]> Raphaël G. Git Repositories - youtubedl/blobdiff - youtube_dl/extractor/youku.py
Prepare to upload
[youtubedl] / youtube_dl / extractor / youku.py
index 2e81d92238307e8914769d6fc48d03befd6af2bf..61d1ab20908bf1a233cf90624170d9ee824ff337 100644 (file)
@@ -1,15 +1,18 @@
 # coding: utf-8
 from __future__ import unicode_literals
 
-import base64
+import random
+import re
+import string
+import time
 
 from .common import InfoExtractor
-from ..utils import ExtractorError
-
-from ..compat import (
-    compat_urllib_parse,
-    compat_ord,
-    compat_urllib_request,
+from ..utils import (
+    ExtractorError,
+    get_element_by_class,
+    js_to_json,
+    str_or_none,
+    strip_jsonp,
 )
 
 
@@ -18,18 +21,26 @@ class YoukuIE(InfoExtractor):
     IE_DESC = '优酷'
     _VALID_URL = r'''(?x)
         (?:
-            http://(?:v|player)\.youku\.com/(?:v_show/id_|player\.php/sid/)|
+            https?://(
+                (?:v|player)\.youku\.com/(?:v_show/id_|player\.php/sid/)|
+                video\.tudou\.com/v/)|
             youku:)
         (?P<id>[A-Za-z0-9]+)(?:\.html|/v\.swf|)
     '''
 
     _TESTS = [{
+        # MD5 is unstable
         'url': 'http://v.youku.com/v_show/id_XMTc1ODE5Njcy.html',
-        'md5': '5f3af4192eabacc4501508d54a8cabd7',
         'info_dict': {
-            'id': 'XMTc1ODE5Njcy_part1',
+            'id': 'XMTc1ODE5Njcy',
             'title': '★Smile﹗♡ Git Fresh -Booty Music舞蹈.',
-            'ext': 'flv'
+            'ext': 'mp4',
+            'duration': 74.73,
+            'thumbnail': r're:^https?://.*',
+            'uploader': '。躲猫猫、',
+            'uploader_id': '36017967',
+            'uploader_url': 'http://i.youku.com/u/UMTQ0MDcxODY4',
+            'tags': list,
         }
     }, {
         'url': 'http://player.youku.com/player.php/sid/XNDgyMDQ2NTQw/v.swf',
@@ -38,216 +49,261 @@ class YoukuIE(InfoExtractor):
         'url': 'http://v.youku.com/v_show/id_XODgxNjg1Mzk2_ev_1.html',
         'info_dict': {
             'id': 'XODgxNjg1Mzk2',
+            'ext': 'mp4',
             'title': '武媚娘传奇 85',
+            'duration': 1999.61,
+            'thumbnail': r're:^https?://.*',
+            'uploader': '疯狂豆花',
+            'uploader_id': '62583473',
+            'uploader_url': 'http://i.youku.com/u/UMjUwMzMzODky',
+            'tags': list,
         },
-        'playlist_count': 11,
     }, {
         'url': 'http://v.youku.com/v_show/id_XMTI1OTczNDM5Mg==.html',
         'info_dict': {
             'id': 'XMTI1OTczNDM5Mg',
+            'ext': 'mp4',
             'title': '花千骨 04',
+            'duration': 2363,
+            'thumbnail': r're:^https?://.*',
+            'uploader': '放剧场-花千骨',
+            'uploader_id': '772849359',
+            'uploader_url': 'http://i.youku.com/u/UMzA5MTM5NzQzNg==',
+            'tags': list,
         },
-        'playlist_count': 13,
-        'skip': 'Available in China only',
     }, {
         'url': 'http://v.youku.com/v_show/id_XNjA1NzA2Njgw.html',
         'note': 'Video protected with password',
         'info_dict': {
             'id': 'XNjA1NzA2Njgw',
+            'ext': 'mp4',
             'title': '邢義田复旦讲座之想象中的胡人—从“左衽孔子”说起',
+            'duration': 7264.5,
+            'thumbnail': r're:^https?://.*',
+            'uploader': 'FoxJin1006',
+            'uploader_id': '322014285',
+            'uploader_url': 'http://i.youku.com/u/UMTI4ODA1NzE0MA==',
+            'tags': list,
         },
-        'playlist_count': 19,
         'params': {
             'videopassword': '100600',
         },
+    }, {
+        # /play/get.json contains streams with "channel_type":"tail"
+        'url': 'http://v.youku.com/v_show/id_XOTUxMzg4NDMy.html',
+        'info_dict': {
+            'id': 'XOTUxMzg4NDMy',
+            'ext': 'mp4',
+            'title': '我的世界☆明月庄主☆车震猎杀☆杀人艺术Minecraft',
+            'duration': 702.08,
+            'thumbnail': r're:^https?://.*',
+            'uploader': '明月庄主moon',
+            'uploader_id': '38465621',
+            'uploader_url': 'http://i.youku.com/u/UMTUzODYyNDg0',
+            'tags': list,
+        },
+    }, {
+        'url': 'http://video.tudou.com/v/XMjIyNzAzMTQ4NA==.html?f=46177805',
+        'info_dict': {
+            'id': 'XMjIyNzAzMTQ4NA',
+            'ext': 'mp4',
+            'title': '卡马乔国足开大脚长传冲吊集锦',
+            'duration': 289,
+            'thumbnail': r're:^https?://.*',
+            'uploader': '阿卜杜拉之星',
+            'uploader_id': '2382249',
+            'uploader_url': 'http://i.youku.com/u/UOTUyODk5Ng==',
+            'tags': list,
+        },
+    }, {
+        'url': 'http://video.tudou.com/v/XMjE4ODI3OTg2MA==.html',
+        'only_matching': True,
     }]
 
