1
1
mirror of https://github.com/ytdl-org/youtube-dl synced 2024-11-18 10:30:03 +01:00
youtube-dl/youtube_dl/extractor/spiegeltv.py

18 lines
504 B
Python

from __future__ import unicode_literals
from .common import InfoExtractor
from .nexx import NexxIE
class SpiegeltvIE(InfoExtractor):
_VALID_URL = r'https?://(?:www\.)?spiegel\.tv/videos/(?P<id>\d+)'
_TEST = {
'url': 'http://www.spiegel.tv/videos/161681-flug-mh370/',
'only_matching': True,
}
def _real_extract(self, url):
return self.url_result(
'https://api.nexx.cloud/v3/748/videos/byid/%s'
% self._match_id(url), ie=NexxIE.ie_key())