]> Raphaël G. Git Repositories - youtubedl/blobdiff - youtube_dl/extractor/youtube.py
New upstream version 2017.02.07
[youtubedl] / youtube_dl / extractor / youtube.py
index bd24a28389bf847f1e72451e232bfab0807809d8..76710931ae5e6a292af767f3f57685ad0be98cac 100644 (file)
@@ -40,6 +40,7 @@ from ..utils import (
     sanitized_Request,
     smuggle_url,
     str_to_int,
+    try_get,
     unescapeHTML,
     unified_strdate,
     unsmuggle_url,
@@ -316,6 +317,7 @@ class YoutubeIE(YoutubeBaseInfoExtractor):
         '137': {'ext': 'mp4', 'height': 1080, 'format_note': 'DASH video', 'vcodec': 'h264', 'preference': -40},
         '138': {'ext': 'mp4', 'format_note': 'DASH video', 'vcodec': 'h264', 'preference': -40},  # Height can vary (https://github.com/rg3/youtube-dl/issues/4559)
         '160': {'ext': 'mp4', 'height': 144, 'format_note': 'DASH video', 'vcodec': 'h264', 'preference': -40},
+        '212': {'ext': 'mp4', 'height': 480, 'format_note': 'DASH video', 'vcodec': 'h264', 'preference': -40},
         '264': {'ext': 'mp4', 'height': 1440, 'format_note': 'DASH video', 'vcodec': 'h264', 'preference': -40},
         '298': {'ext': 'mp4', 'height': 720, 'format_note': 'DASH video', 'vcodec': 'h264', 'fps': 60, 'preference': -40},
         '299': {'ext': 'mp4', 'height': 1080, 'format_note': 'DASH video', 'vcodec': 'h264', 'fps': 60, 'preference': -40},
@@ -327,6 +329,8 @@ class YoutubeIE(YoutubeBaseInfoExtractor):
         '141': {'ext': 'm4a', 'format_note': 'DASH audio', 'acodec': 'aac', 'abr': 256, 'preference': -50, 'container': 'm4a_dash'},
         '256': {'ext': 'm4a', 'format_note': 'DASH audio', 'acodec': 'aac', 'preference': -50, 'container': 'm4a_dash'},
         '258': {'ext': 'm4a', 'format_note': 'DASH audio', 'acodec': 'aac', 'preference': -50, 'container': 'm4a_dash'},
+        '325': {'ext': 'm4a', 'format_note': 'DASH audio', 'acodec': 'dtse', 'preference': -50, 'container': 'm4a_dash'},
+        '328': {'ext': 'm4a', 'format_note': 'DASH audio', 'acodec': 'ec-3', 'preference': -50, 'container': 'm4a_dash'},
 
         # Dash webm
         '167': {'ext': 'webm', 'height': 360, 'width': 640, 'format_note': 'DASH video', 'container': 'webm', 'vcodec': 'vp8', 'preference': -40},
@@ -376,12 +380,13 @@ class YoutubeIE(YoutubeBaseInfoExtractor):
                 'title': 'youtube-dl test video "\'/\\ä↭𝕐',
                 'uploader': 'Philipp Hagemeister',
                 'uploader_id': 'phihag',
-                'uploader_url': 're:https?://(?:www\.)?youtube\.com/user/phihag',
+                'uploader_url': r're:https?://(?:www\.)?youtube\.com/user/phihag',
                 'upload_date': '20121002',
                 'license': 'Standard YouTube License',
                 'description': 'test chars:  "\'/\\ä↭𝕐\ntest URL: https://github.com/rg3/youtube-dl/issues/1892\n\nThis is a test video for youtube-dl.\n\nFor more information, contact phihag@phihag.de .',
                 'categories': ['Science & Technology'],
                 'tags': ['youtube-dl'],
+                'duration': 10,
                 'like_count': int,
                 'dislike_count': int,
                 'start_time': 1,
@@ -401,9 +406,10 @@ class YoutubeIE(YoutubeBaseInfoExtractor):
                 'tags': ['Icona Pop i love it', 'sweden', 'pop music', 'big beat records', 'big beat', 'charli',
                          'xcx', 'charli xcx', 'girls', 'hbo', 'i love it', "i don't care", 'icona', 'pop',
                          'iconic ep', 'iconic', 'love', 'it'],
+                'duration': 180,
                 'uploader': 'Icona Pop',
                 'uploader_id': 'IconaPop',
