 'Nugammalt - 77 händelser som format Sverige', 'season_number': 1, 'episode_number': 1, }, 'params': { 'skip_download': True, }, }, { # geo restricted, via secure api, unsigned download hls URL 'url': 'http://www.dplay.dk/videoer/ted-bundy-mind-of-a-monster/ted-bundy-mind-of-a-monster', 'info_dict': { 'id': '104465', 'display_id': 'ted-bundy-mind-of-a-monster/ted-bundy-mind-of-a-monster', 'ext': 'mp4', 'title': 'Ted Bundy: Mind Of A Monster', 'description': 'md5:8b780f6f18de4dae631668b8a9637995', 'duration': 5290.027, 'timestamp': 1570694400, 'upload_date': '20191010', 'creator': 'ID - Investigation Discovery', 'series': 'Ted Bundy: Mind Of A Monster', 'season_number': 1, 'episode_number': 1, }, 'params': { 'skip_download': True, }, }, { # disco-api 'url': 'https://www.dplay.no/videoer/i-kongens-klr/sesong-1-episode-7', 'info_dict': { 'id': '40206', 'display_id': 'i-kongens-klr/sesong-1-episode-7', 'ext': 'mp4', 'title': 'Episode 7', 'description': 'md5:e3e1411b2b9aebeea36a6ec5d50c60cf', 'duration': 2611.16, 'timestamp': 1516726800, 'upload_date': '20180123', 'series': 'I kongens klær', 'season_number': 1, 'episode_number': 7, }, 'params': { 'skip_download': True, }, 'skip': 'Available for Premium users', }, { 'url': 'http://it.dplay.com/nove/biografie-imbarazzanti/luigi-di-maio-la-psicosi-di-stanislawskij/', 'md5': '2b808ffb00fc47b884a172ca5d13053c', 'info_dict': { 'id': '6918', 'display_id': 'biografie-imbarazzanti/luigi-di-maio-la-psicosi-di-stanislawskij', 'ext': 'mp4', 'title': 'Luigi Di Maio: la psicosi di Stanislawskij', 'description': 'md5:3c7a4303aef85868f867a26f5cc14813', 'thumbnail': r're:^https?://.*\.jpe?g', 'upload_date': '20160524', 'timestamp': 1464076800, 'series': 'Biografie imbarazzanti', 'season_number': 1, 'episode': 'Episode 1', 'episode_number': 1, }, }, { 'url': 'https://es.dplay.com/dmax/la-fiebre-del-oro/temporada-8-episodio-1/', 'info_dict': { 'id': '21652', 'display_id': 'la-fiebre-del-oro/temporada-8-episodio-1', 'ext': 'mp4', 'title': 'Episodio 1', 'description': 'md5:b9dcff2071086e003737485210675f69', 'thumbnail': r're:^https?://.*\.png', 'upload_date': '20180709', 'timestamp': 1531173540, 'series': 'La fiebre del oro', 'season_number': 8, 'episode': 'Episode 1', 'episode_number': 1, }, 'params': { 'skip_download': True, }, }, { 'url': 'https://www.dplay.fi/videot/shifting-gears-with-aaron-kaufman/episode-16', 'only_matching': True, }, { 'url': 'https://www.dplay.jp/video/gold-rush/24086', 'only_matching': True, }, { 'url': 'https://www.discoveryplus.se/videos/nugammalt-77-handelser-som-format-sverige/nugammalt-77-handelser-som-format-sverige-101', 'only_matching': True, }, { 'url': 'https://www.discoveryplus.dk/videoer/ted-bundy-mind-of-a-monster/ted-bundy-mind-of-a-monster', 'only_matching': True, }, { 'url': 'https://www.discoveryplus.no/videoer/i-kongens-klr/sesong-1-episode-7', 'only_matching': True, }, { 'url': 'https://www.discoveryplus.it/videos/biografie-imbarazzanti/luigi-di-maio-la-psicosi-di-stanislawskij', 'only_matching': True, }, { 'url': 'https://www.discoveryplus.es/videos/la-fiebre-del-oro/temporada-8-episodio-1', 'only_matching': True, }, { 'url': 'https://www.discoveryplus.fi/videot/shifting-gears-with-aaron-kaufman/episode-16', 'only_matching': True, }] def _real_extract(self, url): mobj = self._match_valid_url(url) display_id = mobj.group('id') domain = mobj.group('domain').lstrip('www.') country = mobj.group('country') or mobj.group('subdomain_country') or mobj.group('plus_country') host = 'disco-api.' + domain if domain[0] == 'd' else 'eu2-prod.disco-api.com' return self._get_disco_api_info( url, display_id, host, 'dplay' + country, country, domain) class HGTVDeIE(DPlayBaseIE): _VALID_URL = r'https?://de\.hgtv\.com/sendungen' + DPlayBaseIE._PATH_REGEX _TESTS = [{ 'url': 'https://de.hgtv.com/sendungen/tiny-house-klein-aber-oho/wer-braucht-schon-eine-toilette/', 'info_dict': { 'id': '151205', 'display_id': 'tiny-house-klein-aber-oho/wer-braucht-schon-eine-toilette', 'ext': 'mp4', 'title': 'Wer braucht schon eine Toilette', 'description': 'md5:05b40a27e7aed2c9172de34d459134e2', 'duration': 1177.024, 'timestamp': 1595705400, 'upload_date': '20200725', 'creator': 'HGTV', 'series': 'Tiny House - klein, aber oho', 'season_number': 3, 'episode_number': 3, }, }] def _real_extract(self, url): display_id = self._match_id(url) return self._get_disco_api_info( url, display_id, 'eu1-prod.disco-api.com', 'hgtv', 'de') class DiscoveryPlusIE(DPlayBaseIE): _VALID_URL = r'https?://(?:www\.)?discoveryplus\.com/(?:\w{2}/)?video' + DPlayBaseIE._PATH_REGEX _TESTS = [{ 'url': 'https://www.discoveryplus.com/video/property-brothers-forever-home/food-and-family', 'info_dict': { 'id': '1140794', 'display_id': 'property-brothers-forever-home/food-and-family', 'ext': 'mp4', 'title': 'Food and Family', 'description': 'The brothers help a Richmond family expand their single-level home.', 'duration': 2583.113, 'timestamp': 1609304400, 'upload_date': '20201230', 'creator': 'HGTV', 'series': 'Property Brothers: Forever Home', 'season_number': 1, 'episode_number': 1, }, 'skip': 'Available for Premium users', }, { 'url': 'https://discoveryplus.com/ca/video/bering-sea-gold-discovery-ca/goldslingers', 'only_matching': True, }] _PRODUCT = 'dplus_us' _API_URL = 'us1-prod-direct.discoveryplus.com' def _update_disco_api_headers(self, headers, disco_base, display_id, realm): headers['x-disco-client'] = f'WEB:UNKNOWN:{self._PRODUCT}:25.2.6' def _download_video_playback_info(self, disco_base, video_id, headers): return self._download_json( disco_base + 'playback/v3/videoPlaybackInfo', video_id, headers=headers, data=json.dumps({ 'deviceInfo': { 'adBlocker': False, }, 'videoId': video_id, 'wisteriaProperties': { 'platform': 'desktop', 'product': self._PRODUCT, }, }).encode('utf-8'))['data']['attributes']['streaming'] def _real_extract(self, url): display_id = self._match_id(url) return self._get_disco_api_info( url, display_id, self._API_URL, 'go', 'us') class ScienceChannelIE(DiscoveryPlusIE): _VALID_URL = r'https?://(?:www\.)?sciencechannel\.com/video' + DPlayBaseIE._PATH_REGEX _TESTS = [{ 'url': 'https://www.sciencechannel.com/video/strangest-things-science-atve-us/nazi-mystery-machine', 'info_dict': { 'id': '2842849', 'display_id': 'strangest-things-science-atve-us/nazi-mystery-machine', 'ext': 'mp4', 'title': 'Nazi Mystery Machine', 'description': 'Experts investigate the secrets of a revolutionary encryption machine.', 'season_number': 1, 'episode_number': 1, }, 'skip': 'Available for Premium users', }] _PRODUCT = 'sci' _API_URL = 'us1-prod-direct.sciencechannel.com' class DIYNetworkIE(DiscoveryPlusIE): _VALID_URL = r'https?://(?:watch\.)