]> Raphaƫl G. Git Repositories - youtubedl/blobdiff - youtube_dl/extractor/youtube.py
Finalize release.
[youtubedl] / youtube_dl / extractor / youtube.py
index f7471895052ed328465b888ba402eeebaa1351d6..d45545ee490867d8f2e89d5c36e038364281a72d 100644 (file)
@@ -1,50 +1,55 @@
 # coding: utf-8
 
 # coding: utf-8
 
+import collections
+import errno
+import io
+import itertools
 import json
 import json
-import netrc
+import os.path
 import re
 import re
-import socket
-import itertools
+import struct
+import traceback
+import zlib
 
 from .common import InfoExtractor, SearchInfoExtractor
 
 from .common import InfoExtractor, SearchInfoExtractor
+from .subtitles import SubtitlesInfoExtractor
+from ..jsinterp import JSInterpreter
 from ..utils import (
 from ..utils import (
-    compat_http_client,
+    compat_chr,
     compat_parse_qs,
     compat_parse_qs,
-    compat_urllib_error,
     compat_urllib_parse,
     compat_urllib_request,
     compat_urllib_parse,
     compat_urllib_request,
+    compat_urlparse,
     compat_str,
 
     clean_html,
     compat_str,
 
     clean_html,
+    get_cachedir,
     get_element_by_id,
     get_element_by_id,
+    get_element_by_attribute,
     ExtractorError,
     ExtractorError,
+    int_or_none,
+    PagedList,
     unescapeHTML,
     unified_strdate,
     orderedSet,
     unescapeHTML,
     unified_strdate,
     orderedSet,
+    write_json_file,
+    uppercase_escape,
 )
 
 class YoutubeBaseInfoExtractor(InfoExtractor):
     """Provide base functions for Youtube extractors"""
     _LOGIN_URL = 'https://accounts.google.com/ServiceLogin'
     _LANG_URL = r'https://www.youtube.com/?hl=en&persist_hl=1&gl=US&persist_gl=1&opt_out_ackd=1'
 )
 
 class YoutubeBaseInfoExtractor(InfoExtractor):
     """Provide base functions for Youtube extractors"""
     _LOGIN_URL = 'https://accounts.google.com/ServiceLogin'
     _LANG_URL = r'https://www.youtube.com/?hl=en&persist_hl=1&gl=US&persist_gl=1&opt_out_ackd=1'
-    _AGE_URL = 'http://www.youtube.com/verify_age?next_url=/&gl=US&hl=en'
+    _AGE_URL = 'https://www.youtube.com/verify_age?next_url=/&gl=US&hl=en'
     _NETRC_MACHINE = 'youtube'
     # If True it will raise an error if no login info is provided
     _LOGIN_REQUIRED = False
 
     _NETRC_MACHINE = 'youtube'
     # If True it will raise an error if no login info is provided
     _LOGIN_REQUIRED = False
 
-    def report_lang(self):
-        """Report attempt to set language."""
-        self.to_screen(u'Setting language')
-
     def _set_language(self):
     def _set_language(self):
-        request = compat_urllib_request.Request(self._LANG_URL)
-        try:
-            self.report_lang()
-            compat_urllib_request.urlopen(request).read()
-        except (compat_urllib_error.URLError, compat_http_client.HTTPException, socket.error) as err:
-            self._downloader.report_warning(u'unable to set language: %s' % compat_str(err))
-            return False
-        return True
+        return bool(self._download_webpage(
+            self._LANG_URL, None,
+            note=u'Setting language', errnote='unable to set language',
+            fatal=False))
 
     def _login(self):
         (username, password) = self._get_login_info()
 
     def _login(self):
         (username, password) = self._get_login_info()
@@ -54,21 +59,15 @@ class YoutubeBaseInfoExtractor(InfoExtractor):
                 raise ExtractorError(u'No login info available, needed for using %s.' % self.IE_NAME, expected=True)
             return False
 
                 raise ExtractorError(u'No login info available, needed for using %s.' % self.IE_NAME, expected=True)
             return False
 
-        request = compat_urllib_request.Request(self._LOGIN_URL)
-        try:
-            login_page = compat_urllib_request.urlopen(request).read().decode('utf-8')
-        except (compat_urllib_error.URLError, compat_http_client.HTTPException, socket.error) as err:
-            self._downloader.report_warning(u'unable to fetch login page: %s' % compat_str(err))
-            return False
+        login_page = self._download_webpage(
+            self._LOGIN_URL, None,
+            note=u'Downloading login page',
+            errnote=u'unable to fetch login page', fatal=False)
+        if login_page is False:
+            return
 
 
-        galx = None
-        dsh = None
-        match = re.search(re.compile(r'<input.+?name="GALX".+?value="(.+?)"', re.DOTALL), login_page)
-        if match:
-          galx = match.group(1)
-        match = re.search(re.compile(r'<input.+?name="dsh".+?value="(.+?)"', re.DOTALL), login_page)
-        if match:
-          dsh = match.group(1)
+        galx = self._search_regex(r'(?s)<input.+?name="GALX".+?value="(.+?)"',
+                                  login_page, u'Login GALX parameter')
 
         # Log in
         login_form_strs = {
 
         # Log in
         login_form_strs = {
@@ -82,7 +81,6 @@ class YoutubeBaseInfoExtractor(InfoExtractor):
                 u'checkConnection': u'',
                 u'checkedDomains': u'youtube',
                 u'dnConn': u'',
                 u'checkConnection': u'',
                 u'checkedDomains': u'youtube',
                 u'dnConn': u'',
-                u'dsh': dsh,
                 u'pstMsg': u'0',
                 u'rmShown': u'1',
                 u'secTok': u'',
                 u'pstMsg': u'0',
                 u'rmShown': u'1',
                 u'secTok': u'',
@@ -96,29 +94,29 @@ class YoutubeBaseInfoExtractor(InfoExtractor):
         # chokes on unicode
         login_form = dict((k.encode('utf-8'), v.encode('utf-8')) for k,v in login_form_strs.items())
         login_data = compat_urllib_parse.urlencode(login_form).encode('ascii')
         # chokes on unicode
         login_form = dict((k.encode('utf-8'), v.encode('utf-8')) for k,v in login_form_strs.items())
         login_data = compat_urllib_parse.urlencode(login_form).encode('ascii')
-        request = compat_urllib_request.Request(self._LOGIN_URL, login_data)
-        try:
-            self.report_login()
-            login_results = compat_urllib_request.urlopen(request).read().decode('utf-8')
-            if re.search(r'(?i)<form[^>]* id="gaia_loginform"', login_results) is not None:
-                self._downloader.report_warning(u'unable to log in: bad username or password')
-                return False
-        except (compat_urllib_error.URLError, compat_http_client.HTTPException, socket.error) as err:
-            self._downloader.report_warning(u'unable to log in: %s' % compat_str(err))
+
+        req = compat_urllib_request.Request(self._LOGIN_URL, login_data)
+        login_results = self._download_webpage(
+            req, None,
+            note=u'Logging in', errnote=u'unable to log in', fatal=False)
+        if login_results is False:
+            return False
+        if re.search(r'(?i)<form[^>]* id="gaia_loginform"', login_results) is not None:
+            self._downloader.report_warning(u'unable to log in: bad username or password')
             return False
         return True
 
     def _confirm_age(self):
         age_form = {
             return False
         return True
 
     def _confirm_age(self):
         age_form = {
-                'next_url':     '/',
-                'action_confirm':   'Confirm',
-                }
-        request = compat_urllib_request.Request(self._AGE_URL, compat_urllib_parse.urlencode(age_form))
-        try:
-            self.report_age_confirmation()
-            compat_urllib_request.urlopen(request).read().decode('utf-8')
-        except (compat_urllib_error.URLError, compat_http_client.HTTPException, socket.error) as err:
-            raise ExtractorError(u'Unable to confirm age: %s' % compat_str(err))
+            'next_url': '/',
+            'action_confirm': 'Confirm',
+        }
+        req = compat_urllib_request.Request(self._AGE_URL,
+            compat_urllib_parse.urlencode(age_form).encode('ascii'))
+
+        self._download_webpage(
+            req, None,
+            note=u'Confirming age', errnote=u'Unable to confirm age')
         return True
 
     def _real_initialize(self):
         return True
 
     def _real_initialize(self):
@@ -130,98 +128,111 @@ class YoutubeBaseInfoExtractor(InfoExtractor):
             return
         self._confirm_age()
 
             return
         self._confirm_age()
 
-class YoutubeIE(YoutubeBaseInfoExtractor):
+
+class YoutubeIE(YoutubeBaseInfoExtractor, SubtitlesInfoExtractor):
     IE_DESC = u'YouTube.com'
     IE_DESC = u'YouTube.com'
-    _VALID_URL = r"""^
+    _VALID_URL = r"""(?x)^
                      (
                      (
-                         (?:https?://)?                                       # http(s):// (optional)
-                         (?:youtu\.be/|(?:\w+\.)?youtube(?:-nocookie)?\.com/|
-                            tube\.majestyc\.net/)                             # the various hostnames, with wildcard subdomains
+                         (?:https?://|//)?                                    # http(s):// or protocol-independent URL (optional)
+                         (?:(?:(?:(?:\w+\.)?[yY][oO][uU][tT][uU][bB][eE](?:-nocookie)?\.com/|
+                            (?:www\.)?deturl\.com/www\.youtube\.com/|
+                            (?:www\.)?pwnyoutube\.com/|
+                            (?:www\.)?yourepeat\.com/|
+                            tube\.majestyc\.net/|
+                            youtube\.googleapis\.com/)                        # the various hostnames, with wildcard subdomains
                          (?:.*?\#/)?                                          # handle anchor (#/) redirect urls
                          (?:                                                  # the various things that can precede the ID:
                              (?:(?:v|embed|e)/)                               # v/ or embed/ or e/
                              |(?:                                             # or the v= param in all its forms
                          (?:.*?\#/)?                                          # handle anchor (#/) redirect urls
                          (?:                                                  # the various things that can precede the ID:
                              (?:(?:v|embed|e)/)                               # v/ or embed/ or e/
                              |(?:                                             # or the v= param in all its forms
-                                 (?:watch|movie(?:_popup)?(?:\.php)?)?              # preceding watch(_popup|.php) or nothing (like /?v=xxxx)
+                                 (?:(?:watch|movie)(?:_popup)?(?:\.php)?/?)?  # preceding watch(_popup|.php) or nothing (like /?v=xxxx)
                                  (?:\?|\#!?)                                  # the params delimiter ? or # or #!
                                  (?:.*?&)?                                    # any other preceding param (like /?s=tuff&v=xxxx)
                                  v=
                              )
                                  (?:\?|\#!?)                                  # the params delimiter ? or # or #!
                                  (?:.*?&)?                                    # any other preceding param (like /?s=tuff&v=xxxx)
                                  v=
                              )
-                         )?                                                   # optional -> youtube.com/xxxx is OK
+                         ))
+                         |youtu\.be/                                          # just youtu.be/xxxx
+                         |https?://(?:www\.)?cleanvideosearch\.com/media/action/yt/watch\?videoId=
+                         )
                      )?                                                       # all until now is optional -> you can pass the naked ID
                      )?                                                       # all until now is optional -> you can pass the naked ID
-                     ([0-9A-Za-z_-]+)                                         # here is it! the YouTube video ID
+                     ([0-9A-Za-z_-]{11})                                      # here is it! the YouTube video ID
                      (?(1).+)?                                                # if we found the ID, everything can follow
                      $"""
     _NEXT_URL_RE = r'[\?&]next_url=([^&]+)'
                      (?(1).+)?                                                # if we found the ID, everything can follow
                      $"""
     _NEXT_URL_RE = r'[\?&]next_url=([^&]+)'
-    # Listed in order of quality
-    _available_formats = ['38', '37', '46', '22', '45', '35', '44', '34', '18', '43', '6', '5', '17', '13',
-                          '95', '94', '93', '92', '132', '151',
-                          '85', '84', '102', '83', '101', '82', '100',
-                          ]
-    _available_formats_prefer_free = ['38', '46', '37', '45', '22', '44', '35', '43', '34', '18', '6', '5', '17', '13',
-                                      '95', '94', '93', '92', '132', '151',
-                                      '85', '102', '84', '101', '83', '100', '82',
-                                      ]
-    _video_extensions = {
-        '13': '3gp',
-        '17': 'mp4',
-        '18': 'mp4',
-        '22': 'mp4',
-        '37': 'mp4',
-        '38': 'mp4',
-        '43': 'webm',
-        '44': 'webm',
-        '45': 'webm',
-        '46': 'webm',
+    _formats = {
+        '5': {'ext': 'flv', 'width': 400, 'height': 240},
+        '6': {'ext': 'flv', 'width': 450, 'height': 270},
+        '13': {'ext': '3gp'},
+        '17': {'ext': '3gp', 'width': 176, 'height': 144},
+        '18': {'ext': 'mp4', 'width': 640, 'height': 360},
+        '22': {'ext': 'mp4', 'width': 1280, 'height': 720},
+        '34': {'ext': 'flv', 'width': 640, 'height': 360},
+        '35': {'ext': 'flv', 'width': 854, 'height': 480},
+        '36': {'ext': '3gp', 'width': 320, 'height': 240},
+        '37': {'ext': 'mp4', 'width': 1920, 'height': 1080},
+        '38': {'ext': 'mp4', 'width': 4096, 'height': 3072},
+        '43': {'ext': 'webm', 'width': 640, 'height': 360},
+        '44': {'ext': 'webm', 'width': 854, 'height': 480},
+        '45': {'ext': 'webm', 'width': 1280, 'height': 720},
+        '46': {'ext': 'webm', 'width': 1920, 'height': 1080},
+
 
