import re from .common import InfoExtractor from ..compat import compat_str from ..utils import ( ExtractorError, get_element_by_id, int_or_none, parse_iso8601, parse_duration, str_or_none, try_get, update_url_query, urljoin, ) class TVNowBaseIE(InfoExtractor): _VIDEO_FIELDS = ( 'id', 'title', 'free', 'geoblocked', 'articleLong', 'articleShort', 'broadcastStartDate', 'isDrm', 'duration', 'season', 'episode', 'manifest.dashclear', 'manifest.hlsclear', 'manifest.smoothclear', 'format.title', 'format.defaultImage169Format', 'format.defaultImage169Logo') def _call_api(self, path, video_id, query): return self._download_json( 'https://api.tvnow.de/v3/' + path, video_id, query=query) def _extract_video(self, info, display_id): video_id = compat_str(info['id']) title = info['title'] paths = [] for manifest_url in (info.get('manifest') or {}).values(): if not manifest_url: continue manifest_url = update_url_query(manifest_url, {'filter': ''}) path = self._search_regex(r'https?://[^/]+/(.+?)\.ism/', manifest_url, 'path') if path in paths: continue paths.append(path) def url_repl(proto, suffix): return re.sub( r'(?:hls|dash|hss)([.-])', proto + r'\1', re.sub( r'\.ism/(?:[^.]*\.(?:m3u8|mpd)|[Mm]anifest)', '.ism/' + suffix, manifest_url)) def make_urls(proto, suffix): urls = [url_repl(proto, suffix)] hd_url = urls[0].replace('/manifest/', '/ngvod/') if hd_url != urls[0]: urls.append(hd_url) return urls for man_url in make_urls('dash', '.mpd'): formats = self._extract_mpd_formats( man_url, video_id, mpd_id='dash', fatal=False) for man_url in make_urls('hss', 'Manifest'): formats.extend(self._extract_ism_formats( man_url, video_id, ism_id='mss', fatal=False)) for man_url in make_urls('hls', '.m3u8'): formats.extend(self._extract_m3u8_formats( man_url, video_id, 'mp4', 'm3u8_native', m3u8_id='hls', fatal=False)) if formats: break else: if not self.get_param('allow_unplayable_formats') and info.get('isDrm'): raise ExtractorError( 'Video %s is DRM protected' % video_id, expected=True) if info.get('geoblocked'): raise self.raise_geo_restricted() if not info.get('free', True): raise ExtractorError( 'Video %s is not available for free' % video_id, expected=True) self._sort_formats(formats) description = info.get('articleLong') or info.get('articleShort') timestamp = parse_iso8601(info.get('broadcastStartDate'), ' ') duration = parse_duration(info.get('duration')) f = info.get('format', {}) thumbnails = [{ 'url': 'https://aistvnow-a.akamaihd.net/tvnow/movie/%s' % video_id, }] thumbnail = f.get('defaultImage169Format') or f.get('defaultImage169Logo') if thumbnail: thumbnails.append({ 'url': thumbnail, }) return { 'id': video_id, 'display_id': display_id, 'title': title, 'description': description, 'thumbnails': thumbnails, 'timestamp': timestamp, 'duration': duration, 'series': f.get('title'), 'season_number': int_or_none(info.get('season')), 'episode_number': int_or_none(info.get('episode')), 'episode': title, 'formats': formats, } class TVNowIE(TVNowBaseIE): _VALID_URL = r'''(?x) https?:// (?:www\.)?tvnow\.(?:de|at|ch)/(?P[^/]+)/ (?P[^/]+)/ (?!(?:list|jahr)(?:/|$))(?P[^/?