X-Git-Url: https://git.rapsys.eu/youtubedl/blobdiff_plain/33cd347759d6d999325ebf3c69b7ed5692c343b2..89bb9e23b34f43c6563524917f7dffb9841a972e:/youtube_dl/extractor/teamcoco.py?ds=sidebyside diff --git a/youtube_dl/extractor/teamcoco.py b/youtube_dl/extractor/teamcoco.py index c910110..73469cc 100644 --- a/youtube_dl/extractor/teamcoco.py +++ b/youtube_dl/extractor/teamcoco.py @@ -1,46 +1,189 @@ -import re +# coding: utf-8 +from __future__ import unicode_literals -from .common import InfoExtractor +import json + +from .turner import TurnerBaseIE from ..utils import ( + determine_ext, ExtractorError, + int_or_none, + mimetype2ext, + parse_duration, + parse_iso8601, + qualities, ) -class TeamcocoIE(InfoExtractor): - _VALID_URL = r'http://teamcoco\.com/video/(?P.*)' - _TEST = { - u'url': u'http://teamcoco.com/video/louis-ck-interview-george-w-bush', - u'file': u'19705.mp4', - u'md5': u'27b6f7527da5acf534b15f21b032656e', - u'info_dict': { - u"description": u"Louis C.K. got starstruck by George W. Bush, so what? Part one.", - u"title": u"Louis C.K. Interview Pt. 1 11/3/11" +class TeamcocoIE(TurnerBaseIE): + _VALID_URL = r'https?://teamcoco\.com/(?P([^/]+/)*[^/?#]+)' + _TESTS = [ + { + 'url': 'http://teamcoco.com/video/mary-kay-remote', + 'md5': '55d532f81992f5c92046ad02fec34d7d', + 'info_dict': { + 'id': '80187', + 'ext': 'mp4', + 'title': 'Conan Becomes A Mary Kay Beauty Consultant', + 'description': 'Mary Kay is perhaps the most trusted name in female beauty, so of course Conan is a natural choice to sell their products.', + 'duration': 495.0, + 'upload_date': '20140402', + 'timestamp': 1396407600, + } + }, { + 'url': 'http://teamcoco.com/video/louis-ck-interview-george-w-bush', + 'md5': 'cde9ba0fa3506f5f017ce11ead928f9a', + 'info_dict': { + 'id': '19705', + 'ext': 'mp4', + 'description': 'Louis C.K. got starstruck by George W. Bush, so what? Part one.', + 'title': 'Louis C.K. Interview Pt. 1 11/3/11', + 'duration': 288, + 'upload_date': '20111104', + 'timestamp': 1320405840, + } + }, { + 'url': 'http://teamcoco.com/video/timothy-olyphant-drinking-whiskey', + 'info_dict': { + 'id': '88748', + 'ext': 'mp4', + 'title': 'Timothy Olyphant Raises A Toast To “Justified”', + 'description': 'md5:15501f23f020e793aeca761205e42c24', + 'upload_date': '20150415', + 'timestamp': 1429088400, + }, + 'params': { + 'skip_download': True, # m3u8 downloads + } + }, { + 'url': 'http://teamcoco.com/video/full-episode-mon-6-1-joel-mchale-jake-tapper-and-musical-guest-courtney-barnett?playlist=x;eyJ0eXBlIjoidGFnIiwiaWQiOjl9', + 'info_dict': { + 'id': '89341', + 'ext': 'mp4', + 'title': 'Full Episode - Mon. 6/1 - Joel McHale, Jake Tapper, And Musical Guest Courtney Barnett', + 'description': 'Guests: Joel McHale, Jake Tapper, And Musical Guest Courtney Barnett', + }, + 'params': { + 'skip_download': True, # m3u8 downloads + }, + 'skip': 'This video is no longer available.', + }, { + 'url': 'http://teamcoco.com/video/the-conan-audiencey-awards-for-04/25/18', + 'only_matching': True, + }, { + 'url': 'http://teamcoco.com/italy/conan-jordan-schlansky-hit-the-streets-of-florence', + 'only_matching': True, + }, { + 'url': 'http://teamcoco.com/haiti/conan-s-haitian-history-lesson', + 'only_matching': True, + }, { + 