diff --git a/youtube_dl/extractor/go.py b/youtube_dl/extractor/go.py index 878ba14e6..0a40b5b47 100644 --- a/youtube_dl/extractor/go.py +++ b/youtube_dl/extractor/go.py @@ -20,10 +20,12 @@ class GoIE(AdobePassIE): 'abc': { 'brand': '001', 'requestor_id': 'ABC', + 'go': True, }, 'freeform': { 'brand': '002', 'requestor_id': 'ABCFamily', + 'go': True, }, 'watchdisneychannel': { 'brand': '004', @@ -40,6 +42,7 @@ class GoIE(AdobePassIE): 'disneynow': { 'brand': '011', 'resource_id': 'Disney', + 'go': True, }, 'fxnow.fxnetworks': { 'brand': '025', @@ -48,15 +51,19 @@ class GoIE(AdobePassIE): } _VALID_URL = r'''(?x) https?:// - (?: - (?:(?P%s)\.)?go| - (?Pabc|freeform|disneynow|fxnow\.fxnetworks) + (?P + (?:(?:%(sites)s)\.)?go|fxnow\.fxnetworks| + (?:www\.)?(?:%(sites)s) )\.com/ (?: (?:[^/]+/)*(?P[Vv][Dd][Kk][Aa]\w+)| (?:[^/]+/)*(?P[^/?\#]+) ) - ''' % '|'.join(list(_SITE_INFO.keys())) + ''' % { + 'sites': + # keep _SITE_INFO in scope + (lambda s: '|'.join(re.escape(k) for k in s.keys() if s[k].get('go')))(_SITE_INFO), + } _TESTS = [{ 'url': 'http://abc.go.com/shows/designated-survivor/video/most-recent/VDKA3807643', 'info_dict': { @@ -70,14 +77,9 @@ class GoIE(AdobePassIE): 'skip_download': True, }, 'skip': 'This content is no longer available.', - }, { - 'url': 'http://watchdisneyxd.go.com/doraemon', - 'info_dict': { - 'title': 'Doraemon', - 'id': 'SH55574025', - }, - 'playlist_mincount': 51, - }, { + }, + # Disney XD app withdrawn 2018: 'http://watchdisneyxd.go.com/doraemon', + { 'url': 'http://freeform.go.com/shows/shadowhunters/episodes/season-2/1-this-guilty-blood', 'info_dict': { 'id': 'VDKA3609139', @@ -85,23 +87,37 @@ class GoIE(AdobePassIE): 'title': 'This Guilty Blood', 'description': 'md5:f18e79ad1c613798d95fdabfe96cd292', 'age_limit': 14, + 'duration': 2544, + 'episode': 'Episode 1', + 'thumbnail': r're:https?://.*?\.jpg', + 'season_number': 2, + 'episode_number': 1, + 'season': 'Season 2', + 'series': 'Shadowhunters', }, 'params': { - 'geo_bypass_ip_block': '3.244.239.0/24', + 'geo_bypass_ip_block': '6.48.97.0/24', # m3u8 download 'skip_download': True, }, }, { - 'url': 'https://abc.com/shows/the-rookie/episode-guide/season-02/03-the-bet', + 'url': 'https://abc.com/shows/the-rookie/episode-guide/season-04/13-fight-or-flight', 'info_dict': { - 'id': 'VDKA13435179', + 'id': 'VDKA26139487', 'ext': 'mp4', - 'title': 'The Bet', - 'description': 'md5:c66de8ba2e92c6c5c113c3ade84ab404', + 'title': 'Fight or Flight', + 'episode': 'Episode 13', + 'thumbnail': r're:https?://.*?\.jpg', + 'series': 'The Rookie', + 'description': 'md5:8caf91de9806a1c5f79823059fe80267', + 'duration': 2575, + 'episode_number': 13, + 'season_number': 4, + 'season': 'Season 4', 'age_limit': 14, }, 'params': { - 'geo_bypass_ip_block': '3.244.239.0/24', + 'geo_bypass_ip_block': '6.48.97.0/24', # m3u8 download 'skip_download': True, }, @@ -112,9 +128,13 @@ class GoIE(AdobePassIE): 'ext': 'mp4', 'title': 'First Look: Better Things - Season 2', 'description': 'md5:fa73584a95761c605d9d54904e35b407', + 'duration': 161, + 'series': 'Better Things', + 'age_limit': 14, + 'thumbnail': r're:https?://.*?\.jpg', }, 'params': { - 'geo_bypass_ip_block': '3.244.239.0/24', + 'geo_bypass_ip_block': '6.48.97.0/24', # m3u8 download 'skip_download': True, }, @@ -125,6 +145,14 @@ class GoIE(AdobePassIE): 'ext': 'mp4', 'title': 'Pilot', 'description': 'md5:74306df917cfc199d76d061d66bebdb4', + 'season': 'Season 1', + 'episode': 'Episode 1', + 'series': 'Modern Family', + 'season_number': 1, + 'duration': 1301, + 'episode_number': 1, + 'age_limit': 0, + 'thumbnail': r're:https?://.*?