-                'uploader_url': 're:https?://(?:www\.)?youtube\.com/user/IconaPop',
+                'uploader_url': r're:https?://(?:www\.)?youtube\.com/user/IconaPop',
                 'license': 'Standard YouTube License',
                 'creator': 'Icona Pop',
             }
@@ -418,9 +424,10 @@ class YoutubeIE(YoutubeBaseInfoExtractor):
                 'title': 'Justin Timberlake - Tunnel Vision (Explicit)',
                 'alt_title': 'Tunnel Vision',
                 'description': 'md5:64249768eec3bc4276236606ea996373',
+                'duration': 419,
                 'uploader': 'justintimberlakeVEVO',
                 'uploader_id': 'justintimberlakeVEVO',
-                'uploader_url': 're:https?://(?:www\.)?youtube\.com/user/justintimberlakeVEVO',
+                'uploader_url': r're:https?://(?:www\.)?youtube\.com/user/justintimberlakeVEVO',
                 'license': 'Standard YouTube License',
                 'creator': 'Justin Timberlake',
                 'age_limit': 18,
@@ -437,7 +444,7 @@ class YoutubeIE(YoutubeBaseInfoExtractor):
                 'description': 'md5:09b78bd971f1e3e289601dfba15ca4f7',
                 'uploader': 'SET India',
                 'uploader_id': 'setindia',
-                'uploader_url': 're:https?://(?:www\.)?youtube\.com/user/setindia',
+                'uploader_url': r're:https?://(?:www\.)?youtube\.com/user/setindia',
                 'license': 'Standard YouTube License',
                 'age_limit': 18,
             }
@@ -451,12 +458,13 @@ class YoutubeIE(YoutubeBaseInfoExtractor):
                 'title': 'youtube-dl test video "\'/\\ä↭𝕐',
                 'uploader': 'Philipp Hagemeister',
                 'uploader_id': 'phihag',
-                'uploader_url': 're:https?://(?:www\.)?youtube\.com/user/phihag',
+                'uploader_url': r're:https?://(?:www\.)?youtube\.com/user/phihag',
                 'upload_date': '20121002',
                 'license': 'Standard YouTube License',
                 'description': 'test chars:  "\'/\\ä↭𝕐\ntest URL: https://github.com/rg3/youtube-dl/issues/1892\n\nThis is a test video for youtube-dl.\n\nFor more information, contact phihag@phihag.de .',
                 'categories': ['Science & Technology'],
                 'tags': ['youtube-dl'],
+                'duration': 10,
                 'like_count': int,
                 'dislike_count': int,
             },
@@ -472,7 +480,7 @@ class YoutubeIE(YoutubeBaseInfoExtractor):
                 'ext': 'm4a',
                 'upload_date': '20121002',
                 'uploader_id': '8KVIDEO',
-                'uploader_url': 're:https?://(?:www\.)?youtube\.com/user/8KVIDEO',
+                'uploader_url': r're:https?://(?:www\.)?youtube\.com/user/8KVIDEO',
                 'description': '',
                 'uploader': '8KVIDEO',
                 'license': 'Standard YouTube License',
@@ -492,6 +500,7 @@ class YoutubeIE(YoutubeBaseInfoExtractor):
                 'ext': 'm4a',
                 'title': 'Afrojack, Spree Wilson - The Spark ft. Spree Wilson',
                 'description': 'md5:12e7067fa6735a77bdcbb58cb1187d2d',
+                'duration': 244,
                 'uploader': 'AfrojackVEVO',
                 'uploader_id': 'AfrojackVEVO',
                 'upload_date': '20131011',
@@ -511,6 +520,7 @@ class YoutubeIE(YoutubeBaseInfoExtractor):
                 'title': 'Taylor Swift - Shake It Off',
                 'alt_title': 'Shake It Off',
                 'description': 'md5:95f66187cd7c8b2c13eb78e1223b63c3',
+                'duration': 242,
                 'uploader': 'TaylorSwiftVEVO',
                 'uploader_id': 'TaylorSwiftVEVO',
