|
|
@ -0,0 +1,23 @@ |
|
|
|
import re |
|
|
|
|
|
|
|
from .common import InfoExtractor |
|
|
|
|
|
|
|
|
|
|
|
class SlashdotIE(InfoExtractor): |
|
|
|
_VALID_URL = r'https?://tv.slashdot.org/video/\?embed=(?P<id>.*?)(&|$)' |
|
|
|
|
|
|
|
_TEST = { |
|
|
|
u'url': u'http://tv.slashdot.org/video/?embed=JscHMzZDplD0p-yNLOzTfzC3Q3xzJaUz', |
|
|
|
u'file': u'JscHMzZDplD0p-yNLOzTfzC3Q3xzJaUz.mp4', |
|
|
|
u'md5': u'd2222e7a4a4c1541b3e0cf732fb26735', |
|
|
|
u'info_dict': { |
|
|
|
u'title': u' Meet the Stampede Supercomputing Cluster\'s Administrator', |
|
|
|
}, |
|
|
|
} |
|
|
|
|
|
|
|
def _real_extract(self, url): |
|
|
|
mobj = re.match(self._VALID_URL, url) |
|
|
|
video_id = mobj.group('id') |
|
|
|
webpage = self._download_webpage(url, video_id) |
|
|
|
ooyala_url = self._search_regex(r'<script src="(.*?)"', webpage, 'ooyala url') |
|
|
|
return self.url_result(ooyala_url, 'Ooyala') |