summaryrefslogtreecommitdiffstats
path: root/yt_dlp/extractor/mtv.py
diff options
context:
space:
mode:
Diffstat (limited to '')
-rw-r--r--yt_dlp/extractor/mtv.py653
1 files changed, 653 insertions, 0 deletions
diff --git a/yt_dlp/extractor/mtv.py b/yt_dlp/extractor/mtv.py
new file mode 100644
index 0000000..d91be62
--- /dev/null
+++ b/yt_dlp/extractor/mtv.py
@@ -0,0 +1,653 @@
+import re
+
+from .common import InfoExtractor
+from ..compat import compat_str
+from ..utils import (
+ ExtractorError,
+ find_xpath_attr,
+ fix_xml_ampersands,
+ float_or_none,
+ HEADRequest,
+ int_or_none,
+ join_nonempty,
+ RegexNotFoundError,
+ sanitized_Request,
+ strip_or_none,
+ timeconvert,
+ try_get,
+ unescapeHTML,
+ update_url_query,
+ url_basename,
+ xpath_text,
+)
+
+
+def _media_xml_tag(tag):
+ return '{http://search.yahoo.com/mrss/}%s' % tag
+
+
+class MTVServicesInfoExtractor(InfoExtractor):
+ _MOBILE_TEMPLATE = None
+ _LANG = None
+
+ @staticmethod
+ def _id_from_uri(uri):
+ return uri.split(':')[-1]
+
+ @staticmethod
+ def _remove_template_parameter(url):
+ # Remove the templates, like &device={device}
+ return re.sub(r'&[^=]*?={.*?}(?=(&|$))', '', url)
+
+ def _get_feed_url(self, uri, url=None):
+ return self._FEED_URL
+
+ def _get_thumbnail_url(self, uri, itemdoc):
+ search_path = '%s/%s' % (_media_xml_tag('group'), _media_xml_tag('thumbnail'))
+ thumb_node = itemdoc.find(search_path)
+ if thumb_node is None:
+ return None
+ return thumb_node.get('url') or thumb_node.text or None
+
+ def _extract_mobile_video_formats(self, mtvn_id):
+ webpage_url = self._MOBILE_TEMPLATE % mtvn_id
+ req = sanitized_Request(webpage_url)
+ # Otherwise we get a webpage that would execute some javascript
+ req.add_header('User-Agent', 'curl/7')
+ webpage = self._download_webpage(req, mtvn_id,
+ 'Downloading mobile page')
+ metrics_url = unescapeHTML(self._search_regex(r'<a href="(http://metrics.+?)"', webpage, 'url'))
+ req = HEADRequest(metrics_url)
+ response = self._request_webpage(req, mtvn_id, 'Resolving url')
+ url = response.geturl()
+ # Transform the url to get the best quality:
+ url = re.sub(r'.+pxE=mp4', 'http://mtvnmobile.vo.llnwd.net/kip0/_pxn=0+_pxK=18639+_pxE=mp4', url, 1)
+ return [{'url': url, 'ext': 'mp4'}]
+
+ def _extract_video_formats(self, mdoc, mtvn_id, video_id):
+ if re.match(r'.*/(error_country_block\.swf|geoblock\.mp4|copyright_error\.flv(?:\?geo\b.+?)?)$', mdoc.find('.//src').text) is not None:
+ if mtvn_id is not None and self._MOBILE_TEMPLATE is not None:
+ self.to_screen('The normal version is not available from your '
+ 'country, trying with the mobile version')
+ return self._extract_mobile_video_formats(mtvn_id)
+ raise ExtractorError('This video is not available from your country.',
+ expected=True)
+
+ formats = []
+ for rendition in mdoc.findall('.//rendition'):
+ if rendition.get('method') == 'hls':
+ hls_url = rendition.find('./src').text
+ formats.extend(self._extract_m3u8_formats(
+ hls_url, video_id, ext='mp4', entry_protocol='m3u8_native',
+ m3u8_id='hls', fatal=False))
+ else:
+ # fms
+ try:
+ _, _, ext = rendition.attrib['type'].partition('/')
+ rtmp_video_url = rendition.find('./src').text
+ if 'error_not_available.swf' in rtmp_video_url:
+ raise ExtractorError(
+ '%s said: video is not available' % self.IE_NAME,
+ expected=True)
+ if rtmp_video_url.endswith('siteunavail.png'):
+ continue
+ formats.extend([{
+ 'ext': 'flv' if rtmp_video_url.startswith('rtmp') else ext,
+ 'url': rtmp_video_url,
+ 'format_id': join_nonempty(
+ 'rtmp' if rtmp_video_url.startswith('rtmp') else None,
+ rendition.get('bitrate')),
+ 'width': int(rendition.get('width')),
+ 'height': int(rendition.get('height')),
+ }])
+ except (KeyError, TypeError):
+ raise ExtractorError('Invalid rendition field.')