                 'upload_date': '20140818',
@@ -528,10 +538,11 @@ class YoutubeIE(YoutubeBaseInfoExtractor):
             'info_dict': {
                 'id': 'T4XJQO3qol8',
                 'ext': 'mp4',
+                'duration': 219,
                 'upload_date': '20100909',
                 'uploader': 'The Amazing Atheist',
                 'uploader_id': 'TheAmazingAtheist',
-                'uploader_url': 're:https?://(?:www\.)?youtube\.com/user/TheAmazingAtheist',
+                'uploader_url': r're:https?://(?:www\.)?youtube\.com/user/TheAmazingAtheist',
                 'license': 'Standard YouTube License',
                 'title': 'Burning Everyone\'s Koran',
                 'description': 'SUBSCRIBE: http://www.youtube.com/saturninefilms\n\nEven Obama has taken a stand against freedom on this issue: http://www.huffingtonpost.com/2010/09/09/obama-gma-interview-quran_n_710282.html',
@@ -544,10 +555,11 @@ class YoutubeIE(YoutubeBaseInfoExtractor):
                 'id': 'HtVdAasjOgU',
                 'ext': 'mp4',
                 'title': 'The Witcher 3: Wild Hunt - The Sword Of Destiny Trailer',
-                'description': 're:(?s).{100,}About the Game\n.*?The Witcher 3: Wild Hunt.{100,}',
+                'description': r're:(?s).{100,}About the Game\n.*?The Witcher 3: Wild Hunt.{100,}',
+                'duration': 142,
                 'uploader': 'The Witcher',
                 'uploader_id': 'WitcherGame',
-                'uploader_url': 're:https?://(?:www\.)?youtube\.com/user/WitcherGame',
+                'uploader_url': r're:https?://(?:www\.)?youtube\.com/user/WitcherGame',
                 'upload_date': '20140605',
                 'license': 'Standard YouTube License',
                 'age_limit': 18,
@@ -561,9 +573,10 @@ class YoutubeIE(YoutubeBaseInfoExtractor):
                 'ext': 'mp4',
                 'title': 'Dedication To My Ex (Miss That) (Lyric Video)',
                 'description': 'md5:33765bb339e1b47e7e72b5490139bb41',
+                'duration': 247,
                 'uploader': 'LloydVEVO',
                 'uploader_id': 'LloydVEVO',
-                'uploader_url': 're:https?://(?:www\.)?youtube\.com/user/LloydVEVO',
+                'uploader_url': r're:https?://(?:www\.)?youtube\.com/user/LloydVEVO',
                 'upload_date': '20110629',
                 'license': 'Standard YouTube License',
                 'age_limit': 18,
@@ -575,9 +588,10 @@ class YoutubeIE(YoutubeBaseInfoExtractor):
             'info_dict': {
                 'id': '__2ABJjxzNo',
                 'ext': 'mp4',
+                'duration': 266,
                 'upload_date': '20100430',
                 'uploader_id': 'deadmau5',
-                'uploader_url': 're:https?://(?:www\.)?youtube\.com/user/deadmau5',
+                'uploader_url': r're:https?://(?:www\.)?youtube\.com/user/deadmau5',
                 'creator': 'deadmau5',
                 'description': 'md5:12c56784b8032162bb936a5f76d55360',
                 'uploader': 'deadmau5',
@@ -595,9 +609,10 @@ class YoutubeIE(YoutubeBaseInfoExtractor):
             'info_dict': {
                 'id': 'lqQg6PlCWgI',
                 'ext': 'mp4',
+                'duration': 6085,
                 'upload_date': '20150827',
                 'uploader_id': 'olympic',
-                'uploader_url': 're:https?://(?:www\.)?youtube\.com/user/olympic',
+                'uploader_url': r're:https?://(?:www\.)?youtube\.com/user/olympic',
                 'license': 'Standard YouTube License',
                 'description': 'HO09  - Women -  GER-AUS - Hockey - 31 July 2012 - London 2012 Olympic Games',