         # 3d videos
 
         # 3d videos
-        '82': 'mp4',
-        '83': 'mp4',
-        '84': 'mp4',
-        '85': 'mp4',
-        '100': 'webm',
-        '101': 'webm',
-        '102': 'webm',
-        
-        # videos that use m3u8
-        '92': 'mp4',
-        '93': 'mp4',
-        '94': 'mp4',
-        '95': 'mp4',
-        '96': 'mp4',
-        '132': 'mp4',
-        '151': 'mp4',
+        '82': {'ext': 'mp4', 'height': 360, 'format_note': '3D', 'preference': -20},
+        '83': {'ext': 'mp4', 'height': 480, 'format_note': '3D', 'preference': -20},
+        '84': {'ext': 'mp4', 'height': 720, 'format_note': '3D', 'preference': -20},
+        '85': {'ext': 'mp4', 'height': 1080, 'format_note': '3D', 'preference': -20},
+        '100': {'ext': 'webm', 'height': 360, 'format_note': '3D', 'preference': -20},
+        '101': {'ext': 'webm', 'height': 480, 'format_note': '3D', 'preference': -20},
+        '102': {'ext': 'webm', 'height': 720, 'format_note': '3D', 'preference': -20},
+
+        # Apple HTTP Live Streaming
+        '92': {'ext': 'mp4', 'height': 240, 'format_note': 'HLS', 'preference': -10},
+        '93': {'ext': 'mp4', 'height': 360, 'format_note': 'HLS', 'preference': -10},
+        '94': {'ext': 'mp4', 'height': 480, 'format_note': 'HLS', 'preference': -10},
+        '95': {'ext': 'mp4', 'height': 720, 'format_note': 'HLS', 'preference': -10},
+        '96': {'ext': 'mp4', 'height': 1080, 'format_note': 'HLS', 'preference': -10},
+        '132': {'ext': 'mp4', 'height': 240, 'format_note': 'HLS', 'preference': -10},
+        '151': {'ext': 'mp4', 'height': 72, 'format_note': 'HLS', 'preference': -10},
+
+        # DASH mp4 video
+        '133': {'ext': 'mp4', 'height': 240, 'format_note': 'DASH video', 'acodec': 'none', 'preference': -40},
+        '134': {'ext': 'mp4', 'height': 360, 'format_note': 'DASH video', 'acodec': 'none', 'preference': -40},
+        '135': {'ext': 'mp4', 'height': 480, 'format_note': 'DASH video', 'acodec': 'none', 'preference': -40},
+        '136': {'ext': 'mp4', 'height': 720, 'format_note': 'DASH video', 'acodec': 'none', 'preference': -40},
+        '137': {'ext': 'mp4', 'height': 1080, 'format_note': 'DASH video', 'acodec': 'none', 'preference': -40},
+        '138': {'ext': 'mp4', 'height': 2160, 'format_note': 'DASH video', 'acodec': 'none', 'preference': -40},
+        '160': {'ext': 'mp4', 'height': 144, 'format_note': 'DASH video', 'acodec': 'none', 'preference': -40},
+        '264': {'ext': 'mp4', 'height': 1440, 'format_note': 'DASH video', 'acodec': 'none', 'preference': -40},
+
+        # Dash mp4 audio
+        '139': {'ext': 'm4a', 'format_note': 'DASH audio', 'vcodec': 'none', 'abr': 48, 'preference': -50},
+        '140': {'ext': 'm4a', 'format_note': 'DASH audio', 'vcodec': 'none', 'abr': 128, 'preference': -50},
+        '141': {'ext': 'm4a', 'format_note': 'DASH audio', 'vcodec': 'none', 'abr': 256, 'preference': -50},
+
+        # Dash webm
+        '167': {'ext': 'webm', 'height': 360, 'width': 640, 'format_note': 'DASH video', 'acodec': 'none', 'container': 'webm', 'vcodec': 'VP8', 'preference': -40},
+        '168': {'ext': 'webm', 'height': 480, 'width': 854, 'format_note': 'DASH video', 'acodec': 'none', 'container': 'webm', 'vcodec': 'VP8', 'preference': -40},
+        '169': {'ext': 'webm', 'height': 720, 'width': 1280, 'format_note': 'DASH video', 'acodec': 'none', 'container': 'webm', 'vcodec': 'VP8', 'preference': -40},
+        '170': {'ext': 'webm', 'height': 1080, 'width': 1920, 'format_note': 'DASH video', 'acodec': 'none', 'container': 'webm', 'vcodec': 'VP8', 'preference': -40},
+        '218': {'ext': 'webm', 'height': 480, 'width': 854, 'format_note': 'DASH video', 'acodec': 'none', 'container': 'webm', 'vcodec': 'VP8', 'preference': -40},
+        '219': {'ext': 'webm', 'height': 480, 'width': 854, 'format_note': 'DASH video', 'acodec': 'none', 'container': 'webm', 'vcodec': 'VP8', 'preference': -40},
+        '242': {'ext': 'webm', 'height': 240, 'format_note': 'DASH video', 'acodec': 'none', 'preference': -40},
+        '243': {'ext': 'webm', 'height': 360, 'format_note': 'DASH video', 'acodec': 'none', 'preference': -40},
+        '244': {'ext': 'webm', 'height': 480, 'format_note': 'DASH video', 'acodec': 'none', 'preference': -40},
+        '245': {'ext': 'webm', 'height': 480, 'format_note': 'DASH video', 'acodec': 'none', 'preference': -40},
+        '246': {'ext': 'webm', 'height': 480, 'format_note': 'DASH video', 'acodec': 'none', 'preference': -40},
+        '247': {'ext': 'webm', 'height': 720, 'format_note': 'DASH video', 'acodec': 'none', 'preference': -40},
+        '248': {'ext': 'webm', 'height': 1080, 'format_note': 'DASH video', 'acodec': 'none', 'preference': -40},
+        '271': {'ext': 'webm', 'height': 1440, 'format_note': 'DASH video', 'acodec': 'none', 'preference': -40},
+
+        # Dash webm audio
+        '171': {'ext': 'webm', 'vcodec': 'none', 'format_note': 'DASH audio', 'abr': 48, 'preference': -50},
+        '172': {'ext': 'webm', 'vcodec': 'none', 'format_note': 'DASH audio', 'abr': 256, 'preference': -50},
+
+        # RTMP (unnamed)
+        '_rtmp': {'protocol': 'rtmp'},
     }
     }
-    _video_dimensions = {
-        '5': '240x400',
-        '6': '???',
-        '13': '???',
-        '17': '144x176',
-        '18': '360x640',
-        '22': '720x1280',
-        '34': '360x640',
-        '35': '480x854',
-        '37': '1080x1920',
-        '38': '3072x4096',
-        '43': '360x640',
-        '44': '480x854',
-        '45': '720x1280',
-        '46': '1080x1920',
-        '82': '360p',
-        '83': '480p',
-        '84': '720p',
-        '85': '1080p',
-        '92': '240p',
-        '93': '360p',
-        '94': '480p',
-        '95': '720p',
-        '96': '1080p',
-        '100': '360p',
-        '101': '480p',
-        '102': '720p',        
-        '132': '240p',
-        '151': '72p',
-    }
-    _3d_itags = ['85', '84', '102', '83', '101', '82', '100']
+
     IE_NAME = u'youtube'
     _TESTS = [
         {
     IE_NAME = u'youtube'
     _TESTS = [
         {
@@ -232,19 +243,8 @@ class YoutubeIE(YoutubeBaseInfoExtractor):
                 u"uploader": u"Philipp Hagemeister",
                 u"uploader_id": u"phihag",
                 u"upload_date": u"20121002",
                 u"uploader": u"Philipp Hagemeister",
                 u"uploader_id": u"phihag",
                 u"upload_date": u"20121002",
-                u"description": u"test chars:  \"'/\\Ƥā†­š•\n\nThis is a test video for youtube-dl.\n\nFor more information, contact phihag@phihag.de ."
-            }
-        },
-        {
-            u"url":  u"http://www.youtube.com/watch?v=1ltcDfZMA3U",
-            u"file":  u"1ltcDfZMA3U.flv",
-            u"note": u"Test VEVO video (#897)",
-            u"info_dict": {
-                u"upload_date": u"20070518",
-                u"title": u"Maps - It Will Find You",
-                u"description": u"Music video by Maps performing It Will Find You.",
-                u"uploader": u"MuteUSA",
-                u"uploader_id": u"MuteUSA"
+                u"description": u"test chars:  \"'/\\Ƥā†­š•\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 .",
+                u"categories": [u'Science & Technology'],
             }
         },
         {
             }
         },
         {
@@ -254,8 +254,8 @@ class YoutubeIE(YoutubeBaseInfoExtractor):
             u"info_dict": {
                 u"upload_date": u"20120506",
                 u"title": u"Icona Pop - I Love It (feat. Charli XCX) [OFFICIAL VIDEO]",
             u"info_dict": {
                 u"upload_date": u"20120506",
                 u"title": u"Icona Pop - I Love It (feat. Charli XCX) [OFFICIAL VIDEO]",
-                u"description": u"md5:b085c9804f5ab69f4adea963a2dceb3c",
-                u"uploader": u"IconaPop",
+                u"description": u"md5:fea86fda2d5a5784273df5c7cc994d9f",
+                u"uploader": u"Icona Pop",
                 u"uploader_id": u"IconaPop"
             }
         },
                 u"uploader_id": u"IconaPop"
             }
         },
@@ -272,18 +272,48 @@ class YoutubeIE(YoutubeBaseInfoExtractor):
             }
         },
         {
             }
         },
         {
-            u'url': u'https://www.youtube.com/watch?v=TGi3HqYrWHE',
-            u'file': u'TGi3HqYrWHE.mp4',
-            u'note': u'm3u8 video',
+            u"url":  u"//www.YouTube.com/watch?v=yZIXLfi8CZQ",
+            u"file":  u"yZIXLfi8CZQ.mp4",
+            u"note": u"Embed-only video (#1746)",
+            u"info_dict": {
+                u"upload_date": u"20120608",
+                u"title": u"Principal Sexually Assaults A Teacher - Episode 117 - 8th June 2012",
+                u"description": u"md5:09b78bd971f1e3e289601dfba15ca4f7",
+                u"uploader": u"SET India",
+                u"uploader_id": u"setindia"
+            }
+        },
+        {
+            u"url": u"http://www.youtube.com/watch?v=a9LDPn-MO4I",
+            u"file": u"a9LDPn-MO4I.m4a",
+            u"note": u"256k DASH audio (format 141) via DASH manifest",
+            u"info_dict": {
+                u"upload_date": "20121002",
+                u"uploader_id": "8KVIDEO",
+                u"description": "No description available.",
+                u"uploader": "8KVIDEO",
+                u"title": "UHDTV TEST 8K VIDEO.mp4"
+            },
+            u"params": {
+                u"youtube_include_dash_manifest": True,
+                u"format": "141",
+            },
+        },
+        # DASH manifest with encrypted signature
+        {
+            u'url': u'https://www.youtube.com/watch?v=IB3lcPjvWLA',
             u'info_dict': {
             u'info_dict': {
-                u'title': u'Triathlon - Men - London 2012 Olympic Games',
-                u'description': u'- Men -  TR02 - Triathlon - 07 August 2012 - London 2012 Olympic Games',
-                u'uploader': u'olympic',
-                u'upload_date': u'20120807',
-                u'uploader_id': u'olympic',
+                u'id': u'IB3lcPjvWLA',
+                u'ext': u'm4a',
+                u'title': u'Afrojack - The Spark ft. Spree Wilson',
+                u'description': u'md5:9717375db5a9a3992be4668bbf3bc0a8',
+                u'uploader': u'AfrojackVEVO',
+                u'uploader_id': u'AfrojackVEVO',
+                u'upload_date': u'20131011',
             },
             },
-            u'params': {
-                u'skip_download': True,
+            u"params": {
+                u'youtube_include_dash_manifest': True,
+                u'format': '141',
             },
         },
     ]
             },
         },
     ]
@@ -292,30 +322,17 @@ class YoutubeIE(YoutubeBaseInfoExtractor):
     @classmethod
     def suitable(cls, url):
         """Receives a URL and returns True if suitable for this IE."""
     @classmethod
     def suitable(cls, url):
         """Receives a URL and returns True if suitable for this IE."""
-        if YoutubePlaylistIE.suitable(url) or YoutubeSubscriptionsIE.suitable(url): return False
-        return re.match(cls._VALID_URL, url, re.VERBOSE) is not None
+        if YoutubePlaylistIE.suitable(url): return False
+        return re.match(cls._VALID_URL, url) is not None
 
 
-    def report_video_webpage_download(self, video_id):
-        """Report attempt to download video webpage."""
-        self.to_screen(u'%s: Downloading video webpage' % video_id)
+    def __init__(self, *args, **kwargs):
+        super(YoutubeIE, self).__init__(*args, **kwargs)
+        self._player_cache = {}
 
     def report_video_info_webpage_download(self, video_id):
         """Report attempt to download video info webpage."""
         self.to_screen(u'%s: Downloading video info webpage' % video_id)
 
 
     def report_video_info_webpage_download(self, video_id):
         """Report attempt to download video info webpage."""
         self.to_screen(u'%s: Downloading video info webpage' % video_id)
 
-    def report_video_subtitles_download(self, video_id):
-        """Report attempt to download video info webpage."""
-        self.to_screen(u'%s: Checking available subtitles' % video_id)
-
-    def report_video_subtitles_request(self, video_id, sub_lang, format):
-        """Report attempt to download video info webpage."""
-        self.to_screen(u'%s: Downloading video subtitles for %s.%s' % (video_id, sub_lang, format))
-
-    def report_video_subtitles_available(self, video_id, sub_lang_list):
-        """Report available subtitles."""
-        sub_lang = ",".join(list(sub_lang_list.keys()))
-        self.to_screen(u'%s: Available subtitles for video: %s' % (video_id, sub_lang))
-
     def report_information_extraction(self, video_id):
         """Report attempt to extract video information."""
         self.to_screen(u'%s: Extracting video information' % video_id)
     def report_information_extraction(self, video_id):
         """Report attempt to extract video information."""
         self.to_screen(u'%s: Extracting video information' % video_id)
@@ -328,196 +345,672 @@ class YoutubeIE(YoutubeBaseInfoExtractor):
         """Indicate the download will use the RTMP protocol."""
         self.to_screen(u'RTMP download detected')
 
         """Indicate the download will use the RTMP protocol."""
         self.to_screen(u'RTMP download detected')
 