?diynetwork\.com/video' + DPlayBaseIE._PATH_REGEX _TESTS = [{ 'url': 'https://watch.diynetwork.com/video/pool-kings-diy-network/bringing-beach-life-to-texas', 'info_dict': { 'id': '2309730', 'display_id': 'pool-kings-diy-network/bringing-beach-life-to-texas', 'ext': 'mp4', 'title': 'Bringing Beach Life to Texas', 'description': 'The Pool Kings give a family a day at the beach in their own backyard.', 'season_number': 10, 'episode_number': 2, }, 'skip': 'Available for Premium users', }] _PRODUCT = 'diy' _API_URL = 'us1-prod-direct.watch.diynetwork.com' class AnimalPlanetIE(DiscoveryPlusIE): _VALID_URL = r'https?://(?:www\.)?animalplanet\.com/video' + DPlayBaseIE._PATH_REGEX _TESTS = [{ 'url': 'https://www.animalplanet.com/video/north-woods-law-animal-planet/squirrel-showdown', 'info_dict': { 'id': '3338923', 'display_id': 'north-woods-law-animal-planet/squirrel-showdown', 'ext': 'mp4', 'title': 'Squirrel Showdown', 'description': 'A woman is suspected of being in possession of flying squirrel kits.', 'season_number': 16, 'episode_number': 11, }, 'skip': 'Available for Premium users', }] _PRODUCT = 'apl' _API_URL = 'us1-prod-direct.animalplanet.com' class DiscoveryPlusIndiaIE(DPlayBaseIE): _VALID_URL = r'https?://(?:www\.)?discoveryplus\.in/videos?' + DPlayBaseIE._PATH_REGEX _TESTS = [{ 'url': 'https://www.discoveryplus.in/videos/how-do-they-do-it/fugu-and-more?seasonId=8&type=EPISODE', 'info_dict': { 'id': '27104', 'ext': 'mp4', 'display_id': 'how-do-they-do-it/fugu-and-more', 'title': 'Fugu and More', 'description': 'The Japanese catch, prepare and eat the deadliest fish on the planet.', 'duration': 1319, 'timestamp': 1582309800, 'upload_date': '20200221', 'series': 'How Do They Do It?', 'season_number': 8, 'episode_number': 2, 'creator': 'Discovery Channel', }, 'params': { 'skip_download': True, } }] def _update_disco_api_headers(self, headers, disco_base, display_id, realm): headers.update({ 'x-disco-params': 'realm=%s' % realm, 'x-disco-client': 'WEB:UNKNOWN:dplus-india:17.0.0', 'Authorization': self._get_auth(disco_base, display_id, realm), }) def _download_video_playback_info(self, disco_base, video_id, headers): return self._download_json( disco_base + 'playback/v3/videoPlaybackInfo', video_id, headers=headers, data=json.dumps({ 'deviceInfo': { 'adBlocker': False, }, 'videoId': video_id, }).encode('utf-8'))['data']['attributes']['streaming'] def _real_extract(self, url): display_id = self._match_id(url) return self._get_disco_api_info( url, display_id, 'ap2-prod-direct.discoveryplus.in', 'dplusindia', 'in', 'https://www.discoveryplus.in/') class DiscoveryNetworksDeIE(DPlayBaseIE): _VALID_URL = r'https?://(?:www\.)?(?P(?:tlc|dmax)\.de|dplay\.co\.uk)/(?:programme|show|sendungen)/(?P[^/]+)/(?:video/)?