                 'uploader': 'Olympic',
@@ -614,9 +629,10 @@ class YoutubeIE(YoutubeBaseInfoExtractor):
                 'id': '_b-2C3KPAM0',
                 'ext': 'mp4',
                 'stretched_ratio': 16 / 9.,
+                'duration': 85,
                 'upload_date': '20110310',
                 'uploader_id': 'AllenMeow',
-                'uploader_url': 're:https?://(?:www\.)?youtube\.com/user/AllenMeow',
+                'uploader_url': r're:https?://(?:www\.)?youtube\.com/user/AllenMeow',
                 'description': 'made by Wacom from Korea | 字幕&加油添醋 by TY\'s Allen | 感謝heylisa00cavey1001同學熱情提供梗及翻譯',
                 'uploader': '孫艾倫',
                 'license': 'Standard YouTube License',
@@ -648,9 +664,10 @@ class YoutubeIE(YoutubeBaseInfoExtractor):
                 'ext': 'mp4',
                 'title': 'md5:7b81415841e02ecd4313668cde88737a',
                 'description': 'md5:116377fd2963b81ec4ce64b542173306',
+                'duration': 220,
                 'upload_date': '20150625',
                 'uploader_id': 'dorappi2000',
-                'uploader_url': 're:https?://(?:www\.)?youtube\.com/user/dorappi2000',
+                'uploader_url': r're:https?://(?:www\.)?youtube\.com/user/dorappi2000',
                 'uploader': 'dorappi2000',
                 'license': 'Standard YouTube License',
                 'formats': 'mincount:32',
@@ -690,10 +707,11 @@ class YoutubeIE(YoutubeBaseInfoExtractor):
                     'ext': 'mp4',
                     'title': 'teamPGP: Rocket League Noob Stream (Main Camera)',
                     'description': 'md5:dc7872fb300e143831327f1bae3af010',
+                    'duration': 7335,
                     'upload_date': '20150721',
                     'uploader': 'Beer Games Beer',
                     'uploader_id': 'beergamesbeer',
-                    'uploader_url': 're:https?://(?:www\.)?youtube\.com/user/beergamesbeer',
+                    'uploader_url': r're:https?://(?:www\.)?youtube\.com/user/beergamesbeer',
                     'license': 'Standard YouTube License',
                 },
             }, {
@@ -702,10 +720,11 @@ class YoutubeIE(YoutubeBaseInfoExtractor):
                     'ext': 'mp4',
                     'title': 'teamPGP: Rocket League Noob Stream (kreestuh)',
                     'description': 'md5:dc7872fb300e143831327f1bae3af010',
+                    'duration': 7337,
                     'upload_date': '20150721',
                     'uploader': 'Beer Games Beer',
                     'uploader_id': 'beergamesbeer',
-                    'uploader_url': 're:https?://(?:www\.)?youtube\.com/user/beergamesbeer',
+                    'uploader_url': r're:https?://(?:www\.)?youtube\.com/user/beergamesbeer',
                     'license': 'Standard YouTube License',
                 },
             }, {
@@ -714,10 +733,11 @@ class YoutubeIE(YoutubeBaseInfoExtractor):
                     'ext': 'mp4',
                     'title': 'teamPGP: Rocket League Noob Stream (grizzle)',
                     'description': 'md5:dc7872fb300e143831327f1bae3af010',
+                    'duration': 7337,
                     'upload_date': '20150721',
                     'uploader': 'Beer Games Beer',
                     'uploader_id': 'beergamesbeer',
-                    'uploader_url': 're:https?://(?:www\.)?youtube\.com/user/beergamesbeer',
+                    'uploader_url': r're:https?://(?:www\.)?youtube\.com/user/beergamesbeer',
                     'license': 'Standard YouTube License',
                 },
             }, {
@@ -726,10 +746,11 @@ class YoutubeIE(YoutubeBaseInfoExtractor):
