ytdl/youtube_dl/extractor/nextmovie.py
2015-12-31 22:47:18 +01:00

31 lines
966 B
Python

# coding: utf-8
from __future__ import unicode_literals
from .mtv import MTVServicesInfoExtractor
from ..compat import compat_urllib_parse
class NextMovieIE(MTVServicesInfoExtractor):
IE_NAME = 'nextmovie.com'
_VALID_URL = r'https?://(?:www\.)?nextmovie\.com/shows/[^/]+/\d{4}-\d{2}-\d{2}/(?P<id>[^/?#]+)'
_FEED_URL = 'http://lite.dextr.mtvi.com/service1/dispatch.htm'
_TESTS = [{
'url': 'http://www.nextmovie.com/shows/exclusives/2013-03-10/mgid:uma:videolist:nextmovie.com:1715019/',
'md5': '09a9199f2f11f10107d04fcb153218aa',
'info_dict': {
'id': '961726',
'ext': 'mp4',
'title': 'The Muppets\' Gravity',
},
}]
def _get_feed_query(self, uri):
return compat_urllib_parse.urlencode({
'feed': '1505',
'mgid': uri,
})
def _real_extract(self, url):
mgid = self._match_id(url)
return self._get_videos_info(mgid)