]> Raphaël G. Git Repositories - youtubedl/blobdiff - youtube_dl/extractor/vimeo.py
debian/control: Remove mplayer2 from recommends. (Closes: #841187)
[youtubedl] / youtube_dl / extractor / vimeo.py
index 1936755499ceca890c7b4513298a64444293b71d..51c69a80c216889315a4c5fe070572100c13dd36 100644 (file)
@@ -1,4 +1,4 @@
-# encoding: utf-8
+# coding: utf-8
 from __future__ import unicode_literals
 
 import json
 from __future__ import unicode_literals
 
 import json
@@ -6,169 +6,477 @@ import re
 import itertools
 
 from .common import InfoExtractor
 import itertools
 
 from .common import InfoExtractor
+from ..compat import (
+    compat_HTTPError,
+    compat_str,
+    compat_urlparse,
+)
 from ..utils import (
 from ..utils import (
-    compat_urllib_parse,
-    compat_urllib_request,
-
-    clean_html,
-    get_element_by_attribute,
+    determine_ext,
     ExtractorError,
     ExtractorError,
+    InAdvancePagedList,
+    int_or_none,
+    NO_DEFAULT,
     RegexNotFoundError,
     RegexNotFoundError,
+    sanitized_Request,
+    smuggle_url,
     std_headers,
     std_headers,
+    unified_strdate,
     unsmuggle_url,
     unsmuggle_url,
+    urlencode_postdata,
+    unescapeHTML,
+    parse_filesize,
+    try_get,
 )
 
 
 )
 
 
-class VimeoIE(InfoExtractor):
+class VimeoBaseInfoExtractor(InfoExtractor):
+    _NETRC_MACHINE = 'vimeo'
+    _LOGIN_REQUIRED = False
+    _LOGIN_URL = 'https://vimeo.com/log_in'
+
+    def _login(self):
+        (username, password) = self._get_login_info()
+        if username is None:
+            if self._LOGIN_REQUIRED:
+                raise ExtractorError('No login info available, needed for using %s.' % self.IE_NAME, expected=True)
+            return
+        self.report_login()
+        webpage = self._download_webpage(self._LOGIN_URL, None, False)
+        token, vuid = self._extract_xsrft_and_vuid(webpage)
+        data = urlencode_postdata({
+            'action': 'login',
+            'email': username,
+            'password': password,
+            'service': 'vimeo',
+            'token': token,
+        })
+        login_request = sanitized_Request(self._LOGIN_URL, data)
+        login_request.add_header('Content-Type', 'application/x-www-form-urlencoded')
+        login_request.add_header('Referer', self._LOGIN_URL)
+        self._set_vimeo_cookie('vuid', vuid)
+        self._download_webpage(login_request, None, False, 'Wrong login info')
+
+    def _verify_video_password(self, url, video_id, webpage):
+        password = self._downloader.params.get('videopassword')
+        if password is None:
+            raise ExtractorError('This video is protected by a password, use the --video-password option', expected=True)
+        token, vuid = self._extract_xsrft_and_vuid(webpage)
+        data = urlencode_postdata({
+            'password': password,
+            'token': token,
+        })
+        if url.startswith('http://'):
+            # vimeo only supports https now, but the user can give an http url
+            url = url.replace('http://', 'https://')
+        password_request = sanitized_Request(url + '/password', data)
+        password_request.add_header('Content-Type', 'application/x-www-form-urlencoded')
+        password_request.add_header('Referer', url)
+        self._set_vimeo_cookie('vuid', vuid)
+        return self._download_webpage(
+            password_request, video_id,
+            'Verifying the password', 'Wrong password')
+
+    def _extract_xsrft_and_vuid(self, webpage):
+        xsrft = self._search_regex(
+            r'(?:(?P<q1>["\'])xsrft(?P=q1)\s*:|xsrft\s*[=:])\s*(?P<q>["\'])(?P<xsrft>.+?)(?P=q)',
+            webpage, 'login token', group='xsrft')
+        vuid = self._search_regex(
+            r'["\']vuid["\']\s*:\s*(["\'])(?P<vuid>.+?)\1',
+            webpage, 'vuid', group='vuid')
+        return xsrft, vuid
+
+    def _set_vimeo_cookie(self, name, value):
+        self._set_cookie('vimeo.com', name, value)
+
+    def _vimeo_sort_formats(self, formats):
+        # Bitrates are completely broken. Single m3u8 may contain entries in kbps and bps
+        # at the same time without actual units specified. This lead to wrong sorting.
+        self._sort_formats(formats, field_preference=('preference', 'height', 'width', 'fps', 'format_id'))
+
+    def _parse_config(self, config, video_id):
+        # Extract title
+        video_title = config['video']['title']
+
+        # Extract uploader, uploader_url and uploader_id
+        video_uploader = config['video'].get('owner', {}).get('name')
+        video_uploader_url = config['video'].get('owner', {}).get('url')
+        video_uploader_id = video_uploader_url.split('/')[-1] if video_uploader_url else None
+
+        # Extract video thumbnail
+        video_thumbnail = config['video'].get('thumbnail')
+        if video_thumbnail is None:
+            video_thumbs = config['video'].get('thumbs')
+            if video_thumbs and isinstance(video_thumbs, dict):
+                _, video_thumbnail = sorted((int(width if width.isdigit() else 0), t_url) for (width, t_url) in video_thumbs.items())[-1]
+
+        # Extract video duration
+        video_duration = int_or_none(config['video'].get('duration'))
+
+        formats = []
+        config_files = config['video'].get('files') or config['request'].get('files', {})
+        for f in config_files.get('progressive', []):
+            video_url = f.get('url')
+            if not video_url:
+                continue
+            formats.append({
+                'url': video_url,
+                'format_id': 'http-%s' % f.get('quality'),
+                'width': int_or_none(f.get('width')),
+                'height': int_or_none(f.get('height')),
+                'fps': int_or_none(f.get('fps')),
+                'tbr': int_or_none(f.get('bitrate')),
+            })
+        m3u8_url = config_files.get('hls', {}).get('url')
+        if m3u8_url:
+            formats.extend(self._extract_m3u8_formats(
+                m3u8_url, video_id, 'mp4', 'm3u8_native', m3u8_id='hls', fatal=False))
+
+        subtitles = {}
+        text_tracks = config['request'].get('text_tracks')
+        if text_tracks:
+            for tt in text_tracks:
+                subtitles[tt['lang']] = [{
+                    'ext': 'vtt',
+                    'url': 'https://vimeo.com' + tt['url'],
+                }]
+
+        return {
+            'title': video_title,
+            'uploader': video_uploader,
+            'uploader_id': video_uploader_id,
+            'uploader_url': video_uploader_url,
+            'thumbnail': video_thumbnail,
+            'duration': video_duration,
+            'formats': formats,
+            'subtitles': subtitles,
+        }
+
+
+class VimeoIE(VimeoBaseInfoExtractor):
     """Information extractor for vimeo.com."""
 