+ return formats
+
+ def _extract_subtitles(self, mdoc, mtvn_id):
+ subtitles = {}
+ for transcript in mdoc.findall('.//transcript'):
+ if transcript.get('kind') != 'captions':
+ continue
+ lang = transcript.get('srclang')
+ for typographic in transcript.findall('./typographic'):
+ sub_src = typographic.get('src')
+ if not sub_src:
+ continue
+ ext = typographic.get('format')
+ if ext == 'cea-608':
+ ext = 'scc'
+ subtitles.setdefault(lang, []).append({
+ 'url': compat_str(sub_src),
+ 'ext': ext
+ })
+ return subtitles
+
+ def _get_video_info(self, itemdoc, use_hls=True):
+ uri = itemdoc.find('guid').text
+ video_id = self._id_from_uri(uri)
+ self.report_extraction(video_id)
+ content_el = itemdoc.find('%s/%s' % (_media_xml_tag('group'), _media_xml_tag('content')))
+ mediagen_url = self._remove_template_parameter(content_el.attrib['url'])
+ mediagen_url = mediagen_url.replace('device={device}', '')
+ if 'acceptMethods' not in mediagen_url:
+ mediagen_url += '&' if '?' in mediagen_url else '?'
+ mediagen_url += 'acceptMethods='
+ mediagen_url += 'hls' if use_hls else 'fms'
+
+ mediagen_doc = self._download_xml(
+ mediagen_url, video_id, 'Downloading video urls', fatal=False)
+
+ if mediagen_doc is False:
+ return None
+
+ item = mediagen_doc.find('./video/item')
+ if item is not None and item.get('type') == 'text':
+ message = '%s returned error: ' % self.IE_NAME
+ if item.get('code') is not None:
+ message += '%s - ' % item.get('code')
+ message += item.text
+ raise ExtractorError(message, expected=True)
+
+ description = strip_or_none(xpath_text(itemdoc, 'description'))
+
+ timestamp = timeconvert(xpath_text(itemdoc, 'pubDate'))
+
+ title_el = None
+ if title_el is None:
+ title_el = find_xpath_attr(
+ itemdoc, './/{http://search.yahoo.com/mrss/}category',
+ 'scheme', 'urn:mtvn:video_title')
+ if title_el is None:
+ title_el = itemdoc.find('.//{http://search.yahoo.com/mrss/}title')
+ if title_el is None:
+ title_el = itemdoc.find('.//title')
+ if title_el.text is None:
+ title_el = None
+
+ title = title_el.text
+ if title is None:
+ raise ExtractorError('Could not find video title')
+ title = title.strip()
+
+ series = find_xpath_attr(
+ itemdoc, './/{http://search.yahoo.com/mrss/}category',
+ 'scheme', 'urn:mtvn:franchise')
+ season = find_xpath_attr(
+ itemdoc, './/{http://search.yahoo.com/mrss/}category',
+ 'scheme', 'urn:mtvn:seasonN')
+ episode = find_xpath_attr(
+ itemdoc, './/{http://search.yahoo.com/mrss/}category',
+ 'scheme', 'urn:mtvn:episodeN')
+ series = series.text if series is not None else None
+ season = season.text if season is not None else None
+ episode = episode.text if episode is not None else None
+ if season and episode:
+ # episode number includes season, so remove it
+ episode = re.sub(r'^%s' % season, '', episode)
+
+ # This a short id that's used in the webpage urls
+ mtvn_id = None
+ mtvn_id_node = find_xpath_attr(itemdoc, './/{http://search.yahoo.com/mrss/}category',