-    def _decrypt_signature(self, s):
+    def _extract_signature_function(self, video_id, player_url, slen):
+        id_m = re.match(r'.*-(?P<id>[a-zA-Z0-9_-]+)\.(?P<ext>[a-z]+)$',
+                        player_url)
+        player_type = id_m.group('ext')
+        player_id = id_m.group('id')
+
+        # Read from filesystem cache
+        func_id = '%s_%s_%d' % (player_type, player_id, slen)
+        assert os.path.basename(func_id) == func_id
+        cache_dir = get_cachedir(self._downloader.params)
+
+        cache_enabled = cache_dir is not None
+        if cache_enabled:
+            cache_fn = os.path.join(os.path.expanduser(cache_dir),
+                                    u'youtube-sigfuncs',
+                                    func_id + '.json')
+            try:
+                with io.open(cache_fn, 'r', encoding='utf-8') as cachef:
+                    cache_spec = json.load(cachef)
+                return lambda s: u''.join(s[i] for i in cache_spec)
+            except IOError:
+                pass  # No cache available
+
+        if player_type == 'js':
+            code = self._download_webpage(
+                player_url, video_id,
+                note=u'Downloading %s player %s' % (player_type, player_id),
+                errnote=u'Download of %s failed' % player_url)
+            res = self._parse_sig_js(code)
+        elif player_type == 'swf':
+            urlh = self._request_webpage(
+                player_url, video_id,
+                note=u'Downloading %s player %s' % (player_type, player_id),
+                errnote=u'Download of %s failed' % player_url)
+            code = urlh.read()
+            res = self._parse_sig_swf(code)
+        else:
+            assert False, 'Invalid player type %r' % player_type
+
+        if cache_enabled:
+            try:
+                test_string = u''.join(map(compat_chr, range(slen)))
+                cache_res = res(test_string)
+                cache_spec = [ord(c) for c in cache_res]
+                try:
+                    os.makedirs(os.path.dirname(cache_fn))
+                except OSError as ose:
+                    if ose.errno != errno.EEXIST:
+                        raise
+                write_json_file(cache_spec, cache_fn)
+            except Exception:
+                tb = traceback.format_exc()
+                self._downloader.report_warning(
+                    u'Writing cache to %r failed: %s' % (cache_fn, tb))
+
+        return res
+
+    def _print_sig_code(self, func, slen):
+        def gen_sig_code(idxs):
+            def _genslice(start, end, step):
+                starts = u'' if start == 0 else str(start)
+                ends = (u':%d' % (end+step)) if end + step >= 0 else u':'
+                steps = u'' if step == 1 else (u':%d' % step)
+                return u's[%s%s%s]' % (starts, ends, steps)
+
+            step = None
+            start = '(Never used)'  # Quelch pyflakes warnings - start will be
+                                    # set as soon as step is set
+            for i, prev in zip(idxs[1:], idxs[:-1]):
+                if step is not None:
+                    if i - prev == step:
+                        continue
+                    yield _genslice(start, prev, step)
+                    step = None
+                    continue
+                if i - prev in [-1, 1]:
+                    step = i - prev
+                    start = prev
+                    continue
+                else:
+                    yield u's[%d]' % prev
+            if step is None:
+                yield u's[%d]' % i
+            else:
+                yield _genslice(start, i, step)
+
+        test_string = u''.join(map(compat_chr, range(slen)))
+        cache_res = func(test_string)
+        cache_spec = [ord(c) for c in cache_res]
+        expr_code = u' + '.join(gen_sig_code(cache_spec))
+        code = u'if len(s) == %d:\n    return %s\n' % (slen, expr_code)
+        self.to_screen(u'Extracted signature function:\n' + code)
+
+    def _parse_sig_js(self, jscode):
+        funcname = self._search_regex(
+            r'signature=([$a-zA-Z]+)', jscode,
+             u'Initial JS player signature function name')
+
+        jsi = JSInterpreter(jscode)
+        initial_function = jsi.extract_function(funcname)
+        return lambda s: initial_function([s])
+
+    def _parse_sig_swf(self, file_contents):
+        if file_contents[1:3] != b'WS':
+            raise ExtractorError(
+                u'Not an SWF file; header is %r' % file_contents[:3])
+        if file_contents[:1] == b'C':
+            content = zlib.decompress(file_contents[8:])
+        else:
+            raise NotImplementedError(u'Unsupported compression format %r' %
+                                      file_contents[:1])
+
+        def extract_tags(content):
+            pos = 0
+            while pos < len(content):
+                header16 = struct.unpack('<H', content[pos:pos+2])[0]
+                pos += 2
+                tag_code = header16 >> 6
+                tag_len = header16 & 0x3f
+                if tag_len == 0x3f:
+                    tag_len = struct.unpack('<I', content[pos:pos+4])[0]
+                    pos += 4
+                assert pos+tag_len <= len(content)
+                yield (tag_code, content[pos:pos+tag_len])
+                pos += tag_len
+
+        code_tag = next(tag
+                        for tag_code, tag in extract_tags(content)
+                        if tag_code == 82)
+        p = code_tag.index(b'\0', 4) + 1
+        code_reader = io.BytesIO(code_tag[p:])
+
+        # Parse ABC (AVM2 ByteCode)
+        def read_int(reader=None):
+            if reader is None:
+                reader = code_reader
+            res = 0
+            shift = 0
+            for _ in range(5):
+                buf = reader.read(1)
+                assert len(buf) == 1
+                b = struct.unpack('<B', buf)[0]
+                res = res | ((b & 0x7f) << shift)
+                if b & 0x80 == 0:
+                    break
+                shift += 7
+            return res
+
+        def u30(reader=None):
+            res = read_int(reader)
+            assert res & 0xf0000000 == 0
+            return res
+        u32 = read_int
+
+        def s32(reader=None):
+            v = read_int(reader)
+            if v & 0x80000000 != 0:
+                v = - ((v ^ 0xffffffff) + 1)
+            return v
+
+        def read_string(reader=None):
+            if reader is None:
+                reader = code_reader
+            slen = u30(reader)
+            resb = reader.read(slen)
+            assert len(resb) == slen
+            return resb.decode('utf-8')
+
+        def read_bytes(count, reader=None):
+            if reader is None:
+                reader = code_reader
+            resb = reader.read(count)
+            assert len(resb) == count
+            return resb
+
+        def read_byte(reader=None):
+            resb = read_bytes(1, reader=reader)
+            res = struct.unpack('<B', resb)[0]
+            return res
+
+        # minor_version + major_version
+        read_bytes(2 + 2)
+
+        # Constant pool
+        int_count = u30()
+        for _c in range(1, int_count):
+            s32()
+        uint_count = u30()
+        for _c in range(1, uint_count):
+            u32()
+        double_count = u30()
+        read_bytes((double_count-1) * 8)
+        string_count = u30()
+        constant_strings = [u'']
+        for _c in range(1, string_count):
+            s = read_string()
+            constant_strings.append(s)
+        namespace_count = u30()
+        for _c in range(1, namespace_count):
+            read_bytes(1)  # kind
+            u30()  # name
+        ns_set_count = u30()
+        for _c in range(1, ns_set_count):
+            count = u30()
+            for _c2 in range(count):
+                u30()
+        multiname_count = u30()
+        MULTINAME_SIZES = {
+            0x07: 2,  # QName
+            0x0d: 2,  # QNameA
+            0x0f: 1,  # RTQName
+            0x10: 1,  # RTQNameA
+            0x11: 0,  # RTQNameL
+            0x12: 0,  # RTQNameLA
+            0x09: 2,  # Multiname
+            0x0e: 2,  # MultinameA
+            0x1b: 1,  # MultinameL
+            0x1c: 1,  # MultinameLA
+        }
+        multinames = [u'']
+        for _c in range(1, multiname_count):
+            kind = u30()
+            assert kind in MULTINAME_SIZES, u'Invalid multiname kind %r' % kind
+            if kind == 0x07:
+                u30()  # namespace_idx
+                name_idx = u30()
+                multinames.append(constant_strings[name_idx])
+            else:
+                multinames.append('[MULTINAME kind: %d]' % kind)
+                for _c2 in range(MULTINAME_SIZES[kind]):
+                    u30()
+
+        # Methods
+        method_count = u30()
+        MethodInfo = collections.namedtuple(
+            'MethodInfo',
+            ['NEED_ARGUMENTS', 'NEED_REST'])
+        method_infos = []
+        for method_id in range(method_count):
+            param_count = u30()
+            u30()  # return type
+            for _ in range(param_count):
+                u30()  # param type
+            u30()  # name index (always 0 for youtube)
+            flags = read_byte()
+            if flags & 0x08 != 0:
+                # Options present
+                option_count = u30()
+                for c in range(option_count):
+                    u30()  # val
+                    read_bytes(1)  # kind
+            if flags & 0x80 != 0:
+                # Param names present
+                for _ in range(param_count):
+                    u30()  # param name
+            mi = MethodInfo(flags & 0x01 != 0, flags & 0x04 != 0)
+            method_infos.append(mi)
+
+        # Metadata
+        metadata_count = u30()
+        for _c in range(metadata_count):
+            u30()  # name
+            item_count = u30()
+            for _c2 in range(item_count):
+                u30()  # key
+                u30()  # value
+
+        def parse_traits_info():
+            trait_name_idx = u30()
+            kind_full = read_byte()
+            kind = kind_full & 0x0f
+            attrs = kind_full >> 4
+            methods = {}
+            if kind in [0x00, 0x06]:  # Slot or Const
+                u30()  # Slot id
+                u30()  # type_name_idx
+                vindex = u30()
+                if vindex != 0:
+                    read_byte()  # vkind
+            elif kind in [0x01, 0x02, 0x03]:  # Method / Getter / Setter
+                u30()  # disp_id
+                method_idx = u30()
+                methods[multinames[trait_name_idx]] = method_idx
+            elif kind == 0x04:  # Class
+                u30()  # slot_id
+                u30()  # classi
+            elif kind == 0x05:  # Function
+                u30()  # slot_id
+                function_idx = u30()
+                methods[function_idx] = multinames[trait_name_idx]
+            else:
+                raise ExtractorError(u'Unsupported trait kind %d' % kind)
+
+            if attrs & 0x4 != 0:  # Metadata present
+                metadata_count = u30()
+                for _c3 in range(metadata_count):
+                    u30()  # metadata index
+
+            return methods
+
+        # Classes
+        TARGET_CLASSNAME = u'SignatureDecipher'
+        searched_idx = multinames.index(TARGET_CLASSNAME)
+        searched_class_id = None
+        class_count = u30()
+        for class_id in range(class_count):
+            name_idx = u30()
+            if name_idx == searched_idx:
+                # We found the class we're looking for!
+                searched_class_id = class_id
+            u30()  # super_name idx
+            flags = read_byte()
+            if flags & 0x08 != 0:  # Protected namespace is present
+                u30()  # protected_ns_idx
+            intrf_count = u30()
+            for _c2 in range(intrf_count):
+                u30()
+            u30()  # iinit
+            trait_count = u30()
+            for _c2 in range(trait_count):
+                parse_traits_info()
+
+        if searched_class_id is None:
+            raise ExtractorError(u'Target class %r not found' %
+                                 TARGET_CLASSNAME)
+
+        method_names = {}
+        method_idxs = {}
+        for class_id in range(class_count):
+            u30()  # cinit
+            trait_count = u30()
+            for _c2 in range(trait_count):
+                trait_methods = parse_traits_info()
+                if class_id == searched_class_id:
+                    method_names.update(trait_methods.items())
+                    method_idxs.update(dict(
+                        (idx, name)
+                        for name, idx in trait_methods.items()))
+
+        # Scripts
+        script_count = u30()
+        for _c in range(script_count):
+            u30()  # init
+            trait_count = u30()
+            for _c2 in range(trait_count):
+                parse_traits_info()
+
+        # Method bodies
+        method_body_count = u30()
+        Method = collections.namedtuple('Method', ['code', 'local_count'])
+        methods = {}
+        for _c in range(method_body_count):
+            method_idx = u30()
+            u30()  # max_stack
+            local_count = u30()
+            u30()  # init_scope_depth
+            u30()  # max_scope_depth
+            code_length = u30()
+            code = read_bytes(code_length)
+            if method_idx in method_idxs:
+                m = Method(code, local_count)
+                methods[method_idxs[method_idx]] = m
+            exception_count = u30()
+            for _c2 in range(exception_count):
+                u30()  # from
+                u30()  # to
+                u30()  # target
+                u30()  # exc_type
+                u30()  # var_name
+            trait_count = u30()
+            for _c2 in range(trait_count):
+                parse_traits_info()
+
+        assert p + code_reader.tell() == len(code_tag)
+        assert len(methods) == len(method_idxs)
+
+        method_pyfunctions = {}
+
+        def extract_function(func_name):
+            if func_name in method_pyfunctions:
+                return method_pyfunctions[func_name]
+            if func_name not in methods:
+                raise ExtractorError(u'Cannot find function %r' % func_name)
+            m = methods[func_name]
+
+            def resfunc(args):
+                registers = ['(this)'] + list(args) + [None] * m.local_count
+                stack = []
+                coder = io.BytesIO(m.code)
+                while True:
+                    opcode = struct.unpack('!B', coder.read(1))[0]
+                    if opcode == 36:  # pushbyte
+                        v = struct.unpack('!B', coder.read(1))[0]
+                        stack.append(v)
+                    elif opcode == 44:  # pushstring
+                        idx = u30(coder)
+                        stack.append(constant_strings[idx])
+                    elif opcode == 48:  # pushscope
+                        # We don't implement the scope register, so we'll just
+                        # ignore the popped value
+                        stack.pop()
+                    elif opcode == 70:  # callproperty
+                        index = u30(coder)
+                        mname = multinames[index]
+                        arg_count = u30(coder)
+                        args = list(reversed(
+                            [stack.pop() for _ in range(arg_count)]))
+                        obj = stack.pop()
+                        if mname == u'split':
+                            assert len(args) == 1
+                            assert isinstance(args[0], compat_str)
+                            assert isinstance(obj, compat_str)
+                            if args[0] == u'':
+                                res = list(obj)
+                            else:
+                                res = obj.split(args[0])
+                            stack.append(res)
+                        elif mname == u'slice':
+                            assert len(args) == 1
+                            assert isinstance(args[0], int)
+                            assert isinstance(obj, list)
+                            res = obj[args[0]:]
+                            stack.append(res)
+                        elif mname == u'join':
+                            assert len(args) == 1
+                            assert isinstance(args[0], compat_str)
+                            assert isinstance(obj, list)
+                            res = args[0].join(obj)
+                            stack.append(res)
+                        elif mname in method_pyfunctions:
+                            stack.append(method_pyfunctions[mname](args))
+                        else:
+                            raise NotImplementedError(
+                                u'Unsupported property %r on %r'
+                                % (mname, obj))
+                    elif opcode == 72:  # returnvalue
+                        res = stack.pop()
+                        return res
+                    elif opcode == 79:  # callpropvoid
+                        index = u30(coder)
+                        mname = multinames[index]
+                        arg_count = u30(coder)
+                        args = list(reversed(
+                            [stack.pop() for _ in range(arg_count)]))
+                        obj = stack.pop()
+                        if mname == u'reverse':
+                            assert isinstance(obj, list)
+                            obj.reverse()
+                        else:
+                            raise NotImplementedError(
+                                u'Unsupported (void) property %r on %r'
+                                % (mname, obj))
+                    elif opcode == 93:  # findpropstrict
+                        index = u30(coder)
+                        mname = multinames[index]
+                        res = extract_function(mname)
+                        stack.append(res)
+                    elif opcode == 97:  # setproperty
+                        index = u30(coder)
+                        value = stack.pop()
+                        idx = stack.pop()
+                        obj = stack.pop()
+                        assert isinstance(obj, list)
+                        assert isinstance(idx, int)
+                        obj[idx] = value
+                    elif opcode == 98:  # getlocal
+                        index = u30(coder)
+                        stack.append(registers[index])
+                    elif opcode == 99:  # setlocal
+                        index = u30(coder)
+                        value = stack.pop()
+                        registers[index] = value
+                    elif opcode == 102:  # getproperty
+                        index = u30(coder)
+                        pname = multinames[index]
+                        if pname == u'length':
+                            obj = stack.pop()
+                            assert isinstance(obj, list)
+                            stack.append(len(obj))
+                        else:  # Assume attribute access
+                            idx = stack.pop()
+                            assert isinstance(idx, int)
+                            obj = stack.pop()
+                            assert isinstance(obj, list)
+                            stack.append(obj[idx])
+                    elif opcode == 128:  # coerce
+                        u30(coder)
+                    elif opcode == 133:  # coerce_s
+                        assert isinstance(stack[-1], (type(None), compat_str))
+                    elif opcode == 164:  # modulo
+                        value2 = stack.pop()
+                        value1 = stack.pop()
+                        res = value1 % value2
+                        stack.append(res)
+                    elif opcode == 208:  # getlocal_0
+                        stack.append(registers[0])
+                    elif opcode == 209:  # getlocal_1
+                        stack.append(registers[1])
+                    elif opcode == 210:  # getlocal_2
+                        stack.append(registers[2])
+                    elif opcode == 211:  # getlocal_3
+                        stack.append(registers[3])
+                    elif opcode == 214:  # setlocal_2
+                        registers[2] = stack.pop()
+                    elif opcode == 215:  # setlocal_3
+                        registers[3] = stack.pop()
+                    else:
+                        raise NotImplementedError(
+                            u'Unsupported opcode %d' % opcode)
+
+            method_pyfunctions[func_name] = resfunc
+            return resfunc
+
+        initial_function = extract_function(u'decipher')
+        return lambda s: initial_function([s])
+
+    def _decrypt_signature(self, s, video_id, player_url, age_gate=False):
         """Turn the encrypted s field into a working signature"""
 