     # _VALID_URL matches Vimeo URLs
     _VALID_URL = r'''(?x)
     """Information extractor for vimeo.com."""
 
     # _VALID_URL matches Vimeo URLs
     _VALID_URL = r'''(?x)
-        (?P<proto>(?:https?:)?//)?
-        (?:(?:www|(?P<player>player))\.)?
-        vimeo(?P<pro>pro)?\.com/
-        (?:.*?/)?
-        (?:(?:play_redirect_hls|moogaloop\.swf)\?clip_id=)?
-        (?:videos?/)?
-        (?P<id>[0-9]+)
-        /?(?:[?&].*)?(?:[#].*)?$'''
-    _NETRC_MACHINE = 'vimeo'
+                    https?://
+                        (?:
+                            (?:
+                                www|
+                                (?P<player>player)
+                            )
+                            \.
+                        )?
+                        vimeo(?P<pro>pro)?\.com/
+                        (?!(?:channels|album)/[^/?#]+/?(?:$|[?#])|[^/]+/review/|ondemand/)
+                        (?:.*?/)?
+                        (?:
+                            (?:
+                                play_redirect_hls|
+                                moogaloop\.swf)\?clip_id=
+                            )?
+                        (?:videos?/)?
+                        (?P<id>[0-9]+)
+                        (?:/[\da-f]+)?
+                        /?(?:[?&].*)?(?:[#].*)?$
+                    '''
     IE_NAME = 'vimeo'
     _TESTS = [
         {
             'url': 'http://vimeo.com/56015672#at=0',
     IE_NAME = 'vimeo'
     _TESTS = [
         {
             'url': 'http://vimeo.com/56015672#at=0',
-            'file': '56015672.mp4',
             'md5': '8879b6cc097e987f02484baf890129e5',
             'info_dict': {
             'md5': '8879b6cc097e987f02484baf890129e5',
             'info_dict': {
-                "upload_date": "20121220", 
-                "description": "This is a test case for youtube-dl.\nFor more information, see github.com/rg3/youtube-dl\nTest chars: \u2605 \" ' \u5e78 / \\ \u00e4 \u21ad \U0001d550", 
-                "uploader_id": "user7108434", 
-                "uploader": "Filippo Valsorda", 
-                "title": "youtube-dl test video - \u2605 \" ' \u5e78 / \\ \u00e4 \u21ad \U0001d550",
+                'id': '56015672',
+                'ext': 'mp4',
+                'title': "youtube-dl test video - \u2605 \" ' \u5e78 / \\ \u00e4 \u21ad \U0001d550",
+                'description': 'md5:2d3305bad981a06ff79f027f19865021',
+                'upload_date': '20121220',
+                'uploader_url': 're:https?://(?:www\.)?vimeo\.com/user7108434',
+                'uploader_id': 'user7108434',
+                'uploader': 'Filippo Valsorda',
+                'duration': 10,
             },
         },
         {
             'url': 'http://vimeopro.com/openstreetmapus/state-of-the-map-us-2013/video/68093876',
             },
         },
         {
             'url': 'http://vimeopro.com/openstreetmapus/state-of-the-map-us-2013/video/68093876',
-            'file': '68093876.mp4',
             'md5': '3b5ca6aa22b60dfeeadf50b72e44ed82',
             'note': 'Vimeo Pro video (#1197)',
             'info_dict': {
             'md5': '3b5ca6aa22b60dfeeadf50b72e44ed82',
             'note': 'Vimeo Pro video (#1197)',
             'info_dict': {
+                'id': '68093876',
+                'ext': 'mp4',
+                'uploader_url': 're:https?://(?:www\.)?vimeo\.com/openstreetmapus',
                 'uploader_id': 'openstreetmapus',
                 'uploader': 'OpenStreetMap US',
                 'title': 'Andy Allan - Putting the Carto into OpenStreetMap Cartography',
                 'uploader_id': 'openstreetmapus',
                 'uploader': 'OpenStreetMap US',
                 'title': 'Andy Allan - Putting the Carto into OpenStreetMap Cartography',
+                'description': 'md5:fd69a7b8d8c34a4e1d2ec2e4afd6ec30',
+                'duration': 1595,
             },
         },
         {
             'url': 'http://player.vimeo.com/video/54469442',
             },
         },
         {
             'url': 'http://player.vimeo.com/video/54469442',
-            'file': '54469442.mp4',
             'md5': '619b811a4417aa4abe78dc653becf511',
             'note': 'Videos that embed the url in the player page',
             'info_dict': {
             'md5': '619b811a4417aa4abe78dc653becf511',
             'note': 'Videos that embed the url in the player page',
             'info_dict': {
-                'title': 'Kathy Sierra: Building the minimum Badass User, Business of Software',
+                'id': '54469442',
+                'ext': 'mp4',
+                'title': 'Kathy Sierra: Building the minimum Badass User, Business of Software 2012',
                 'uploader': 'The BLN & Business of Software',
                 'uploader': 'The BLN & Business of Software',
+                'uploader_url': 're:https?://(?:www\.)?vimeo\.com/theblnbusinessofsoftware',
                 'uploader_id': 'theblnbusinessofsoftware',
                 'uploader_id': 'theblnbusinessofsoftware',
+                'duration': 3610,
+                'description': None,
             },
         },
         {
             'url': 'http://vimeo.com/68375962',
             },
         },
         {
             'url': 'http://vimeo.com/68375962',
-            'file': '68375962.mp4',
             'md5': 'aaf896bdb7ddd6476df50007a0ac0ae7',
             'note': 'Video protected with password',
             'info_dict': {
             'md5': 'aaf896bdb7ddd6476df50007a0ac0ae7',
             'note': 'Video protected with password',
             'info_dict': {
+                'id': '68375962',
+                'ext': 'mp4',
                 'title': 'youtube-dl password protected test video',
                 'upload_date': '20130614',
                 'title': 'youtube-dl password protected test video',
                 'upload_date': '20130614',
+                'uploader_url': 're:https?://(?:www\.)?vimeo\.com/user18948128',
                 'uploader_id': 'user18948128',
                 'uploader': 'Jaime Marquínez Ferrándiz',
                 'uploader_id': 'user18948128',
                 'uploader': 'Jaime Marquínez Ferrándiz',
+                'duration': 10,
+                'description': 'This is "youtube-dl password protected test video" by  on Vimeo, the home for high quality videos and the people who love them.',
             },
             'params': {
                 'videopassword': 'youtube-dl',
             },
         },
             },
             'params': {
                 'videopassword': 'youtube-dl',
             },
         },
+        {
+            'url': 'http://vimeo.com/channels/keypeele/75629013',
+            'md5': '2f86a05afe9d7abc0b9126d229bbe15d',
+            'info_dict': {
+                'id': '75629013',
+                'ext': 'mp4',
+                'title': 'Key & Peele: Terrorist Interrogation',
+                'description': 'md5:8678b246399b070816b12313e8b4eb5c',
+                'uploader_url': 're:https?://(?:www\.)?vimeo\.com/atencio',
+                'uploader_id': 'atencio',
+                'uploader': 'Peter Atencio',
+                'upload_date': '20130927',
+                'duration': 187,
+            },
+        },
+        {
+            'url': 'http://vimeo.com/76979871',
+            'note': 'Video with subtitles',
+            'info_dict': {
+                'id': '76979871',
+                'ext': 'mp4',
+                'title': 'The New Vimeo Player (You Know, For Videos)',
+                'description': 'md5:2ec900bf97c3f389378a96aee11260ea',
+                'upload_date': '20131015',
+                'uploader_url': 're:https?://(?:www\.)?vimeo\.com/staff',
+                'uploader_id': 'staff',
+                'uploader': 'Vimeo Staff',
+                'duration': 62,
+            }
+        },
+        {
+            # from https://www.ouya.tv/game/Pier-Solar-and-the-Great-Architects/
+            'url': 'https://player.vimeo.com/video/98044508',
+            'note': 'The js code contains assignments to the same variable as the config',
+            'info_dict': {
+                'id': '98044508',
+                'ext': 'mp4',
+                'title': 'Pier Solar OUYA Official Trailer',
+                'uploader': 'Tulio Gonçalves',
+                'uploader_url': 're:https?://(?:www\.)?vimeo\.com/user28849593',
+                'uploader_id': 'user28849593',
+            },
+        },
+        {
+            # contains original format
+            'url': 'https://vimeo.com/33951933',
+            'md5': '2d9f5475e0537f013d0073e812ab89e6',
+            'info_dict': {
+                'id': '33951933',
+                'ext': 'mp4',
+                'title': 'FOX CLASSICS - Forever Classic ID - A Full Minute',
+                'uploader': 'The DMCI',
+                'uploader_url': 're:https?://(?:www\.)?vimeo\.com/dmci',
+                'uploader_id': 'dmci',
+                'upload_date': '20111220',
+                'description': 'md5:ae23671e82d05415868f7ad1aec21147',
+            },
+        },
+        {
+            # only available via https://vimeo.com/channels/tributes/6213729 and
+            # not via https://vimeo.com/6213729
+            'url': 'https://vimeo.com/channels/tributes/6213729',
+            'info_dict': {
+                'id': '6213729',
+                'ext': 'mp4',
+                'title': 'Vimeo Tribute: The Shining',
+                'uploader': 'Casey Donahue',
+                'uploader_url': 're:https?://(?:www\.)?vimeo\.com/caseydonahue',
+                'uploader_id': 'caseydonahue',
+                'upload_date': '20090821',
+                'description': 'md5:bdbf314014e58713e6e5b66eb252f4a6',
+            },
+            'params': {
+                'skip_download': True,
+            },
+            'expected_warnings': ['Unable to download JSON metadata'],
+        },
+        {
+            # redirects to ondemand extractor and should be passed throught it
+            # for successful extraction
+            'url': 'https://vimeo.com/73445910',
+            'info_dict': {
+                'id': '73445910',
+                'ext': 'mp4',
+                'title': 'The Reluctant Revolutionary',
+                'uploader': '10Ft Films',
+                'uploader_url': 're:https?://(?:www\.)?vimeo\.com/tenfootfilms',
+                'uploader_id': 'tenfootfilms',
+            },
+            'params': {
+                'skip_download': True,
+            },
+        },
+        {
+            'url': 'http://vimeo.com/moogaloop.swf?clip_id=2539741',
+            'only_matching': True,
+        },
+        {
+            'url': 'https://vimeo.com/109815029',
+            'note': 'Video not completely processed, "failed" seed status',
+            'only_matching': True,
+        },
+        {
+            'url': 'https://vimeo.com/groups/travelhd/videos/22439234',
+            'only_matching': True,
+        },
+        {
+            'url': 'https://vimeo.com/album/2632481/video/79010983',
+            'only_matching': True,
+        },
+        {
+            # source file returns 403: Forbidden
+            'url': 'https://vimeo.com/7809605',
+            'only_matching': True,
+        },
+        {
+            'url': 'https://vimeo.com/160743502/abd0e13fb4',
+            'only_matching': True,
+        }
     ]
 
