]> Raphaƫl G. Git Repositories - youtubedl/blobdiff - youtube_dl/extractor/ign.py
debian/gbp.conf: Change to new git-buildpackage syntax.
[youtubedl] / youtube_dl / extractor / ign.py
index 381af91e42d4c9f642b35643107f5dafd026aad9..3aade9e740673da3193324add6a8a3ac4eff8b1f 100644 (file)
@@ -1,10 +1,8 @@
+from __future__ import unicode_literals
+
 import re
-import json
 
 from .common import InfoExtractor
-from ..utils import (
-    determine_ext,
-)
 
 
 class IGNIE(InfoExtractor):
@@ -14,52 +12,76 @@ class IGNIE(InfoExtractor):
     """
 
     _VALID_URL = r'https?://.+?\.ign\.com/(?P<type>videos|show_videos|articles|(?:[^/]*/feature))(/.+)?/(?P<name_or_id>.+)'
-    IE_NAME = u'ign.com'
+    IE_NAME = 'ign.com'
 
     _CONFIG_URL_TEMPLATE = 'http://www.ign.com/videos/configs/id/%s.config'
-    _DESCRIPTION_RE = [r'<span class="page-object-description">(.+?)</span>',
-                       r'id="my_show_video">.*?<p>(.*?)</p>',
-                       ]
+    _DESCRIPTION_RE = [
+        r'<span class="page-object-description">(.+?)</span>',
+        r'id="my_show_video">.*?<p>(.*?)</p>',
+        r'<meta name="description" content="(.*?)"',
+    ]
 
     _TESTS = [
         {
-            u'url': u'http://www.ign.com/videos/2013/06/05/the-last-of-us-review',
-            u'file': u'8f862beef863986b2785559b9e1aa599.mp4',
-            u'md5': u'eac8bdc1890980122c3b66f14bdd02e9',
-            u'info_dict': {
-                u'title': u'The Last of Us Review',
-                u'description': u'md5:c8946d4260a4d43a00d5ae8ed998870c',
+            'url': 'http://www.ign.com/videos/2013/06/05/the-last-of-us-review',
+            'md5': 'eac8bdc1890980122c3b66f14bdd02e9',
+            'info_dict': {
+                'id': '8f862beef863986b2785559b9e1aa599',
+                'ext': 'mp4',
+                'title': 'The Last of Us Review',
+                'description': 'md5:c8946d4260a4d43a00d5ae8ed998870c',
             }
         },
         {
-            u'url': u'http://me.ign.com/en/feature/15775/100-little-things-in-gta-5-that-will-blow-your-mind',
-            u'playlist': [
+            'url': 'http://me.ign.com/en/feature/15775/100-little-things-in-gta-5-that-will-blow-your-mind',
+            'info_dict': {
+                'id': '100-little-things-in-gta-5-that-will-blow-your-mind',
+            },
+            'playlist': [
                 {
-                    u'file': u'5ebbd138523268b93c9141af17bec937.mp4',
-                    u'info_dict': {
-                        u'title': u'GTA 5 Video Review',
-                        u'description': u'Rockstar drops the mic on this generation of games. Watch our review of the masterly Grand Theft Auto V.',
+                    'info_dict': {
+                        'id': '5ebbd138523268b93c9141af17bec937',
+                        'ext': 'mp4',
+                        'title': 'GTA 5 Video Review',
+                        'description': 'Rockstar drops the mic on this generation of games. Watch our review of the masterly Grand Theft Auto V.',
                     },
                 },
                 {
-                    u'file': u'638672ee848ae4ff108df2a296418ee2.mp4',
-                    u'info_dict': {
-                        u'title': u'26 Twisted Moments from GTA 5 in Slow Motion',
-                        u'description': u'The twisted beauty of GTA 5 in stunning slow motion.',
+                    'info_dict': {
+                        'id': '638672ee848ae4ff108df2a296418ee2',
+                        'ext': 'mp4',
+                        'title': '26 Twisted Moments from GTA 5 in Slow Motion',
+                        'description': 'The twisted beauty of GTA 5 in stunning slow motion.',
                     },
                 },
             ],
-            u'params': {
-                u'skip_download': True,
+            'params': {
+                'skip_download': True,
+            },
+        },
+        {
+            'url': 'http://www.ign.com/articles/2014/08/15/rewind-theater-wild-trailer-gamescom-2014?watch',
+            'md5': '4e9a0bda1e5eebd31ddcf86ec0b9b3c7',
+            'info_dict': {
+                'id': '078fdd005f6d3c02f63d795faa1b984f',
+                'ext': 'mp4',
+                'title': 'Rewind Theater - Wild Trailer Gamescom 2014',
+                'description': (
+                    'Giant skeletons, bloody hunts, and captivating'
+                    ' natural beauty take our breath away.'
+                ),
             },
         },
     ]
 
