]> Raphaƫl G. Git Repositories - youtubedl/blobdiff - youtube_dl/extractor/ted.py
debian/copyright: Fix missing-license-paragraph-in-dep5-copyright.
[youtubedl] / youtube_dl / extractor / ted.py
index d260c91c2172deabb697c3ad242cfda24f395d41..f8a87afdaf4d27c59b4b29491569b243331b2322 100644 (file)
@@ -27,17 +27,18 @@ class TEDIE(SubtitlesInfoExtractor):
         '''
     _TESTS = [{
         'url': 'http://www.ted.com/talks/dan_dennett_on_our_consciousness.html',
-        'md5': '4ea1dada91e4174b53dac2bb8ace429d',
+        'md5': 'fc94ac279feebbce69f21c0c6ee82810',
         'info_dict': {
             'id': '102',
             'ext': 'mp4',
             'title': 'The illusion of consciousness',
             'description': ('Philosopher Dan Dennett makes a compelling '
-                'argument that not only don\'t we understand our own '
-                'consciousness, but that half the time our brains are '
-                'actively fooling us.'),
+                            'argument that not only don\'t we understand our own '
+                            'consciousness, but that half the time our brains are '
+                            'actively fooling us.'),
             'uploader': 'Dan Dennett',
             'width': 854,
+            'duration': 1308,
         }
     }, {
         'url': 'http://www.ted.com/watch/ted-institute/ted-bcg/vishal-sikka-the-beauty-and-power-of-algorithms',
@@ -51,13 +52,36 @@ class TEDIE(SubtitlesInfoExtractor):
         }
     }, {
         'url': 'http://www.ted.com/talks/gabby_giffords_and_mark_kelly_be_passionate_be_courageous_be_your_best',
-        'md5': '49144e345a899b8cb34d315f3b9cfeeb',
         'info_dict': {
             'id': '1972',
             'ext': 'mp4',
             'title': 'Be passionate. Be courageous. Be your best.',
             'uploader': 'Gabby Giffords and Mark Kelly',
             'description': 'md5:5174aed4d0f16021b704120360f72b92',
+            'duration': 1128,
+        },
+    }, {
+        'url': 'http://www.ted.com/playlists/who_are_the_hackers',
+        'info_dict': {
+            'id': '10',
+            'title': 'Who are the hackers?',
+        },
+        'playlist_mincount': 6,
+    }, {
+        # contains a youtube video
+        'url': 'https://www.ted.com/talks/douglas_adams_parrots_the_universe_and_everything',
+        'add_ie': ['Youtube'],
+        'info_dict': {
+            'id': '_ZG8HBuDjgc',
+            'ext': 'mp4',
+            'title': 'Douglas Adams: Parrots the Universe and Everything',
+            'description': 'md5:01ad1e199c49ac640cb1196c0e9016af',
+            'uploader': 'University of California Television (UCTV)',
+            'uploader_id': 'UCtelevision',
+            'upload_date': '20080522',
+        },
+        'params': {
+            'skip_download': True,
         },
     }]
 
@@ -69,7 +93,7 @@ class TEDIE(SubtitlesInfoExtractor):
 
     def _extract_info(self, webpage):
         info_json = self._search_regex(r'q\("\w+.init",({.+})\)</script>',
-            webpage, 'info json')
+                                       webpage, 'info json')
         return json.loads(info_json)
 
     def _real_extract(self, url):
@@ -89,7 +113,7 @@ class TEDIE(SubtitlesInfoExtractor):
         '''Returns the videos of the playlist'''
 
         webpage = self._download_webpage(url, name,
-            'Downloading playlist webpage')
+                                         'Downloading playlist webpage')
         info = self._extract_info(webpage)
         playlist_info = info['playlist']
 
@@ -108,6 +132,13 @@ class TEDIE(SubtitlesInfoExtractor):
 
         talk_info = self._extract_info(webpage)['talks'][0]
 
+        if talk_info.get('external') is not None:
+            self.to_screen('Found video from %s' % talk_info['external']['service'])
+            return {
+                '_type': 'url',
+                'url': talk_info['external']['uri'],
+            }
+
         formats = [{
             'url': format_url,
             'format_id': format_id,
@@ -143,12 +174,13 @@ class TEDIE(SubtitlesInfoExtractor):
             thumbnail = 'http://' + thumbnail
         return {
             'id': video_id,
-            'title': talk_info['title'],
+            'title': talk_info['title'].strip(),
             'uploader': talk_info['speaker'],
             'thumbnail': thumbnail,
             'description': self._og_search_description(webpage),
             'subtitles': video_subtitles,
             'formats': formats,
+            'duration': talk_info.get('duration'),
         }
 
     def _get_available_subtitles(self, video_id, talk_info):