     ]
 
-    def _login(self):
-        (username, password) = self._get_login_info()
-        if username is None:
-            return
-        self.report_login()
-        login_url = 'https://vimeo.com/log_in'
-        webpage = self._download_webpage(login_url, None, False)
-        token = self._search_regex(r'xsrft: \'(.*?)\'', webpage, 'login token')
-        data = compat_urllib_parse.urlencode({'email': username,
-                                              'password': password,
-                                              'action': 'login',
-                                              'service': 'vimeo',
-                                              'token': token,
-                                              })
-        login_request = compat_urllib_request.Request(login_url, data)
-        login_request.add_header('Content-Type', 'application/x-www-form-urlencoded')
-        login_request.add_header('Cookie', 'xsrft=%s' % token)
-        self._download_webpage(login_request, None, False, 'Wrong login info')
-
-    def _verify_video_password(self, url, video_id, webpage):
-        password = self._downloader.params.get('videopassword', None)
-        if password is None:
-            raise ExtractorError('This video is protected by a password, use the --video-password option')
-        token = self._search_regex(r'xsrft: \'(.*?)\'', webpage, 'login token')
-        data = compat_urllib_parse.urlencode({'password': password,
-                                              'token': token})
-        # I didn't manage to use the password with https
-        if url.startswith('https'):
-            pass_url = url.replace('https','http')
-        else:
-            pass_url = url
-        password_request = compat_urllib_request.Request(pass_url+'/password', data)
-        password_request.add_header('Content-Type', 'application/x-www-form-urlencoded')
-        password_request.add_header('Cookie', 'xsrft=%s' % token)
-        self._download_webpage(password_request, video_id,
-                               'Verifying the password',
-                               'Wrong password')
+    @staticmethod
+    def _smuggle_referrer(url, referrer_url):
+        return smuggle_url(url, {'http_headers': {'Referer': referrer_url}})
+
+    @staticmethod
+    def _extract_urls(url, webpage):
+        urls = []
+        # Look for embedded (iframe) Vimeo player
+        for mobj in re.finditer(
+                r'<iframe[^>]+?src=(["\'])(?P<url>(?:https?:)?//player\.vimeo\.com/video/.+?)\1',
+                webpage):
+            urls.append(VimeoIE._smuggle_referrer(unescapeHTML(mobj.group('url')), url))
+        PLAIN_EMBED_RE = (
+            # Look for embedded (swf embed) Vimeo player
+            r'<embed[^>]+?src=(["\'])(?P<url>(?:https?:)?//(?:www\.)?vimeo\.com/moogaloop\.swf.+?)\1',
+            # Look more for non-standard embedded Vimeo player
+            r'<video[^>]+src=(["\'])(?P<url>(?:https?:)?//(?:www\.)?vimeo\.com/[0-9]+)\1',
+        )
+        for embed_re in PLAIN_EMBED_RE:
+            for mobj in re.finditer(embed_re, webpage):
+                urls.append(mobj.group('url'))
+        return urls
+
+    @staticmethod
+    def _extract_url(url, webpage):
+        urls = VimeoIE._extract_urls(url, webpage)
+        return urls[0] if urls else None
 