(?P[^/]+)' _TESTS = [{ 'url': 'https://www.tlc.de/programme/breaking-amish/video/die-welt-da-drauen/DCB331270001100', 'info_dict': { 'id': '78867', 'ext': 'mp4', 'title': 'Die Welt da draußen', 'description': 'md5:61033c12b73286e409d99a41742ef608', 'timestamp': 1554069600, 'upload_date': '20190331', }, 'params': { 'skip_download': True, }, }, { 'url': 'https://www.dmax.de/programme/dmax-highlights/video/tuning-star-sidney-hoffmann-exklusiv-bei-dmax/191023082312316', 'only_matching': True, }, { 'url': 'https://www.dplay.co.uk/show/ghost-adventures/video/hotel-leger-103620/EHD_280313B', 'only_matching': True, }, { 'url': 'https://tlc.de/sendungen/breaking-amish/die-welt-da-drauen/', 'only_matching': True, }] def _real_extract(self, url): domain, programme, alternate_id = self._match_valid_url(url).groups() country = 'GB' if domain == 'dplay.co.uk' else 'DE' realm = 'questuk' if country == 'GB' else domain.replace('.', '') return self._get_disco_api_info( url, '%s/%s' % (programme, alternate_id), 'sonic-eu1-prod.disco-api.com', realm, country) class DiscoveryPlusShowBaseIE(DPlayBaseIE): def _entries(self, show_name): headers = { 'x-disco-client': self._X_CLIENT, 'x-disco-params': f'realm={self._REALM}', 'referer': self._DOMAIN, 'Authentication': self._get_auth(self._BASE_API, None, self._REALM), } show_json = self._download_json( f'{self._BASE_API}cms/routes/{self._SHOW_STR}/{show_name}?include=default', video_id=show_name, headers=headers)['included'][self._INDEX]['attributes']['component'] show_id = show_json['mandatoryParams'].split('=')[-1] season_url = self._BASE_API + 'content/videos?sort=episodeNumber&filter[seasonNumber]={}&filter[show.id]={}&page[size]=100&page[number]={}' for season in show_json['filters'][0]['options']: season_id = season['id'] total_pages, page_num = 1, 0 while page_num < total_pages: season_json = self._download_json( season_url.format(season_id, show_id, str(page_num + 1)), show_name, headers=headers, note='Downloading season %s JSON metadata%s' % (season_id, ' page %d' % page_num if page_num else '')) if page_num == 0: total_pages = try_get(season_json, lambda x: x['meta']['totalPages'], int) or 1 episodes_json = season_json['data'] for episode in episodes_json: video_path = episode['attributes']['path'] yield self.url_result( '%svideos/%s' % (self._DOMAIN, video_path), ie=self._VIDEO_IE.ie_key(), video_id=episode.get('id') or video_path) page_num += 1 def _real_extract(self, url): show_name = self._match_valid_url(url).group('show_name') return self.playlist_result(self._entries(show_name), playlist_id=show_name) class DiscoveryPlusItalyShowIE(DiscoveryPlusShowBaseIE): _VALID_URL = r'https?://(?:www\.)?discoveryplus\.it/programmi/(?P[^/]+)/?(?:[?#]|$)' _TESTS = [{ 'url': 'https://www.discoveryplus.it/programmi/deal-with-it-stai-al-gioco', 'playlist_mincount': 168, 'info_dict': { 'id': 'deal-with-it-stai-al-gioco', }, }] _BASE_API = 'https://disco-api.discoveryplus.it/' _DOMAIN = 'https://www.discoveryplus.it/' _X_CLIENT = 'WEB:UNKNOWN:dplay-client:2.6.0' _REALM = 'dplayit' _SHOW_STR = 'programmi' _INDEX = 1 _VIDEO_IE = DPlayIE class DiscoveryPlusIndiaShowIE(DiscoveryPlusShowBaseIE): _VALID_URL = r'https?://(?:www\.)?discoveryplus\.in/show/(?P[^/]+)/?(?:[?#]|$)' _TESTS = [{ 'url': 'https://www.discoveryplus.in/show/how-do-they-do-it', 'playlist_mincount': 140, 'info_dict': { 'id': 'how-do-they-do-it', }, }] _BASE_API = 'https://ap2-prod-direct.discoveryplus.in/' _DOMAIN = 'https://www.discoveryplus.in/' _X_CLIENT = 'WEB:UNKNOWN:dplus-india:prod' _REALM = 'dplusindia' _SHOW_STR = 'show' _INDEX = 4 _VIDEO_IE = DiscoveryPlusIndiaIE