[ynet] Add new extractor
authornet <net@h-MacBook-Pro-sl-net.local>
Sat, 27 Sep 2014 17:11:22 +0000 (20:11 +0300)
committernet <net@h-MacBook-Pro-sl-net.local>
Sat, 27 Sep 2014 17:11:22 +0000 (20:11 +0300)
youtube_dl/extractor/__init__.py
youtube_dl/extractor/ynet.py [new file with mode: 0644]

index 1f1fc0eb269d5c34821692193da1c93849ff8276..944e356ae410b58deb28abc0140411a76618ed37 100644 (file)
@@ -450,6 +450,7 @@ from .yahoo import (
     YahooNewsIE,
     YahooSearchIE,
 )
+from .ynet import YnetIE
 from .youjizz import YouJizzIE
 from .youku import YoukuIE
 from .youporn import YouPornIE
diff --git a/youtube_dl/extractor/ynet.py b/youtube_dl/extractor/ynet.py
new file mode 100644 (file)
index 0000000..94d2536
--- /dev/null
@@ -0,0 +1,47 @@
+# coding: utf-8
+from __future__ import unicode_literals
+
+import re
+import base64
+import json
+
+from .common import InfoExtractor
+from youtube_dl.utils import compat_urllib_parse_urlparse, compat_urllib_parse
+
+class YnetIE(InfoExtractor):
+    _VALID_URL = r'http://.*ynet\.co\.il/.*/0,7340,(?P<id>L(-[0-9]+)+),00\.html'
+    _TEST = {
+        'url': 'http://hot.ynet.co.il/home/0,7340,L-11659-99244,00.html',
+        'info_dict': {
+            'id': 'L-11659-99244',
+            'ext': 'flv',
+            'title': 'md5:3dba12d2837ee2ad9652cc64af652b16',
+            'thumbnail': 'http://hot.ynet.co.il/PicServer4/2014/09/23/5606015/AMERICAN_COMMUNE1_T.jpg',
+        }
+    }
+
+    def _real_extract(self, url):
+        mobj = re.match(self._VALID_URL, url)
+        
+        id = mobj.group('id')
+        
+        webpage = self._download_webpage(url, id)
+
+        content = compat_urllib_parse.unquote_plus(self._og_search_video_url(webpage).decode('utf-8'))
+
+        player_url = re.match('(http.*\.swf)\?' ,content).group(1)
+        
+        config = json.loads(re.match('.*config\=(.*)' ,content).group(1))
+                
+        f4m_url = config['clip']['url']    
+
+        title = re.sub(': Video$', '', self._og_search_title(webpage))
+
+        return {
+            'id': id,
+            'title': title,
+            'formats': self._extract_f4m_formats(f4m_url, id),
+            'thumbnail': self._og_search_thumbnail(webpage),
+            'player_url': player_url,
+        }
+