     def _verify_player_video_password(self, url, video_id):
 
     def _verify_player_video_password(self, url, video_id):
-        password = self._downloader.params.get('videopassword', None)
+        password = self._downloader.params.get('videopassword')
         if password is None:
             raise ExtractorError('This video is protected by a password, use the --video-password option')
         if password is None:
             raise ExtractorError('This video is protected by a password, use the --video-password option')
-        data = compat_urllib_parse.urlencode({'password': password})
+        data = urlencode_postdata({'password': password})
         pass_url = url + '/check-password'
         pass_url = url + '/check-password'
-        password_request = compat_urllib_request.Request(pass_url, data)
+        password_request = sanitized_Request(pass_url, data)
         password_request.add_header('Content-Type', 'application/x-www-form-urlencoded')
         password_request.add_header('Content-Type', 'application/x-www-form-urlencoded')
+        password_request.add_header('Referer', url)
         return self._download_json(
             password_request, video_id,
         return self._download_json(
             password_request, video_id,
-            'Verifying the password',
-            'Wrong password')
+            'Verifying the password', 'Wrong password')
 
     def _real_initialize(self):
         self._login()
 
     def _real_extract(self, url):
 
     def _real_initialize(self):
         self._login()
 
     def _real_extract(self, url):
-        url, data = unsmuggle_url(url)
-        headers = std_headers
-        if data is not None:
-            headers = headers.copy()
-            headers.update(data)
+        url, data = unsmuggle_url(url, {})
+        headers = std_headers.copy()
+        if 'http_headers' in data:
+            headers.update(data['http_headers'])
+        if 'Referer' not in headers:
+            headers['Referer'] = url
 
         # Extract ID from URL
         mobj = re.match(self._VALID_URL, url)
         video_id = mobj.group('id')
 
         # Extract ID from URL
         mobj = re.match(self._VALID_URL, url)
         video_id = mobj.group('id')
+        orig_url = url
         if mobj.group('pro') or mobj.group('player'):
         if mobj.group('pro') or mobj.group('player'):
-            url = 'http://player.vimeo.com/video/' + video_id
-        else:
+            url = 'https://player.vimeo.com/video/' + video_id
+        elif any(p in url for p in ('play_redirect_hls', 'moogaloop.swf')):
             url = 'https://vimeo.com/' + video_id
 
         # Retrieve video webpage to extract further information
             url = 'https://vimeo.com/' + video_id
 
         # Retrieve video webpage to extract further information
-        request = compat_urllib_request.Request(url, None, headers)
-        webpage = self._download_webpage(request, video_id)
+        request = sanitized_Request(url, headers=headers)
+        try:
+            webpage, urlh = self._download_webpage_handle(request, video_id)
+            # Some URLs redirect to ondemand can't be extracted with
+            # this extractor right away thus should be passed through
+            # ondemand extractor (e.g. https://vimeo.com/73445910)
+            if VimeoOndemandIE.suitable(urlh.geturl()):
+                return self.url_result(urlh.geturl(), VimeoOndemandIE.ie_key())
+        except ExtractorError as ee:
+            if isinstance(ee.cause, compat_HTTPError) and ee.cause.code == 403:
+                errmsg = ee.cause.read()
+                if b'Because of its privacy settings, this video cannot be played here' in errmsg:
+                    raise ExtractorError(
+                        'Cannot download embed-only video without embedding '
+                        'URL. Please call youtube-dl with the URL of the page '
+                        'that embeds this video.',
+                        expected=True)
+            raise
 
         # Now we begin extracting as much information as we can from what we
         # retrieved. First we extract the information common to all extractors,
         # and latter we extract those that are Vimeo specific.
         self.report_extraction(video_id)
 
 
         # Now we begin extracting as much information as we can from what we
         # retrieved. First we extract the information common to all extractors,
         # and latter we extract those that are Vimeo specific.
         self.report_extraction(video_id)
 
+        vimeo_config = self._search_regex(
+            r'vimeo\.config\s*=\s*(?:({.+?})|_extend\([^,]+,\s+({.+?})\));', webpage,
+            'vimeo config', default=None)
+        if vimeo_config:
+            seed_status = self._parse_json(vimeo_config, video_id).get('seed_status', {})
+            if seed_status.get('state') == 'failed':
+                raise ExtractorError(
+                    '%s said: %s' % (self.IE_NAME, seed_status['title']),
+                    expected=True)
+
         # Extract the config JSON
         try:
             try:
                 config_url = self._html_search_regex(
         # Extract the config JSON
         try:
             try:
                 config_url = self._html_search_regex(
-                    r' data-config-url="(.+?)"', webpage, 'config URL')
+                    r' data-config-url="(.+?)"', webpage,
+                    'config URL', default=None)
+                if not config_url:
+                    # Sometimes new react-based page is served instead of old one that require
+                    # different config URL extraction approach (see
+                    # https://github.com/rg3/youtube-dl/pull/7209)
+                    vimeo_clip_page_config = self._search_regex(
+                        r'vimeo\.clip_page_config\s*=\s*({.+?});', webpage,
+                        'vimeo clip page config')
+                    config_url = self._parse_json(
+                        vimeo_clip_page_config, video_id)['player']['config_url']
                 config_json = self._download_webpage(config_url, video_id)
                 config = json.loads(config_json)
             except RegexNotFoundError:
                 config_json = self._download_webpage(config_url, video_id)
                 config = json.loads(config_json)
             except RegexNotFoundError:
@@ -176,19 +484,22 @@ class VimeoIE(InfoExtractor):
                 # We try to find out to which variable is assigned the config dic
                 m_variable_name = re.search('(\w)\.video\.id', webpage)
                 if m_variable_name is not None:
                 # We try to find out to which variable is assigned the config dic
                 m_variable_name = re.search('(\w)\.video\.id', webpage)
                 if m_variable_name is not None:
-                    config_re = r'%s=({.+?});' % re.escape(m_variable_name.group(1))
+                    config_re = r'%s=({[^}].+?});' % re.escape(m_variable_name.group(1))
                 else:
                     config_re = [r' = {config:({.+?}),assets:', r'(?:[abc])=({.+?});']
                 config = self._search_regex(config_re, webpage, 'info section',
                 else:
                     config_re = [r' = {config:({.+?}),assets:', r'(?:[abc])=({.+?});']
                 config = self._search_regex(config_re, webpage, 'info section',
-                    flags=re.DOTALL)
+                                            flags=re.DOTALL)
                 config = json.loads(config)
         except Exception as e:
             if re.search('The creator of this video has not given you permission to embed it on this domain.', webpage):
                 raise ExtractorError('The author has restricted the access to this video, try with the "--referer" option')
 
                 config = json.loads(config)
         except Exception as e:
             if re.search('The creator of this video has not given you permission to embed it on this domain.', webpage):
                 raise ExtractorError('The author has restricted the access to this video, try with the "--referer" option')
 