                     'ext': 'mp4',
                     'title': 'teamPGP: Rocket League Noob Stream (zim)',
                     'description': 'md5:dc7872fb300e143831327f1bae3af010',
+                    'duration': 7334,
                     'upload_date': '20150721',
                     'uploader': 'Beer Games Beer',
                     'uploader_id': 'beergamesbeer',
-                    'uploader_url': 're:https?://(?:www\.)?youtube\.com/user/beergamesbeer',
+                    'uploader_url': r're:https?://(?:www\.)?youtube\.com/user/beergamesbeer',
                     'license': 'Standard YouTube License',
                 },
             }],
@@ -767,9 +788,10 @@ class YoutubeIE(YoutubeBaseInfoExtractor):
                 'title': '{dark walk}; Loki/AC/Dishonored; collab w/Elflover21',
                 'alt_title': 'Dark Walk',
                 'description': 'md5:8085699c11dc3f597ce0410b0dcbb34a',
+                'duration': 133,
                 'upload_date': '20151119',
                 'uploader_id': 'IronSoulElf',
-                'uploader_url': 're:https?://(?:www\.)?youtube\.com/user/IronSoulElf',
+                'uploader_url': r're:https?://(?:www\.)?youtube\.com/user/IronSoulElf',
                 'uploader': 'IronSoulElf',
                 'license': 'Standard YouTube License',
                 'creator': 'Todd Haberman, Daniel Law Heath & Aaron Kaplan',
@@ -808,10 +830,11 @@ class YoutubeIE(YoutubeBaseInfoExtractor):
                 'ext': 'mp4',
                 'title': 'md5:e41008789470fc2533a3252216f1c1d1',
                 'description': 'md5:a677553cf0840649b731a3024aeff4cc',
+                'duration': 721,
                 'upload_date': '20150127',
                 'uploader_id': 'BerkmanCenter',
-                'uploader_url': 're:https?://(?:www\.)?youtube\.com/user/BerkmanCenter',
-                'uploader': 'BerkmanCenter',
+                'uploader_url': r're:https?://(?:www\.)?youtube\.com/user/BerkmanCenter',
+                'uploader': 'The Berkman Klein Center for Internet & Society',
                 'license': 'Creative Commons Attribution license (reuse allowed)',
             },
             'params': {
@@ -826,10 +849,11 @@ class YoutubeIE(YoutubeBaseInfoExtractor):
                 'ext': 'mp4',
                 'title': 'Democratic Socialism and Foreign Policy | Bernie Sanders',
                 'description': 'md5:dda0d780d5a6e120758d1711d062a867',
+                'duration': 4060,
                 'upload_date': '20151119',
                 'uploader': 'Bernie 2016',
                 'uploader_id': 'UCH1dpzjCEiGAt8CXkryhkZg',
-                'uploader_url': 're:https?://(?:www\.)?youtube\.com/channel/UCH1dpzjCEiGAt8CXkryhkZg',
+                'uploader_url': r're:https?://(?:www\.)?youtube\.com/channel/UCH1dpzjCEiGAt8CXkryhkZg',
                 'license': 'Creative Commons Attribution license (reuse allowed)',
             },
             'params': {
@@ -856,12 +880,42 @@ class YoutubeIE(YoutubeBaseInfoExtractor):
                 'upload_date': '20150811',
                 'uploader': 'FlixMatrix',
                 'uploader_id': 'FlixMatrixKaravan',
-                'uploader_url': 're:https?://(?:www\.)?youtube\.com/user/FlixMatrixKaravan',
+                'uploader_url': r're:https?://(?:www\.)?youtube\.com/user/FlixMatrixKaravan',
                 'license': 'Standard YouTube License',
             },
             'params': {
                 'skip_download': True,
             },
+        },
+        {
+            # YouTube Red video with episode data
+            'url': 'https://www.youtube.com/watch?v=iqKdEhx-dD4',