\.jpg', }, 'params': { # m3u8 download @@ -147,17 +175,48 @@ class GoIE(AdobePassIE): }, { 'url': 'https://disneynow.com/shows/minnies-bow-toons/video/happy-campers/vdka4872013', 'only_matching': True, + }, { + 'url': 'https://www.freeform.com/shows/cruel-summer/episode-guide/season-01/01-happy-birthday-jeanette-turner', + 'only_matching': True, + }, { + 'url': 'https://disneynow.com/shows/ducktales-disney-channel/season-02/episode-09-the-outlaw-scrooge-mcduck/vdka9680773', + 'info_dict': { + 'id': 'VDKA9680773', + 'ext': 'mp4', + 'season_number': 2, + 'title': 'The Outlaw Scrooge McDuck!', + 'episode': 'Episode 9', + 'season': 'Season 2', + 'episode_number': 9, + 'age_limit': 7, + 'series': 'DuckTales', + 'description': 'md5:3d450c1094a5e7544ac094f1e6da2827', + 'duration': 1348, + 'thumbnail': r're:https?://.*?\.jpg', + }, + 'params': { + # AES-128: ffmpeg + 'skip_download': True, + }, + }, { + 'url': 'https://disneynow.com/shows/ducktales-disney-channel', + 'info_dict': { + 'title': 'DuckTales TV Show', + 'id': 'SH553923438', + }, + 'playlist_mincount': 33, + 'expected_warnings': ['Ignoring subtitle tracks', ], }] def _extract_videos(self, brand, video_id='-1', show_id='-1'): display_id = video_id if video_id != '-1' else show_id return self._download_json( - 'http://api.contents.watchabc.go.com/vp2/ws/contents/3000/videos/%s/001/-1/%s/-1/%s/-1/-1.json' % (brand, show_id, video_id), + 'http://api.contents.watchabc.go.com/vp2/ws/contents/4000/videos/%s/001/-1/%s/-1/%s/-1/-1.json' % (brand, show_id, video_id), display_id)['video'] def _real_extract(self, url): mobj = re.match(self._VALID_URL, url) - sub_domain = mobj.group('sub_domain') or mobj.group('sub_domain_2') + sub_domain = re.sub(r'^(?:www\.)?(.+?)(?:\.go)?$', r'\1', mobj.group('sub_domain') or '') video_id, display_id = mobj.group('id', 'display_id') site_info = self._SITE_INFO.get(sub_domain, {}) brand = site_info.get('brand') @@ -181,26 +240,33 @@ class GoIE(AdobePassIE): ( # There may be inner quotes, e.g. data-video-id="'VDKA3609139'" # from http://freeform.go.com/shows/shadowhunters/episodes/season-2/1-this-guilty-blood - r'data-video-id=["\']*(VDKA\w+)', + r'data-video-id\s*=\s*["\']{,2}(VDKA\w+)', # page.analytics.videoIdCode r'\bvideoIdCode["\']\s*:\s*["\']((?:vdka|VDKA)\w+)', + # This obsolete pattern, if matched, overrides detection of DisneyNow show pages # https://abc.com/shows/the-rookie/episode-guide/season-02/03-the-bet - r'\b(?:video)?id["\']\s*:\s*["\'](VDKA\w+)' + # r'\b(?:video)?id["\']\s*:\s*["\'](VDKA\w+)' ), webpage, 'video id', default=video_id) if not site_info: brand = self._search_regex( (r'data-brand=\s*["\']\s*(\d+)', - r'data-page-brand=\s*["\']\s*(\d+)'), webpage, 'brand', - default='004') + r'data-page-brand=\s*["\']\s*(\d+)', + r'"brand"\s*:\s*"(\d+)"'), webpage, 'brand', + # default to a brand that has an active domain + default='011') site_info = next( - si for _, si in self._SITE_INFO.items() + si for si in self._SITE_INFO.values() if si.get('brand') == brand) if not video_id: # show extraction works for Disney, DisneyJunior and DisneyXD # ABC and Freeform has different layout show_id = self._search_regex(r'data-show-id=["\']*(SH\d+)', webpage, 'show id') videos = self._extract_videos(brand, show_id=show_id) - show_title = self._search_regex(r'data-show-title="([^"]+)"', webpage, 'show title', fatal=False) + show_title = ( + self._search_regex(r'data-show-title="([^"]+)"', webpage, 'show title', default=None) + or self._og_search_title(webpage, fatal=False)) + if show_title: + show_title = re.sub(r'^(?:Watch\s+)?(.+?)(?:\s*\|\s*Disney.*)?$', r'\1', show_title) entries = [] for video in videos: entries.append(self.url_result( @@ -262,7 +328,8 @@ class GoIE(AdobePassIE): if re.search(r'(?:/mp4/source/|_source\.mp4)', asset_url): f.update({ 'format_id': ('%s-' % format_id if format_id else '') + 'SOURCE', - 'preference': 1, + 'quality': 1, + # 'preference': 1, }) else: mobj = re.search(r'/(\d+)x(\d+)/', asset_url)