         """Turn the encrypted s field into a working signature"""
 
-        if len(s) == 92:
+        if player_url is not None:
+            if player_url.startswith(u'//'):
+                player_url = u'https:' + player_url
+            try:
+                player_id = (player_url, len(s))
+                if player_id not in self._player_cache:
+                    func = self._extract_signature_function(
+                        video_id, player_url, len(s)
+                    )
+                    self._player_cache[player_id] = func
+                func = self._player_cache[player_id]
+                if self._downloader.params.get('youtube_print_sig_code'):
+                    self._print_sig_code(func, len(s))
+                return func(s)
+            except Exception:
+                tb = traceback.format_exc()
+                self._downloader.report_warning(
+                    u'Automatic signature extraction failed: ' + tb)
+
+            self._downloader.report_warning(
+                u'Warning: Falling back to static signature algorithm')
+
+        return self._static_decrypt_signature(
+            s, video_id, player_url, age_gate)
+
+    def _static_decrypt_signature(self, s, video_id, player_url, age_gate):
+        if age_gate:
+            # The videos with age protection use another player, so the
+            # algorithms can be different.
+            if len(s) == 86:
+                return s[2:63] + s[82] + s[64:82] + s[63]
+
+        if len(s) == 93:
+            return s[86:29:-1] + s[88] + s[28:5:-1]
+        elif len(s) == 92:
             return s[25] + s[3:25] + s[0] + s[26:42] + s[79] + s[43:79] + s[91] + s[80:83]
             return s[25] + s[3:25] + s[0] + s[26:42] + s[79] + s[43:79] + s[91] + s[80:83]
+        elif len(s) == 91:
+            return s[84:27:-1] + s[86] + s[26:5:-1]
         elif len(s) == 90:
             return s[25] + s[3:25] + s[2] + s[26:40] + s[77] + s[41:77] + s[89] + s[78:81]
         elif len(s) == 89:
             return s[84:78:-1] + s[87] + s[77:60:-1] + s[0] + s[59:3:-1]
         elif len(s) == 88:
         elif len(s) == 90:
             return s[25] + s[3:25] + s[2] + s[26:40] + s[77] + s[41:77] + s[89] + s[78:81]
         elif len(s) == 89:
             return s[84:78:-1] + s[87] + s[77:60:-1] + s[0] + s[59:3:-1]
         elif len(s) == 88:
-            return s[48] + s[81:67:-1] + s[82] + s[66:62:-1] + s[85] + s[61:48:-1] + s[67] + s[47:12:-1] + s[3] + s[11:3:-1] + s[2] + s[12]
+            return s[7:28] + s[87] + s[29:45] + s[55] + s[46:55] + s[2] + s[56:87] + s[28]
         elif len(s) == 87:
             return s[6:27] + s[4] + s[28:39] + s[27] + s[40:59] + s[2] + s[60:]
         elif len(s) == 86:
         elif len(s) == 87:
             return s[6:27] + s[4] + s[28:39] + s[27] + s[40:59] + s[2] + s[60:]
         elif len(s) == 86:
-            return s[5:20] + s[2] + s[21:]
+            return s[80:72:-1] + s[16] + s[71:39:-1] + s[72] + s[38:16:-1] + s[82] + s[15::-1]
         elif len(s) == 85:
         elif len(s) == 85:
-            return s[83:34:-1] + s[0] + s[33:27:-1] + s[3] + s[26:19:-1] + s[34] + s[18:3:-1] + s[27]
+            return s[3:11] + s[0] + s[12:55] + s[84] + s[56:84]
         elif len(s) == 84:
         elif len(s) == 84:
-            return s[83:27:-1] + s[0] + s[26:5:-1] + s[2:0:-1] + s[27]
+            return s[78:70:-1] + s[14] + s[69:37:-1] + s[70] + s[36:14:-1] + s[80] + s[:14][::-1]
         elif len(s) == 83:
         elif len(s) == 83:
-            return s[81:64:-1] + s[82] + s[63:52:-1] + s[45] + s[51:45:-1] + s[1] + s[44:1:-1] + s[0]
+            return s[80:63:-1] + s[0] + s[62:0:-1] + s[63]
         elif len(s) == 82:
         elif len(s) == 82:
-            return s[36] + s[79:67:-1] + s[81] + s[66:40:-1] + s[33] + s[39:36:-1] + s[40] + s[35] + s[0] + s[67] + s[32:0:-1] + s[34]
+            return s[80:37:-1] + s[7] + s[36:7:-1] + s[0] + s[6:0:-1] + s[37]
         elif len(s) == 81:
             return s[56] + s[79:56:-1] + s[41] + s[55:41:-1] + s[80] + s[40:34:-1] + s[0] + s[33:29:-1] + s[34] + s[28:9:-1] + s[29] + s[8:0:-1] + s[9]
         elif len(s) == 81:
             return s[56] + s[79:56:-1] + s[41] + s[55:41:-1] + s[80] + s[40:34:-1] + s[0] + s[33:29:-1] + s[34] + s[28:9:-1] + s[29] + s[8:0:-1] + s[9]
+        elif len(s) == 80:
+            return s[1:19] + s[0] + s[20:68] + s[19] + s[69:80]
         elif len(s) == 79:
             return s[54] + s[77:54:-1] + s[39] + s[53:39:-1] + s[78] + s[38:34:-1] + s[0] + s[33:29:-1] + s[34] + s[28:9:-1] + s[29] + s[8:0:-1] + s[9]
 
         else:
             raise ExtractorError(u'Unable to decrypt signature, key length %d not supported; retrying might work' % (len(s)))
 
         elif len(s) == 79:
             return s[54] + s[77:54:-1] + s[39] + s[53:39:-1] + s[78] + s[38:34:-1] + s[0] + s[33:29:-1] + s[34] + s[28:9:-1] + s[29] + s[8:0:-1] + s[9]
 
         else:
             raise ExtractorError(u'Unable to decrypt signature, key length %d not supported; retrying might work' % (len(s)))
 
-    def _decrypt_signature_age_gate(self, s):
-        # The videos with age protection use another player, so the algorithms
-        # can be different.
-        if len(s) == 86:
-            return s[2:63] + s[82] + s[64:82] + s[63]
-        else:
-            # Fallback to the other algortihms
-            return self._decrypt_signature(s)
-
-
-    def _get_available_subtitles(self, video_id):
-        self.report_video_subtitles_download(video_id)
-        request = compat_urllib_request.Request('http://video.google.com/timedtext?hl=en&type=list&v=%s' % video_id)
+    def _get_available_subtitles(self, video_id, webpage):
         try:
         try:
-            sub_list = compat_urllib_request.urlopen(request).read().decode('utf-8')
-        except (compat_urllib_error.URLError, compat_http_client.HTTPException, socket.error) as err:
-            return (u'unable to download video subtitles: %s' % compat_str(err), None)
-        sub_lang_list = re.findall(r'name="([^"]*)"[^>]+lang_code="([\w\-]+)"', sub_list)
-        sub_lang_list = dict((l[1], l[0]) for l in sub_lang_list)
+            sub_list = self._download_webpage(
+                'https://video.google.com/timedtext?hl=en&type=list&v=%s' % video_id,
+                video_id, note=False)
+        except ExtractorError as err:
+            self._downloader.report_warning(u'unable to download video subtitles: %s' % compat_str(err))
+            return {}
+        lang_list = re.findall(r'name="([^"]*)"[^>]+lang_code="([\w\-]+)"', sub_list)
+
+        sub_lang_list = {}
+        for l in lang_list:
+            lang = l[1]
+            params = compat_urllib_parse.urlencode({
+                'lang': lang,
+                'v': video_id,
+                'fmt': self._downloader.params.get('subtitlesformat', 'srt'),
+                'name': unescapeHTML(l[0]).encode('utf-8'),
+            })
+            url = u'https://www.youtube.com/api/timedtext?' + params
+            sub_lang_list[lang] = url
         if not sub_lang_list:
         if not sub_lang_list:
-            return (u'video doesn\'t have subtitles', None)
+            self._downloader.report_warning(u'video doesn\'t have subtitles')
+            return {}
         return sub_lang_list
 
         return sub_lang_list
 
-    def _list_available_subtitles(self, video_id):
-        sub_lang_list = self._get_available_subtitles(video_id)
-        self.report_video_subtitles_available(video_id, sub_lang_list)
-
-    def _request_subtitle(self, sub_lang, sub_name, video_id, format):
-        """
-        Return tuple:
-        (error_message, sub_lang, sub)
-        """
-        self.report_video_subtitles_request(video_id, sub_lang, format)
-        params = compat_urllib_parse.urlencode({
-            'lang': sub_lang,
-            'name': sub_name,
-            'v': video_id,
-            'fmt': format,
-        })
-        url = 'http://www.youtube.com/api/timedtext?' + params
-        try:
-            sub = compat_urllib_request.urlopen(url).read().decode('utf-8')
-        except (compat_urllib_error.URLError, compat_http_client.HTTPException, socket.error) as err:
-            return (u'unable to download video subtitles: %s' % compat_str(err), None, None)
-        if not sub:
-            return (u'Did not fetch video subtitles', None, None)
-        return (None, sub_lang, sub)
-
-    def _request_automatic_caption(self, video_id, webpage):
+    def _get_available_automatic_caption(self, video_id, webpage):
         """We need the webpage for getting the captions url, pass it as an
            argument to speed up the process."""
         """We need the webpage for getting the captions url, pass it as an
            argument to speed up the process."""
-        sub_lang = self._downloader.params.get('subtitleslang') or 'en'
-        sub_format = self._downloader.params.get('subtitlesformat')
+        sub_format = self._downloader.params.get('subtitlesformat', 'srt')
         self.to_screen(u'%s: Looking for automatic captions' % video_id)
         mobj = re.search(r';ytplayer.config = ({.*?});', webpage)
         self.to_screen(u'%s: Looking for automatic captions' % video_id)
         mobj = re.search(r';ytplayer.config = ({.*?});', webpage)
-        err_msg = u'Couldn\'t find automatic captions for "%s"' % sub_lang
+        err_msg = u'Couldn\'t find automatic captions for %s' % video_id
         if mobj is None:
         if mobj is None:
-            return [(err_msg, None, None)]
+            self._downloader.report_warning(err_msg)
+            return {}
         player_config = json.loads(mobj.group(1))
         try:
             args = player_config[u'args']
             caption_url = args[u'ttsurl']
             timestamp = args[u'timestamp']
         player_config = json.loads(mobj.group(1))
         try:
             args = player_config[u'args']
             caption_url = args[u'ttsurl']
             timestamp = args[u'timestamp']
-            params = compat_urllib_parse.urlencode({
-                'lang': 'en',
-                'tlang': sub_lang,
-                'fmt': sub_format,
-                'ts': timestamp,
-                'kind': 'asr',
+            # We get the available subtitles
+            list_params = compat_urllib_parse.urlencode({
+                'type': 'list',
+                'tlangs': 1,
+                'asrs': 1,
             })
             })
-            subtitles_url = caption_url + '&' + params
-            sub = self._download_webpage(subtitles_url, video_id, u'Downloading automatic captions')
-            return [(None, sub_lang, sub)]
-        except KeyError:
-            return [(err_msg, None, None)]
-
-    def _extract_subtitle(self, video_id):
-        """
-        Return a list with a tuple:
-        [(error_message, sub_lang, sub)]
-        """
-        sub_lang_list = self._get_available_subtitles(video_id)
-        sub_format = self._downloader.params.get('subtitlesformat')
-        if  isinstance(sub_lang_list,tuple): #There was some error, it didn't get the available subtitles
-            return [(sub_lang_list[0], None, None)]
-        if self._downloader.params.get('subtitleslang', False):
-            sub_lang = self._downloader.params.get('subtitleslang')
-        elif 'en' in sub_lang_list:
-            sub_lang = 'en'
-        else:
-            sub_lang = list(sub_lang_list.keys())[0]
-        if not sub_lang in sub_lang_list:
-            return [(u'no closed captions found in the specified language "%s"' % sub_lang, None, None)]
-
-        subtitle = self._request_subtitle(sub_lang, sub_lang_list[sub_lang].encode('utf-8'), video_id, sub_format)
-        return [subtitle]
-
-    def _extract_all_subtitles(self, video_id):
-        sub_lang_list = self._get_available_subtitles(video_id)
-        sub_format = self._downloader.params.get('subtitlesformat')
-        if  isinstance(sub_lang_list,tuple): #There was some error, it didn't get the available subtitles
-            return [(sub_lang_list[0], None, None)]
-        subtitles = []
-        for sub_lang in sub_lang_list:
-            subtitle = self._request_subtitle(sub_lang, sub_lang_list[sub_lang].encode('utf-8'), video_id, sub_format)
-            subtitles.append(subtitle)
-        return subtitles
-
-    def _print_formats(self, formats):
-        print('Available formats:')
-        for x in formats:
-            print('%s\t:\t%s\t[%s]%s' %(x, self._video_extensions.get(x, 'flv'),
-                                        self._video_dimensions.get(x, '???'),
-                                        ' (3D)' if x in self._3d_itags else ''))
-
-    def _extract_id(self, url):
-        mobj = re.match(self._VALID_URL, url, re.VERBOSE)
+            list_url = caption_url + '&' + list_params
+            caption_list = self._download_xml(list_url, video_id)
+            original_lang_node = caption_list.find('track')
+            if original_lang_node is None or original_lang_node.attrib.get('kind') != 'asr' :
+                self._downloader.report_warning(u'Video doesn\'t have automatic captions')
+                return {}
+            original_lang = original_lang_node.attrib['lang_code']
+
+            sub_lang_list = {}
+            for lang_node in caption_list.findall('target'):
+                sub_lang = lang_node.attrib['lang_code']
+                params = compat_urllib_parse.urlencode({
+                    'lang': original_lang,
+                    'tlang': sub_lang,
+                    'fmt': sub_format,
+                    'ts': timestamp,
+                    'kind': 'asr',
+                })
+                sub_lang_list[sub_lang] = caption_url + '&' + params
+            return sub_lang_list
+        # An extractor error can be raise by the download process if there are
+        # no automatic captions but there are subtitles
+        except (KeyError, ExtractorError):
+            self._downloader.report_warning(err_msg)
+            return {}
+
+    @classmethod
+    def extract_id(cls, url):
+        mobj = re.match(cls._VALID_URL, url, re.VERBOSE)
         if mobj is None:
             raise ExtractorError(u'Invalid URL: %s' % url)
         video_id = mobj.group(2)
         return video_id
 
         if mobj is None:
             raise ExtractorError(u'Invalid URL: %s' % url)
         video_id = mobj.group(2)
         return video_id
 