     def _find_video_id(self, webpage):
-        res_id = [r'data-video-id="(.+?)"',
-                  r'<object id="vid_(.+?)"',
-                  r'<meta name="og:image" content=".*/(.+?)-(.+?)/.+.jpg"',
-                  ]
+        res_id = [
+            r'"video_id"\s*:\s*"(.*?)"',
+            r'data-video-id="(.+?)"',
+            r'<object id="vid_(.+?)"',
+            r'<meta name="og:image" content=".*/(.+?)-(.+?)/.+.jpg"',
+            r'class="hero-poster[^"]*?"[^>]*id="(.+?)"',
+        ]
         return self._search_regex(res_id, webpage, 'video id')
 
     def _real_extract(self, url):
@@ -67,63 +89,57 @@ class IGNIE(InfoExtractor):
         name_or_id = mobj.group('name_or_id')
         page_type = mobj.group('type')
         webpage = self._download_webpage(url, name_or_id)
-        if page_type == 'articles':
-            video_url = self._search_regex(r'var videoUrl = "(.+?)"', webpage, u'video url')
-            return self.url_result(video_url, ie='IGN')
-        elif page_type != 'video':
+        if page_type != 'video':
             multiple_urls = re.findall(
-                '<param name="flashvars" value="[^"]*?url=(https?://www\.ign\.com/videos/.*?)["&]',
+                '<param name="flashvars"[^>]*value="[^"]*?url=(https?://www\.ign\.com/videos/.*?)["&]',
                 webpage)
             if multiple_urls:
-                return [self.url_result(u, ie='IGN') for u in multiple_urls]
+                entries = [self.url_result(u, ie='IGN') for u in multiple_urls]
+                return {
+                    '_type': 'playlist',
+                    'id': name_or_id,
+                    'entries': entries,
+                }
 
         video_id = self._find_video_id(webpage)
         result = self._get_video_info(video_id)
         description = self._html_search_regex(self._DESCRIPTION_RE,
-                                              webpage, 'video description',
-                                              flags=re.DOTALL)
+                                              webpage, 'video description', flags=re.DOTALL)
         result['description'] = description
         return result
 
     def _get_video_info(self, video_id):
         config_url = self._CONFIG_URL_TEMPLATE % video_id
-        config = json.loads(self._download_webpage(config_url, video_id,
-                            u'Downloading video info'))
+        config = self._download_json(config_url, video_id)
         media = config['playlist']['media']
-        video_url = media['url']
 
-        return {'id': media['metadata']['videoId'],
-                'url': video_url,
-                'ext': determine_ext(video_url),
-                'title': media['metadata']['title'],
-                'thumbnail': media['poster'][0]['url'].replace('{size}', 'grande'),
-                }
+        return {
+            'id': media['metadata']['videoId'],
+            'url': media['url'],
+            'title': media['metadata']['title'],
+            'thumbnail': media['poster'][0]['url'].replace('{size}', 'grande'),
+        }
 
 
 class OneUPIE(IGNIE):
-    """Extractor for 1up.com, it uses the ign videos system."""
-
-    _VALID_URL = r'https?://gamevideos\.1up\.com/(?P<type>video)/id/(?P<name_or_id>.+)'
+    _VALID_URL = r'https?://gamevideos\.1up\.com/(?P<type>video)/id/(?P<name_or_id>.+)\.html'
     IE_NAME = '1up.com'
 
     _DESCRIPTION_RE = r'<div id="vid_summary">(.+?)</div>'
 
-    _TEST = {
-        u'url': u'http://gamevideos.1up.com/video/id/34976',
-        u'file': u'34976.mp4',
-        u'md5': u'68a54ce4ebc772e4b71e3123d413163d',
-        u'info_dict': {
-            u'title': u'Sniper Elite V2 - Trailer',
-            u'description': u'md5:5d289b722f5a6d940ca3136e9dae89cf',
+    _TESTS = [{
+        'url': 'http://gamevideos.1up.com/video/id/34976.html',
+        'md5': '68a54ce4ebc772e4b71e3123d413163d',
+        'info_dict': {
+            'id': '34976',
+            'ext': 'mp4',
+            'title': 'Sniper Elite V2 - Trailer',
+            'description': 'md5:5d289b722f5a6d940ca3136e9dae89cf',
         }
-    }
-
-    # Override IGN tests
-    _TESTS = []
+    }]
 
     def _real_extract(self, url):
         mobj = re.match(self._VALID_URL, url)
-        id = mobj.group('name_or_id')
         result = super(OneUPIE, self)._real_extract(url)
-        result['id'] = id
+        result['id'] = mobj.group('name_or_id')
         return result