]> gitweb @ CieloNegro.org - youtube-dl.git/commitdiff
merge upstream
authorFilippo Valsorda <filippo.valsorda@gmail.com>
Tue, 1 May 2012 16:22:08 +0000 (18:22 +0200)
committerFilippo Valsorda <filippo.valsorda@gmail.com>
Tue, 1 May 2012 16:22:08 +0000 (18:22 +0200)
youtube_dl/InfoExtractors.py

index 256d106b10409c6fbba199c038b4651bc0e2d36e..01201a145404deadfb74066d47a7296bf3ad2b5b 100644 (file)
@@ -103,8 +103,8 @@ class YoutubeIE(InfoExtractor):
        _NEXT_URL_RE = r'[\?&]next_url=([^&]+)'
        _NETRC_MACHINE = 'youtube'
        # Listed in order of quality
-       _available_formats = ['38', '37', '22', '45', '35', '44', '34', '18', '43', '6', '5', '17', '13']
-       _available_formats_prefer_free = ['38', '37', '45', '22', '44', '35', '43', '34', '18', '6', '5', '17', '13']
+       _available_formats = ['38', '37', '46', '22', '45', '35', '44', '34', '18', '43', '6', '5', '17', '13']
+       _available_formats_prefer_free = ['38', '46', '37', '45', '22', '44', '35', '43', '34', '18', '6', '5', '17', '13']
        _video_extensions = {
                '13': '3gp',
                '17': 'mp4',
@@ -115,6 +115,7 @@ class YoutubeIE(InfoExtractor):
                '43': 'webm',
                '44': 'webm',
                '45': 'webm',
+               '46': 'webm',
        }
        _video_dimensions = {
                '5': '240x400',
@@ -130,6 +131,7 @@ class YoutubeIE(InfoExtractor):
                '43': '360x640',
                '44': '480x854',
                '45': '720x1280',
+               '46': '1080x1920',
        }       
        IE_NAME = u'youtube'
 
@@ -1977,14 +1979,14 @@ class BlipTVIE(InfoExtractor):
                                        data = json_data['Post']
                                else:
                                        data = json_data
-       
+
                                upload_date = datetime.datetime.strptime(data['datestamp'], '%m-%d-%y %H:%M%p').strftime('%Y%m%d')
                                video_url = data['media']['url']
                                umobj = re.match(self._URL_EXT, video_url)
                                if umobj is None:
                                        raise ValueError('Can not determine filename extension')
                                ext = umobj.group(1)
-       
+
                                info = {
                                        'id': data['item_id'],
                                        'url': video_url,
@@ -2077,7 +2079,7 @@ class ComedyCentralIE(InfoExtractor):
 
        def report_extraction(self, episode_id):
                self._downloader.to_screen(u'[comedycentral] %s: Extracting information' % episode_id)
-       
+
        def report_config_download(self, episode_id):
                self._downloader.to_screen(u'[comedycentral] %s: Downloading configuration' % episode_id)
 
@@ -2605,7 +2607,6 @@ class MixcloudIE(InfoExtractor):
                        url_list = jsonData[fmt][bitrate]
                except TypeError: # we have no bitrate info.
                        url_list = jsonData[fmt]
-                               
                return url_list
 
        def check_urls(self, url_list):
@@ -2719,7 +2720,7 @@ class StanfordOpenClassroomIE(InfoExtractor):
                        info = {
                                'id': simplify_title(course + '_' + video),
                        }
-       
+
                        self.report_extraction(info['id'])
                        baseUrl = 'http://openclassroom.stanford.edu/MainFolder/courses/' + course + '/videos/'
                        xmlUrl = baseUrl + video + '.xml'