+            'info_dict': {
+                'id': 'iqKdEhx-dD4',
+                'ext': 'mp4',
+                'title': 'Isolation - Mind Field (Ep 1)',
+                'description': 'md5:8013b7ddea787342608f63a13ddc9492',
+                'duration': 2085,
+                'upload_date': '20170118',
+                'uploader': 'Vsauce',
+                'uploader_id': 'Vsauce',
+                'uploader_url': r're:https?://(?:www\.)?youtube\.com/user/Vsauce',
+                'license': 'Standard YouTube License',
+                'series': 'Mind Field',
+                'season_number': 1,
+                'episode_number': 1,
+            },
+            'params': {
+                'skip_download': True,
+            },
+            'expected_warnings': [
+                'Skipping DASH manifest',
+            ],
+        },
+        {
+            # itag 212
+            'url': '1t24XAntNCY',
+            'only_matching': True,
         }
     ]
 
@@ -976,8 +1030,9 @@ class YoutubeIE(YoutubeBaseInfoExtractor):
 
     def _parse_sig_js(self, jscode):
         funcname = self._search_regex(
-            r'\.sig\|\|([a-zA-Z0-9$]+)\(', jscode,
-            'Initial JS player signature function name')
+            (r'(["\'])signature\1\s*,\s*(?P<sig>[a-zA-Z0-9$]+)\(',
+             r'\.sig\|\|(?P<sig>[a-zA-Z0-9$]+)\('),
+            jscode, 'Initial JS player signature function name', group='sig')
 
         jsi = JSInterpreter(jscode)
         initial_function = jsi.extract_function(funcname)
@@ -998,6 +1053,9 @@ class YoutubeIE(YoutubeBaseInfoExtractor):
 
         if player_url.startswith('//'):
             player_url = 'https:' + player_url
+        elif not re.match(r'https?://', player_url):
+            player_url = compat_urlparse.urljoin(
+                'https://www.youtube.com', player_url)
         try:
             player_id = (player_url, self._signature_cache_id(s))
             if player_id not in self._player_cache:
@@ -1448,6 +1506,16 @@ class YoutubeIE(YoutubeBaseInfoExtractor):
         else:
             video_alt_title = video_creator = None
 
+        m_episode = re.search(
+            r'<div[^>]+id="watch7-headline"[^>]*>\s*<span[^>]*>.*?>(?P<series>[^<]+)</a></b>\s*S(?P<season>\d+)\s*•\s*E(?P<episode>\d+)</span>',
+            video_webpage)
+        if m_episode:
+            series = m_episode.group('series')
+            season_number = int(m_episode.group('season'))
+            episode_number = int(m_episode.group('episode'))
+        else:
+            series = season_number = episode_number = None
+
         m_cat_container = self._search_regex(
             r'(?s)<h4[^>]*>\s*Category\s*</h4>\s*<ul[^>]*>(.*?)</ul>',
             video_webpage, 'categories', default=None)
@@ -1476,11 +1544,11 @@ class YoutubeIE(YoutubeBaseInfoExtractor):
         video_subtitles = self.extract_subtitles(video_id, video_webpage)
         automatic_captions = self.extract_automatic_captions(video_id, video_webpage)
 
-        if 'length_seconds' not in video_info:
-            self._downloader.report_warning('unable to extract video duration')
-            video_duration = None
-        else:
-            video_duration = int(compat_urllib_parse_unquote_plus(video_info['length_seconds'][0]))
+        video_duration = try_get(
+            video_info, lambda x: int_or_none(x['length_seconds'][0]))
+        if not video_duration:
+            video_duration = parse_duration(self._html_search_meta(
+                'duration', video_webpage, 'video duration'))
 
         # annotations
         video_annotations = None
@@ -1737,6 +1805,9 @@ class YoutubeIE(YoutubeBaseInfoExtractor):