-            if re.search('<form[^>]+?id="pw_form"', webpage) is not None:
+            if re.search(r'<form[^>]+?id="pw_form"', webpage) is not None:
+                if '_video_password_verified' in data:
+                    raise ExtractorError('video password verification failed!')
                 self._verify_video_password(url, video_id, webpage)
                 self._verify_video_password(url, video_id, webpage)
-                return self._real_extract(url)
+                return self._real_extract(
+                    smuggle_url(url, {'_video_password_verified': 'verified'}))
             else:
                 raise ExtractorError('Unable to extract info section',
                                      cause=e)
             else:
                 raise ExtractorError('Unable to extract info section',
                                      cause=e)
@@ -196,36 +507,48 @@ class VimeoIE(InfoExtractor):
             if config.get('view') == 4:
                 config = self._verify_player_video_password(url, video_id)
 
             if config.get('view') == 4:
                 config = self._verify_player_video_password(url, video_id)
 
-        # Extract title
-        video_title = config["video"]["title"]
+        def is_rented():
+            if '>You rented this title.<' in webpage:
+                return True
+            if config.get('user', {}).get('purchased'):
+                return True
+            label = try_get(
+                config, lambda x: x['video']['vod']['purchase_options'][0]['label_string'], compat_str)
+            if label and label.startswith('You rented this'):
+                return True
+            return False
 
 
-        # Extract uploader and uploader_id
-        video_uploader = config["video"]["owner"]["name"]
-        video_uploader_id = config["video"]["owner"]["url"].split('/')[-1] if config["video"]["owner"]["url"] else None
-
-        # Extract video thumbnail
-        video_thumbnail = config["video"].get("thumbnail")
-        if video_thumbnail is None:
-            _, video_thumbnail = sorted((int(width), t_url) for (width, t_url) in config["video"]["thumbs"].items())[-1]
+        if is_rented():
+            feature_id = config.get('video', {}).get('vod', {}).get('feature_id')
+            if feature_id and not data.get('force_feature_id', False):
+                return self.url_result(smuggle_url(
+                    'https://player.vimeo.com/player/%s' % feature_id,
+                    {'force_feature_id': True}), 'Vimeo')
 
         # Extract video description
 
         # Extract video description
-        video_description = None
-        try:
-            video_description = get_element_by_attribute("itemprop", "description", webpage)
-            if video_description: video_description = clean_html(video_description)
-        except AssertionError as err:
-            # On some pages like (http://player.vimeo.com/video/54469442) the
-            # html tags are not closed, python 2.6 cannot handle it
-            if err.args[0] == 'we should not get here!':
-                pass
-            else:
-                raise
+
+        video_description = self._html_search_regex(
+            r'(?s)<div\s+class="[^"]*description[^"]*"[^>]*>(.*?)</div>',
+            webpage, 'description', default=None)
+        if not video_description:
+            video_description = self._html_search_meta(
+                'description', webpage, default=None)
+        if not video_description and mobj.group('pro'):
+            orig_webpage = self._download_webpage(
+                orig_url, video_id,
+                note='Downloading webpage for description',
+                fatal=False)
+            if orig_webpage:
+                video_description = self._html_search_meta(
+                    'description', orig_webpage, default=None)
+        if not video_description and not mobj.group('player'):
+            self._downloader.report_warning('Cannot find video description')
 
         # Extract upload date
         video_upload_date = None
 
         # Extract upload date
         video_upload_date = None
-        mobj = re.search(r'<meta itemprop="dateCreated" content="(\d{4})-(\d{2})-(\d{2})T', webpage)
+        mobj = re.search(r'<time[^>]+datetime="([^"]+)"', webpage)
         if mobj is not None:
         if mobj is not None:
-            video_upload_date = mobj.group(1) + mobj.group(2) + mobj.group(3)
+            video_upload_date = unified_strdate(mobj.group(1))
 
         try:
             view_count = int(self._search_regex(r'UserPlays:(\d+)', webpage, 'view count'))
 
         try:
             view_count = int(self._search_regex(r'UserPlays:(\d+)', webpage, 'view count'))
@@ -237,128 +560,254 @@ class VimeoIE(InfoExtractor):
             like_count = None
             comment_count = None
 
             like_count = None
             comment_count = None
 
-        # Vimeo specific: extract request signature and timestamp
-        sig = config['request']['signature']
-        timestamp = config['request']['timestamp']
-
-        # Vimeo specific: extract video codec and quality information
-        # First consider quality, then codecs, then take everything
-        codecs = [('vp6', 'flv'), ('vp8', 'flv'), ('h264', 'mp4')]
-        files = {'hd': [], 'sd': [], 'other': []}
-        config_files = config["video"].get("files") or config["request"].get("files")
-        for codec_name, codec_extension in codecs:
-            for quality in config_files.get(codec_name, []):
-                format_id = '-'.join((codec_name, quality)).lower()
-                key = quality if quality in files else 'other'
-                video_url = None
-                if isinstance(config_files[codec_name], dict):
-                    file_info = config_files[codec_name][quality]
-                    video_url = file_info.get('url')
-                else:
-                    file_info = {}
-                if video_url is None:
-                    video_url = "http://player.vimeo.com/play_redirect?clip_id=%s&sig=%s&time=%s&quality=%s&codecs=%s&type=moogaloop_local&embed_location=" \
-                        %(video_id, sig, timestamp, quality, codec_name.upper())
-
-                files[key].append({
-                    'ext': codec_extension,
-                    'url': video_url,
-                    'format_id': format_id,
-                    'width': file_info.get('width'),
-                    'height': file_info.get('height'),
-                })
         formats = []
         formats = []
-        for key in ('other', 'sd', 'hd'):
-            formats += files[key]
-        if len(formats) == 0:
-            raise ExtractorError('No known codec found')
-
-        return {
-            'id':       video_id,
-            'uploader': video_uploader,
-            'uploader_id': video_uploader_id,
-            'upload_date':  video_upload_date,
-            'title':    video_title,
-            'thumbnail':    video_thumbnail,
-            'description':  video_description,
+        download_request = sanitized_Request('https://vimeo.com/%s?action=load_download_config' % video_id, headers={
+            'X-Requested-With': 'XMLHttpRequest'})
+        download_data = self._download_json(download_request, video_id, fatal=False)
+        if download_data:
+            source_file = download_data.get('source_file')
+            if isinstance(source_file, dict):
+                download_url = source_file.get('download_url')
+                if download_url and not source_file.get('is_cold') and not source_file.get('is_defrosting'):
+                    source_name = source_file.get('public_name', 'Original')
+                    if self._is_valid_url(download_url, video_id, '%s video' % source_name):
+                        ext = source_file.get('extension', determine_ext(download_url)).lower()
+                        formats.append({
+                            'url': download_url,
+                            'ext': ext,
+                            'width': int_or_none(source_file.get('width')),
+                            'height': int_or_none(source_file.get('height')),
+                            'filesize': parse_filesize(source_file.get('size')),
+                            'format_id': source_name,
+                            'preference': 1,
+                        })
+
+        info_dict = self._parse_config(config, video_id)
+        formats.extend(info_dict['formats'])
+        self._vimeo_sort_formats(formats)
+        info_dict.update({
+            'id': video_id,
             'formats': formats,
             'formats': formats,
+            'upload_date': video_upload_date,
+            'description': video_description,
             'webpage_url': url,
             'view_count': view_count,
             'like_count': like_count,
             'comment_count': comment_count,
             'webpage_url': url,
             'view_count': view_count,
             'like_count': like_count,
             'comment_count': comment_count,
-        }
+        })
+
+        return info_dict
+
 
