]> Raphaƫl G. Git Repositories - youtubedl/blobdiff - youtube_dl/extractor/syfy.py
Prepare for release.
[youtubedl] / youtube_dl / extractor / syfy.py
index 8809a57fe31d23ea081e553302ef38222fb02221..f76b6e2b22c7fa391664218f9e59fa62908c4c08 100644 (file)
@@ -6,9 +6,9 @@ from .common import InfoExtractor
 
 
 class SyfyIE(InfoExtractor):
-    _VALID_URL = r'https?://www\.syfy\.com/videos/.+?vid:(?P<id>\d+)'
+    _VALID_URL = r'https?://www\.syfy\.com/(?:videos/.+?vid:(?P<id>[0-9]+)|(?!videos)(?P<video_name>[^/]+)(?:$|[?#]))'
 
-    _TEST = {
+    _TESTS = [{
         'url': 'http://www.syfy.com/videos/Robot%20Combat%20League/Behind%20the%20Scenes/vid:2631458',
         'md5': 'e07de1d52c7278adbb9b9b1c93a66849',
         'info_dict': {
@@ -18,10 +18,30 @@ class SyfyIE(InfoExtractor):
             'description': 'Listen to what insights George Lucas give his daughter Amanda.',
         },
         'add_ie': ['ThePlatform'],
-    }
+    }, {
+        'url': 'http://www.syfy.com/wilwheaton',
+        'md5': '94dfa54ee3ccb63295b276da08c415f6',
+        'info_dict': {
+            'id': '4yoffOOXC767',
+            'ext': 'flv',
+            'title': 'The Wil Wheaton Project - Premiering May 27th at 10/9c.',
+            'description': 'The Wil Wheaton Project premieres May 27th at 10/9c. Don\'t miss it.',
+        },
+        'add_ie': ['ThePlatform'],
+        'skip': 'Blocked outside the US',
+    }]
 
     def _real_extract(self, url):
         mobj = re.match(self._VALID_URL, url)
-        video_id = mobj.group('id')
+        video_name = mobj.group('video_name')
+        if video_name:
+            generic_webpage = self._download_webpage(url, video_name)
+            video_id = self._search_regex(
+                r'<iframe.*?class="video_iframe_page"\s+src="/_utils/video/thP_video_controller.php.*?_vid([0-9]+)">',
+                generic_webpage, 'video ID')
+            url = 'http://www.syfy.com/videos/%s/%s/vid:%s' % (
+                video_name, video_name, video_id)
+        else:
+            video_id = mobj.group('id')
         webpage = self._download_webpage(url, video_id)
         return self.url_result(self._og_search_video_url(webpage))