]> Raphaƫl G. Git Repositories - youtubedl/blobdiff - youtube_dl/extractor/rottentomatoes.py
Update upstream source from tag 'upstream/2019.06.08'
[youtubedl] / youtube_dl / extractor / rottentomatoes.py
index f9cd48790c3b4a92b82bf1880020d53a074b1434..14c8e823698174f60890d9c27535e1dce40c9ce6 100644 (file)
@@ -1,31 +1,32 @@
 from __future__ import unicode_literals
 
 from .common import InfoExtractor
-from ..compat import compat_urlparse
 from .internetvideoarchive import InternetVideoArchiveIE
 
 
 class RottenTomatoesIE(InfoExtractor):
-    _VALID_URL = r'https?://www\.rottentomatoes\.com/m/[^/]+/trailers/(?P<id>\d+)'
+    _VALID_URL = r'https?://(?:www\.)?rottentomatoes\.com/m/[^/]+/trailers/(?P<id>\d+)'
 
     _TEST = {
         'url': 'http://www.rottentomatoes.com/m/toy_story_3/trailers/11028566/',
         'info_dict': {
-            'id': '613340',
+            'id': '11028566',
             'ext': 'mp4',
             'title': 'Toy Story 3',
+            'description': 'From the creators of the beloved TOY STORY films, comes a story that will reunite the gang in a whole new way.',
+            'thumbnail': r're:^https?://.*\.jpg$',
         },
     }
 
     def _real_extract(self, url):
         video_id = self._match_id(url)
         webpage = self._download_webpage(url, video_id)
-        og_video = self._og_search_video_url(webpage)
-        query = compat_urlparse.urlparse(og_video).query
+        iva_id = self._search_regex(r'publishedid=(\d+)', webpage, 'internet video archive id')
 
         return {
             '_type': 'url_transparent',
-            'url': InternetVideoArchiveIE._build_xml_url(query),
+            'url': 'http://video.internetvideoarchive.net/player/6/configuration.ashx?domain=www.videodetective.com&customerid=69249&playerid=641&publishedid=' + iva_id,
             'ie_key': InternetVideoArchiveIE.ie_key(),
+            'id': video_id,
             'title': self._og_search_title(webpage),
         }