\#&]+) ''' @classmethod def suitable(cls, url): return (False if TVNowNewIE.suitable(url) or TVNowSeasonIE.suitable(url) or TVNowAnnualIE.suitable(url) or TVNowShowIE.suitable(url) else super(TVNowIE, cls).suitable(url)) _TESTS = [{ 'url': 'https://www.tvnow.de/rtl2/grip-das-motormagazin/der-neue-porsche-911-gt-3/player', 'info_dict': { 'id': '331082', 'display_id': 'grip-das-motormagazin/der-neue-porsche-911-gt-3', 'ext': 'mp4', 'title': 'Der neue Porsche 911 GT 3', 'description': 'md5:6143220c661f9b0aae73b245e5d898bb', 'timestamp': 1495994400, 'upload_date': '20170528', 'duration': 5283, 'series': 'GRIP - Das Motormagazin', 'season_number': 14, 'episode_number': 405, 'episode': 'Der neue Porsche 911 GT 3', }, }, { # rtl2 'url': 'https://www.tvnow.de/rtl2/armes-deutschland/episode-0008/player', 'only_matching': True, }, { # rtlnitro 'url': 'https://www.tvnow.de/nitro/alarm-fuer-cobra-11-die-autobahnpolizei/auf-eigene-faust-pilot/player', 'only_matching': True, }, { # superrtl 'url': 'https://www.tvnow.de/superrtl/die-lustigsten-schlamassel-der-welt/u-a-ketchup-effekt/player', 'only_matching': True, }, { # ntv 'url': 'https://www.tvnow.de/ntv/startup-news/goetter-in-weiss/player', 'only_matching': True, }, { # vox 'url': 'https://www.tvnow.de/vox/auto-mobil/neues-vom-automobilmarkt-2017-11-19-17-00-00/player', 'only_matching': True, }, { # rtlplus 'url': 'https://www.tvnow.de/rtlplus/op-ruft-dr-bruckner/die-vernaehte-frau/player', 'only_matching': True, }, { 'url': 'https://www.tvnow.de/rtl2/grip-das-motormagazin/der-neue-porsche-911-gt-3', 'only_matching': True, }] def _real_extract(self, url): mobj = self._match_valid_url(url) display_id = '%s/%s' % mobj.group(2, 3) info = self._call_api( 'movies/' + display_id, display_id, query={ 'fields': ','.join(self._VIDEO_FIELDS), }) return self._extract_video(info, display_id) class TVNowNewIE(InfoExtractor): _VALID_URL = r'''(?x) (?Phttps?:// (?:www\.)?tvnow\.(?:de|at|ch)/ (?:shows|serien))/ (?P[^/]+)-\d+/ [^/]+/ episode-\d+-(?P[^/?$&]+)-(?P\d+) ''' _TESTS = [{ 'url': 'https://www.tvnow.de/shows/grip-das-motormagazin-1669/2017-05/episode-405-der-neue-porsche-911-gt-3-331082', 'only_matching': True, }] def _real_extract(self, url): mobj = self._match_valid_url(url) base_url = re.sub(r'(?:shows|serien)', '_', mobj.group('base_url')) show, episode = mobj.group('show', 'episode') return self.url_result( # Rewrite new URLs to the old format and use extraction via old API # at api.tvnow.de as a loophole for bypassing premium content checks '%s/%s/%s' % (base_url, show, episode), ie=TVNowIE.ie_key(), video_id=mobj.group('id')) class TVNowFilmIE(TVNowBaseIE): _VALID_URL = r'''(?x) (?Phttps?:// (?:www\.)?tvnow\.(?:de|at|ch)/ (?:filme))/ (?P[^/?$&]+)-(?P<id>\d+) ''' _TESTS = [{ 'url': 'https://www.tvnow.de/filme/lord-of-war-haendler-des-todes-7959', 'info_dict': { 'id': '1426690', 'display_id': 'lord-of-war-haendler-des-todes', 'ext': 'mp4', 'title': 'Lord of War', 'description': 'md5:5eda15c0d5b8cb70dac724c8a0ff89a9', 'timestamp': 1550010000, 'upload_date': '20190212', 'duration': 7016, }, }, { 'url': 'https://www.tvnow.de/filme/the-machinist-12157', 'info_dict': { 'id': '328160', 'display_id': 'the-machinist', 'ext': 'mp4', 'title': 'The Machinist', 'description': 'md5:9a0e363fdd74b3a9e1cdd9e21d0ecc28', 'timestamp': 1496469720, 'upload_date': '20170603', 'duration': 5836, }, }, { 'url': 'https://www.tvnow.de/filme/horst-schlaemmer-isch-kandidiere-17777', 'only_matching': True, # DRM protected }] def _real_extract(self, url): mobj = self._match_valid_url(url) display_id = mobj.group('title') webpage = self._download_webpage(url, display_id, fatal=False) if not webpage: raise ExtractorError('Cannot download "%s"' % url, expected=True) json_text = get_element_by_id('now-web-state', webpage) if not json_text: raise ExtractorError('Cannot read video data', expected=True) json_data = self._parse_json( json_text, display_id, transform_source=lambda x: x.replace('&q;', '"'), fatal=False) if not json_data: raise