 
+class VimeoOndemandIE(VimeoBaseInfoExtractor):
+    IE_NAME = 'vimeo:ondemand'
+    _VALID_URL = r'https?://(?:www\.)?vimeo\.com/ondemand/(?P<id>[^/?#&]+)'
+    _TESTS = [{
+        # ondemand video not available via https://vimeo.com/id
+        'url': 'https://vimeo.com/ondemand/20704',
+        'md5': 'c424deda8c7f73c1dfb3edd7630e2f35',
+        'info_dict': {
+            'id': '105442900',
+            'ext': 'mp4',
+            'title': 'המעבדה - במאי יותם פלדמן',
+            'uploader': 'גם סרטים',
+            'uploader_url': 're:https?://(?:www\.)?vimeo\.com/gumfilms',
+            'uploader_id': 'gumfilms',
+        },
+    }, {
+        # requires Referer to be passed along with og:video:url
+        'url': 'https://vimeo.com/ondemand/36938/126682985',
+        'info_dict': {
+            'id': '126682985',
+            'ext': 'mp4',
+            'title': 'Rävlock, rätt läte på rätt plats',
+            'uploader': 'Lindroth & Norin',
+            'uploader_url': 're:https?://(?:www\.)?vimeo\.com/user14430847',
+            'uploader_id': 'user14430847',
+        },
+        'params': {
+            'skip_download': True,
+        },
+    }, {
+        'url': 'https://vimeo.com/ondemand/nazmaalik',
+        'only_matching': True,
+    }, {
+        'url': 'https://vimeo.com/ondemand/141692381',
+        'only_matching': True,
+    }, {
+        'url': 'https://vimeo.com/ondemand/thelastcolony/150274832',
+        'only_matching': True,
+    }]
 
 
-class VimeoChannelIE(InfoExtractor):
+    def _real_extract(self, url):
+        video_id = self._match_id(url)
+        webpage = self._download_webpage(url, video_id)
+        return self.url_result(
+            # Some videos require Referer to be passed along with og:video:url
+            # similarly to generic vimeo embeds (e.g.
+            # https://vimeo.com/ondemand/36938/126682985).
+            VimeoIE._smuggle_referrer(self._og_search_video_url(webpage), url),
+            VimeoIE.ie_key())
+
+
+class VimeoChannelIE(VimeoBaseInfoExtractor):
     IE_NAME = 'vimeo:channel'
     IE_NAME = 'vimeo:channel'
-    _VALID_URL = r'(?:https?://)?vimeo.\com/channels/(?P<id>[^/]+)'
+    _VALID_URL = r'https://vimeo\.com/channels/(?P<id>[^/?#]+)/?(?:$|[?#])'
     _MORE_PAGES_INDICATOR = r'<a.+?rel="next"'
     _MORE_PAGES_INDICATOR = r'<a.+?rel="next"'
+    _TITLE = None
     _TITLE_RE = r'<link rel="alternate"[^>]+?title="(.*?)"'
     _TITLE_RE = r'<link rel="alternate"[^>]+?title="(.*?)"'
+    _TESTS = [{
+        'url': 'https://vimeo.com/channels/tributes',
+        'info_dict': {
+            'id': 'tributes',
+            'title': 'Vimeo Tributes',
+        },
+        'playlist_mincount': 25,
+    }]
 
     def _page_url(self, base_url, pagenum):
         return '%s/videos/page:%d/' % (base_url, pagenum)
 
     def _extract_list_title(self, webpage):
 
     def _page_url(self, base_url, pagenum):
         return '%s/videos/page:%d/' % (base_url, pagenum)
 
     def _extract_list_title(self, webpage):
-        return self._html_search_regex(self._TITLE_RE, webpage, 'list title')
+        return self._TITLE or self._html_search_regex(self._TITLE_RE, webpage, 'list title')
 
 
-    def _extract_videos(self, list_id, base_url):
-        video_ids = []
+    def _login_list_password(self, page_url, list_id, webpage):
+        login_form = self._search_regex(
+            r'(?s)<form[^>]+?id="pw_form"(.*?)</form>',
+            webpage, 'login form', default=None)
+        if not login_form:
+            return webpage
+
+        password = self._downloader.params.get('videopassword')
+        if password is None:
+            raise ExtractorError('This album is protected by a password, use the --video-password option', expected=True)
+        fields = self._hidden_inputs(login_form)
+        token, vuid = self._extract_xsrft_and_vuid(webpage)
+        fields['token'] = token
+        fields['password'] = password
+        post = urlencode_postdata(fields)
+        password_path = self._search_regex(
+            r'action="([^"]+)"', login_form, 'password URL')
+        password_url = compat_urlparse.urljoin(page_url, password_path)
+        password_request = sanitized_Request(password_url, post)
+        password_request.add_header('Content-type', 'application/x-www-form-urlencoded')
+        self._set_vimeo_cookie('vuid', vuid)
+        self._set_vimeo_cookie('xsrft', token)
+
+        return self._download_webpage(
+            password_request, list_id,
+            'Verifying the password', 'Wrong password')
+
+    def _title_and_entries(self, list_id, base_url):
         for pagenum in itertools.count(1):
         for pagenum in itertools.count(1):
+            page_url = self._page_url(base_url, pagenum)
             webpage = self._download_webpage(
             webpage = self._download_webpage(
-                self._page_url(base_url, pagenum) ,list_id,
+                page_url, list_id,
                 'Downloading page %s' % pagenum)
                 'Downloading page %s' % pagenum)
-            video_ids.extend(re.findall(r'id="clip_(\d+?)"', webpage))
+
+            if pagenum == 1:
+                webpage = self._login_list_password(page_url, list_id, webpage)
+                yield self._extract_list_title(webpage)
+
+            # Try extracting href first since not all videos are available via
+            # short https://vimeo.com/id URL (e.g. https://vimeo.com/channels/tributes/6213729)
+            clips = re.findall(
+                r'id="clip_(\d+)"[^>]*>\s*<a[^>]+href="(/(?:[^/]+/)*\1)', webpage)
+            if clips:
+                for video_id, video_url in clips:
+                    yield self.url_result(
+                        compat_urlparse.urljoin(base_url, video_url),
+                        VimeoIE.ie_key(), video_id=video_id)
+            # More relaxed fallback
+            else:
+                for video_id in re.findall(r'id=["\']clip_(\d+)', webpage):
+                    yield self.url_result(
+                        'https://vimeo.com/%s' % video_id,
+                        VimeoIE.ie_key(), video_id=video_id)
+
             if re.search(self._MORE_PAGES_INDICATOR, webpage, re.DOTALL) is None:
                 break
 
             if re.search(self._MORE_PAGES_INDICATOR, webpage, re.DOTALL) is None:
                 break
 