-    def _get_video_url_list(self, url_map):
-        """
-        Transform a dictionary in the format {itag:url} to a list of (itag, url)
-        with the requested formats.
-        """
-        req_format = self._downloader.params.get('format', None)
-        format_limit = self._downloader.params.get('format_limit', None)
-        available_formats = self._available_formats_prefer_free if self._downloader.params.get('prefer_free_formats', False) else self._available_formats
-        if format_limit is not None and format_limit in available_formats:
-            format_list = available_formats[available_formats.index(format_limit):]
-        else:
-            format_list = available_formats
-        existing_formats = [x for x in format_list if x in url_map]
-        if len(existing_formats) == 0:
-            raise ExtractorError(u'no known formats available for video')
-        if self._downloader.params.get('listformats', None):
-            self._print_formats(existing_formats)
-            return
-        if req_format is None or req_format == 'best':
-            video_url_list = [(existing_formats[0], url_map[existing_formats[0]])] # Best quality
-        elif req_format == 'worst':
-            video_url_list = [(existing_formats[-1], url_map[existing_formats[-1]])] # worst quality
-        elif req_format in ('-1', 'all'):
-            video_url_list = [(f, url_map[f]) for f in existing_formats] # All formats
-        else:
-            # Specific formats. We pick the first in a slash-delimeted sequence.
-            # For example, if '1/2/3/4' is requested and '2' and '4' are available, we pick '2'.
-            req_formats = req_format.split('/')
-            video_url_list = None
-            for rf in req_formats:
-                if rf in url_map:
-                    video_url_list = [(rf, url_map[rf])]
-                    break
-            if video_url_list is None:
-                raise ExtractorError(u'requested format not available')
-        return video_url_list
-
     def _extract_from_m3u8(self, manifest_url, video_id):
         url_map = {}
         def _get_urls(_manifest):
     def _extract_from_m3u8(self, manifest_url, video_id):
         url_map = {}
         def _get_urls(_manifest):
@@ -532,29 +1025,27 @@ class YoutubeIE(YoutubeBaseInfoExtractor):
             url_map[itag] = format_url
         return url_map
 
             url_map[itag] = format_url
         return url_map
 
+    def _extract_annotations(self, video_id):
+        url = 'https://www.youtube.com/annotations_invideo?features=1&legacy=1&video_id=%s' % video_id
+        return self._download_webpage(url, video_id, note=u'Searching for annotations.', errnote=u'Unable to download video annotations.')
+
     def _real_extract(self, url):
     def _real_extract(self, url):
-        if re.match(r'(?:https?://)?[^/]+/watch\?feature=[a-z_]+$', url):
-            self._downloader.report_warning(u'Did you forget to quote the URL? Remember that & is a meta-character in most shells, so you want to put the URL in quotes, like  youtube-dl \'http://www.youtube.com/watch?feature=foo&v=BaW_jenozKc\' (or simply  youtube-dl BaW_jenozKc  ).')
+        proto = (
+            u'http' if self._downloader.params.get('prefer_insecure', False)
+            else u'https')
 
         # Extract original video URL from URL with redirection, like age verification, using next_url parameter
         mobj = re.search(self._NEXT_URL_RE, url)
         if mobj:
 
         # Extract original video URL from URL with redirection, like age verification, using next_url parameter
         mobj = re.search(self._NEXT_URL_RE, url)
         if mobj:
-            url = 'https://www.youtube.com/' + compat_urllib_parse.unquote(mobj.group(1)).lstrip('/')
-        video_id = self._extract_id(url)
+            url = proto + '://www.youtube.com/' + compat_urllib_parse.unquote(mobj.group(1)).lstrip('/')
+        video_id = self.extract_id(url)
 
         # Get video webpage
 
         # Get video webpage
-        self.report_video_webpage_download(video_id)
-        url = 'https://www.youtube.com/watch?v=%s&gl=US&hl=en&has_verified=1' % video_id
-        request = compat_urllib_request.Request(url)
-        try:
-            video_webpage_bytes = compat_urllib_request.urlopen(request).read()
-        except (compat_urllib_error.URLError, compat_http_client.HTTPException, socket.error) as err:
-            raise ExtractorError(u'Unable to download video webpage: %s' % compat_str(err))
-
-        video_webpage = video_webpage_bytes.decode('utf-8', 'ignore')
+        url = proto + '://www.youtube.com/watch?v=%s&gl=US&hl=en&has_verified=1' % video_id
+        video_webpage = self._download_webpage(url, video_id)
 
         # Attempt to extract SWF player URL
 
         # Attempt to extract SWF player URL
-        mobj = re.search(r'swfConfig.*?"(http:\\/\\/.*?watch.*?-.*?\.swf)"', video_webpage)
+        mobj = re.search(r'swfConfig.*?"(https?:\\/\\/.*?watch.*?-.*?\.swf)"', video_webpage)
         if mobj is not None:
             player_url = re.sub(r'\\(.)', r'\1', mobj.group(1))
         else:
         if mobj is not None:
             player_url = re.sub(r'\\(.)', r'\1', mobj.group(1))
         else:
@@ -568,14 +1059,14 @@ class YoutubeIE(YoutubeBaseInfoExtractor):
             # We simulate the access to the video from www.youtube.com/v/{video_id}
             # this can be viewed without login into Youtube
             data = compat_urllib_parse.urlencode({'video_id': video_id,
             # We simulate the access to the video from www.youtube.com/v/{video_id}
             # this can be viewed without login into Youtube
             data = compat_urllib_parse.urlencode({'video_id': video_id,
-                                                  'el': 'embedded',
+                                                  'el': 'player_embedded',
                                                   'gl': 'US',
                                                   'hl': 'en',
                                                   'eurl': 'https://youtube.googleapis.com/v/' + video_id,
                                                   'asv': 3,
                                                   'sts':'1588',
                                                   })
                                                   'gl': 'US',
                                                   'hl': 'en',
                                                   'eurl': 'https://youtube.googleapis.com/v/' + video_id,
                                                   'asv': 3,
                                                   'sts':'1588',
                                                   })
-            video_info_url = 'https://www.youtube.com/get_video_info?' + data
+            video_info_url = proto + '://www.youtube.com/get_video_info?' + data
             video_info_webpage = self._download_webpage(video_info_url, video_id,
                                     note=False,
                                     errnote='unable to download video info webpage')
             video_info_webpage = self._download_webpage(video_info_url, video_id,
                                     note=False,
                                     errnote='unable to download video info webpage')
@@ -583,7 +1074,7 @@ class YoutubeIE(YoutubeBaseInfoExtractor):
         else:
             age_gate = False
             for el_type in ['&el=embedded', '&el=detailpage', '&el=vevo', '']:
         else:
             age_gate = False
             for el_type in ['&el=embedded', '&el=detailpage', '&el=vevo', '']:
-                video_info_url = ('https://www.youtube.com/get_video_info?&video_id=%s%s&ps=default&eurl=&gl=US&hl=en'
+                video_info_url = (proto + '://www.youtube.com/get_video_info?&video_id=%s%s&ps=default&eurl=&gl=US&hl=en'
                         % (video_id, el_type))
                 video_info_webpage = self._download_webpage(video_info_url, video_id,
                                         note=False,
                         % (video_id, el_type))
                 video_info_webpage = self._download_webpage(video_info_url, video_id,
                                         note=False,
@@ -593,9 +1084,18 @@ class YoutubeIE(YoutubeBaseInfoExtractor):
                     break
         if 'token' not in video_info:
             if 'reason' in video_info:
                     break
         if 'token' not in video_info:
             if 'reason' in video_info:
-                raise ExtractorError(u'YouTube said: %s' % video_info['reason'][0], expected=True)
+                raise ExtractorError(
+                    u'YouTube said: %s' % video_info['reason'][0],
+                    expected=True, video_id=video_id)
             else:
             else:
-                raise ExtractorError(u'"token" parameter not in video info for unknown reason')
+                raise ExtractorError(
+                    u'"token" parameter not in video info for unknown reason',
+                    video_id=video_id)
+
+        if 'view_count' in video_info:
+            view_count = int(video_info['view_count'][0])
+        else:
+            view_count = None
 
         # Check for "rental" videos
         if 'ypc_video_rental_bar_text' in video_info and 'author' not in video_info:
 
         # Check for "rental" videos
         if 'ypc_video_rental_bar_text' in video_info and 'author' not in video_info:
@@ -618,9 +1118,11 @@ class YoutubeIE(YoutubeBaseInfoExtractor):
             self._downloader.report_warning(u'unable to extract uploader nickname')
 
         # title
             self._downloader.report_warning(u'unable to extract uploader nickname')
 
         # title
-        if 'title' not in video_info:
-            raise ExtractorError(u'Unable to extract video title')
-        video_title = compat_urllib_parse.unquote_plus(video_info['title'][0])
+        if 'title' in video_info:
+            video_title = video_info['title'][0]
+        else:
+            self._downloader.report_warning(u'Unable to extract video title')
+            video_title = u'_'
 
         # thumbnail image
         # We try first to get a high quality image:
 
         # thumbnail image
         # We try first to get a high quality image:
@@ -630,20 +1132,42 @@ class YoutubeIE(YoutubeBaseInfoExtractor):
             video_thumbnail = m_thumb.group(1)
         elif 'thumbnail_url' not in video_info:
             self._downloader.report_warning(u'unable to extract video thumbnail')
             video_thumbnail = m_thumb.group(1)
         elif 'thumbnail_url' not in video_info:
             self._downloader.report_warning(u'unable to extract video thumbnail')
-            video_thumbnail = ''
+            video_thumbnail = None
         else:   # don't panic if we can't find it
             video_thumbnail = compat_urllib_parse.unquote_plus(video_info['thumbnail_url'][0])
 
         # upload date
         upload_date = None
         else:   # don't panic if we can't find it
             video_thumbnail = compat_urllib_parse.unquote_plus(video_info['thumbnail_url'][0])
 
         # upload date
         upload_date = None
-        mobj = re.search(r'id="eow-date.*?>(.*?)</span>', video_webpage, re.DOTALL)
+        mobj = re.search(r'(?s)id="eow-date.*?>(.*?)</span>', video_webpage)
+        if mobj is None:
+            mobj = re.search(
+                r'(?s)id="watch-uploader-info".*?>.*?(?:Published|Uploaded|Streamed live) on (.*?)</strong>',
+                video_webpage)
         if mobj is not None:
             upload_date = ' '.join(re.sub(r'[/,-]', r' ', mobj.group(1)).split())
             upload_date = unified_strdate(upload_date)
 
         if mobj is not None:
             upload_date = ' '.join(re.sub(r'[/,-]', r' ', mobj.group(1)).split())
             upload_date = unified_strdate(upload_date)
 
+        m_cat_container = get_element_by_id("eow-category", video_webpage)
+        if m_cat_container:
+            category = self._html_search_regex(
+                r'(?s)<a[^<]+>(.*?)</a>', m_cat_container, 'category',
+                default=None)
+            video_categories = None if category is None else [category]
+        else:
+            video_categories = None
+
         # description
         video_description = get_element_by_id("eow-description", video_webpage)
         if video_description:
         # description
         video_description = get_element_by_id("eow-description", video_webpage)
         if video_description:
+            video_description = re.sub(r'''(?x)
+                <a\s+
+                    (?:[a-zA-Z-]+="[^"]+"\s+)*?
+                    title="([^"]+)"\s+
+                    (?:[a-zA-Z-]+="[^"]+"\s+)*?
+                    class="yt-uix-redirect-link"\s*>
+                [^<]+
+                </a>
+            ''', r'\1', video_description)
             video_description = clean_html(video_description)
         else:
             fd_mobj = re.search(r'<meta name="description" content="([^"]+)"', video_webpage)
             video_description = clean_html(video_description)
         else:
             fd_mobj = re.search(r'<meta name="description" content="([^"]+)"', video_webpage)
@@ -652,134 +1176,207 @@ class YoutubeIE(YoutubeBaseInfoExtractor):
             else:
                 video_description = u''
 
             else:
                 video_description = u''
 
+        def _extract_count(klass):
+            count = self._search_regex(
+                r'class="%s">([\d,]+)</span>' % re.escape(klass),
+                video_webpage, klass, default=None)
+            if count is not None:
+                return int(count.replace(',', ''))
+            return None
+        like_count = _extract_count(u'likes-count')
+        dislike_count = _extract_count(u'dislikes-count')
+
         # subtitles
         # subtitles
-        video_subtitles = None
-
-        if self._downloader.params.get('writesubtitles', False):
-            video_subtitles = self._extract_subtitle(video_id)
-            if video_subtitles:
-                (sub_error, sub_lang, sub) = video_subtitles[0]
-                if sub_error:
-                    self._downloader.report_warning(sub_error)
-        
-        if self._downloader.params.get('writeautomaticsub', False):
-            video_subtitles = self._request_automatic_caption(video_id, video_webpage)
-            (sub_error, sub_lang, sub) = video_subtitles[0]
-            if sub_error:
-                self._downloader.report_warning(sub_error)
-
-        if self._downloader.params.get('allsubtitles', False):
-            video_subtitles = self._extract_all_subtitles(video_id)
-            for video_subtitle in video_subtitles:
-                (sub_error, sub_lang, sub) = video_subtitle
-                if sub_error:
-                    self._downloader.report_warning(sub_error)
+        video_subtitles = self.extract_subtitles(video_id, video_webpage)
 
         if self._downloader.params.get('listsubtitles', False):
 
         if self._downloader.params.get('listsubtitles', False):
-            self._list_available_subtitles(video_id)
+            self._list_available_subtitles(video_id, video_webpage)
             return
 
         if 'length_seconds' not in video_info:
             self._downloader.report_warning(u'unable to extract video duration')
             return
 
         if 'length_seconds' not in video_info:
             self._downloader.report_warning(u'unable to extract video duration')
-            video_duration = ''
+            video_duration = None
         else:
         else:
-            video_duration = compat_urllib_parse.unquote_plus(video_info['length_seconds'][0])
+            video_duration = int(compat_urllib_parse.unquote_plus(video_info['length_seconds'][0]))
 
