'ext': 'vtt',
}]
return res
+
+
+class VikiShowIE(InfoExtractor):
+ IE_NAME = 'viki:show'
+ _VALID_URL = r'^https?://(?:www\.)?viki\.com/tv/(?P<id>[0-9]+c)'
+ _TESTS = [{
+ 'url': 'http://www.viki.com/tv/50c-boys-over-flowers',
+ 'info_dict': {
+ 'id': '50c',
+ 'title': 'Boys Over Flowers',
+ 'description': 'md5:ecd3cff47967fe193cff37c0bec52790',
+ },
+ 'playlist_count': 25,
+ }]
+
+ def _real_extract(self, url):
+ show_id = self._match_id(url)
+ show_page = self._download_webpage(url, show_id, 'Download show page')
+
+ title = self._og_search_title(show_page)
+ description = self._og_search_description(show_page)
+
+ show_json = self._download_json(
+ 'http://api.viki.io/v4/containers/%s/episodes.json?app=100000a&per_page=999&sort=number&direction=asc' % show_id,
+ show_id, note='Retrieve show json', errnote='Unable to get show json'
+ )
+ entries = []
+ for video in show_json['response']:
+ video_id = video['id']
+ entries.append(self.url_result(
+ 'http://www.viki.com/videos/%s' % video_id, 'Viki', video_id))
+
+ return self.playlist_result(entries, show_id, title, description)