]> Raphaël G. Git Repositories - youtubedl/blobdiff - youtube_dl/extractor/tudou.py
Update upstream source from tag 'upstream/2018.03.14'
[youtubedl] / youtube_dl / extractor / tudou.py
index f56b66d06f2f7323b60d2fc4788d26cb3274ef9e..7421378a8c771543b3418a7d1f54b40cacb324e1 100644 (file)
@@ -3,109 +3,6 @@
 from __future__ import unicode_literals
 
 from .common import InfoExtractor
-from ..compat import compat_str
-from ..utils import (
-    int_or_none,
-    float_or_none,
-    unescapeHTML,
-)
-
-
-class TudouIE(InfoExtractor):
-    IE_NAME = 'tudou'
-    _VALID_URL = r'https?://(?:www\.)?tudou\.com/(?:(?:programs|wlplay)/view|(?:listplay|albumplay)/[\w-]{11})/(?P<id>[\w-]{11})'
-    _TESTS = [{
-        'url': 'http://www.tudou.com/listplay/zzdE77v6Mmo/2xN2duXMxmw.html',
-        'md5': '140a49ed444bd22f93330985d8475fcb',
-        'info_dict': {
-            'id': '159448201',
-            'ext': 'f4v',
-            'title': '卡马乔国足开大脚长传冲吊集锦',
-            'thumbnail': 're:^https?://.*\.jpg$',
-            'timestamp': 1372113489000,
-            'description': '卡马乔卡家军,开大脚先进战术不完全集锦!',
-            'duration': 289.04,
-            'view_count': int,
-            'filesize': int,
-        }
-    }, {
-        'url': 'http://www.tudou.com/programs/view/ajX3gyhL0pc/',
-        'info_dict': {
-            'id': '117049447',
-            'ext': 'f4v',
-            'title': 'La Sylphide-Bolshoi-Ekaterina Krysanova & Vyacheslav Lopatin 2012',
-            'thumbnail': 're:^https?://.*\.jpg$',
-            'timestamp': 1349207518000,
-            'description': 'md5:294612423894260f2dcd5c6c04fe248b',
-            'duration': 5478.33,
-            'view_count': int,
-            'filesize': int,
-        }
-    }]
-
-    _PLAYER_URL = 'http://js.tudouui.com/bin/lingtong/PortalPlayer_177.swf'
-
-    def _url_for_id(self, video_id, quality=None):
-        info_url = 'http://v2.tudou.com/f?id=' + compat_str(video_id)
-        if quality:
-            info_url += '&hd' + quality
-        xml_data = self._download_xml(info_url, video_id, 'Opening the info XML page')
-        final_url = xml_data.text
-        return final_url
-
-    def _real_extract(self, url):
-        video_id = self._match_id(url)
-        item_data = self._download_json(
-            'http://www.tudou.com/tvp/getItemInfo.action?ic=%s' % video_id, video_id)
-
-        youku_vcode = item_data.get('vcode')
-        if youku_vcode:
-            return self.url_result('youku:' + youku_vcode, ie='Youku')
-
-        title = unescapeHTML(item_data['kw'])
-        description = item_data.get('desc')
-        thumbnail_url = item_data.get('pic')
-        view_count = int_or_none(item_data.get('playTimes'))
-        timestamp = int_or_none(item_data.get('pt'))
-
-        segments = self._parse_json(item_data['itemSegs'], video_id)
-        # It looks like the keys are the arguments that have to be passed as
-        # the hd field in the request url, we pick the higher
-        # Also, filter non-number qualities (see issue #3643).
-        quality = sorted(filter(lambda k: k.isdigit(), segments.keys()),
-                         key=lambda k: int(k))[-1]
-        parts = segments[quality]
-        result = []
-        len_parts = len(parts)
-        if len_parts > 1:
-            self.to_screen('%s: found %s parts' % (video_id, len_parts))
-        for part in parts:
-            part_id = part['k']
-            final_url = self._url_for_id(part_id, quality)
-            ext = (final_url.split('?')[0]).split('.')[-1]
-            part_info = {
-                'id': '%s' % part_id,
-                'url': final_url,
-                'ext': ext,
-                'title': title,
-                'thumbnail': thumbnail_url,
-                'description': description,
-                'view_count': view_count,
-                'timestamp': timestamp,
-                'duration': float_or_none(part.get('seconds'), 1000),
-                'filesize': int_or_none(part.get('size')),
-                'http_headers': {
-                    'Referer': self._PLAYER_URL,
-                },
-            }
-            result.append(part_info)
-
-        return {
-            '_type': 'multi_video',
-            'entries': result,
-            'id': video_id,
-            'title': title,
-        }
 
 
 class TudouPlaylistIE(InfoExtractor):