 
-        # Decide which formats to download
+        # annotations
+        video_annotations = None
+        if self._downloader.params.get('writeannotations', False):
+                video_annotations = self._extract_annotations(video_id)
 
 
+        # Decide which formats to download
         try:
         try:
-            mobj = re.search(r';ytplayer.config = ({.*?});', video_webpage)
+            mobj = re.search(r';ytplayer\.config\s*=\s*({.*?});', video_webpage)
             if not mobj:
                 raise ValueError('Could not find vevo ID')
             if not mobj:
                 raise ValueError('Could not find vevo ID')
-            info = json.loads(mobj.group(1))
-            args = info['args']
+            json_code = uppercase_escape(mobj.group(1))
+            ytplayer_config = json.loads(json_code)
+            args = ytplayer_config['args']
             # Easy way to know if the 's' value is in url_encoded_fmt_stream_map
             # this signatures are encrypted
             # Easy way to know if the 's' value is in url_encoded_fmt_stream_map
             # this signatures are encrypted
-            m_s = re.search(r'[&,]s=', args['url_encoded_fmt_stream_map'])
+            if 'url_encoded_fmt_stream_map' not in args:
+                raise ValueError(u'No stream_map present')  # caught below
+            re_signature = re.compile(r'[&,]s=')
+            m_s = re_signature.search(args['url_encoded_fmt_stream_map'])
             if m_s is not None:
                 self.to_screen(u'%s: Encrypted signatures detected.' % video_id)
                 video_info['url_encoded_fmt_stream_map'] = [args['url_encoded_fmt_stream_map']]
             if m_s is not None:
                 self.to_screen(u'%s: Encrypted signatures detected.' % video_id)
                 video_info['url_encoded_fmt_stream_map'] = [args['url_encoded_fmt_stream_map']]
+            m_s = re_signature.search(args.get('adaptive_fmts', u''))
+            if m_s is not None:
+                if 'adaptive_fmts' in video_info:
+                    video_info['adaptive_fmts'][0] += ',' + args['adaptive_fmts']
+                else:
+                    video_info['adaptive_fmts'] = [args['adaptive_fmts']]
         except ValueError:
             pass
 
         except ValueError:
             pass
 
+        def _map_to_format_list(urlmap):
+            formats = []
+            for itag, video_real_url in urlmap.items():
+                dct = {
+                    'format_id': itag,
+                    'url': video_real_url,
+                    'player_url': player_url,
+                }
+                if itag in self._formats:
+                    dct.update(self._formats[itag])
+                formats.append(dct)
+            return formats
+
         if 'conn' in video_info and video_info['conn'][0].startswith('rtmp'):
             self.report_rtmp_download()
         if 'conn' in video_info and video_info['conn'][0].startswith('rtmp'):
             self.report_rtmp_download()
-            video_url_list = [(None, video_info['conn'][0])]
-        elif 'url_encoded_fmt_stream_map' in video_info and len(video_info['url_encoded_fmt_stream_map']) >= 1:
-            if 'rtmpe%3Dyes' in video_info['url_encoded_fmt_stream_map'][0]:
+            formats = [{
+                'format_id': '_rtmp',
+                'protocol': 'rtmp',
+                'url': video_info['conn'][0],
+                'player_url': player_url,
+            }]
+        elif len(video_info.get('url_encoded_fmt_stream_map', [])) >= 1 or len(video_info.get('adaptive_fmts', [])) >= 1:
+            encoded_url_map = video_info.get('url_encoded_fmt_stream_map', [''])[0] + ',' + video_info.get('adaptive_fmts',[''])[0]
+            if 'rtmpe%3Dyes' in encoded_url_map:
                 raise ExtractorError('rtmpe downloads are not supported, see https://github.com/rg3/youtube-dl/issues/343 for more information.', expected=True)
             url_map = {}
                 raise ExtractorError('rtmpe downloads are not supported, see https://github.com/rg3/youtube-dl/issues/343 for more information.', expected=True)
             url_map = {}
-            for url_data_str in video_info['url_encoded_fmt_stream_map'][0].split(','):
+            for url_data_str in encoded_url_map.split(','):
                 url_data = compat_parse_qs(url_data_str)
                 if 'itag' in url_data and 'url' in url_data:
                     url = url_data['url'][0]
                     if 'sig' in url_data:
                         url += '&signature=' + url_data['sig'][0]
                     elif 's' in url_data:
                 url_data = compat_parse_qs(url_data_str)
                 if 'itag' in url_data and 'url' in url_data:
                     url = url_data['url'][0]
                     if 'sig' in url_data:
                         url += '&signature=' + url_data['sig'][0]
                     elif 's' in url_data:
+                        encrypted_sig = url_data['s'][0]
                         if self._downloader.params.get('verbose'):
                         if self._downloader.params.get('verbose'):
-                            s = url_data['s'][0]
                             if age_gate:
                             if age_gate:
-                                player_version = self._search_regex(r'ad3-(.+?)\.swf',
-                                    video_info['ad3_module'][0] if 'ad3_module' in video_info else 'NOT FOUND',
-                                    'flash player', fatal=False)
-                                player = 'flash player %s' % player_version
+                                if player_url is None:
+                                    player_version = 'unknown'
+                                else:
+                                    player_version = self._search_regex(
+                                        r'-(.+)\.swf$', player_url,
+                                        u'flash player', fatal=False)
+                                player_desc = 'flash player %s' % player_version
                             else:
                             else:
-                                player = u'html5 player %s' % self._search_regex(r'html5player-(.+?)\.js', video_webpage,
+                                player_version = self._search_regex(
+                                    r'html5player-(.+?)\.js', video_webpage,
                                     'html5 player', fatal=False)
                                     'html5 player', fatal=False)
-                            parts_sizes = u'.'.join(compat_str(len(part)) for part in s.split('.'))
+                                player_desc = u'html5 player %s' % player_version
+
+                            parts_sizes = u'.'.join(compat_str(len(part)) for part in encrypted_sig.split('.'))
                             self.to_screen(u'encrypted signature length %d (%s), itag %s, %s' %
                             self.to_screen(u'encrypted signature length %d (%s), itag %s, %s' %
-                                (len(s), parts_sizes, url_data['itag'][0], player))
-                        encrypted_sig = url_data['s'][0]
-                        if age_gate:
-                            signature = self._decrypt_signature_age_gate(encrypted_sig)
-                        else:
-                            signature = self._decrypt_signature(encrypted_sig)
+                                (len(encrypted_sig), parts_sizes, url_data['itag'][0], player_desc))
+
+                        if not age_gate:
+                            jsplayer_url_json = self._search_regex(
+                                r'"assets":.+?"js":\s*("[^"]+")',
+                                video_webpage, u'JS player URL')
+                            player_url = json.loads(jsplayer_url_json)
+
+                        signature = self._decrypt_signature(
+                            encrypted_sig, video_id, player_url, age_gate)
                         url += '&signature=' + signature
                     if 'ratebypass' not in url:
                         url += '&ratebypass=yes'
                     url_map[url_data['itag'][0]] = url
                         url += '&signature=' + signature
                     if 'ratebypass' not in url:
                         url += '&ratebypass=yes'
                     url_map[url_data['itag'][0]] = url
-            video_url_list = self._get_video_url_list(url_map)
-            if not video_url_list:
-                return
+            formats = _map_to_format_list(url_map)
         elif video_info.get('hlsvp'):
             manifest_url = video_info['hlsvp'][0]
             url_map = self._extract_from_m3u8(manifest_url, video_id)
         elif video_info.get('hlsvp'):
             manifest_url = video_info['hlsvp'][0]
             url_map = self._extract_from_m3u8(manifest_url, video_id)
-            video_url_list = self._get_video_url_list(url_map)
-            if not video_url_list:
-                return
-
+            formats = _map_to_format_list(url_map)
         else:
         else:
-            raise ExtractorError(u'no conn or url_encoded_fmt_stream_map information found in video info')
-
-        results = []
-        for format_param, video_real_url in video_url_list:
-            # Extension
-            video_extension = self._video_extensions.get(format_param, 'flv')
-
-            video_format = '{0} - {1}{2}'.format(format_param if format_param else video_extension,
-                                              self._video_dimensions.get(format_param, '???'),
-                                              ' (3D)' if format_param in self._3d_itags else '')
-
-            results.append({
-                'id':       video_id,
-                'url':      video_real_url,
-                'uploader': video_uploader,
-                'uploader_id': video_uploader_id,
-                'upload_date':  upload_date,
-                'title':    video_title,
-                'ext':      video_extension,
-                'format':   video_format,
-                'thumbnail':    video_thumbnail,
-                'description':  video_description,
-                'player_url':   player_url,
-                'subtitles':    video_subtitles,
-                'duration':     video_duration
-            })
-        return results
+            raise ExtractorError(u'no conn, hlsvp or url_encoded_fmt_stream_map information found in video info')
+
+        # Look for the DASH manifest
+        if (self._downloader.params.get('youtube_include_dash_manifest', False)):
+            try:
+                # The DASH manifest used needs to be the one from the original video_webpage.
+                # The one found in get_video_info seems to be using different signatures.
+                # However, in the case of an age restriction there won't be any embedded dashmpd in the video_webpage.
+                # Luckily, it seems, this case uses some kind of default signature (len == 86), so the
+                # combination of get_video_info and the _static_decrypt_signature() decryption fallback will work here.
+                if age_gate:
+                    dash_manifest_url = video_info.get('dashmpd')[0]
+                else:
+                    dash_manifest_url = ytplayer_config['args']['dashmpd']
+                def decrypt_sig(mobj):
+                    s = mobj.group(1)
+                    dec_s = self._decrypt_signature(s, video_id, player_url, age_gate)
+                    return '/signature/%s' % dec_s
+                dash_manifest_url = re.sub(r'/s/([\w\.]+)', decrypt_sig, dash_manifest_url)
+                dash_doc = self._download_xml(
+                    dash_manifest_url, video_id,
+                    note=u'Downloading DASH manifest',
+                    errnote=u'Could not download DASH manifest')
+                for r in dash_doc.findall(u'.//{urn:mpeg:DASH:schema:MPD:2011}Representation'):
+                    url_el = r.find('{urn:mpeg:DASH:schema:MPD:2011}BaseURL')
+                    if url_el is None:
+                        continue
+                    format_id = r.attrib['id']
+                    video_url = url_el.text
+                    filesize = int_or_none(url_el.attrib.get('{http://youtube.com/yt/2012/10/10}contentLength'))
+                    f = {
+                        'format_id': format_id,
+                        'url': video_url,
+                        'width': int_or_none(r.attrib.get('width')),
+                        'tbr': int_or_none(r.attrib.get('bandwidth'), 1000),
+                        'asr': int_or_none(r.attrib.get('audioSamplingRate')),
+                        'filesize': filesize,
+                    }
+                    try:
+                        existing_format = next(
+                            fo for fo in formats
+                            if fo['format_id'] == format_id)
+                    except StopIteration:
+                        f.update(self._formats.get(format_id, {}))
+                        formats.append(f)
+                    else:
+                        existing_format.update(f)
+
+            except (ExtractorError, KeyError) as e:
+                self.report_warning(u'Skipping DASH manifest: %s' % e, video_id)
+
+        self._sort_formats(formats)
+
+        return {
+            'id':           video_id,
+            'uploader':     video_uploader,
+            'uploader_id':  video_uploader_id,
+            'upload_date':  upload_date,
+            'title':        video_title,
+            'thumbnail':    video_thumbnail,
+            'description':  video_description,
+            'categories':   video_categories,
+            'subtitles':    video_subtitles,
+            'duration':     video_duration,
+            'age_limit':    18 if age_gate else 0,
+            'annotations':  video_annotations,
+            'webpage_url': proto + '://www.youtube.com/watch?v=%s' % video_id,
+            'view_count':   view_count,
+            'like_count': like_count,
+            'dislike_count': dislike_count,
+            'formats':      formats,
+        }
 
 
-class YoutubePlaylistIE(InfoExtractor):
+class YoutubePlaylistIE(YoutubeBaseInfoExtractor):
     IE_DESC = u'YouTube.com playlists'
     IE_DESC = u'YouTube.com playlists'
-    _VALID_URL = r"""(?:
+    _VALID_URL = r"""(?x)(?:
                         (?:https?://)?
                         (?:\w+\.)?
                         youtube\.com/
                         (?:https?://)?
                         (?:\w+\.)?
                         youtube\.com/
@@ -788,67 +1385,144 @@ class YoutubePlaylistIE(InfoExtractor):
                            \? (?:.*?&)*? (?:p|a|list)=
                         |  p/
                         )
                            \? (?:.*?&)*? (?:p|a|list)=
                         |  p/
                         )
-                        ((?:PL|EC|UU|FL)?[0-9A-Za-z-_]{10,})
+                        (
+                            (?:PL|LL|EC|UU|FL|RD)?[0-9A-Za-z-_]{10,}
+                            # Top tracks, they can also include dots 
+                            |(?:MC)[\w\.]*
+                        )
                         .*
                      |
                         .*
                      |
-                        ((?:PL|EC|UU|FL)[0-9A-Za-z-_]{10,})
+                        ((?:PL|LL|EC|UU|FL|RD)[0-9A-Za-z-_]{10,})
                      )"""
                      )"""
-    _TEMPLATE_URL = 'https://gdata.youtube.com/feeds/api/playlists/%s?max-results=%i&start-index=%i&v=2&alt=json&safeSearch=none'
-    _MAX_RESULTS = 50
+    _TEMPLATE_URL = 'https://www.youtube.com/playlist?list=%s'
+    _MORE_PAGES_INDICATOR = r'data-link-type="next"'
+    _VIDEO_RE = r'href="\s*/watch\?v=(?P<id>[0-9A-Za-z_-]{11})&amp;[^"]*?index=(?P<index>\d+)'
     IE_NAME = u'youtube:playlist'
 
