X-Git-Url: https://git.rapsys.eu/youtubedl/blobdiff_plain/54477c858fa2504360964e675faf953d78ac57e6..3e696c728febc53e74c912c40ccd3b9504d536df:/youtube_dl/extractor/vevo.py diff --git a/youtube_dl/extractor/vevo.py b/youtube_dl/extractor/vevo.py index 1c1cc41..d82261e 100644 --- a/youtube_dl/extractor/vevo.py +++ b/youtube_dl/extractor/vevo.py @@ -1,77 +1,462 @@ +from __future__ import unicode_literals + import re -import json -import xml.etree.ElementTree -import datetime from .common import InfoExtractor +from ..compat import ( + compat_etree_fromstring, + compat_str, + compat_urlparse, +) from ..utils import ( - determine_ext, ExtractorError, + int_or_none, + sanitized_Request, + parse_iso8601, ) -class VevoIE(InfoExtractor): - """ +class VevoBaseIE(InfoExtractor): + def _extract_json(self, webpage, video_id, item): + return self._parse_json( + self._search_regex( + r'window\.__INITIAL_STORE__\s*=\s*({.+?});\s*', + webpage, 'initial store'), + video_id)['default'][item] + + +class VevoIE(VevoBaseIE): + ''' Accepts urls from vevo.com or in the format 'vevo:{id}' - (currently used by MTVIE) - """ - _VALID_URL = r'((http://www.vevo.com/watch/.*?/.*?/)|(vevo:))(?P.*?)(\?|$)' - _TEST = { - u'url': u'http://www.vevo.com/watch/hurts/somebody-to-die-for/GB1101300280', - u'file': u'GB1101300280.mp4', - u'info_dict': { - u"upload_date": u"20130624", - u"uploader": u"Hurts", - u"title": u"Somebody to Die For", - u'duration': 230, - } + (currently used by MTVIE and MySpaceIE) + ''' + _VALID_URL = r'''(?x) + (?:https?://(?:www\.)?vevo\.com/watch/(?!playlist|genre)(?:[^/]+/(?:[^/]+/)?)?| + https?://cache\.vevo\.com/m/html/embed\.html\?video=| + https?://videoplayer\.vevo\.com/embed/embedded\?videoId=| + vevo:) + (?P[^&?#]+)''' + + _TESTS = [{ + 'url': 'http://www.vevo.com/watch/hurts/somebody-to-die-for/GB1101300280', + 'md5': '95ee28ee45e70130e3ab02b0f579ae23', + 'info_dict': { + 'id': 'GB1101300280', + 'ext': 'mp4', + 'title': 'Hurts - Somebody to Die For', + 'timestamp': 1372057200, + 'upload_date': '20130624', + 'uploader': 'Hurts', + 'track': 'Somebody to Die For', + 'artist': 'Hurts', + 'genre': 'Pop', + }, + 'expected_warnings': ['Unable to download SMIL file', 'Unable to download info'], + }, { + 'note': 'v3 SMIL format', + 'url': 'http://www.vevo.com/watch/cassadee-pope/i-wish-i-could-break-your-heart/USUV71302923', + 'md5': 'f6ab09b034f8c22969020b042e5ac7fc', + 'info_dict': { + 'id': 'USUV71302923', + 'ext': 'mp4', + 'title': 'Cassadee Pope - I Wish I Could Break Your Heart', + 'timestamp': 1392796919, + 'upload_date': '20140219', + 'uploader': 'Cassadee Pope', + 'track': 'I Wish I Could Break Your Heart', + 'artist': 'Cassadee Pope', + 'genre': 'Country', + }, + 'expected_warnings': ['Unable to download SMIL file', 'Unable to download info'], + }, { + 'note': 'Age-limited video', + 'url': 'https://www.vevo.com/watch/justin-timberlake/tunnel-vision-explicit/USRV81300282', + 'info_dict': { + 'id': 'USRV81300282', + 'ext': 'mp4', + 'title': 'Justin Timberlake - Tunnel Vision (Explicit)', + 'age_limit': 18, + 'timestamp': 1372888800, + 'upload_date': '20130703', + 'uploader': 'Justin Timberlake', + 'track': 'Tunnel Vision (Explicit)', + 'artist': 'Justin Timberlake', + 'genre': 'Pop', + }, + 'expected_warnings': ['Unable to download SMIL file', 'Unable to download info'], + }, { + 'note': 'No video_info', + 'url': 'http://www.vevo.com/watch/k-camp-1/Till-I-Die/USUV71503000', + 'md5': '8b83cc492d72fc9cf74a02acee7dc1b0', + 'info_dict': { + 'id': 'USUV71503000', + 'ext': 'mp4', + 'title': 'K Camp ft. T.I. - Till I Die', + 'age_limit': 18, + 'timestamp': 1449468000, + 'upload_date': '20151207', + 'uploader': 'K Camp', + 'track': 'Till I Die', + 'artist': 'K Camp', + 'genre': 'Hip-Hop', + }, + 'expected_warnings': ['Unable to download SMIL file', 'Unable to download info'], + }, { + 'note': 'Featured test', + 'url': 'https://www.vevo.com/watch/lemaitre/Wait/USUV71402190', + 'md5': 'd28675e5e8805035d949dc5cf161071d', + 