X-Git-Url: http://git.bitcoin.ninja/index.cgi?a=blobdiff_plain;f=youtube_dl%2Fextractor%2Fcnn.py;h=774b7105580f69dedac4318c82aee2ab62c27381;hb=HEAD;hp=cee78765b1cb0a3fc65655ed366604e4cb8e938c;hpb=1a582dd49d628914fa6a056b490914738f15c56d;p=youtube-dl diff --git a/youtube_dl/extractor/cnn.py b/youtube_dl/extractor/cnn.py index cee78765b..774b71055 100644 --- a/youtube_dl/extractor/cnn.py +++ b/youtube_dl/extractor/cnn.py @@ -1,47 +1,144 @@ +from __future__ import unicode_literals + import re -import xml.etree.ElementTree from .common import InfoExtractor -from ..utils import determine_ext +from .turner import TurnerBaseIE +from ..utils import url_basename -class CNNIE(InfoExtractor): - _VALID_URL = r'https?://(edition\.)?cnn\.com/video/(data/.+?|\?)/(?P.+?/(?P[^/]+?)\.cnn)' - _TEST = { - u'url': u'http://edition.cnn.com/video/?/video/sports/2013/06/09/nadal-1-on-1.cnn', - u'file': u'sports_2013_06_09_nadal-1-on-1.cnn.mp4', - u'md5': u'3e6121ea48df7e2259fe73a0628605c4', - u'info_dict': { - u'title': u'Nadal wins 8th French Open title', - u'description': u'World Sport\'s Amanda Davies chats with 2013 French Open champion Rafael Nadal.', +class CNNIE(TurnerBaseIE): + _VALID_URL = r'''(?x)https?://(?:(?P<sub_domain>edition|www|money)\.)?cnn\.com/(?:video/(?:data/.+?|\?)/)?videos?/ + (?P<path>.+?/(?P<title>[^/]+?)(?:\.(?:[a-z\-]+)|(?=&)))''' + + _TESTS = [{ + 'url': 'http://edition.cnn.com/video/?/video/sports/2013/06/09/nadal-1-on-1.cnn', + 'md5': '3e6121ea48df7e2259fe73a0628605c4', + 'info_dict': { + 'id': 'sports/2013/06/09/nadal-1-on-1.cnn', + 'ext': 'mp4', + 'title': 'Nadal wins 8th French Open title', + 'description': 'World Sport\'s Amanda Davies chats with 2013 French Open champion Rafael Nadal.', + 'duration': 135, + 'upload_date': '20130609', + }, + 'expected_warnings': ['Failed to download m3u8 information'], + }, { + 'url': 'http://edition.cnn.com/video/?/video/us/2013/08/21/sot-student-gives-epic-speech.georgia-institute-of-technology&utm_source=feedburner&utm_medium=feed&utm_campaign=Feed%3A+rss%2Fcnn_topstories+%28RSS%3A+Top+Stories%29', + 'md5': 'b5cc60c60a3477d185af8f19a2a26f4e', + 'info_dict': { + 'id': 'us/2013/08/21/sot-student-gives-epic-speech.georgia-institute-of-technology', + 'ext': 'mp4', + 'title': "Student's epic speech stuns new freshmen", + 'description': "A Georgia Tech student welcomes the incoming freshmen with an epic speech backed by music from \"2001: A Space Odyssey.\"", + 'upload_date': '20130821', + }, + 'expected_warnings': ['Failed to download m3u8 information'], + }, { + 'url': 'http://www.cnn.com/video/data/2.0/video/living/2014/12/22/growing-america-nashville-salemtown-board-episode-1.hln.html', + 'md5': 'f14d02ebd264df951feb2400e2c25a1b', + 'info_dict': { + 'id': 'living/2014/12/22/growing-america-nashville-salemtown-board-episode-1.hln', + 'ext': 'mp4', + 'title': 'Nashville Ep. 1: Hand crafted skateboards', + 'description': 'md5:e7223a503315c9f150acac52e76de086', + 'upload_date': '20141222', + }, + 'expected_warnings': ['Failed to download m3u8 information'], + }, { + 'url': 'http://money.cnn.com/video/news/2016/08/19/netflix-stunning-stats.cnnmoney/index.html', + 'md5': '52a515dc1b0f001cd82e4ceda32be9d1', + 'info_dict': { + 'id': '/video/news/2016/08/19/netflix-stunning-stats.cnnmoney', + 'ext': 'mp4', + 'title': '5 stunning stats about Netflix', + 'description': 'Did you know that Netflix has more than 80 million members? Here are five facts about the online video distributor that you probably didn\'t know.', + 'upload_date': '20160819', + }, + 'params': { + # m3u8 download + 'skip_download': True, + }, + }, { + 'url': 'http://cnn.com/video/?/video/politics/2015/03/27/pkg-arizona-senator-church-attendance-mandatory.ktvk', + 'only_matching': True, + }, { + 'url': 'http://cnn.com/video/?