ExtractorError('Cannot read video data', expected=True) player_key = next( (key for key in json_data.keys() if 'module/player' in key), None) page_key = next( (key for key in json_data.keys() if 'page/filme' in key), None) movie_id = try_get( json_data, [ lambda x: x[player_key]['body']['id'], lambda x: x[page_key]['body']['modules'][0]['id'], lambda x: x[page_key]['body']['modules'][1]['id']], int) if not movie_id: raise ExtractorError('Cannot extract movie ID', expected=True) info = self._call_api( 'movies/%d' % movie_id, display_id, query={'fields': ','.join(self._VIDEO_FIELDS)}) return self._extract_video(info, display_id) class TVNowNewBaseIE(InfoExtractor): def _call_api(self, path, video_id, query={}): result = self._download_json( 'https://apigw.tvnow.de/module/' + path, video_id, query=query) error = result.get('error') if error: raise ExtractorError( '%s said: %s' % (self.IE_NAME, error), expected=True) return result r""" TODO: new apigw.tvnow.de based version of TVNowIE. Replace old TVNowIE with it when api.tvnow.de is shut down. This version can't bypass premium checks though. class TVNowIE(TVNowNewBaseIE): _VALID_URL = r'''(?x) https?:// (?:www\.)?tvnow\.(?:de|at|ch)/ (?:shows|serien)/[^/]+/ (?:[^/]+/)+ (?P<display_id>[^/?$&]+)-(?P<id>\d+) ''' _TESTS = [{ # episode with annual navigation 'url': 'https://www.tvnow.de/shows/grip-das-motormagazin-1669/2017-05/episode-405-der-neue-porsche-911-gt-3-331082', 'info_dict': { 'id': '331082', 'display_id': 'grip-das-motormagazin/der-neue-porsche-911-gt-3', 'ext': 'mp4', 'title': 'Der neue Porsche 911 GT 3', 'description': 'md5:6143220c661f9b0aae73b245e5d898bb', 'thumbnail': r're:^https?://.*\.jpg$', 'timestamp': 1495994400, 'upload_date': '20170528', 'duration': 5283, 'series': 'GRIP - Das Motormagazin', 'season_number': 14, 'episode_number': 405, 'episode': 'Der neue Porsche 911 GT 3', }, }, { # rtl2, episode with season navigation 'url': 'https://www.tvnow.de/shows/armes-deutschland-11471/staffel-3/episode-14-bernd-steht-seit-der-trennung-von-seiner-frau-allein-da-526124', 'only_matching': True, }, { # rtlnitro 'url': 'https://www.tvnow.de/serien/alarm-fuer-cobra-11-die-autobahnpolizei-1815/staffel-13/episode-5-auf-eigene-faust-pilot-366822', 'only_matching': True, }, { # superrtl 'url': 'https://www.tvnow.de/shows/die-lustigsten-schlamassel-der-welt-1221/staffel-2/episode-14-u-a-ketchup-effekt-364120', 'only_matching': True, }, { # ntv 'url': 'https://www.tvnow.de/shows/startup-news-10674/staffel-2/episode-39-goetter-in-weiss-387630', 'only_matching': True, }, { # vox 'url': 'https://www.tvnow.de/shows/auto-mobil-174/2017-11/episode-46-neues-vom-automobilmarkt-2017-11-19-17-00-00-380072', 'only_matching': True, }, { 'url': 'https://www.tvnow.de/shows/grip-das-motormagazin-1669/2017-05/episode-405-der-neue-porsche-911-gt-3-331082', 'only_matching': True, }] def _extract_video(self, info, url, display_id): config = info['config'] source = config['source'] video_id = compat_str(info.get('id') or source['videoId']) title = source['title'].strip() paths = [] for manifest_url in (info.get('manifest') or {}).values(): if not manifest_url: continue manifest_url = update_url_query(manifest_url, {'filter': ''}) path = self._search_regex(r'https?://[^/]+/(.+?)\.ism/', manifest_url, 'path') if path in paths: continue paths.append(path) def url_repl(proto, suffix): return re.sub( r'(?:hls|dash|hss)([.-])', proto + r'\1', re.sub( r'\.ism/(?:[^.]*\.