aboutsummaryrefslogtreecommitdiff
diff options
context:
space:
mode:
-rw-r--r--youtube_dl/extractor/youku.py111
1 files changed, 64 insertions, 47 deletions
diff --git a/youtube_dl/extractor/youku.py b/youtube_dl/extractor/youku.py
index 69ecc837a..5110efcb4 100644
--- a/youtube_dl/extractor/youku.py
+++ b/youtube_dl/extractor/youku.py
@@ -2,6 +2,7 @@
from __future__ import unicode_literals
import base64
+import json
from .common import InfoExtractor
from ..compat import (
@@ -63,7 +64,7 @@ class YoukuIE(InfoExtractor):
},
}]
- def construct_video_urls(self, data1, data2):
+ def construct_video_urls(self, data):
# get sid, token
def yk_t(s1, s2):
ls = list(range(256))
@@ -81,34 +82,28 @@ class YoukuIE(InfoExtractor):
return bytes(s)
sid, token = yk_t(
- b'becaf9be', base64.b64decode(data2['ep'].encode('ascii'))
+ b'becaf9be', base64.b64decode(data['security']['encrypt_string'].encode('ascii'))
).decode('ascii').split('_')
# get oip
- oip = data2['ip']
+ oip = data['security']['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:]
+ for stream in data['stream']:
+ format = stream.get('stream_type')
+ fileid = stream['stream_fileid']
+ fileid_dict[format] = fileid
def get_fileid(format, n):
- fileid = fileid_dict[format] % hex(int(n))[2:].upper().zfill(2)
+ number = hex(int(str(n), 10))[2:].upper()
+ if len(number) == 1:
+ number = '0' + number
+ streamfileids = fileid_dict[format]
+ fileid = streamfileids[0:8] + number + streamfileids[10:]
return fileid
# get ep
@@ -123,15 +118,17 @@ class YoukuIE(InfoExtractor):
# generate video_urls
video_urls_dict = {}
- for format in data1['streamtypes']:
+ for stream in data['stream']:
+ format = stream.get('stream_type')
video_urls = []
- for dt in data1['segs'][format]:
- n = str(int(dt['no']))
+ for dt in stream['segs']:
+ #n = str(int(dt['size']))
+ n = str(stream['segs'].index(dt))
param = {
- 'K': dt['k'],
+ 'K': dt['key'],
'hd': self.get_hd(format),
'myp': 0,
- 'ts': dt['seconds'],
+ #'ts': dt['total_milliseconds_video'],
'ypp': 0,
'ctype': 12,
'ev': 1,
@@ -142,7 +139,7 @@ class YoukuIE(InfoExtractor):
video_url = \
'http://k.youku.com/player/getFlvPath/' + \
'sid/' + sid + \
- '_' + str(int(n) + 1).zfill(2) + \
+ '_00'+ \
'/st/' + self.parse_ext_l(format) + \
'/fileid/' + get_fileid(format, n) + '?' + \
compat_urllib_parse.urlencode(param)
@@ -158,7 +155,10 @@ class YoukuIE(InfoExtractor):
'hd2': '2',
'hd3': '3',
'3gp': '0',
- '3gphd': '1'
+ '3gphd': '1',
+ 'flvhd': '0',
+ 'mp4hd': '1',
+ 'mp4hd2': '1'
}
return hd_id_dict[fm]
@@ -166,10 +166,14 @@ class YoukuIE(InfoExtractor):
ext_dict = {
'flv': 'flv',
'mp4': 'mp4',
+ 'mp4hd': 'mp4',
+ 'mp4hd2': 'flv',
+ 'mp4hd3': 'flv',
'hd2': 'flv',
'hd3': 'flv',
'3gp': 'flv',
- '3gphd': 'mp4'
+ '3gphd': 'mp4',
+ 'flvhd': 'flv'
}
return ext_dict[fm]
@@ -177,10 +181,14 @@ class YoukuIE(InfoExtractor):
_dict = {
'3gp': 'h6',
'3gphd': 'h5',
+ 'flvhd': 'h4',
'flv': 'h4',
'mp4': 'h3',
'hd2': 'h2',
- 'hd3': 'h1'
+ 'hd3': 'h1',
+ 'mp4hd': 'h3',
+ 'mp4hd3': 'h4',
+ 'mp4hd2': 'h4'
}
return _dict[fm]
@@ -188,45 +196,53 @@ class YoukuIE(InfoExtractor):
video_id = self._match_id(url)
def retrieve_data(req_url, note):
- req = sanitized_Request(req_url)
+
+
+ headers = {
+ 'Referer': req_url,
+ }
+ self._set_cookie('youku.com','xreferrer','http://www.youku.com')
+ req = sanitized_Request(req_url,headers=headers)
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]
+ js = json.dumps(raw_data)
+
+ return raw_data['data']
+
video_password = self._downloader.params.get('videopassword', None)
# request basic data
- basic_data_url = 'http://v.youku.com/player/getPlayList/VideoIDS/%s' % video_id
+ basic_data_url = "http://play.youku.com/play/get.json?vid=%s&ct=12" % video_id
if video_password:
- basic_data_url += '?password=%s' % video_password
+ basic_data_url += '&pwd=%s' % video_password
- data1 = retrieve_data(
+ data = 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:
+
+ 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)
else:
- msg = 'Youku server reported error %i' % error_code
+ msg = 'Youku server reported error %i' % error.get('code')
if error is not None:
- msg += ': ' + error
+ msg += ': ' + error_note
raise ExtractorError(msg)
- title = data1['title']
+ #get video title
+ title = data['video']['title']
+
# generate video_urls_dict
- video_urls_dict = self.construct_video_urls(data1, data2)
+ video_urls_dict = self.construct_video_urls(data)
# construct info
entries = [{
@@ -235,10 +251,11 @@ class YoukuIE(InfoExtractor):
'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']:
+ } for i in range(max(len(v.get('segs')) for v in data['stream']))]
+ for stream in data['stream']:
+ fm = stream.get('stream_type')
video_urls = video_urls_dict[fm]
- for video_url, seg, entry in zip(video_urls, data1['segs'][fm], entries):
+ for video_url, seg, entry in zip(video_urls, stream['segs'], entries):
entry['formats'].append({
'url': video_url,
'format_id': self.get_format_name(fm),