             'is_live': is_live,
             'start_time': start_time,
             'end_time': end_time,
+            'series': series,
+            'season_number': season_number,
+            'episode_number': episode_number,
         }
 
 
@@ -1788,13 +1859,13 @@ class YoutubePlaylistIE(YoutubePlaylistBaseInfoExtractor):
                             youtu\.be/[0-9A-Za-z_-]{11}\?.*?\blist=
                         )
                         (
-                            (?:PL|LL|EC|UU|FL|RD|UL)?[0-9A-Za-z-_]{10,}
+                            (?:PL|LL|EC|UU|FL|RD|UL|TL)?[0-9A-Za-z-_]{10,}
                             # Top tracks, they can also include dots
                             |(?:MC)[\w\.]*
                         )
                         .*
                      |
-                        ((?:PL|LL|EC|UU|FL|RD|UL)[0-9A-Za-z-_]{10,})
+                        ((?:PL|LL|EC|UU|FL|RD|UL|TL)[0-9A-Za-z-_]{10,})
                      )"""
     _TEMPLATE_URL = 'https://www.youtube.com/playlist?list=%s&disable_polymer=true'
     _VIDEO_RE = r'href="\s*/watch\?v=(?P<id>[0-9A-Za-z_-]{11})&amp;[^"]*?index=(?P<index>\d+)(?:[^>]+>(?P<title>[^<]+))?'
@@ -1813,6 +1884,7 @@ class YoutubePlaylistIE(YoutubePlaylistBaseInfoExtractor):
             'title': 'YDL_Empty_List',
         },
         'playlist_count': 0,
+        'skip': 'This playlist is private',
     }, {
         'note': 'Playlist with deleted videos (#651). As a bonus, the video #51 is also twice in this list.',
         'url': 'https://www.youtube.com/playlist?list=PLwP_SiAcdui0KVebT0mU9Apz359a4ubsC',
@@ -1844,6 +1916,7 @@ class YoutubePlaylistIE(YoutubePlaylistBaseInfoExtractor):
             'id': 'PLtPgu7CB4gbY9oDN3drwC3cMbJggS7dKl',
         },
         'playlist_count': 2,
+        'skip': 'This playlist is private',
     }, {
         'note': 'embedded',
         'url': 'https://www.youtube.com/embed/videoseries?list=PL6IaIsEjSbf96XFRuNccS_RuEXwNdsoEu',
@@ -1877,7 +1950,7 @@ class YoutubePlaylistIE(YoutubePlaylistBaseInfoExtractor):
             'title': "Smiley's People 01 detective, Adventure Series, Action",
             'uploader': 'STREEM',
             'uploader_id': 'UCyPhqAZgwYWZfxElWVbVJng',
-            'uploader_url': 're:https?://(?:www\.)?youtube\.com/channel/UCyPhqAZgwYWZfxElWVbVJng',
+            'uploader_url': r're:https?://(?:www\.)?youtube\.com/channel/UCyPhqAZgwYWZfxElWVbVJng',
             'upload_date': '20150526',
             'license': 'Standard YouTube License',
             'description': 'md5:507cdcb5a49ac0da37a920ece610be80',
@@ -1898,7 +1971,7 @@ class YoutubePlaylistIE(YoutubePlaylistBaseInfoExtractor):
             'title': 'Small Scale Baler and Braiding Rugs',
             'uploader': 'Backus-Page House Museum',
             'uploader_id': 'backuspagemuseum',
-            'uploader_url': 're:https?://(?:www\.)?youtube\.com/user/backuspagemuseum',
+            'uploader_url': r're:https?://(?:www\.)?youtube\.com/user/backuspagemuseum',
             'upload_date': '20161008',
             'license': 'Standard YouTube License',
             'description': 'md5:800c0c78d5eb128500bffd4f0b4f2e8a',
@@ -1914,6 +1987,9 @@ class YoutubePlaylistIE(YoutubePlaylistBaseInfoExtractor):
     }, {
         'url': 'https://youtu.be/uWyaPkt-VOI?list=PL9D9FC436B881BA21',
         'only_matching': True,
+    }, {
+        'url': 'TLGGrESM50VT6acwMjAyMjAxNw',
+        'only_matching': True,
     }]
 
     def _real_initialize(self):
@@ -1955,14 +2031,18 @@ class YoutubePlaylistIE(YoutubePlaylistBaseInfoExtractor):
         url = self._TEMPLATE_URL % playlist_id
         page = self._download_webpage(url, playlist_id)
 
