[cbssports] Add extractor (closes #4996)

pull/8/head
Jaime Marquínez Ferrándiz 10 years ago
parent b88ba05356
commit a3b9157f49

@ -58,6 +58,7 @@ from .canalplus import CanalplusIE
from .canalc2 import Canalc2IE from .canalc2 import Canalc2IE
from .cbs import CBSIE from .cbs import CBSIE
from .cbsnews import CBSNewsIE from .cbsnews import CBSNewsIE
from .cbssports import CBSSportsIE
from .ccc import CCCIE from .ccc import CCCIE
from .ceskatelevize import CeskaTelevizeIE from .ceskatelevize import CeskaTelevizeIE
from .channel9 import Channel9IE from .channel9 import Channel9IE

@ -0,0 +1,30 @@
from __future__ import unicode_literals
import re
from .common import InfoExtractor
class CBSSportsIE(InfoExtractor):
_VALID_URL = r'http://www\.cbssports\.com/video/player/(?P<section>[^/]+)/(?P<id>[^/]+)'
_TEST = {
'url': 'http://www.cbssports.com/video/player/tennis/318462531970/0/us-open-flashbacks-1990s',
'info_dict': {
'id': '_d5_GbO8p1sT',
'ext': 'flv',
'title': 'US Open flashbacks: 1990s',
'description': 'Bill Macatee relives the best moments in US Open history from the 1990s.',
},
}
def _real_extract(self, url):
mobj = re.match(self._VALID_URL, url)
section = mobj.group('section')
video_id = mobj.group('id')
all_videos = self._download_json(
'http://www.cbssports.com/data/video/player/getVideos/%s?as=json' % section,
video_id)
# The json file contains the info of all the videos in the section
video_info = next(v for v in all_videos if v['pcid'] == video_id)
return self.url_result('theplatform:%s' % video_info['pid'], 'ThePlatform')
Loading…
Cancel
Save