     IE_NAME = u'youtube:playlist'
 
-    @classmethod
-    def suitable(cls, url):
-        """Receives a URL and returns True if suitable for this IE."""
-        return re.match(cls._VALID_URL, url, re.VERBOSE) is not None
+    def _real_initialize(self):
+        self._login()
+
+    def _ids_to_results(self, ids):
+        return [self.url_result(vid_id, 'Youtube', video_id=vid_id)
+                       for vid_id in ids]
+
+    def _extract_mix(self, playlist_id):
+        # The mixes are generated from a a single video
+        # the id of the playlist is just 'RD' + video_id
+        url = 'https://youtube.com/watch?v=%s&list=%s' % (playlist_id[-11:], playlist_id)
+        webpage = self._download_webpage(url, playlist_id, u'Downloading Youtube mix')
+        search_title = lambda class_name: get_element_by_attribute('class', class_name, webpage)
+        title_span = (search_title('playlist-title') or
+            search_title('title long-title') or search_title('title'))
+        title = clean_html(title_span)
+        video_re = r'''(?x)data-video-username=".*?".*?
+                       href="/watch\?v=([0-9A-Za-z_-]{11})&amp;[^"]*?list=%s''' % re.escape(playlist_id)
+        ids = orderedSet(re.findall(video_re, webpage, flags=re.DOTALL))
+        url_results = self._ids_to_results(ids)
+
+        return self.playlist_result(url_results, playlist_id, title)
 
     def _real_extract(self, url):
         # Extract playlist id
 
     def _real_extract(self, url):
         # Extract playlist id
-        mobj = re.match(self._VALID_URL, url, re.VERBOSE)
+        mobj = re.match(self._VALID_URL, url)
         if mobj is None:
             raise ExtractorError(u'Invalid URL: %s' % url)
         if mobj is None:
             raise ExtractorError(u'Invalid URL: %s' % url)
-
-        # Download playlist videos from API
         playlist_id = mobj.group(1) or mobj.group(2)
         playlist_id = mobj.group(1) or mobj.group(2)
-        videos = []
+
+        # Check if it's a video-specific URL
+        query_dict = compat_urlparse.parse_qs(compat_urlparse.urlparse(url).query)
+        if 'v' in query_dict:
+            video_id = query_dict['v'][0]
+            if self._downloader.params.get('noplaylist'):
+                self.to_screen(u'Downloading just video %s because of --no-playlist' % video_id)
+                return self.url_result(video_id, 'Youtube', video_id=video_id)
+            else:
+                self.to_screen(u'Downloading playlist %s - add --no-playlist to just download video %s' % (playlist_id, video_id))
+
+        if playlist_id.startswith('RD'):
+            # Mixes require a custom extraction process
+            return self._extract_mix(playlist_id)
+        if playlist_id.startswith('TL'):
+            raise ExtractorError(u'For downloading YouTube.com top lists, use '
+                u'the "yttoplist" keyword, for example "youtube-dl \'yttoplist:music:Top Tracks\'"', expected=True)
+
+        url = self._TEMPLATE_URL % playlist_id
+        page = self._download_webpage(url, playlist_id)
+        more_widget_html = content_html = page
+
+        # Check if the playlist exists or is private
+        if re.search(r'<div class="yt-alert-message">[^<]*?(The|This) playlist (does not exist|is private)[^<]*?</div>', page) is not None:
+            raise ExtractorError(
+                u'The playlist doesn\'t exist or is private, use --username or '
+                '--netrc to access it.',
+                expected=True)
+
+        # Extract the video ids from the playlist pages
+        ids = []
 
         for page_num in itertools.count(1):
 
         for page_num in itertools.count(1):
-            start_index = self._MAX_RESULTS * (page_num - 1) + 1
-            if start_index >= 1000:
-                self._downloader.report_warning(u'Max number of results reached')
+            matches = re.finditer(self._VIDEO_RE, content_html)
+            # We remove the duplicates and the link with index 0
+            # (it's not the first video of the playlist)
+            new_ids = orderedSet(m.group('id') for m in matches if m.group('index') != '0')
+            ids.extend(new_ids)
+
+            mobj = re.search(r'data-uix-load-more-href="/?(?P<more>[^"]+)"', more_widget_html)
+            if not mobj:
                 break
                 break
-            url = self._TEMPLATE_URL % (playlist_id, self._MAX_RESULTS, start_index)
-            page = self._download_webpage(url, playlist_id, u'Downloading page #%s' % page_num)
 
 
-            try:
-                response = json.loads(page)
-            except ValueError as err:
-                raise ExtractorError(u'Invalid JSON in API response: ' + compat_str(err))
+            more = self._download_json(
+                'https://youtube.com/%s' % mobj.group('more'), playlist_id,
+                'Downloading page #%s' % page_num,
+                transform_source=uppercase_escape)
+            content_html = more['content_html']
+            more_widget_html = more['load_more_widget_html']
 
 
-            if 'feed' not in response:
-                raise ExtractorError(u'Got a malformed response from YouTube API')
-            playlist_title = response['feed']['title']['$t']
-            if 'entry' not in response['feed']:
-                # Number of videos is a multiple of self._MAX_RESULTS
-                break
+        playlist_title = self._html_search_regex(
+            r'(?s)<h1 class="pl-header-title[^"]*">\s*(.*?)\s*</h1>',
+            page, u'title')
+
+        url_results = self._ids_to_results(ids)
+        return self.playlist_result(url_results, playlist_id, playlist_title)
 
 
-            for entry in response['feed']['entry']:
-                index = entry['yt$position']['$t']
-                if 'media$group' in entry and 'media$player' in entry['media$group']:
-                    videos.append((index, entry['media$group']['media$player']['url']))
 
 
-        videos = [v[1] for v in sorted(videos)]
+class YoutubeTopListIE(YoutubePlaylistIE):
+    IE_NAME = u'youtube:toplist'
+    IE_DESC = (u'YouTube.com top lists, "yttoplist:{channel}:{list title}"'
+        u' (Example: "yttoplist:music:Top Tracks")')
+    _VALID_URL = r'yttoplist:(?P<chann>.*?):(?P<title>.*?)$'
 
 
-        url_results = [self.url_result(vurl, 'Youtube') for vurl in videos]
-        return [self.playlist_result(url_results, playlist_id, playlist_title)]
+    def _real_extract(self, url):
+        mobj = re.match(self._VALID_URL, url)
+        channel = mobj.group('chann')
+        title = mobj.group('title')
+        query = compat_urllib_parse.urlencode({'title': title})
+        playlist_re = 'href="([^"]+?%s.*?)"' % re.escape(query)
+        channel_page = self._download_webpage('https://www.youtube.com/%s' % channel, title)
+        link = self._html_search_regex(playlist_re, channel_page, u'list')
+        url = compat_urlparse.urljoin('https://www.youtube.com/', link)
+        
+        video_re = r'data-index="\d+".*?data-video-id="([0-9A-Za-z_-]{11})"'
+        ids = []
+        # sometimes the webpage doesn't contain the videos
+        # retry until we get them
+        for i in itertools.count(0):
+            msg = u'Downloading Youtube mix'
+            if i > 0:
+                msg += ', retry #%d' % i
+            webpage = self._download_webpage(url, title, msg)
+            ids = orderedSet(re.findall(video_re, webpage))
+            if ids:
+                break
+        url_results = self._ids_to_results(ids)
+        return self.playlist_result(url_results, playlist_title=title)
 
 
 class YoutubeChannelIE(InfoExtractor):
     IE_DESC = u'YouTube.com channels'
     _VALID_URL = r"^(?:https?://)?(?:youtu\.be|(?:\w+\.)?youtube(?:-nocookie)?\.com)/channel/([0-9A-Za-z_-]+)"
 
 
 class YoutubeChannelIE(InfoExtractor):
     IE_DESC = u'YouTube.com channels'
     _VALID_URL = r"^(?:https?://)?(?:youtu\.be|(?:\w+\.)?youtube(?:-nocookie)?\.com)/channel/([0-9A-Za-z_-]+)"
-    _TEMPLATE_URL = 'http://www.youtube.com/channel/%s/videos?sort=da&flow=list&view=0&page=%s&gl=US&hl=en'
     _MORE_PAGES_INDICATOR = 'yt-uix-load-more'
     _MORE_PAGES_INDICATOR = 'yt-uix-load-more'
-    _MORE_PAGES_URL = 'http://www.youtube.com/c4_browse_ajax?action_load_more_videos=1&flow=list&paging=%s&view=0&sort=da&channel_id=%s'
+    _MORE_PAGES_URL = 'https://www.youtube.com/c4_browse_ajax?action_load_more_videos=1&flow=list&paging=%s&view=0&sort=da&channel_id=%s'
     IE_NAME = u'youtube:channel'
 
     def extract_videos_from_page(self, page):
     IE_NAME = u'youtube:channel'
 
     def extract_videos_from_page(self, page):
@@ -867,47 +1541,55 @@ class YoutubeChannelIE(InfoExtractor):
         # Download channel page
         channel_id = mobj.group(1)
         video_ids = []
         # Download channel page
         channel_id = mobj.group(1)
         video_ids = []
-        pagenum = 1
-
-        url = self._TEMPLATE_URL % (channel_id, pagenum)
-        page = self._download_webpage(url, channel_id,
-                                      u'Downloading page #%s' % pagenum)
-
-        # Extract video identifiers
-        ids_in_page = self.extract_videos_from_page(page)
-        video_ids.extend(ids_in_page)
-
-        # Download any subsequent channel pages using the json-based channel_ajax query
-        if self._MORE_PAGES_INDICATOR in page:
+        url = 'https://www.youtube.com/channel/%s/videos' % channel_id
+        channel_page = self._download_webpage(url, channel_id)
+        autogenerated = re.search(r'''(?x)
+                class="[^"]*?(?:
+                    channel-header-autogenerated-label|
+                    yt-channel-title-autogenerated
+                )[^"]*"''', channel_page) is not None
+
+        if autogenerated:
+            # The videos are contained in a single page
+            # the ajax pages can't be used, they are empty
+            video_ids = self.extract_videos_from_page(channel_page)
+        else:
+            # Download all channel pages using the json-based channel_ajax query
             for pagenum in itertools.count(1):
                 url = self._MORE_PAGES_URL % (pagenum, channel_id)
             for pagenum in itertools.count(1):
                 url = self._MORE_PAGES_URL % (pagenum, channel_id)
-                page = self._download_webpage(url, channel_id,
-                                              u'Downloading page #%s' % pagenum)
-
-                page = json.loads(page)
+                page = self._download_json(
+                    url, channel_id, note=u'Downloading page #%s' % pagenum,
+                    transform_source=uppercase_escape)
 
                 ids_in_page = self.extract_videos_from_page(page['content_html'])
                 video_ids.extend(ids_in_page)
 
                 ids_in_page = self.extract_videos_from_page(page['content_html'])
                 video_ids.extend(ids_in_page)
-
-                if self._MORE_PAGES_INDICATOR  not in page['load_more_widget_html']:
+    
+                if self._MORE_PAGES_INDICATOR not in page['load_more_widget_html']:
                     break
 
         self._downloader.to_screen(u'[youtube] Channel %s: Found %i videos' % (channel_id, len(video_ids)))
 
                     break
 
         self._downloader.to_screen(u'[youtube] Channel %s: Found %i videos' % (channel_id, len(video_ids)))
 
-        urls = ['http://www.youtube.com/watch?v=%s' % id for id in video_ids]
-        url_entries = [self.url_result(eurl, 'Youtube') for eurl in urls]
-        return [self.playlist_result(url_entries, channel_id)]
+        url_entries = [self.url_result(video_id, 'Youtube', video_id=video_id)
+                       for video_id in video_ids]
+        return self.playlist_result(url_entries, channel_id)
 
 
 class YoutubeUserIE(InfoExtractor):
     IE_DESC = u'YouTube.com user videos (URL or "ytuser" keyword)'
 
 
 class YoutubeUserIE(InfoExtractor):
     IE_DESC = u'YouTube.com user videos (URL or "ytuser" keyword)'
-    _VALID_URL = r'(?:(?:(?:https?://)?(?:\w+\.)?youtube\.com/user/)|ytuser:)([A-Za-z0-9_-]+)'
-    _TEMPLATE_URL = 'http://gdata.youtube.com/feeds/api/users/%s'
+    _VALID_URL = r'(?:(?:(?:https?://)?(?:\w+\.)?youtube\.com/(?:user/)?(?!(?:attribution_link|watch|results)(?:$|[^a-z_A-Z0-9-])))|ytuser:)(?!feed/)([A-Za-z0-9_-]+)'
+    _TEMPLATE_URL = 'https://gdata.youtube.com/feeds/api/users/%s'
     _GDATA_PAGE_SIZE = 50
     _GDATA_PAGE_SIZE = 50
-    _GDATA_URL = 'http://gdata.youtube.com/feeds/api/users/%s/uploads?max-results=%d&start-index=%d'
-    _VIDEO_INDICATOR = r'/watch\?v=(.+?)[\<&]'
+    _GDATA_URL = 'https://gdata.youtube.com/feeds/api/users/%s/uploads?max-results=%d&start-index=%d&alt=json'
     IE_NAME = u'youtube:user'
 
     IE_NAME = u'youtube:user'
 
+    @classmethod
+    def suitable(cls, url):
+        # Don't return True if the url can be extracted with other youtube
+        # extractor, the regex would is too permissive and it would match.
+        other_ies = iter(klass for (name, klass) in globals().items() if name.endswith('IE') and klass is not cls)
+        if any(ie.suitable(url) for ie in other_ies): return False
+        else: return super(YoutubeUserIE, cls).suitable(url)
+
     def _real_extract(self, url):
         # Extract username
         mobj = re.match(self._VALID_URL, url)
     def _real_extract(self, url):
         # Extract username
         mobj = re.match(self._VALID_URL, url)
@@ -921,67 +1603,68 @@ class YoutubeUserIE(InfoExtractor):
         # page by page until there are no video ids - it means we got
         # all of them.
 