+ 'scheme', 'urn:mtvn:id')
+ if mtvn_id_node is not None:
+ mtvn_id = mtvn_id_node.text
+
+ formats = self._extract_video_formats(mediagen_doc, mtvn_id, video_id)
+
+ # Some parts of complete video may be missing (e.g. missing Act 3 in
+ # http://www.southpark.de/alle-episoden/s14e01-sexual-healing)
+ if not formats:
+ return None
+
+ return {
+ 'title': title,
+ 'formats': formats,
+ 'subtitles': self._extract_subtitles(mediagen_doc, mtvn_id),
+ 'id': video_id,
+ 'thumbnail': self._get_thumbnail_url(uri, itemdoc),
+ 'description': description,
+ 'duration': float_or_none(content_el.attrib.get('duration')),
+ 'timestamp': timestamp,
+ 'series': series,
+ 'season_number': int_or_none(season),
+ 'episode_number': int_or_none(episode),
+ }
+
+ def _get_feed_query(self, uri):
+ data = {'uri': uri}
+ if self._LANG:
+ data['lang'] = self._LANG
+ return data
+
+ def _get_videos_info(self, uri, use_hls=True, url=None):
+ video_id = self._id_from_uri(uri)
+ feed_url = self._get_feed_url(uri, url)
+ info_url = update_url_query(feed_url, self._get_feed_query(uri))
+ return self._get_videos_info_from_url(info_url, video_id, use_hls)
+
+ def _get_videos_info_from_url(self, url, video_id, use_hls=True):
+ idoc = self._download_xml(
+ url, video_id,
+ 'Downloading info', transform_source=fix_xml_ampersands)
+
+ title = xpath_text(idoc, './channel/title')
+ description = xpath_text(idoc, './channel/description')
+
+ entries = []
+ for item in idoc.findall('.//item'):
+ info = self._get_video_info(item, use_hls)
+ if info:
+ entries.append(info)
+
+ # TODO: should be multi-video
+ return self.playlist_result(
+ entries, playlist_title=title, playlist_description=description)
+
+ def _extract_triforce_mgid(self, webpage, data_zone=None, video_id=None):
+ triforce_feed = self._parse_json(self._search_regex(
+ r'triforceManifestFeed\s*=\s*({.+?})\s*;\s*\n', webpage,
+ 'triforce feed', default='{}'), video_id, fatal=False)
+
+ data_zone = self._search_regex(
+ r'data-zone=(["\'])(?P<zone>.+?_lc_promo.*?)\1', webpage,
+ 'data zone', default=data_zone, group='zone')
+
+ feed_url = try_get(
+ triforce_feed, lambda x: x['manifest']['zones'][data_zone]['feed'],
+ compat_str)
+ if not feed_url:
+ return
+
+ feed = self._download_json(feed_url, video_id, fatal=False)
+ if not feed:
+ return
+
+ return try_get(feed, lambda x: x['result']['data']['id'], compat_str)
+
+ @staticmethod
+ def _extract_child_with_type(parent, t):
+ for c in parent['children']:
+ if c.get('type') == t:
+ return c
+
+ def _extract_mgid(self, webpage):
+ try:
+ # the url can be http://media.mtvnservices.com/fb/{mgid}.swf
+ # or http://media.mtvnservices.com/{mgid}
+ og_url = self._og_search_video_url(webpage)
+ mgid = url_basename(og_url)
+ if mgid.endswith('.swf'):
+ mgid = mgid[:-4]
+ except RegexNotFoundError:
+ mgid = None
+
+ if mgid is None or ':' not in mgid:
+ mgid = self._search_regex(