-        entries = [self.url_result('http://vimeo.com/%s' % video_id, 'Vimeo')
-                   for video_id in video_ids]
-        return {'_type': 'playlist',
-                'id': list_id,
-                'title': self._extract_list_title(webpage),
-                'entries': entries,
-                }
+    def _extract_videos(self, list_id, base_url):
+        title_and_entries = self._title_and_entries(list_id, base_url)
+        list_title = next(title_and_entries)
+        return self.playlist_result(title_and_entries, list_id, list_title)
 
     def _real_extract(self, url):
         mobj = re.match(self._VALID_URL, url)
 
     def _real_extract(self, url):
         mobj = re.match(self._VALID_URL, url)
-        channel_id =  mobj.group('id')
-        return self._extract_videos(channel_id, 'http://vimeo.com/channels/%s' % channel_id)
+        channel_id = mobj.group('id')
+        return self._extract_videos(channel_id, 'https://vimeo.com/channels/%s' % channel_id)
 
 
 class VimeoUserIE(VimeoChannelIE):
     IE_NAME = 'vimeo:user'
 
 
 class VimeoUserIE(VimeoChannelIE):
     IE_NAME = 'vimeo:user'
-    _VALID_URL = r'(?:https?://)?vimeo.\com/(?P<name>[^/]+)(?:/videos|[#?]|$)'
+    _VALID_URL = r'https://vimeo\.com/(?!(?:[0-9]+|watchlater)(?:$|[?#/]))(?P<name>[^/]+)(?:/videos|[#?]|$)'
     _TITLE_RE = r'<a[^>]+?class="user">([^<>]+?)</a>'
     _TITLE_RE = r'<a[^>]+?class="user">([^<>]+?)</a>'
-
-    @classmethod
-    def suitable(cls, url):
-        if VimeoChannelIE.suitable(url) or VimeoIE.suitable(url) or VimeoAlbumIE.suitable(url) or VimeoGroupsIE.suitable(url):
-            return False
-        return super(VimeoUserIE, cls).suitable(url)
+    _TESTS = [{
+        'url': 'https://vimeo.com/nkistudio/videos',
+        'info_dict': {
+            'title': 'Nki',
+            'id': 'nkistudio',
+        },
+        'playlist_mincount': 66,
+    }]
 
     def _real_extract(self, url):
         mobj = re.match(self._VALID_URL, url)
         name = mobj.group('name')
 
     def _real_extract(self, url):
         mobj = re.match(self._VALID_URL, url)
         name = mobj.group('name')
-        return self._extract_videos(name, 'http://vimeo.com/%s' % name)
+        return self._extract_videos(name, 'https://vimeo.com/%s' % name)
 
 
 class VimeoAlbumIE(VimeoChannelIE):
     IE_NAME = 'vimeo:album'
 
 
 class VimeoAlbumIE(VimeoChannelIE):
     IE_NAME = 'vimeo:album'
-    _VALID_URL = r'(?:https?://)?vimeo.\com/album/(?P<id>\d+)'
+    _VALID_URL = r'https://vimeo\.com/album/(?P<id>\d+)(?:$|[?#]|/(?!video))'
     _TITLE_RE = r'<header id="page_header">\n\s*<h1>(.*?)</h1>'
     _TITLE_RE = r'<header id="page_header">\n\s*<h1>(.*?)</h1>'
+    _TESTS = [{
+        'url': 'https://vimeo.com/album/2632481',
+        'info_dict': {
+            'id': '2632481',
+            'title': 'Staff Favorites: November 2013',
+        },
+        'playlist_mincount': 13,
+    }, {
+        'note': 'Password-protected album',
+        'url': 'https://vimeo.com/album/3253534',
+        'info_dict': {
+            'title': 'test',
+            'id': '3253534',
+        },
+        'playlist_count': 1,
+        'params': {
+            'videopassword': 'youtube-dl',
+        }
+    }, {
+        'url': 'https://vimeo.com/album/2632481/sort:plays/format:thumbnail',
+        'only_matching': True,
+    }, {
+        # TODO: respect page number
+        'url': 'https://vimeo.com/album/2632481/page:2/sort:plays/format:thumbnail',
+        'only_matching': True,
+    }]
 
     def _page_url(self, base_url, pagenum):
         return '%s/page:%d/' % (base_url, pagenum)
 
     def _real_extract(self, url):
 
     def _page_url(self, base_url, pagenum):
         return '%s/page:%d/' % (base_url, pagenum)
 
     def _real_extract(self, url):
-        mobj = re.match(self._VALID_URL, url)
-        album_id = mobj.group('id')
-        return self._extract_videos(album_id, 'http://vimeo.com/album/%s' % album_id)
+        album_id = self._match_id(url)
+        return self._extract_videos(album_id, 'https://vimeo.com/album/%s' % album_id)
 
 
 class VimeoGroupsIE(VimeoAlbumIE):
     IE_NAME = 'vimeo:group'
 
 
 class VimeoGroupsIE(VimeoAlbumIE):
     IE_NAME = 'vimeo:group'
-    _VALID_URL = r'(?:https?://)?vimeo.\com/groups/(?P<name>[^/]+)'
+    _VALID_URL = r'https://vimeo\.com/groups/(?P<name>[^/]+)(?:/(?!videos?/\d+)|$)'
+    _TESTS = [{
+        'url': 'https://vimeo.com/groups/rolexawards',
+        'info_dict': {
+            'id': 'rolexawards',
+            'title': 'Rolex Awards for Enterprise',
+        },
+        'playlist_mincount': 73,
+    }]
 
     def _extract_list_title(self, webpage):
         return self._og_search_title(webpage)
 
     def _extract_list_title(self, webpage):
         return self._og_search_title(webpage)
@@ -366,25 +815,154 @@ class VimeoGroupsIE(VimeoAlbumIE):
     def _real_extract(self, url):
         mobj = re.match(self._VALID_URL, url)
         name = mobj.group('name')
     def _real_extract(self, url):
         mobj = re.match(self._VALID_URL, url)
         name = mobj.group('name')
-        return self._extract_videos(name, 'http://vimeo.com/groups/%s' % name)
+        return self._extract_videos(name, 'https://vimeo.com/groups/%s' % name)
 
 
 