'info_dict': { + 'id': 'USUV71402190', + 'ext': 'mp4', + 'title': 'Lemaitre ft. LoLo - Wait', + 'age_limit': 0, + 'timestamp': 1413432000, + 'upload_date': '20141016', + 'uploader': 'Lemaitre', + 'track': 'Wait', + 'artist': 'Lemaitre', + 'genre': 'Electronic', + }, + 'expected_warnings': ['Unable to download SMIL file', 'Unable to download info'], + }, { + 'note': 'Only available via webpage', + 'url': 'http://www.vevo.com/watch/GBUV71600656', + 'md5': '67e79210613865b66a47c33baa5e37fe', + 'info_dict': { + 'id': 'GBUV71600656', + 'ext': 'mp4', + 'title': 'ABC - Viva Love', + 'age_limit': 0, + 'timestamp': 1461830400, + 'upload_date': '20160428', + 'uploader': 'ABC', + 'track': 'Viva Love', + 'artist': 'ABC', + 'genre': 'Pop', + }, + 'expected_warnings': ['Failed to download video versions info'], + }, { + # no genres available + 'url': 'http://www.vevo.com/watch/INS171400764', + 'only_matching': True, + }] + _SMIL_BASE_URL = 'http://smil.lvl3.vevo.com' + _SOURCE_TYPES = { + 0: 'youtube', + 1: 'brightcove', + 2: 'http', + 3: 'hls_ios', + 4: 'hls', + 5: 'smil', # http + 7: 'f4m_cc', + 8: 'f4m_ak', + 9: 'f4m_l3', + 10: 'ism', + 13: 'smil', # rtmp + 18: 'dash', + } + _VERSIONS = { + 0: 'youtube', # only in AuthenticateVideo videoVersions + 1: 'level3', + 2: 'akamai', + 3: 'level3', + 4: 'amazon', } - def _real_extract(self, url): - mobj = re.match(self._VALID_URL, url) - video_id = mobj.group('id') - - json_url = 'http://videoplayer.vevo.com/VideoService/AuthenticateVideo?isrc=%s' % video_id - info_json = self._download_webpage(json_url, video_id, u'Downloading json info') - - self.report_extraction(video_id) - video_info = json.loads(info_json)['video'] - last_version = {'version': -1} - for version in video_info['videoVersions']: - # These are the HTTP downloads, other types are for different manifests - if version['sourceType'] == 2: - if version['version'] > last_version['version']: - last_version = version - if last_version['version'] == -1: - raise ExtractorError(u'Unable to extract last version of the video') - - renditions = xml.etree.ElementTree.fromstring(last_version['data']) + def _parse_smil_formats(self, smil, smil_url, video_id, namespace=None, f4m_params=None, transform_rtmp_url=None): formats = [] - # Already sorted from worst to best quality - for rend in renditions.findall('rendition'): - attr = rend.attrib - f_url = attr['url'] + els = smil.findall('.//{http://www.w3.org/2001/SMIL20/Language}video') + for el in els: + src = el.attrib['src'] + m = re.match(r'''(?xi) + (?P[a-z0-9]+): + (?P + [/a-z0-9]+ # The directory and main part of the URL + _(?P[0-9]+)k + _(?P[0-9]+)x(?P[0-9]+) + _(?P[a-z0-9]+) + _(?P[0-9]+) + _(?P[a-z0-9]+) + _(?P[0-9]+) + \.[a-z0-9]+ # File extension + )''', src) + if not m: + continue + + format_url = self._SMIL_BASE_URL + m.group('path') formats.append({ - 'url': f_url, - 'ext': determine_ext(f_url), - 'height': int(attr['frameheight']), - 'width': int(attr['frameWidth']), + 'url': format_url, + 'format_id': 'smil_' + m.group('tbr'), + 'vcodec': m.group('vcodec'), + 'acodec': m.group('acodec'), + 'tbr': int(m.group('tbr')), + 'vbr': int(m.group('vbr')), + 'abr': int(m.group('abr')), + 'ext': m.group('ext'), + 'width': int(m.group('width')), + 'height': int(m.group('height')), }) + return formats + + def _initialize_api(self, video_id): + req = sanitized_Request( + 'http://www.vevo.com/auth', data=b'') + webpage = self._download_webpage( + req, None, + note='Retrieving oauth token', + errnote='Unable to retrieve oauth token') + + if 'THIS PAGE IS CURRENTLY UNAVAILABLE IN YOUR REGION' in webpage: + self.raise_geo_restricted( + '%s said: This page is currently unavailable in your region' % self.IE_NAME) + + auth_info = self._parse_json(webpage, video_id) + self._api_url_template = self.http_scheme() + '//apiv2.vevo.com/%s?token=' + auth_info['access_token'] + + def _call_api(self, path, *args, **kwargs): + return self._download_json(self._api_url_template % path, *args, **kwargs) + + def _real_extract(self, url): + video_id = self._match_id(url) + + json_url = 'http://api.vevo.com/VideoService/AuthenticateVideo?isrc=%s' % video_id + response = self._download_json( + json_url, video_id, 'Downloading video info', + 'Unable to download info', fatal=False) or {} + video_info = response.get('video') or {} + artist = None + featured_artist = None + uploader = None + view_count = None + formats = [] + + if not video_info: + try: + self._initialize_api(video_id) + except ExtractorError: + ytid = response.get('errorInfo', {}).get('ytid') + if ytid: + self.report_warning( + 'Video is geoblocked, trying with the YouTube video %s' % ytid) + return self.url_result(ytid, 'Youtube', ytid) + + raise + + video_info = self._call_api( + 'video/%s' % video_id, video_id, 'Downloading api video info', + 'Failed to download video info') + + video_versions = self._call_api( + 'video/%s/streams' % video_id, video_id, + 'Downloading video versions info', + 'Failed to download video versions info', + fatal=False) - date_epoch = int(self._search_regex( - r'/Date\((\d+)\)/', video_info['launchDate'], u'launch date'))/1000 - upload_date = datetime.datetime.fromtimestamp(date_epoch) - info = { + # Some videos are only available via webpage (e.g. + # https://github.com/rg3/youtube-dl/issues/9366) + if not video_versions: + webpage = self._download_webpage(url, video_id) + video_versions = self._extract_json(webpage, video_id, 'streams')[video_id][0] + + timestamp = parse_iso8601(video_info.get('releaseDate')) + artists = video_info.get('artists') + for curr_artist in artists: + if curr_artist.get('role') == 'Featured': + featured_artist = curr_artist['name'] + else: + artist = uploader = curr_artist['name'] + view_count = int_or_none(video_info.get('views', {}).get('total')) + + for video_version in video_versions: + version = self._VERSIONS.get(video_version['version']) + version_url = video_version.get('url') + if not version_url: + continue + + if '.ism' in version_url: + continue + elif '.mpd' in version_url: + formats.extend(self._extract_mpd_formats( + version_url, video_id, mpd_id='dash-%s' % version, + note='Downloading %s MPD information' % version, + errnote='Failed to download %s MPD information' % version, + fatal=False)) + elif '.m3u8' in version_url: + formats.extend(self._extract_m3u8_formats( + version_url, video_id, 'mp4', 'm3u8_native', + m3u8_id='hls-%s' % version, + note='Downloading %s m3u8 information' % version, + errnote='Failed to download %s m3u8 information' % version, + fatal=False)) + else: + m = re.search(r'''(?xi) + _(?P[0-9]+)x(?P[0-9]+) + _(?P[a-z0-9]+) + _(?P[0-9]+) + _(?P[a-z0-9]+) + _(?P[0-9]+) + \.(?P[a-z0-9]+)''', version_url) + if not m: + continue + + formats.append({ + 'url': version_url, + 'format_id': 'http-%s-%s' % (version, video_version['quality']), + 'vcodec': m.group('vcodec'), + 'acodec': m.group('acodec'), + 'vbr': int(m.group('vbr')), + 'abr': int(m.group('abr')), + 'ext': m.group('ext'), + 'width': int(m.group('width')), + 'height': int(m.group('height')), + }) + else: + timestamp = int_or_none(self._search_regex( + r'/Date\((\d+)\)/', + video_info['releaseDate'], 'release date', fatal=False), + scale=1000) + artists = video_info.get('mainArtists') + if artists: + artist = uploader = artists[0]['artistName'] + + featured_artists = video_info.get('featuredArtists') + if featured_artists: + featured_artist = featured_artists[0]['artistName'] + + smil_parsed = False + for video_version in video_info['videoVersions']: + version = self._VERSIONS.get(video_version['version']) + if version == 'youtube': + continue + else: + source_type = self._SOURCE_TYPES.get(video_version['sourceType']) + renditions = compat_etree_fromstring(video_version['data']) + if source_type == 'http': + for rend in renditions.findall('rendition'): + attr = rend.attrib + formats.append({ + 'url': attr['url'], + 'format_id': 'http-%s-%s' % (version, attr['name']), + 