+ [r'data-mgid="(.*?)"', r'swfobject\.embedSWF\(".*?(mgid:.*?)"'],
+ webpage, 'mgid', default=None)
+
+ if not mgid:
+ sm4_embed = self._html_search_meta(
+ 'sm4:video:embed', webpage, 'sm4 embed', default='')
+ mgid = self._search_regex(
+ r'embed/(mgid:.+?)["\'&?/]', sm4_embed, 'mgid', default=None)
+
+ if not mgid:
+ mgid = self._extract_triforce_mgid(webpage)
+
+ if not mgid:
+ data = self._parse_json(self._search_regex(
+ r'__DATA__\s*=\s*({.+?});', webpage, 'data'), None)
+ main_container = self._extract_child_with_type(data, 'MainContainer')
+ ab_testing = self._extract_child_with_type(main_container, 'ABTesting')
+ video_player = self._extract_child_with_type(ab_testing or main_container, 'VideoPlayer')
+ if video_player:
+ mgid = try_get(video_player, lambda x: x['props']['media']['video']['config']['uri'])
+ else:
+ flex_wrapper = self._extract_child_with_type(ab_testing or main_container, 'FlexWrapper')
+ auth_suite_wrapper = self._extract_child_with_type(flex_wrapper, 'AuthSuiteWrapper')
+ player = self._extract_child_with_type(auth_suite_wrapper or flex_wrapper, 'Player')
+ if player:
+ mgid = try_get(player, lambda x: x['props']['videoDetail']['mgid'])
+
+ if not mgid:
+ raise ExtractorError('Could not extract mgid')
+
+ return mgid
+
+ def _real_extract(self, url):
+ title = url_basename(url)
+ webpage = self._download_webpage(url, title)
+ mgid = self._extract_mgid(webpage)
+ videos_info = self._get_videos_info(mgid, url=url)
+ return videos_info
+
+
+class MTVServicesEmbeddedIE(MTVServicesInfoExtractor):
+ IE_NAME = 'mtvservices:embedded'
+ _VALID_URL = r'https?://media\.mtvnservices\.com/embed/(?P<mgid>.+?)(\?|/|$)'
+ _EMBED_REGEX = [r'<iframe[^>]+?src=(["\'])(?P<url>(?:https?:)?//media\.mtvnservices\.com/embed/.+?)\1']
+
+ _TEST = {
+ # From http://www.thewrap.com/peter-dinklage-sums-up-game-of-thrones-in-45-seconds-video/
+ 'url': 'http://media.mtvnservices.com/embed/mgid:uma:video:mtv.com:1043906/cp~vid%3D1043906%26uri%3Dmgid%3Auma%3Avideo%3Amtv.com%3A1043906',
+ 'md5': 'cb349b21a7897164cede95bd7bf3fbb9',
+ 'info_dict': {
+ 'id': '1043906',
+ 'ext': 'mp4',
+ 'title': 'Peter Dinklage Sums Up \'Game Of Thrones\' In 45 Seconds',
+ 'description': '"Sexy sexy sexy, stabby stabby stabby, beautiful language," says Peter Dinklage as he tries summarizing "Game of Thrones" in under a minute.',
+ 'timestamp': 1400126400,
+ 'upload_date': '20140515',
+ },
+ }
+
+ def _get_feed_url(self, uri, url=None):
+ video_id = self._id_from_uri(uri)
+ config = self._download_json(
+ 'http://media.mtvnservices.com/pmt/e1/access/index.html?uri=%s&configtype=edge' % uri, video_id)
+ return self._remove_template_parameter(config['feedWithQueryParams'])
+
+ def _real_extract(self, url):
+ mobj = self._match_valid_url(url)
+ mgid = mobj.group('mgid')
+ return self._get_videos_info(mgid)
+
+
+class MTVIE(MTVServicesInfoExtractor):
+ IE_NAME = 'mtv'
+ _VALID_URL = r'https?://(?:www\.)?mtv\.com/(?:video-clips|(?:full-)?episodes)/(?P<id>[^/?#.]+)'