-    def construct_video_urls(self, data1, data2):
-        # get sid, token
-        def yk_t(s1, s2):
-            ls = list(range(256))
-            t = 0
-            for i in range(256):
-                t = (t + ls[i] + compat_ord(s1[i % len(s1)])) % 256
-                ls[i], ls[t] = ls[t], ls[i]
-            s = bytearray()
-            x, y = 0, 0
-            for i in range(len(s2)):
-                y = (y + 1) % 256
-                x = (x + ls[y]) % 256
-                ls[x], ls[y] = ls[y], ls[x]
-                s.append(compat_ord(s2[i]) ^ ls[(ls[x] + ls[y]) % 256])
-            return bytes(s)
-
-        sid, token = yk_t(
-            b'becaf9be', base64.b64decode(data2['ep'].encode('ascii'))
-        ).decode('ascii').split('_')
-
-        # get oip
-        oip = data2['ip']
-
-        # get fileid
-        string_ls = list(
-            'abcdefghijklmnopqrstuvwxyzABCDEFGHIJKLMNOPQRSTUVWXYZ/\:._-1234567890')
-        shuffled_string_ls = []
-        seed = data1['seed']
-        N = len(string_ls)
-        for ii in range(N):
-            seed = (seed * 0xd3 + 0x754f) % 0x10000
-            idx = seed * len(string_ls) // 0x10000
-            shuffled_string_ls.append(string_ls[idx])
-            del string_ls[idx]
-
-        fileid_dict = {}
-        for format in data1['streamtypes']:
-            streamfileid = [
-                int(i) for i in data1['streamfileids'][format].strip('*').split('*')]
-            fileid = ''.join(
-                [shuffled_string_ls[i] for i in streamfileid])
-            fileid_dict[format] = fileid[:8] + '%s' + fileid[10:]
-
-        def get_fileid(format, n):
-            fileid = fileid_dict[format] % hex(int(n))[2:].upper().zfill(2)
-            return fileid
-
-        # get ep
-        def generate_ep(format, n):
-            fileid = get_fileid(format, n)
-            ep_t = yk_t(
-                b'bf7e5f01',
-                ('%s_%s_%s' % (sid, fileid, token)).encode('ascii')
-            )
-            ep = base64.b64encode(ep_t).decode('ascii')
-            return ep
-
-        # generate video_urls
-        video_urls_dict = {}
-        for format in data1['streamtypes']:
-            video_urls = []
-            for dt in data1['segs'][format]:
-                n = str(int(dt['no']))
-                param = {
-                    'K': dt['k'],
-                    'hd': self.get_hd(format),
-                    'myp': 0,
-                    'ts': dt['seconds'],
-                    'ypp': 0,
-                    'ctype': 12,
-                    'ev': 1,
-                    'token': token,
-                    'oip': oip,
-                    'ep': generate_ep(format, n)
-                }
-                video_url = \
-                    'http://k.youku.com/player/getFlvPath/' + \
-                    'sid/' + sid + \
-                    '_' + str(int(n) + 1).zfill(2) + \
-                    '/st/' + self.parse_ext_l(format) + \
-                    '/fileid/' + get_fileid(format, n) + '?' + \
-                    compat_urllib_parse.urlencode(param)
-                video_urls.append(video_url)
-            video_urls_dict[format] = video_urls
-
-        return video_urls_dict
-
-    def get_hd(self, fm):
-        hd_id_dict = {
-            'flv': '0',
-            'mp4': '1',
-            'hd2': '2',
-            'hd3': '3',
-            '3gp': '0',
-            '3gphd': '1'
-        }
-        return hd_id_dict[fm]
-
-    def parse_ext_l(self, fm):
-        ext_dict = {
-            'flv': 'flv',
-            'mp4': 'mp4',
-            'hd2': 'flv',
-            'hd3': 'flv',
-            '3gp': 'flv',
-            '3gphd': 'mp4'
-        }
-        return ext_dict[fm]
+    @staticmethod
+    def get_ysuid():
+        return '%d%s' % (int(time.time()), ''.join([
+            random.choice(string.ascii_letters) for i in range(3)]))
 
