using re.search

This commit is contained in:
Johny Mo Swag 2013-07-29 14:39:14 -07:00
parent 579e2691fe
commit d75654c15e

View file

@ -21,17 +21,17 @@ def _real_extract(self, url):
webpage_src = self._download_webpage(url, video_id)
video_url = self._search_regex(r'videoId=(.*?)&amp?',
webpage_src, u'video URL', fatal=False)
video_url = re.search(r'videoId=(.*?)&amp?',
webpage_src)
if video_url:
self.to_screen(u'Vevo video detected:')
return self.url_result('vevo:%s' % video_url, ie='Vevo')
return self.url_result('vevo:%s' % video_url.group(1), ie='Vevo')
video_url = self._search_regex(r'so\.addVariable\("file","(.*?)"\)',
webpage_src, u'video URL')
if video_url == None:
if video_url is None:
video_url = self._search_regex(r'videoId=(.*?)&amp?',
webpage_src, u'video URL')
self.to_screen(u'Vevo video detected:')