+ _FEED_URL = 'http://www.mtv.com/feeds/mrss/'
+
+ _TESTS = [{
+ 'url': 'http://www.mtv.com/video-clips/vl8qof/unlocking-the-truth-trailer',
+ 'md5': '1edbcdf1e7628e414a8c5dcebca3d32b',
+ 'info_dict': {
+ 'id': '5e14040d-18a4-47c4-a582-43ff602de88e',
+ 'ext': 'mp4',
+ 'title': 'Unlocking The Truth|July 18, 2016|1|101|Trailer',
+ 'description': '"Unlocking the Truth" premieres August 17th at 11/10c.',
+ 'timestamp': 1468846800,
+ 'upload_date': '20160718',
+ },
+ }, {
+ 'url': 'http://www.mtv.com/full-episodes/94tujl/unlocking-the-truth-gates-of-hell-season-1-ep-101',
+ 'only_matching': True,
+ }, {
+ 'url': 'http://www.mtv.com/episodes/g8xu7q/teen-mom-2-breaking-the-wall-season-7-ep-713',
+ 'only_matching': True,
+ }]
+
+
+class MTVJapanIE(MTVServicesInfoExtractor):
+ IE_NAME = 'mtvjapan'
+ _VALID_URL = r'https?://(?:www\.)?mtvjapan\.com/videos/(?P<id>[0-9a-z]+)'
+
+ _TEST = {
+ 'url': 'http://www.mtvjapan.com/videos/prayht/fresh-info-cadillac-escalade',
+ 'info_dict': {
+ 'id': 'bc01da03-6fe5-4284-8880-f291f4e368f5',
+ 'ext': 'mp4',
+ 'title': '【Fresh Info】Cadillac ESCALADE Sport Edition',
+ },
+ 'params': {
+ 'skip_download': True,
+ },
+ }
+ _GEO_COUNTRIES = ['JP']
+ _FEED_URL = 'http://feeds.mtvnservices.com/od/feed/intl-mrss-player-feed'
+
+ def _get_feed_query(self, uri):
+ return {
+ 'arcEp': 'mtvjapan.com',
+ 'mgid': uri,
+ }
+
+
+class MTVVideoIE(MTVServicesInfoExtractor):
+ IE_NAME = 'mtv:video'
+ _VALID_URL = r'''(?x)^https?://
+ (?:(?:www\.)?mtv\.com/videos/.+?/(?P<videoid>[0-9]+)/[^/]+$|
+ m\.mtv\.com/videos/video\.rbml\?.*?id=(?P<mgid>[^&]+))'''
+
+ _FEED_URL = 'http://www.mtv.com/player/embed/AS3/rss/'
+
+ _TESTS = [
+ {
+ 'url': 'http://www.mtv.com/videos/misc/853555/ours-vh1-storytellers.jhtml',
+ 'md5': '850f3f143316b1e71fa56a4edfd6e0f8',
+ 'info_dict': {
+ 'id': '853555',
+ 'ext': 'mp4',
+ 'title': 'Taylor Swift - "Ours (VH1 Storytellers)"',
+ 'description': 'Album: Taylor Swift performs "Ours" for VH1 Storytellers at Harvey Mudd College.',
+ 'timestamp': 1352610000,
+ 'upload_date': '20121111',
+ },
+ },
+ ]
+
+ def _get_thumbnail_url(self, uri, itemdoc):
+ return 'http://mtv.mtvnimages.com/uri/' + uri
+
+ def _real_extract(self, url):
+ mobj = self._match_valid_url(url)
+ video_id = mobj.group('videoid')
+ uri = mobj.groupdict().get('mgid')
+ if uri is None:
+ webpage = self._download_webpage(url, video_id)
+
+ # Some videos come from Vevo.com
+ m_vevo = re.search(
+ r'(?s)isVevoVideo = true;.*?vevoVideoId = "(.*?)";', webpage)
+ if m_vevo:
+ vevo_id = m_vevo.group(1)
+ self.to_screen('Vevo video detected: %s' % vevo_id)
+ return self.url_result('vevo:%s' % vevo_id, ie='Vevo')
+
+ uri = self._html_search_regex(r'/uri/(.*?)\?', webpage, 'uri')
+ return self._get_videos_info(uri)
+
+
+class MTVDEIE(MTVServicesInfoExtractor):
+ IE_NAME = 'mtv.de'
+ _VALID_URL = r'https?://(?:www\.)?mtv\.de/(?:musik/videoclips|folgen|news)/(?P<id>[0-9a-z]+)'