'height': int_or_none(attr.get('frameheight')), + 'width': int_or_none(attr.get('frameWidth')), + 'tbr': int_or_none(attr.get('totalBitrate')), + 'vbr': int_or_none(attr.get('videoBitrate')), + 'abr': int_or_none(attr.get('audioBitrate')), + 'vcodec': attr.get('videoCodec'), + 'acodec': attr.get('audioCodec'), + }) + elif source_type == 'hls': + formats.extend(self._extract_m3u8_formats( + renditions.find('rendition').attrib['url'], video_id, + 'mp4', 'm3u8_native', m3u8_id='hls-%s' % version, + note='Downloading %s m3u8 information' % version, + errnote='Failed to download %s m3u8 information' % version, + fatal=False)) + elif source_type == 'smil' and version == 'level3' and not smil_parsed: + formats.extend(self._extract_smil_formats( + renditions.find('rendition').attrib['url'], video_id, False)) + smil_parsed = True + self._sort_formats(formats) + + track = video_info['title'] + if featured_artist: + artist = '%s ft. %s' % (artist, featured_artist) + title = '%s - %s' % (artist, track) if artist else track + + genres = video_info.get('genres') + genre = ( + genres[0] if genres and isinstance(genres, list) and + isinstance(genres[0], compat_str) else None) + + is_explicit = video_info.get('isExplicit') + if is_explicit is True: + age_limit = 18 + elif is_explicit is False: + age_limit = 0 + else: + age_limit = None + + duration = video_info.get('duration') + + return { 'id': video_id, - 'title': video_info['title'], + 'title': title, 'formats': formats, - 'thumbnail': video_info['imageUrl'], - 'upload_date': upload_date.strftime('%Y%m%d'), - 'uploader': video_info['mainArtists'][0]['artistName'], - 'duration': video_info['duration'], + 'thumbnail': video_info.get('imageUrl') or video_info.get('thumbnailUrl'), + 'timestamp': timestamp, + 'uploader': uploader, + 'duration': duration, + 'view_count': view_count, + 'age_limit': age_limit, + 'track': track, + 'artist': uploader, + 'genre': genre, } - # TODO: Remove when #980 has been merged - info.update(formats[-1]) - return info +class VevoPlaylistIE(VevoBaseIE): + _VALID_URL = r'https?://(?:www\.)?vevo\.com/watch/(?Pplaylist|genre)/(?P[^/?#&]+)' + + _TESTS = [{ + 'url': 'http://www.vevo.com/watch/playlist/dadbf4e7-b99f-4184-9670-6f0e547b6a29', + 'info_dict': { + 'id': 'dadbf4e7-b99f-4184-9670-6f0e547b6a29', + 'title': 'Best-Of: Birdman', + }, + 'playlist_count': 10, + }, { + 'url': 'http://www.vevo.com/watch/genre/rock', + 'info_dict': { + 'id': 'rock', + 'title': 'Rock', + }, + 'playlist_count': 20, + }, { + 'url': 'http://www.vevo.com/watch/playlist/dadbf4e7-b99f-4184-9670-6f0e547b6a29?index=0', + 'md5': '32dcdfddddf9ec6917fc88ca26d36282', + 'info_dict': { + 'id': 'USCMV1100073', + 'ext': 'mp4', + 'title': 'Birdman - Y.U. MAD', + 'timestamp': 1323417600, + 'upload_date': '20111209', + 'uploader': 'Birdman', + 'track': 'Y.U. MAD', + 'artist': 'Birdman', + 'genre': 'Rap/Hip-Hop', + }, + 'expected_warnings': ['Unable to download SMIL file'], + }, { + 'url': 'http://www.vevo.com/watch/genre/rock?index=0', + 'only_matching': True, + }] + + def _real_extract(self, url): + mobj = re.match(self._VALID_URL, url) + playlist_id = mobj.group('id') + playlist_kind = mobj.group('kind') + + webpage = self._download_webpage(url, playlist_id) + + qs = compat_urlparse.parse_qs(compat_urlparse.urlparse(url).query) + index = qs.get('index', [None])[0] + + if index: + video_id = self._search_regex( + r']+content=(["\'])vevo://video/(?P.+?)\1[^>]*>', + webpage, 'video id', default=None, group='id') + if video_id: + return self.url_result('vevo:%s' % video_id, VevoIE.ie_key()) + + playlists = self._extract_json(webpage, playlist_id, '%ss' % playlist_kind) + + playlist = (list(playlists.values())[0] + if playlist_kind == 'playlist' else playlists[playlist_id]) + + entries = [ + self.url_result('vevo:%s' % src, VevoIE.ie_key()) + for src in playlist['isrcs']] + + return self.playlist_result( + entries, playlist.get('playlistId') or playlist_id, + playlist.get('name'), playlist.get('description'))