-        for match in re.findall(r'<div class="yt-alert-message">([^<]+)</div>', page):
+        # the yt-alert-message now has tabindex attribute (see https://github.com/rg3/youtube-dl/issues/11604)
+        for match in re.findall(r'<div class="yt-alert-message"[^>]*>([^<]+)</div>', page):
             match = match.strip()
             # Check if the playlist exists or is private
-            if re.match(r'[^<]*(The|This) playlist (does not exist|is private)[^<]*', match):
-                raise ExtractorError(
-                    'The playlist doesn\'t exist or is private, use --username or '
-                    '--netrc to access it.',
-                    expected=True)
+            mobj = re.match(r'[^<]*(?:The|This) playlist (?P<reason>does not exist|is private)[^<]*', match)
+            if mobj:
+                reason = mobj.group('reason')
+                message = 'This playlist %s' % reason
+                if 'private' in reason:
+                    message += ', use --username or --netrc to access it'
+                message += '.'
+                raise ExtractorError(message, expected=True)
             elif re.match(r'[^<]*Invalid parameters[^<]*', match):
                 raise ExtractorError(
                     'Invalid parameters. Maybe URL is incorrect.',
@@ -2186,7 +2266,7 @@ class YoutubeLiveIE(YoutubeBaseInfoExtractor):
             'title': 'The Young Turks - Live Main Show',
             'uploader': 'The Young Turks',
             'uploader_id': 'TheYoungTurks',
-            'uploader_url': 're:https?://(?:www\.)?youtube\.com/user/TheYoungTurks',
+            'uploader_url': r're:https?://(?:www\.)?youtube\.com/user/TheYoungTurks',
             'upload_date': '20150715',
             'license': 'Standard YouTube License',
             'description': 'md5:438179573adcdff3c97ebb1ee632b891',
@@ -2270,18 +2350,18 @@ class YoutubeSearchIE(SearchInfoExtractor, YoutubePlaylistIE):
         videos = []
         limit = n
 
+        url_query = {
+            'search_query': query.encode('utf-8'),
+        }
+        url_query.update(self._EXTRA_QUERY_ARGS)
+        result_url = 'https://www.youtube.com/results?' + compat_urllib_parse_urlencode(url_query)
+
         for pagenum in itertools.count(1):
-            url_query = {
-                'search_query': query.encode('utf-8'),
-                'page': pagenum,
-                'spf': 'navigate',
-            }
-            url_query.update(self._EXTRA_QUERY_ARGS)
-            result_url = 'https://www.youtube.com/results?' + compat_urllib_parse_urlencode(url_query)
             data = self._download_json(
                 result_url, video_id='query "%s"' % query,
                 note='Downloading page %s' % pagenum,
-                errnote='Unable to download API page')
+                errnote='Unable to download API page',
+                query={'spf': 'navigate'})
             html_content = data[1]['body']['content']
 
             if 'class="search-message' in html_content:
@@ -2293,6 +2373,12 @@ class YoutubeSearchIE(SearchInfoExtractor, YoutubePlaylistIE):
             videos += new_videos
             if not new_videos or len(videos) > limit:
                 break
+            next_link = self._html_search_regex(
+                r'href="(/results\?[^"]*\bsp=[^"]+)"[^>]*>\s*<span[^>]+class="[^"]*\byt-uix-button-content\b[^"]*"[^>]*>Next',
+                html_content, 'next link', default=None)
+            if next_link is None:
+                break
+            result_url = compat_urlparse.urljoin('https://www.youtube.com/', next_link)
 
         if len(videos) > n:
             videos = videos[:n]