+ _TESTS = [{
+ 'url': 'http://www.mtv.de/musik/videoclips/2gpnv7/Traum',
+ 'info_dict': {
+ 'id': 'd5d472bc-f5b7-11e5-bffd-a4badb20dab5',
+ 'ext': 'mp4',
+ 'title': 'Traum',
+ 'description': 'Traum',
+ },
+ 'params': {
+ # rtmp download
+ 'skip_download': True,
+ },
+ 'skip': 'Blocked at Travis CI',
+ }, {
+ # mediagen URL without query (e.g. http://videos.mtvnn.com/mediagen/e865da714c166d18d6f80893195fcb97)
+ 'url': 'http://www.mtv.de/folgen/6b1ylu/teen-mom-2-enthuellungen-S5-F1',
+ 'info_dict': {
+ 'id': '1e5a878b-31c5-11e7-a442-0e40cf2fc285',
+ 'ext': 'mp4',
+ 'title': 'Teen Mom 2',
+ 'description': 'md5:dc65e357ef7e1085ed53e9e9d83146a7',
+ },
+ 'params': {
+ # rtmp download
+ 'skip_download': True,
+ },
+ 'skip': 'Blocked at Travis CI',
+ }, {
+ 'url': 'http://www.mtv.de/news/glolix/77491-mtv-movies-spotlight--pixels--teil-3',
+ 'info_dict': {
+ 'id': 'local_playlist-4e760566473c4c8c5344',
+ 'ext': 'mp4',
+ 'title': 'Article_mtv-movies-spotlight-pixels-teil-3_short-clips_part1',
+ 'description': 'MTV Movies Supercut',
+ },
+ 'params': {
+ # rtmp download
+ 'skip_download': True,
+ },
+ 'skip': 'Das Video kann zur Zeit nicht abgespielt werden.',
+ }]
+ _GEO_COUNTRIES = ['DE']
+ _FEED_URL = 'http://feeds.mtvnservices.com/od/feed/intl-mrss-player-feed'
+
+ def _get_feed_query(self, uri):
+ return {
+ 'arcEp': 'mtv.de',
+ 'mgid': uri,
+ }
+
+
+class MTVItaliaIE(MTVServicesInfoExtractor):
+ IE_NAME = 'mtv.it'
+ _VALID_URL = r'https?://(?:www\.)?mtv\.it/(?:episodi|video|musica)/(?P<id>[0-9a-z]+)'
+ _TESTS = [{
+ 'url': 'http://www.mtv.it/episodi/24bqab/mario-una-serie-di-maccio-capatonda-cavoli-amario-episodio-completo-S1-E1',
+ 'info_dict': {
+ 'id': '0f0fc78e-45fc-4cce-8f24-971c25477530',
+ 'ext': 'mp4',
+ 'title': 'Cavoli amario (episodio completo)',
+ 'description': 'md5:4962bccea8fed5b7c03b295ae1340660',
+ 'series': 'Mario - Una Serie Di Maccio Capatonda',
+ 'season_number': 1,
+ 'episode_number': 1,
+ },
+ 'params': {
+ 'skip_download': True,
+ },
+ }]
+ _GEO_COUNTRIES = ['IT']
+ _FEED_URL = 'http://feeds.mtvnservices.com/od/feed/intl-mrss-player-feed'
+
+ def _get_feed_query(self, uri):
+ return {
+ 'arcEp': 'mtv.it',
+ 'mgid': uri,
+ }
+
+
+class MTVItaliaProgrammaIE(MTVItaliaIE): # XXX: Do not subclass from concrete IE
+ IE_NAME = 'mtv.it:programma'
+ _VALID_URL = r'https?://(?:www\.)?mtv\.it/(?:programmi|playlist)/(?P<id>[0-9a-z]+)'
+ _TESTS = [{
+ # program page: general
+ 'url': 'http://www.mtv.it/programmi/s2rppv/mario-una-serie-di-maccio-capatonda',
+ 'info_dict': {
+ 'id': 'a6f155bc-8220-4640-aa43-9b95f64ffa3d',
+ 'title': 'Mario - Una Serie Di Maccio Capatonda',
+ 'description': 'md5:72fbffe1f77ccf4e90757dd4e3216153',
+ },
+ 'playlist_count': 2,
+ 'params': {
+ 'skip_download': True,
+ },
+ }, {
+ # program page: specific season
+ 'url': 'http://www.mtv.it/programmi/d9ncjf/mario-una-serie-di-maccio-capatonda-S2',
+ 'info_dict': {
+ 'id': '4deeb5d8-f272-490c-bde2-ff8d261c6dd1',
