|
|
@ -3483,7 +3483,7 @@ class UstreamIE(InfoExtractor):
|
|
|
|
return [info]
|
|
|
|
return [info]
|
|
|
|
|
|
|
|
|
|
|
|
class WorldStarHipHopIE(InfoExtractor):
|
|
|
|
class WorldStarHipHopIE(InfoExtractor):
|
|
|
|
_VALID_URL = r'http://(?:www|m)\.worldstar(?:candy|hiphop)\.com/videos/video\.php\?v=(?P<id>.*)'
|
|
|
|
_VALID_URL = r'https?://(?:www|m)\.worldstar(?:candy|hiphop)\.com/videos/video\.php\?v=(?P<id>.*)'
|
|
|
|
IE_NAME = u'WorldStarHipHop'
|
|
|
|
IE_NAME = u'WorldStarHipHop'
|
|
|
|
|
|
|
|
|
|
|
|
def _real_extract(self, url):
|
|
|
|
def _real_extract(self, url):
|
|
|
@ -3503,21 +3503,15 @@ class WorldStarHipHopIE(InfoExtractor):
|
|
|
|
else:
|
|
|
|
else:
|
|
|
|
ext = 'flv'
|
|
|
|
ext = 'flv'
|
|
|
|
else:
|
|
|
|
else:
|
|
|
|
self._downloader.report_error(u'Cannot find video url for %s' % video_id)
|
|
|
|
raise ExtractorError(u'Cannot find video url for %s' % video_id)
|
|
|
|
return
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
_title = r"""<title>(.*)</title>"""
|
|
|
|
mobj = re.search(r"<title>(.*)</title>", webpage_src)
|
|
|
|
|
|
|
|
|
|
|
|
mobj = re.search(_title, webpage_src)
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
if mobj is not None:
|
|
|
|
if mobj is None:
|
|
|
|
|
|
|
|
raise ExtractorError(u'Cannot determine title')
|
|
|
|
title = mobj.group(1)
|
|
|
|
title = mobj.group(1)
|
|
|
|
else:
|
|
|
|
|
|
|
|
title = 'World Start Hip Hop - %s' % time.ctime()
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
_thumbnail = r"""rel="image_src" href="(.*)" />"""
|
|
|
|
|
|
|
|
mobj = re.search(_thumbnail, webpage_src)
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
mobj = re.search(r'rel="image_src" href="(.*)" />', webpage_src)
|
|
|
|
# Getting thumbnail and if not thumbnail sets correct title for WSHH candy video.
|
|
|
|
# Getting thumbnail and if not thumbnail sets correct title for WSHH candy video.
|
|
|
|
if mobj is not None:
|
|
|
|
if mobj is not None:
|
|
|
|
thumbnail = mobj.group(1)
|
|
|
|
thumbnail = mobj.group(1)
|
|
|
|