 
-class VimeoReviewIE(InfoExtractor):
+class VimeoReviewIE(VimeoBaseInfoExtractor):
     IE_NAME = 'vimeo:review'
     IE_DESC = 'Review pages on vimeo'
     IE_NAME = 'vimeo:review'
     IE_DESC = 'Review pages on vimeo'
-    _VALID_URL = r'(?:https?://)?vimeo.\com/[^/]+/review/(?P<id>[^/]+)'
-    _TEST = {
+    _VALID_URL = r'https://vimeo\.com/[^/]+/review/(?P<id>[^/]+)'
+    _TESTS = [{
         'url': 'https://vimeo.com/user21297594/review/75524534/3c257a1b5d',
         'url': 'https://vimeo.com/user21297594/review/75524534/3c257a1b5d',
-        'file': '75524534.mp4',
         'md5': 'c507a72f780cacc12b2248bb4006d253',
         'info_dict': {
         'md5': 'c507a72f780cacc12b2248bb4006d253',
         'info_dict': {
+            'id': '75524534',
+            'ext': 'mp4',
             'title': "DICK HARDWICK 'Comedian'",
             'uploader': 'Richard Hardwick',
             'title': "DICK HARDWICK 'Comedian'",
             'uploader': 'Richard Hardwick',
+            'uploader_id': 'user21297594',
         }
         }
+    }, {
+        'note': 'video player needs Referer',
+        'url': 'https://vimeo.com/user22258446/review/91613211/13f927e053',
+        'md5': '6295fdab8f4bf6a002d058b2c6dce276',
+        'info_dict': {
+            'id': '91613211',
+            'ext': 'mp4',
+            'title': 're:(?i)^Death by dogma versus assembling agile . Sander Hoogendoorn',
+            'uploader': 'DevWeek Events',
+            'duration': 2773,
+            'thumbnail': 're:^https?://.*\.jpg$',
+            'uploader_id': 'user22258446',
+        }
+    }, {
+        'note': 'Password protected',
+        'url': 'https://vimeo.com/user37284429/review/138823582/c4d865efde',
+        'info_dict': {
+            'id': '138823582',
+            'ext': 'mp4',
+            'title': 'EFFICIENT PICKUP MASTERCLASS MODULE 1',
+            'uploader': 'TMB',
+            'uploader_id': 'user37284429',
+        },
+        'params': {
+            'videopassword': 'holygrail',
+        },
+        'skip': 'video gone',
+    }]
+
+    def _real_initialize(self):
+        self._login()
+
+    def _get_config_url(self, webpage_url, video_id, video_password_verified=False):
+        webpage = self._download_webpage(webpage_url, video_id)
+        data = self._parse_json(self._search_regex(
+            r'window\s*=\s*_extend\(window,\s*({.+?})\);', webpage, 'data',
+            default=NO_DEFAULT if video_password_verified else '{}'), video_id)
+        config_url = data.get('vimeo_esi', {}).get('config', {}).get('configUrl')
+        if config_url is None:
+            self._verify_video_password(webpage_url, video_id, webpage)
+            config_url = self._get_config_url(
+                webpage_url, video_id, video_password_verified=True)
+        return config_url
+
+    def _real_extract(self, url):
+        video_id = self._match_id(url)
+        config_url = self._get_config_url(url, video_id)
+        config = self._download_json(config_url, video_id)
+        info_dict = self._parse_config(config, video_id)
+        self._vimeo_sort_formats(info_dict['formats'])
+        info_dict['id'] = video_id
+        return info_dict
+
+
+class VimeoWatchLaterIE(VimeoChannelIE):
+    IE_NAME = 'vimeo:watchlater'
+    IE_DESC = 'Vimeo watch later list, "vimeowatchlater" keyword (requires authentication)'
+    _VALID_URL = r'https://vimeo\.com/(?:home/)?watchlater|:vimeowatchlater'
+    _TITLE = 'Watch Later'
+    _LOGIN_REQUIRED = True
+    _TESTS = [{
+        'url': 'https://vimeo.com/watchlater',
+        'only_matching': True,
+    }]
+
+    def _real_initialize(self):
+        self._login()
+
+    def _page_url(self, base_url, pagenum):
+        url = '%s/page:%d/' % (base_url, pagenum)
+        request = sanitized_Request(url)
+        # Set the header to get a partial html page with the ids,
+        # the normal page doesn't contain them.
+        request.add_header('X-Requested-With', 'XMLHttpRequest')
+        return request
+
+    def _real_extract(self, url):
+        return self._extract_videos('watchlater', 'https://vimeo.com/watchlater')
+
+
+class VimeoLikesIE(InfoExtractor):
+    _VALID_URL = r'https://(?:www\.)?vimeo\.com/user(?P<id>[0-9]+)/likes/?(?:$|[?#]|sort:)'
+    IE_NAME = 'vimeo:likes'
+    IE_DESC = 'Vimeo user likes'
+    _TEST = {
+        'url': 'https://vimeo.com/user755559/likes/',
+        'playlist_mincount': 293,
+        'info_dict': {
+            'id': 'user755559_likes',
+            'description': 'See all the videos urza likes',
+            'title': 'Videos urza likes',
+        },
     }
 
     def _real_extract(self, url):
     }
 
     def _real_extract(self, url):
-        mobj = re.match(self._VALID_URL, url)
-        video_id = mobj.group('id')
-        player_url = 'https://player.vimeo.com/player/' + video_id
-        return self.url_result(player_url, 'Vimeo', video_id)
+        user_id = self._match_id(url)
+        webpage = self._download_webpage(url, user_id)
+        page_count = self._int(
+            self._search_regex(
+                r'''(?x)<li><a\s+href="[^"]+"\s+data-page="([0-9]+)">
+                    .*?</a></li>\s*<li\s+class="pagination_next">
+                ''', webpage, 'page count'),
+            'page count', fatal=True)
+        PAGE_SIZE = 12
+        title = self._html_search_regex(
+            r'(?s)<h1>(.+?)</h1>', webpage, 'title', fatal=False)
+        description = self._html_search_meta('description', webpage)
+
+        def _get_page(idx):
+            page_url = 'https://vimeo.com/user%s/likes/page:%d/sort:date' % (
+                user_id, idx + 1)
+            webpage = self._download_webpage(
+                page_url, user_id,
+                note='Downloading page %d/%d' % (idx + 1, page_count))
+            video_list = self._search_regex(
+                r'(?s)<ol class="js-browse_list[^"]+"[^>]*>(.*?)</ol>',
+                webpage, 'video content')
+            paths = re.findall(
+                r'<li[^>]*>\s*<a\s+href="([^"]+)"', video_list)
+            for path in paths:
+                yield {
+                    '_type': 'url',
+                    'url': compat_urlparse.urljoin(page_url, path),
+                }
+
+        pl = InAdvancePagedList(_get_page, page_count, PAGE_SIZE)
+
+        return {
+            '_type': 'playlist',
+            'id': 'user%s_likes' % user_id,
+            'title': title,
+            'description': description,
+            'entries': pl,
+        }