(?:m3u8|mpd)|[Mm]anifest)', '.ism/' + suffix, manifest_url)) formats = self._extract_mpd_formats( url_repl('dash', '.mpd'), video_id, mpd_id='dash', fatal=False) formats.extend(self._extract_ism_formats( url_repl('hss', 'Manifest'), video_id, ism_id='mss', fatal=False)) formats.extend(self._extract_m3u8_formats( url_repl('hls', '.m3u8'), video_id, 'mp4', 'm3u8_native', m3u8_id='hls', fatal=False)) if formats: break else: if try_get(info, lambda x: x['rights']['isDrm']): raise ExtractorError( 'Video %s is DRM protected' % video_id, expected=True) if try_get(config, lambda x: x['boards']['geoBlocking']['block']): raise self.raise_geo_restricted() if not info.get('free', True): raise ExtractorError( 'Video %s is not available for free' % video_id, expected=True) self._sort_formats(formats) description = source.get('description') thumbnail = url_or_none(source.get('poster')) timestamp = unified_timestamp(source.get('previewStart')) duration = parse_duration(source.get('length')) series = source.get('format') season_number = int_or_none(self._search_regex( r'staffel-(\d+)', url, 'season number', default=None)) episode_number = int_or_none(self._search_regex( r'episode-(\d+)', url, 'episode number', default=None)) return { 'id': video_id, 'display_id': display_id, 'title': title, 'description': description, 'thumbnail': thumbnail, 'timestamp': timestamp, 'duration': duration, 'series': series, 'season_number': season_number, 'episode_number': episode_number, 'episode': title, 'formats': formats, } def _real_extract(self, url): display_id, video_id = self._match_valid_url(url).groups() info = self._call_api('player/' + video_id, video_id) return self._extract_video(info, video_id, display_id) class TVNowFilmIE(TVNowIE): _VALID_URL = r'''(?x) (?P<base_url>https?:// (?:www\.)?tvnow\.(?:de|at|ch)/ (?:filme))/ (?P<title>[^/?$&]+)-(?P<id>\d+) ''' _TESTS = [{ 'url': 'https://www.tvnow.de/filme/lord-of-war-haendler-des-todes-7959', 'info_dict': { 'id': '1426690', 'display_id': 'lord-of-war-haendler-des-todes', 'ext': 'mp4', 'title': 'Lord of War', 'description': 'md5:5eda15c0d5b8cb70dac724c8a0ff89a9', 'timestamp': 1550010000, 'upload_date': '20190212', 'duration': 7016, }, }, { 'url': 'https://www.tvnow.de/filme/the-machinist-12157', 'info_dict': { 'id': '328160', 'display_id': 'the-machinist', 'ext': 'mp4', 'title': 'The Machinist', 'description': 'md5:9a0e363fdd74b3a9e1cdd9e21d0ecc28', 'timestamp': 1496469720, 'upload_date': '20170603', 'duration': 5836, }, }, { 'url': 'https://www.tvnow.de/filme/horst-schlaemmer-isch-kandidiere-17777', 'only_matching': True, # DRM protected }] def _real_extract(self, url): mobj = self._match_valid_url(url) display_id = mobj.group('title') webpage = self._download_webpage(url, display_id, fatal=False) if not webpage: raise ExtractorError('Cannot download "%s"' % url, expected=True) json_text = get_element_by_id('now-web-state', webpage) if not json_text: raise ExtractorError('Cannot read video data', expected=True) json_data = self._parse_json( json_text, display_id, transform_source=lambda x: x.replace('&q;', '"'), fatal=False) if not json_data: raise ExtractorError('Cannot read video data', expected=True) player_key = next( (key for key in json_data.keys() if 'module/player' in key), None) page_key = next( (key for key in json_data.keys() if 'page/filme' in key), None) movie_id = try_get( json_data, [ lambda x: x[player_key]['body']['id'], lambda x: x[page_key]['body']['modules'][0]['id'], lambda x: x[page_key]['body']['modules'][1]['id']], int) if not movie_id: raise ExtractorError('Cannot extract movie ID', expected=True) info = self._call_api('player/%d' % movie_id, display_id) return self._extract_video(info, url, display_id) """ class TVNowListBaseIE(TVNowNewBaseIE): _SHOW_VALID_URL = r'''(?x) (?P<base_url> https?:// (?:www\.)