     def get_format_name(self, fm):
         _dict = {
             '3gp': 'h6',
             '3gphd': 'h5',
             'flv': 'h4',
+            'flvhd': 'h4',
             'mp4': 'h3',
+            'mp4hd': 'h3',
+            'mp4hd2': 'h4',
+            'mp4hd3': 'h4',
             'hd2': 'h2',
-            'hd3': 'h1'
+            'hd3': 'h1',
         }
-        return _dict[fm]
+        return _dict.get(fm)
 
     def _real_extract(self, url):
         video_id = self._match_id(url)
 
-        def retrieve_data(req_url, note):
-            req = compat_urllib_request.Request(req_url)
+        self._set_cookie('youku.com', '__ysuid', self.get_ysuid())
+        self._set_cookie('youku.com', 'xreferrer', 'http://www.youku.com')
 
-            cn_verification_proxy = self._downloader.params.get('cn_verification_proxy')
-            if cn_verification_proxy:
-                req.add_header('Ytdl-request-proxy', cn_verification_proxy)
-
-            raw_data = self._download_json(req, video_id, note=note)
-            return raw_data['data'][0]
-
-        video_password = self._downloader.params.get('videopassword', None)
+        _, urlh = self._download_webpage_handle(
+            'https://log.mmstat.com/eg.js', video_id, 'Retrieving cna info')
+        # The etag header is '"foobar"'; let's remove the double quotes
+        cna = urlh.headers['etag'][1:-1]
 
         # request basic data
-        basic_data_url = 'http://v.youku.com/player/getPlayList/VideoIDS/%s' % video_id
+        basic_data_params = {
+            'vid': video_id,
+            'ccode': '0590',
+            'client_ip': '192.168.1.1',
+            'utid': cna,
+            'client_ts': time.time() / 1000,
+        }
+
+        video_password = self._downloader.params.get('videopassword')
         if video_password:
-            basic_data_url += '?password=%s' % video_password
-
-        data1 = retrieve_data(
-            basic_data_url,
-            'Downloading JSON metadata 1')
-        data2 = retrieve_data(
-            'http://v.youku.com/player/getPlayList/VideoIDS/%s/Pf/4/ctype/12/ev/1' % video_id,
-            'Downloading JSON metadata 2')
-
-        error_code = data1.get('error_code')
-        if error_code:
-            error = data1.get('error')
-            if error is not None and '因版权原因无法观看此视频' in error:
+            basic_data_params['password'] = video_password
+
+        headers = {
+            'Referer': url,
+        }
+        headers.update(self.geo_verification_headers())
+        data = self._download_json(
+            'https://ups.youku.com/ups/get.json', video_id,
+            'Downloading JSON metadata',
+            query=basic_data_params, headers=headers)['data']
+
+        error = data.get('error')
+        if error:
+            error_note = error.get('note')
+            if error_note is not None and '因版权原因无法观看此视频' in error_note:
                 raise ExtractorError(
                     'Youku said: Sorry, this video is available in China only', expected=True)
+            elif error_note and '该视频被设为私密' in error_note:
+                raise ExtractorError(
+                    'Youku said: Sorry, this video is private', expected=True)
             else:
-                msg = 'Youku server reported error %i' % error_code
-                if error is not None:
-                    msg += ': ' + error
+                msg = 'Youku server reported error %i' % error.get('code')
+                if error_note is not None:
+                    msg += ': ' + error_note
                 raise ExtractorError(msg)
 
-        title = data1['title']
-
-        # generate video_urls_dict
-        video_urls_dict = self.construct_video_urls(data1, data2)
+        # get video title
+        video_data = data['video']
+        title = video_data['title']
 
-        # construct info
-        entries = [{
-            'id': '%s_part%d' % (video_id, i + 1),
-            'title': title,
-            'formats': [],
-            # some formats are not available for all parts, we have to detect
-            # which one has all
-        } for i in range(max(len(v) for v in data1['segs'].values()))]
-        for fm in data1['streamtypes']:
-            video_urls = video_urls_dict[fm]
-            for video_url, seg, entry in zip(video_urls, data1['segs'][fm], entries):
-                entry['formats'].append({
-                    'url': video_url,
-                    'format_id': self.get_format_name(fm),
-                    'ext': self.parse_ext_l(fm),
-                    'filesize': int(seg['size']),
-                })
+        formats = [{
+            'url': stream['m3u8_url'],
+            'format_id': self.get_format_name(stream.get('stream_type')),
+            'ext': 'mp4',
+            'protocol': 'm3u8_native',
+            'filesize': int(stream.get('size')),
+            'width': stream.get('width'),
+            'height': stream.get('height'),
+        } for stream in data['stream'] if stream.get('channel_type') != 'tail']
+        self._sort_formats(formats)
 
