]> Raphaƫl G. Git Repositories - youtubedl/blobdiff - youtube_dl/extractor/funnyordie.py
Finalize release.
[youtubedl] / youtube_dl / extractor / funnyordie.py
index 7c40e675398f7738bda09827031cfb99c620a17a..6e6b6666003d0837bffd6c25ddd12fe1ce892e50 100644 (file)
@@ -1,24 +1,35 @@
 from __future__ import unicode_literals
 
+import json
 import re
 
 from .common import InfoExtractor
+from ..utils import ExtractorError
 
 
 class FunnyOrDieIE(InfoExtractor):
-    _VALID_URL = r'^(?:https?://)?(?:www\.)?funnyordie\.com/videos/(?P<id>[0-9a-f]+)/.*$'
-    _TEST = {
+    _VALID_URL = r'https?://(?:www\.)?funnyordie\.com/(?P<type>embed|videos)/(?P<id>[0-9a-f]+)(?:$|[?#/])'
+    _TESTS = [{
         'url': 'http://www.funnyordie.com/videos/0732f586d7/heart-shaped-box-literal-video-version',
-        'file': '0732f586d7.mp4',
-        'md5': 'f647e9e90064b53b6e046e75d0241fbd',
+        'md5': 'bcd81e0c4f26189ee09be362ad6e6ba9',
         'info_dict': {
-            'description': ('Lyrics changed to match the video. Spoken cameo '
-                'by Obscurus Lupa (from ThatGuyWithTheGlasses.com). Based on a '
-                'concept by Dustin McLean (DustFilms.com). Performed, edited, '
-                'and written by David A. Scott.'),
+            'id': '0732f586d7',
+            'ext': 'mp4',
             'title': 'Heart-Shaped Box: Literal Video Version',
+            'description': 'md5:ea09a01bc9a1c46d9ab696c01747c338',
+            'thumbnail': 're:^http:.*\.jpg$',
         },
-    }
+    }, {
+        'url': 'http://www.funnyordie.com/embed/e402820827',
+        'md5': 'ff4d83318f89776ed0250634cfaa8d36',
+        'info_dict': {
+            'id': 'e402820827',
+            'ext': 'mp4',
+            'title': 'Please Use This Song (Jon Lajoie)',
+            'description': 'md5:2ed27d364f5a805a6dba199faaf6681d',
+            'thumbnail': 're:^http:.*\.jpg$',
+        },
+    }]
 
     def _real_extract(self, url):
         mobj = re.match(self._VALID_URL, url)
@@ -26,14 +37,34 @@ class FunnyOrDieIE(InfoExtractor):
         video_id = mobj.group('id')
         webpage = self._download_webpage(url, video_id)
 
-        video_url = self._search_regex(
-            [r'type="video/mp4" src="(.*?)"', r'src="([^>]*?)" type=\'video/mp4\''],
-            webpage, 'video URL', flags=re.DOTALL)
+        links = re.findall(r'<source src="([^"]+/v)\d+\.([^"]+)" type=\'video', webpage)
+        if not links:
+            raise ExtractorError('No media links available for %s' % video_id)
+
+        links.sort(key=lambda link: 1 if link[1] == 'mp4' else 0)
+
+        bitrates = self._html_search_regex(r'<source src="[^"]+/v,((?:\d+,)+)\.mp4\.csmil', webpage, 'video bitrates')
+        bitrates = [int(b) for b in bitrates.rstrip(',').split(',')]
+        bitrates.sort()
+
+        formats = []
+
+        for bitrate in bitrates:
+            for link in links:
+                formats.append({
+                    'url': '%s%d.%s' % (link[0], bitrate, link[1]),
+                    'format_id': '%s-%d' % (link[1], bitrate),
+                    'vbr': bitrate,
+                })
+
+        post_json = self._search_regex(
+            r'fb_post\s*=\s*(\{.*?\});', webpage, 'post details')
+        post = json.loads(post_json)
 
         return {
             'id': video_id,
-            'url': video_url,
-            'ext': 'mp4',
-            'title': self._og_search_title(webpage),
-            'description': self._og_search_description(webpage),
+            'title': post['name'],
+            'description': post.get('description'),
+            'thumbnail': post.get('picture'),
+            'formats': formats,
         }