]> Raphaƫl G. Git Repositories - youtubedl/blobdiff - youtube_dl/extractor/ooyala.py
debian/gbp.conf: Change to new git-buildpackage syntax.
[youtubedl] / youtube_dl / extractor / ooyala.py
index 13f12824c99aa71c357047ff62a866365bbc49fb..d5b05c18febb580a448263b4f7b2876ef3234957 100644 (file)
@@ -3,23 +3,36 @@ import re
 import json
 
 from .common import InfoExtractor
-from ..utils import unescapeHTML
+from ..utils import (
+    unescapeHTML,
+    ExtractorError,
+)
 
 
 class OoyalaIE(InfoExtractor):
     _VALID_URL = r'(?:ooyala:|https?://.+?\.ooyala\.com/.*?(?:embedCode|ec)=)(?P<id>.+?)(&|$)'
 
-    _TEST = {
-        # From http://it.slashdot.org/story/13/04/25/178216/recovering-data-from-broken-hard-drives-and-ssds-video
-        'url': 'http://player.ooyala.com/player.js?embedCode=pxczE2YjpfHfn1f3M-ykG_AmJRRn0PD8',
-        'md5': '3f5cceb3a7bf461d6c29dc466cf8033c',
-        'info_dict': {
-            'id': 'pxczE2YjpfHfn1f3M-ykG_AmJRRn0PD8',
-            'ext': 'mp4',
-            'title': 'Explaining Data Recovery from Hard Drives and SSDs',
-            'description': 'How badly damaged does a drive have to be to defeat Russell and his crew? Apparently, smashed to bits.',
+    _TESTS = [
+        {
+            # From http://it.slashdot.org/story/13/04/25/178216/recovering-data-from-broken-hard-drives-and-ssds-video
+            'url': 'http://player.ooyala.com/player.js?embedCode=pxczE2YjpfHfn1f3M-ykG_AmJRRn0PD8',
+            'info_dict': {
+                'id': 'pxczE2YjpfHfn1f3M-ykG_AmJRRn0PD8',
+                'ext': 'mp4',
+                'title': 'Explaining Data Recovery from Hard Drives and SSDs',
+                'description': 'How badly damaged does a drive have to be to defeat Russell and his crew? Apparently, smashed to bits.',
+            },
+        }, {
+            # Only available for ipad
+            'url': 'http://player.ooyala.com/player.js?embedCode=x1b3lqZDq9y_7kMyC2Op5qo-p077tXD0',
+            'info_dict': {
+                'id': 'x1b3lqZDq9y_7kMyC2Op5qo-p077tXD0',
+                'ext': 'mp4',
+                'title': 'Simulation Overview - Levels of Simulation',
+                'description': '',
+            },
         },
-    }
+    ]
 
     @staticmethod
     def _url_for_embed_code(embed_code):
@@ -28,7 +41,7 @@ class OoyalaIE(InfoExtractor):
     @classmethod
     def _build_url_result(cls, embed_code):
         return cls.url_result(cls._url_for_embed_code(embed_code),
-            ie=cls.ie_key())
+                              ie=cls.ie_key())
 
     def _extract_result(self, info, more_info):
         return {
@@ -47,13 +60,30 @@ class OoyalaIE(InfoExtractor):
         player = self._download_webpage(player_url, embedCode)
         mobile_url = self._search_regex(r'mobile_player_url="(.+?)&device="',
                                         player, 'mobile player url')
-        mobile_player = self._download_webpage(mobile_url, embedCode)
-        videos_info = self._search_regex(
-            r'var streams=window.oo_testEnv\?\[\]:eval\("\((\[{.*?}\])\)"\);',
-            mobile_player, 'info').replace('\\"','"')
-        videos_more_info = self._search_regex(r'eval\("\(({.*?\\"promo\\".*?})\)"', mobile_player, 'more info').replace('\\"','"')
+        # Looks like some videos are only available for particular devices
+        # (e.g. http://player.ooyala.com/player.js?embedCode=x1b3lqZDq9y_7kMyC2Op5qo-p077tXD0
+        # is only available for ipad)
+        # Working around with fetching URLs for all the devices found starting with 'unknown'
+        # until we succeed or eventually fail for each device.
+        devices = re.findall(r'device\s*=\s*"([^"]+)";', player)
+        devices.remove('unknown')
+        devices.insert(0, 'unknown')
+        for device in devices:
+            mobile_player = self._download_webpage(
+                '%s&device=%s' % (mobile_url, device), embedCode,
+                'Downloading mobile player JS for %s device' % device)
+            videos_info = self._search_regex(
+                r'var streams=window.oo_testEnv\?\[\]:eval\("\((\[{.*?}\])\)"\);',
+                mobile_player, 'info', fatal=False, default=None)
+            if videos_info:
+                break
+        if not videos_info:
+            raise ExtractorError('Unable to extract info')
+        videos_info = videos_info.replace('\\"', '"')
+        videos_more_info = self._search_regex(
+            r'eval\("\(({.*?\\"promo\\".*?})\)"', mobile_player, 'more info').replace('\\"', '"')
         videos_info = json.loads(videos_info)
-        videos_more_info =json.loads(videos_more_info)
+        videos_more_info = json.loads(videos_more_info)
 
         if videos_more_info.get('lineup'):
             videos = [self._extract_result(info, more_info) for (info, more_info) in zip(videos_info, videos_more_info['lineup'])]
@@ -65,4 +95,3 @@ class OoyalaIE(InfoExtractor):
             }
         else:
             return self._extract_result(videos_info[0], videos_more_info)
-