]> Raphaël G. Git Repositories - youtubedl/blobdiff - youtube_dl/extractor/aparat.py
Merge tag 'upstream/2016.02.22'
[youtubedl] / youtube_dl / extractor / aparat.py
index 74860882628017c5ab7a44f22bd9b05286ad556e..63429780e8abf528165daf7e50a6317bce9a6c7d 100644 (file)
@@ -1,5 +1,4 @@
-#coding: utf-8
-
+# coding: utf-8
 from __future__ import unicode_literals
 
 import re
@@ -21,13 +20,13 @@ class AparatIE(InfoExtractor):
             'id': 'wP8On',
             'ext': 'mp4',
             'title': 'تیم گلکسی 11 - زومیت',
+            'age_limit': 0,
         },
         # 'skip': 'Extremely unreliable',
     }
 
     def _real_extract(self, url):
-        m = re.match(self._VALID_URL, url)
-        video_id = m.group('id')
+        video_id = self._match_id(url)
 
         # Note: There is an easier-to-parse configuration at
         # http://www.aparat.com/video/video/config/videohash/%video_id
@@ -36,19 +35,20 @@ class AparatIE(InfoExtractor):
                      video_id + '/vt/frame')
         webpage = self._download_webpage(embed_url, video_id)
 
-        video_urls = re.findall(r'fileList\[[0-9]+\]\s*=\s*"([^"]+)"', webpage)
+        video_urls = [video_url.replace('\\/', '/') for video_url in re.findall(
+            r'(?:fileList\[[0-9]+\]\s*=|"file"\s*:)\s*"([^"]+)"', webpage)]
         for i, video_url in enumerate(video_urls):
             req = HEADRequest(video_url)
             res = self._request_webpage(
-                req, video_id, note=u'Testing video URL %d' % i, errnote=False)
+                req, video_id, note='Testing video URL %d' % i, errnote=False)
             if res:
                 break
         else:
-            raise ExtractorError(u'No working video URLs found')
+            raise ExtractorError('No working video URLs found')
 
-        title = self._search_regex(r'\s+title:\s*"([^"]+)"', webpage, u'title')
+        title = self._search_regex(r'\s+title:\s*"([^"]+)"', webpage, 'title')
         thumbnail = self._search_regex(
-            r'\s+image:\s*"([^"]+)"', webpage, u'thumbnail', fatal=False)
+            r'image:\s*"([^"]+)"', webpage, 'thumbnail', fatal=False)
 
         return {
             'id': video_id,
@@ -56,4 +56,5 @@ class AparatIE(InfoExtractor):
             'url': video_url,
             'ext': 'mp4',
             'thumbnail': thumbnail,
+            'age_limit': self._family_friendly_search(webpage),
         }