+ 'title': 'Mario - Una Serie Di Maccio Capatonda - Stagione 2',
+ },
+ 'playlist_count': 34,
+ 'params': {
+ 'skip_download': True,
+ },
+ }, {
+ # playlist page + redirect
+ 'url': 'http://www.mtv.it/playlist/sexy-videos/ilctal',
+ 'info_dict': {
+ 'id': 'dee8f9ee-756d-493b-bf37-16d1d2783359',
+ 'title': 'Sexy Videos',
+ },
+ 'playlist_mincount': 145,
+ 'params': {
+ 'skip_download': True,
+ },
+ }]
+ _GEO_COUNTRIES = ['IT']
+ _FEED_URL = 'http://www.mtv.it/feeds/triforce/manifest/v8'
+
+ def _get_entries(self, title, url):
+ while True:
+ pg = self._search_regex(r'/(\d+)$', url, 'entries', '1')
+ entries = self._download_json(url, title, 'page %s' % pg)
+ url = try_get(
+ entries, lambda x: x['result']['nextPageURL'], compat_str)
+ entries = try_get(
+ entries, (
+ lambda x: x['result']['data']['items'],
+ lambda x: x['result']['data']['seasons']),
+ list)
+ for entry in entries or []:
+ if entry.get('canonicalURL'):
+ yield self.url_result(entry['canonicalURL'])
+ if not url:
+ break
+
+ def _real_extract(self, url):
+ query = {'url': url}
+ info_url = update_url_query(self._FEED_URL, query)
+ video_id = self._match_id(url)
+ info = self._download_json(info_url, video_id).get('manifest')
+
+ redirect = try_get(
+ info, lambda x: x['newLocation']['url'], compat_str)
+ if redirect:
+ return self.url_result(redirect)
+
+ title = info.get('title')
+ video_id = try_get(
+ info, lambda x: x['reporting']['itemId'], compat_str)
+ parent_id = try_get(
+ info, lambda x: x['reporting']['parentId'], compat_str)
+
+ playlist_url = current_url = None
+ for z in (info.get('zones') or {}).values():
+ if z.get('moduleName') in ('INTL_M304', 'INTL_M209'):
+ info_url = z.get('feed')
+ if z.get('moduleName') in ('INTL_M308', 'INTL_M317'):
+ playlist_url = playlist_url or z.get('feed')
+ if z.get('moduleName') in ('INTL_M300',):
+ current_url = current_url or z.get('feed')
+
+ if not info_url:
+ raise ExtractorError('No info found')
+
+ if video_id == parent_id:
+ video_id = self._search_regex(
+ r'([^\/]+)/[^\/]+$', info_url, 'video_id')
+
+ info = self._download_json(info_url, video_id, 'Show infos')
+ info = try_get(info, lambda x: x['result']['data'], dict)
+ title = title or try_get(
+ info, (
+ lambda x: x['title'],
+ lambda x: x['headline']),
+ compat_str)
+ description = try_get(info, lambda x: x['content'], compat_str)
+
+ if current_url:
+ season = try_get(
+ self._download_json(playlist_url, video_id, 'Seasons info'),
+ lambda x: x['result']['data'], dict)
+ current = try_get(
+ season, lambda x: x['currentSeason'], compat_str)
+ seasons = try_get(
+ season, lambda x: x['seasons'], list) or []
+
+ if current in [s.get('eTitle') for s in seasons]:
+ playlist_url = current_url
+
+ title = re.sub(
+ r'[-|]\s*(?:mtv\s*italia|programma|playlist)',
+ '', title, flags=re.IGNORECASE).strip()
+
+ return self.playlist_result(
+ self._get_entries(title, playlist_url),
+ video_id, title, description)