         return {
-            '_type': 'multi_video',
             'id': video_id,
             'title': title,
-            'entries': entries,
+            'formats': formats,
+            'duration': video_data.get('seconds'),
+            'thumbnail': video_data.get('logo'),
+            'uploader': video_data.get('username'),
+            'uploader_id': str_or_none(video_data.get('userid')),
+            'uploader_url': data.get('uploader', {}).get('homepage'),
+            'tags': video_data.get('tags'),
         }
+
+
+class YoukuShowIE(InfoExtractor):
+    _VALID_URL = r'https?://list\.youku\.com/show/id_(?P<id>[0-9a-z]+)\.html'
+    IE_NAME = 'youku:show'
+
+    _TESTS = [{
+        'url': 'http://list.youku.com/show/id_zc7c670be07ff11e48b3f.html',
+        'info_dict': {
+            'id': 'zc7c670be07ff11e48b3f',
+            'title': '花千骨 DVD版',
+            'description': 'md5:a1ae6f5618571bbeb5c9821f9c81b558',
+        },
+        'playlist_count': 50,
+    }, {
+        # Episode number not starting from 1
+        'url': 'http://list.youku.com/show/id_zefbfbd70efbfbd780bef.html',
+        'info_dict': {
+            'id': 'zefbfbd70efbfbd780bef',
+            'title': '超级飞侠3',
+            'description': 'md5:275715156abebe5ccc2a1992e9d56b98',
+        },
+        'playlist_count': 24,
+    }, {
+        # Ongoing playlist. The initial page is the last one
+        'url': 'http://list.youku.com/show/id_za7c275ecd7b411e1a19e.html',
+        'only_matching': True,
+    }, {
+        #  No data-id value.
+        'url': 'http://list.youku.com/show/id_zefbfbd61237fefbfbdef.html',
+        'only_matching': True,
+    }, {
+        #  Wrong number of reload_id.
+        'url': 'http://list.youku.com/show/id_z20eb4acaf5c211e3b2ad.html',
+        'only_matching': True,
+    }]
+
+    def _extract_entries(self, playlist_data_url, show_id, note, query):
+        query['callback'] = 'cb'
+        playlist_data = self._download_json(
+            playlist_data_url, show_id, query=query, note=note,
+            transform_source=lambda s: js_to_json(strip_jsonp(s))).get('html')
+        if playlist_data is None:
+            return [None, None]
+        drama_list = (get_element_by_class('p-drama-grid', playlist_data)
+                      or get_element_by_class('p-drama-half-row', playlist_data))
+        if drama_list is None:
+            raise ExtractorError('No episodes found')
+        video_urls = re.findall(r'<a[^>]+href="([^"]+)"', drama_list)
+        return playlist_data, [
+            self.url_result(self._proto_relative_url(video_url, 'http:'), YoukuIE.ie_key())
+            for video_url in video_urls]
+
+    def _real_extract(self, url):
+        show_id = self._match_id(url)
+        webpage = self._download_webpage(url, show_id)
+
+        entries = []
+        page_config = self._parse_json(self._search_regex(
+            r'var\s+PageConfig\s*=\s*({.+});', webpage, 'page config'),
+            show_id, transform_source=js_to_json)
+        first_page, initial_entries = self._extract_entries(
+            'http://list.youku.com/show/module', show_id,
+            note='Downloading initial playlist data page',
+            query={
+                'id': page_config['showid'],
+                'tab': 'showInfo',
+            })
+        first_page_reload_id = self._html_search_regex(
+            r'<div[^>]+id="(reload_\d+)', first_page, 'first page reload id')
+        # The first reload_id has the same items as first_page
+        reload_ids = re.findall('<li[^>]+data-id="([^"]+)">', first_page)
+        entries.extend(initial_entries)
+        for idx, reload_id in enumerate(reload_ids):
+            if reload_id == first_page_reload_id:
+                continue
+            _, new_entries = self._extract_entries(
+                'http://list.youku.com/show/episode', show_id,
+                note='Downloading playlist data page %d' % (idx + 1),
+                query={
+                    'id': page_config['showid'],
+                    'stage': reload_id,
+                })
+            if new_entries is not None:
+                entries.extend(new_entries)
+        desc = self._html_search_meta('description', webpage, fatal=False)
+        playlist_title = desc.split(',')[0] if desc else None
+        detail_li = get_element_by_class('p-intro', webpage)
+        playlist_description = get_element_by_class(
+            'intro-more', detail_li) if detail_li else None
+
+        return self.playlist_result(
+            entries, show_id, playlist_title, playlist_description)