?tvnow\.(?:de|at|ch)/(?:shows|serien)/ [^/?#&]+-(?P<show_id>\d+) ) ''' @classmethod def suitable(cls, url): return (False if TVNowNewIE.suitable(url) else super(TVNowListBaseIE, cls).suitable(url)) def _extract_items(self, url, show_id, list_id, query): items = self._call_api( 'teaserrow/format/episode/' + show_id, list_id, query=query)['items'] entries = [] for item in items: if not isinstance(item, dict): continue item_url = urljoin(url, item.get('url')) if not item_url: continue video_id = str_or_none(item.get('id') or item.get('videoId')) item_title = item.get('subheadline') or item.get('text') entries.append(self.url_result( item_url, ie=TVNowNewIE.ie_key(), video_id=video_id, video_title=item_title)) return self.playlist_result(entries, '%s/%s' % (show_id, list_id)) class TVNowSeasonIE(TVNowListBaseIE): _VALID_URL = r'%s/staffel-(?P<id>\d+)' % TVNowListBaseIE._SHOW_VALID_URL _TESTS = [{ 'url': 'https://www.tvnow.de/serien/alarm-fuer-cobra-11-die-autobahnpolizei-1815/staffel-13', 'info_dict': { 'id': '1815/13', }, 'playlist_mincount': 22, }] def _real_extract(self, url): _, show_id, season_id = self._match_valid_url(url).groups() return self._extract_items( url, show_id, season_id, {'season': season_id}) class TVNowAnnualIE(TVNowListBaseIE): _VALID_URL = r'%s/(?P<year>\d{4})-(?P<month>\d{2})' % TVNowListBaseIE._SHOW_VALID_URL _TESTS = [{ 'url': 'https://www.tvnow.de/shows/grip-das-motormagazin-1669/2017-05', 'info_dict': { 'id': '1669/2017-05', }, 'playlist_mincount': 2, }] def _real_extract(self, url): _, show_id, year, month = self._match_valid_url(url).groups() return self._extract_items( url, show_id, '%s-%s' % (year, month), { 'year': int(year), 'month': int(month), }) class TVNowShowIE(TVNowListBaseIE): _VALID_URL = TVNowListBaseIE._SHOW_VALID_URL _TESTS = [{ # annual navigationType 'url': 'https://www.tvnow.de/shows/grip-das-motormagazin-1669', 'info_dict': { 'id': '1669', }, 'playlist_mincount': 73, }, { # season navigationType 'url': 'https://www.tvnow.de/shows/armes-deutschland-11471', 'info_dict': { 'id': '11471', }, 'playlist_mincount': 3, }] @classmethod def suitable(cls, url): return (False if TVNowNewIE.suitable(url) or TVNowSeasonIE.suitable(url) or TVNowAnnualIE.suitable(url) else super(TVNowShowIE, cls).suitable(url)) def _real_extract(self, url): base_url, show_id = self._match_valid_url(url).groups() result = self._call_api( 'teaserrow/format/navigation/' + show_id, show_id) items = result['items'] entries = [] navigation = result.get('navigationType') if navigation == 'annual': for item in items: if not isinstance(item, dict): continue year = int_or_none(item.get('year')) if year is None: continue months = item.get('months') if not isinstance(months, list): continue for month_dict in months: if not isinstance(month_dict, dict) or not month_dict: continue month_number = int_or_none(list(month_dict.keys())[0]) if month_number is None: continue entries.append(self.url_result( '%s/%04d-%02d' % (base_url, year, month_number), ie=TVNowAnnualIE.ie_key())) elif navigation == 'season': for item in items: if not isinstance(item, dict): continue season_number = int_or_none(item.get('season')) if season_number is None: continue entries.append(self.url_result( '%s/staffel-%d' % (base_url, season_number), ie=TVNowSeasonIE.ie_key())) else: raise ExtractorError('Unknown navigationType') return self.playlist_result(entries, show_id)