/video/us/2015/04/06/dnt-baker-refuses-anti-gay-order.wkmg', + 'only_matching': True, + }, { + 'url': 'http://edition.cnn.com/videos/arts/2016/04/21/olympic-games-cultural-a-z-brazil.cnn', + 'only_matching': True, + }] + + _CONFIG = { + # http://edition.cnn.com/.element/apps/cvp/3.0/cfg/spider/cnn/expansion/config.xml + 'edition': { + 'data_src': 'http://edition.cnn.com/video/data/3.0/video/%s/index.xml', + 'media_src': 'http://pmd.cdn.turner.com/cnn/big', + }, + # http://money.cnn.com/.element/apps/cvp2/cfg/config.xml + 'money': { + 'data_src': 'http://money.cnn.com/video/data/4.0/video/%s.xml', + 'media_src': 'http://ht3.cdn.turner.com/money/big', }, } + def _extract_timestamp(self, video_data): + # TODO: fix timestamp extraction + return None + def _real_extract(self, url): - mobj = re.match(self._VALID_URL, url) - path = mobj.group('path') - page_title = mobj.group('title') - info_xml = self._download_webpage( - 'http://cnn.com/video/data/3.0/%s/index.xml' % path, page_title) - info = xml.etree.ElementTree.fromstring(info_xml.encode('utf-8')) - - formats = [] - for f in info.findall('files/file'): - mf = re.match(r'(\d+)x(\d+)(?:_(.*)k)?',f.attrib['bitrate']) - if mf is not None: - formats.append((int(mf.group(1)), int(mf.group(2)), int(mf.group(3) or 0), f.text)) - formats = sorted(formats) - (_,_,_, video_path) = formats[-1] - video_url = 'http://ht.cdn.turner.com/cnn/big%s' % video_path - - thumbnails = sorted([((int(t.attrib['height']),int(t.attrib['width'])), t.text) for t in info.findall('images/image')]) - thumbs_dict = [{'resolution': res, 'url': t_url} for (res, t_url) in thumbnails] - - return {'id': info.attrib['id'], - 'title': info.find('headline').text, - 'url': video_url, - 'ext': determine_ext(video_url), - 'thumbnail': thumbnails[-1][1], - 'thumbnails': thumbs_dict, - 'description': info.find('description').text, + sub_domain, path, page_title = re.match(self._VALID_URL, url).groups() + if sub_domain not in ('money', 'edition'): + sub_domain = 'edition' + config = self._CONFIG[sub_domain] + return self._extract_cvp_info( + config['data_src'] % path, page_title, { + 'default': { + 'media_src': config['media_src'], } + }) + + +class CNNBlogsIE(InfoExtractor): + _VALID_URL = r'https?://[^\.]+\.blogs\.cnn\.com/.+' + _TEST = { + 'url': 'http://reliablesources.blogs.cnn.com/2014/02/09/criminalizing-journalism/', + 'md5': '3e56f97b0b6ffb4b79f4ea0749551084', + 'info_dict': { + 'id': 'bestoftv/2014/02/09/criminalizing-journalism.cnn', + 'ext': 'mp4', + 'title': 'Criminalizing journalism?', + 'description': 'Glenn Greenwald responds to comments made this week on Capitol Hill that journalists could be criminal accessories.', + 'upload_date': '20140209', + }, + 'expected_warnings': ['Failed to download m3u8 information'], + 'add_ie': ['CNN'], + } + + def _real_extract(self, url): + webpage = self._download_webpage(url, url_basename(url)) + cnn_url = self._html_search_regex(r'data-url="(.+?)"', webpage, 'cnn url') + return self.url_result(cnn_url, CNNIE.ie_key()) + + +class CNNArticleIE(InfoExtractor): + _VALID_URL = r'https?://(?:(?:edition|www)\.)?cnn\.com/(?!videos?/)' + _TEST = { + 'url': 'http://www.cnn.com/2014/12/21/politics/obama-north-koreas-hack-not-war-but-cyber-vandalism/', + 'md5': '689034c2a3d9c6dc4aa72d65a81efd01', + 'info_dict': { + 'id': 'bestoftv/2014/12/21/ip-north-korea-obama.cnn', + 'ext': 'mp4', + 'title': 'Obama: Cyberattack not an act of war', + 'description': 'md5:0a802a40d2376f60e6b04c8d5bcebc4b', + 'upload_date': '20141221', + }, + 'expected_warnings': ['Failed to download m3u8 information'], + 'add_ie': ['CNN'], + } + + def _real_extract(self, url): + webpage = self._download_webpage(url, url_basename(url)) + cnn_url = self._html_search_regex(r"video:\s*'([^']+)'", webpage, 'cnn url') + return self.url_result('http://cnn.com/video/?/video/' + cnn_url, CNNIE.ie_key())