'url': 'http://teamcoco.com/israel/conan-hits-the-streets-beaches-of-tel-aviv', + 'only_matching': True, } - } + ] + + def _graphql_call(self, query_template, object_type, object_id): + find_object = 'find' + object_type + return self._download_json( + 'http://teamcoco.com/graphql/', object_id, data=json.dumps({ + 'query': query_template % (find_object, object_id) + }))['data'][find_object] def _real_extract(self, url): - mobj = re.match(self._VALID_URL, url) - if mobj is None: - raise ExtractorError(u'Invalid URL: %s' % url) - url_title = mobj.group('url_title') - webpage = self._download_webpage(url, url_title) - - video_id = self._html_search_regex(r'
]*type="high".*?>(.*?)', - data, u'video URL') - - return [{ - 'id': video_id, - 'url': video_url, - 'ext': 'mp4', - 'title': self._og_search_title(webpage), - 'thumbnail': self._og_search_thumbnail(webpage), - 'description': self._og_search_description(webpage), - }] + display_id = self._match_id(url) + + response = self._graphql_call('''{ + %s(slug: "%s") { + ... on RecordSlug { + record { + id + title + teaser + publishOn + thumb { + preview + } + file { + url + } + tags { + name + } + duration + turnerMediaId + turnerMediaAuthToken + } + } + ... on NotFoundSlug { + status + } + } +}''', 'Slug', display_id) + if response.get('status'): + raise ExtractorError('This video is no longer available.', expected=True) + + record = response['record'] + video_id = record['id'] + + info = { + 'id': video_id, + 'display_id': display_id, + 'title': record['title'], + 'thumbnail': record.get('thumb', {}).get('preview'), + 'description': record.get('teaser'), + 'duration': parse_duration(record.get('duration')), + 'timestamp': parse_iso8601(record.get('publishOn')), + } + + media_id = record.get('turnerMediaId') + if media_id: + self._initialize_geo_bypass({ + 'countries': ['US'], + }) + info.update(self._extract_ngtv_info(media_id, { + 'accessToken': record['turnerMediaAuthToken'], + 'accessTokenType': 'jws', + })) + else: + video_sources = self._graphql_call('''{ + %s(id: "%s") { + src + } +}''', 'RecordVideoSource', video_id) or {} + + formats = [] + get_quality = qualities(['low', 'sd', 'hd', 'uhd']) + for format_id, src in video_sources.get('src', {}).items(): + if not isinstance(src, dict): + continue + src_url = src.get('src') + if not src_url: + continue + ext = determine_ext(src_url, mimetype2ext(src.get('type'))) + if format_id == 'hls' or ext == 'm3u8': + # compat_urllib_parse.urljoin does not work here + if src_url.startswith('/'): + src_url = 'http://ht.cdn.turner.com/tbs/big/teamcoco' + src_url + formats.extend(self._extract_m3u8_formats( + src_url, video_id, 'mp4', m3u8_id=format_id, fatal=False)) + else: + if src_url.startswith('/mp4:protected/'): + # TODO Correct extraction for these files + continue + tbr = int_or_none(self._search_regex( + r'(\d+)k\.mp4', src_url, 'tbr', default=None)) + + formats.append({ + 'url': src_url, + 'ext': ext, + 'tbr': tbr, + 'format_id': format_id, + 'quality': get_quality(format_id), + }) + if not formats: + formats = self._extract_m3u8_formats( + record['file']['url'], video_id, 'mp4', fatal=False) + self._sort_formats(formats) + info['formats'] = formats + + return info