         # page by page until there are no video ids - it means we got
         # all of them.
 
-        video_ids = []
-
-        for pagenum in itertools.count(0):
+        def download_page(pagenum):
             start_index = pagenum * self._GDATA_PAGE_SIZE + 1
 
             gdata_url = self._GDATA_URL % (username, self._GDATA_PAGE_SIZE, start_index)
             start_index = pagenum * self._GDATA_PAGE_SIZE + 1
 
             gdata_url = self._GDATA_URL % (username, self._GDATA_PAGE_SIZE, start_index)
-            page = self._download_webpage(gdata_url, username,
-                                          u'Downloading video ids from %d to %d' % (start_index, start_index + self._GDATA_PAGE_SIZE))
-
-            # Extract video identifiers
-            ids_in_page = []
+            page = self._download_webpage(
+                gdata_url, username,
+                u'Downloading video ids from %d to %d' % (
+                    start_index, start_index + self._GDATA_PAGE_SIZE))
 
 
-            for mobj in re.finditer(self._VIDEO_INDICATOR, page):
-                if mobj.group(1) not in ids_in_page:
-                    ids_in_page.append(mobj.group(1))
-
-            video_ids.extend(ids_in_page)
+            try:
+                response = json.loads(page)
+            except ValueError as err:
+                raise ExtractorError(u'Invalid JSON in API response: ' + compat_str(err))
+            if 'entry' not in response['feed']:
+                return
 
 
-            # A little optimization - if current page is not
-            # "full", ie. does not contain PAGE_SIZE video ids then
-            # we can assume that this page is the last one - there
-            # are no more ids on further pages - no need to query
-            # again.
+            # Extract video identifiers
+            entries = response['feed']['entry']
+            for entry in entries:
+                title = entry['title']['$t']
+                video_id = entry['id']['$t'].split('/')[-1]
+                yield {
+                    '_type': 'url',
+                    'url': video_id,
+                    'ie_key': 'Youtube',
+                    'id': video_id,
+                    'title': title,
+                }
+        url_results = PagedList(download_page, self._GDATA_PAGE_SIZE)
 
 
-            if len(ids_in_page) < self._GDATA_PAGE_SIZE:
-                break
+        return self.playlist_result(url_results, playlist_title=username)
 
 
-        urls = ['http://www.youtube.com/watch?v=%s' % video_id for video_id in video_ids]
-        url_results = [self.url_result(rurl, 'Youtube') for rurl in urls]
-        return [self.playlist_result(url_results, playlist_title = username)]
 
 class YoutubeSearchIE(SearchInfoExtractor):
     IE_DESC = u'YouTube.com searches'
 
 class YoutubeSearchIE(SearchInfoExtractor):
     IE_DESC = u'YouTube.com searches'
-    _API_URL = 'https://gdata.youtube.com/feeds/api/videos?q=%s&start-index=%i&max-results=50&v=2&alt=jsonc'
+    _API_URL = u'https://gdata.youtube.com/feeds/api/videos?q=%s&start-index=%i&max-results=50&v=2&alt=jsonc'
     _MAX_RESULTS = 1000
     IE_NAME = u'youtube:search'
     _SEARCH_KEY = 'ytsearch'
 
     _MAX_RESULTS = 1000
     IE_NAME = u'youtube:search'
     _SEARCH_KEY = 'ytsearch'
 
-    def report_download_page(self, query, pagenum):
-        """Report attempt to download search page with given number."""
-        self._downloader.to_screen(u'[youtube] query "%s": Downloading page %s' % (query, pagenum))
-
     def _get_n_results(self, query, n):
         """Get a specified number of results for a query"""
 
         video_ids = []
         pagenum = 0
         limit = n
     def _get_n_results(self, query, n):
         """Get a specified number of results for a query"""
 
         video_ids = []
         pagenum = 0
         limit = n
-
-        while (50 * pagenum) < limit:
-            self.report_download_page(query, pagenum+1)
-            result_url = self._API_URL % (compat_urllib_parse.quote_plus(query), (50*pagenum)+1)
-            request = compat_urllib_request.Request(result_url)
-            try:
-                data = compat_urllib_request.urlopen(request).read().decode('utf-8')
-            except (compat_urllib_error.URLError, compat_http_client.HTTPException, socket.error) as err:
-                raise ExtractorError(u'Unable to download API page: %s' % compat_str(err))
-            api_response = json.loads(data)['data']
-
-            if not 'items' in api_response:
-                raise ExtractorError(u'[youtube] No video results')
+        PAGE_SIZE = 50
+
+        while (PAGE_SIZE * pagenum) < limit:
+            result_url = self._API_URL % (
+                compat_urllib_parse.quote_plus(query.encode('utf-8')),
+                (PAGE_SIZE * pagenum) + 1)
+            data_json = self._download_webpage(
+                result_url, video_id=u'query "%s"' % query,
+                note=u'Downloading page %s' % (pagenum + 1),
+                errnote=u'Unable to download API page')
+            data = json.loads(data_json)
+            api_response = data['data']
+
+            if 'items' not in api_response:
+                raise ExtractorError(
+                    u'[youtube] No video results', expected=True)
 
             new_ids = list(video['id'] for video in api_response['items'])
             video_ids += new_ids
 
             new_ids = list(video['id'] for video in api_response['items'])
             video_ids += new_ids
@@ -991,10 +1674,54 @@ class YoutubeSearchIE(SearchInfoExtractor):
 
         if len(video_ids) > n:
             video_ids = video_ids[:n]
 
         if len(video_ids) > n:
             video_ids = video_ids[:n]
-        videos = [self.url_result('http://www.youtube.com/watch?v=%s' % id, 'Youtube') for id in video_ids]
+        videos = [self.url_result(video_id, 'Youtube', video_id=video_id)
+                  for video_id in video_ids]
         return self.playlist_result(videos, query)
 
 
         return self.playlist_result(videos, query)
 
 
+class YoutubeSearchDateIE(YoutubeSearchIE):
+    IE_NAME = YoutubeSearchIE.IE_NAME + ':date'
+    _API_URL = 'https://gdata.youtube.com/feeds/api/videos?q=%s&start-index=%i&max-results=50&v=2&alt=jsonc&orderby=published'
+    _SEARCH_KEY = 'ytsearchdate'
+    IE_DESC = u'YouTube.com searches, newest videos first'
+
+
+class YoutubeSearchURLIE(InfoExtractor):
+    IE_DESC = u'YouTube.com search URLs'
+    IE_NAME = u'youtube:search_url'
+    _VALID_URL = r'https?://(?:www\.)?youtube\.com/results\?(.*?&)?search_query=(?P<query>[^&]+)(?:[&]|$)'
+
+    def _real_extract(self, url):
+        mobj = re.match(self._VALID_URL, url)
+        query = compat_urllib_parse.unquote_plus(mobj.group('query'))
+
+        webpage = self._download_webpage(url, query)
+        result_code = self._search_regex(
+            r'(?s)<ol id="search-results"(.*?)</ol>', webpage, u'result HTML')
+
+        part_codes = re.findall(
+            r'(?s)<h3 class="yt-lockup-title">(.*?)</h3>', result_code)
+        entries = []
+        for part_code in part_codes:
+            part_title = self._html_search_regex(
+                r'(?s)title="([^"]+)"', part_code, 'item title', fatal=False)
+            part_url_snippet = self._html_search_regex(
+                r'(?s)href="([^"]+)"', part_code, 'item URL')
+            part_url = compat_urlparse.urljoin(
+                'https://www.youtube.com/', part_url_snippet)
+            entries.append({
+                '_type': 'url',
+                'url': part_url,
+                'title': part_title,
+            })
+
+        return {
+            '_type': 'playlist',
+            'entries': entries,
+            'title': query,
+        }
+
+
 class YoutubeShowIE(InfoExtractor):
     IE_DESC = u'YouTube.com (multi-season) shows'
     _VALID_URL = r'https?://www\.youtube\.com/show/(.*)'
 class YoutubeShowIE(InfoExtractor):
     IE_DESC = u'YouTube.com (multi-season) shows'
     _VALID_URL = r'https?://www\.youtube\.com/show/(.*)'
@@ -1017,7 +1744,6 @@ class YoutubeFeedsInfoExtractor(YoutubeBaseInfoExtractor):
     Subclasses must define the _FEED_NAME and _PLAYLIST_TITLE properties.
     """
     _LOGIN_REQUIRED = True
     Subclasses must define the _FEED_NAME and _PLAYLIST_TITLE properties.
     """
     _LOGIN_REQUIRED = True
-    _PAGING_STEP = 30
     # use action_load_personal_feed instead of action_load_system_feed
     _PERSONAL_FEED = False
 
     # use action_load_personal_feed instead of action_load_system_feed
     _PERSONAL_FEED = False
 
@@ -1026,7 +1752,7 @@ class YoutubeFeedsInfoExtractor(YoutubeBaseInfoExtractor):
         action = 'action_load_system_feed'
         if self._PERSONAL_FEED:
             action = 'action_load_personal_feed'
         action = 'action_load_system_feed'
         if self._PERSONAL_FEED:
             action = 'action_load_personal_feed'
-        return 'http://www.youtube.com/feed_ajax?%s=1&feed_name=%s&paging=%%s' % (action, self._FEED_NAME)
+        return 'https://www.youtube.com/feed_ajax?%s=1&feed_name=%s&paging=%%s' % (action, self._FEED_NAME)
 
     @property
     def IE_NAME(self):
 
     @property
     def IE_NAME(self):
@@ -1037,23 +1763,27 @@ class YoutubeFeedsInfoExtractor(YoutubeBaseInfoExtractor):
 
     def _real_extract(self, url):
         feed_entries = []
 
     def _real_extract(self, url):
         feed_entries = []
-        # The step argument is available only in 2.7 or higher
-        for i in itertools.count(0):
-            paging = i*self._PAGING_STEP
-            info = self._download_webpage(self._FEED_TEMPLATE % paging,
+        paging = 0
+        for i in itertools.count(1):
+            info = self._download_json(self._FEED_TEMPLATE % paging,
                                           u'%s feed' % self._FEED_NAME,
                                           u'Downloading page %s' % i)
                                           u'%s feed' % self._FEED_NAME,
                                           u'Downloading page %s' % i)
-            info = json.loads(info)
-            feed_html = info['feed_html']
+            feed_html = info.get('feed_html') or info.get('content_html')
             m_ids = re.finditer(r'"/watch\?v=(.*?)["&]', feed_html)
             ids = orderedSet(m.group(1) for m in m_ids)
             m_ids = re.finditer(r'"/watch\?v=(.*?)["&]', feed_html)
             ids = orderedSet(m.group(1) for m in m_ids)
-            feed_entries.extend(self.url_result(id, 'Youtube') for id in ids)
-            if info['paging'] is None:
+            feed_entries.extend(
+                self.url_result(video_id, 'Youtube', video_id=video_id)
+                for video_id in ids)
+            mobj = re.search(
+                r'data-uix-load-more-href="/?[^"]+paging=(?P<paging>\d+)',
+                feed_html)
+            if mobj is None:
                 break
                 break
+            paging = mobj.group('paging')
         return self.playlist_result(feed_entries, playlist_title=self._PLAYLIST_TITLE)
 
 class YoutubeSubscriptionsIE(YoutubeFeedsInfoExtractor):
         return self.playlist_result(feed_entries, playlist_title=self._PLAYLIST_TITLE)
 
 class YoutubeSubscriptionsIE(YoutubeFeedsInfoExtractor):
-    IE_DESC = u'YouTube.com subscriptions feed, "ytsubs" keyword(requires authentication)'
+    IE_DESC = u'YouTube.com subscriptions feed, "ytsubs" keyword (requires authentication)'
     _VALID_URL = r'https?://www\.youtube\.com/feed/subscriptions|:ytsubs(?:criptions)?'
     _FEED_NAME = 'subscriptions'
     _PLAYLIST_TITLE = u'Youtube Subscriptions'
     _VALID_URL = r'https?://www\.youtube\.com/feed/subscriptions|:ytsubs(?:criptions)?'
     _FEED_NAME = 'subscriptions'
     _PLAYLIST_TITLE = u'Youtube Subscriptions'
@@ -1069,16 +1799,40 @@ class YoutubeWatchLaterIE(YoutubeFeedsInfoExtractor):
     _VALID_URL = r'https?://www\.youtube\.com/feed/watch_later|:ytwatchlater'
     _FEED_NAME = 'watch_later'
     _PLAYLIST_TITLE = u'Youtube Watch Later'
     _VALID_URL = r'https?://www\.youtube\.com/feed/watch_later|:ytwatchlater'
     _FEED_NAME = 'watch_later'
     _PLAYLIST_TITLE = u'Youtube Watch Later'
-    _PAGING_STEP = 100
     _PERSONAL_FEED = True
 
     _PERSONAL_FEED = True
 
+class YoutubeHistoryIE(YoutubeFeedsInfoExtractor):
+    IE_DESC = u'Youtube watch history, "ythistory" keyword (requires authentication)'
+    _VALID_URL = u'https?://www\.youtube\.com/feed/history|:ythistory'
+    _FEED_NAME = 'history'
+    _PERSONAL_FEED = True
+    _PLAYLIST_TITLE = u'Youtube Watch History'
+
 class YoutubeFavouritesIE(YoutubeBaseInfoExtractor):
     IE_NAME = u'youtube:favorites'
     IE_DESC = u'YouTube.com favourite videos, "ytfav" keyword (requires authentication)'
 class YoutubeFavouritesIE(YoutubeBaseInfoExtractor):
     IE_NAME = u'youtube:favorites'
     IE_DESC = u'YouTube.com favourite videos, "ytfav" keyword (requires authentication)'
-    _VALID_URL = r'https?://www\.youtube\.com/my_favorites|:ytfav(?:o?rites)?'
+    _VALID_URL = r'https?://www\.youtube\.com/my_favorites|:ytfav(?:ou?rites)?'
     _LOGIN_REQUIRED = True
 
     def _real_extract(self, url):
         webpage = self._download_webpage('https://www.youtube.com/my_favorites', 'Youtube Favourites videos')
         playlist_id = self._search_regex(r'list=(.+?)["&]', webpage, u'favourites playlist id')
         return self.url_result(playlist_id, 'YoutubePlaylist')
     _LOGIN_REQUIRED = True
 
     def _real_extract(self, url):
         webpage = self._download_webpage('https://www.youtube.com/my_favorites', 'Youtube Favourites videos')
         playlist_id = self._search_regex(r'list=(.+?)["&]', webpage, u'favourites playlist id')
         return self.url_result(playlist_id, 'YoutubePlaylist')
+
+
+class YoutubeTruncatedURLIE(InfoExtractor):
+    IE_NAME = 'youtube:truncated_url'
+    IE_DESC = False  # Do not list
+    _VALID_URL = r'''(?x)
+        (?:https?://)?[^/]+/watch\?(?:feature=[a-z_]+)?$|
+        (?:https?://)?(?:www\.)?youtube\.com/attribution_link\?a=[^&]+$
+    '''
+
+    def _real_extract(self, url):
+        raise ExtractorError(
+            u'Did you forget to quote the URL? Remember that & is a meta '
+            u'character in most shells, so you want to put the URL in quotes, '
+            u'like  youtube-dl '
+            u'"http://www.youtube.com/watch?feature=foo&v=BaW_jenozKc" '
+            u' or simply  youtube-dl BaW_jenozKc  .',
+            expected=True)