from __future__ import unicode_literals
import re
+
from .common import InfoExtractor
-from ..utils import (
+from ..compat import (
compat_urllib_parse,
compat_urllib_request,
)
}]
def _real_extract(self, url):
- mobj = re.match(self._VALID_URL, url)
- video_id = mobj.group('id')
+ video_id = self._match_id(url)
webpage = self._download_webpage(url, video_id)
fields = dict(re.findall(r'''(?x)<input\s+