# coding: utf-8
from __future__ import unicode_literals
-import json
import base64
from .common import InfoExtractor
from ..compat import (
compat_urllib_parse,
- compat_cookiejar,
- compat_cookies,
- compat_urllib_request,
compat_ord,
)
from ..utils import (
).decode('ascii').split('_')
# get oip
- oip = data2['security']['ip']
+ oip = data1['security']['ip']
# get fileid
string_ls = list(
format = stream.get('stream_type')
video_urls = []
for dt in stream['segs']:
- n = str(int(dt['size']))
+ #n = str(int(dt['size']))
+ n = str(stream['segs'].index(dt))
param = {
'K': dt['key'],
'hd': self.get_hd(format),
ext_dict = {
'flv': 'flv',
'mp4': 'mp4',
+ 'mp4hd': 'mp4',
+ 'mp4hd2': 'mp4',
'hd2': 'flv',
'hd3': 'flv',
'3gp': 'flv',
req.add_header('Ytdl-request-proxy', cn_verification_proxy)
raw_data = self._download_json(req, video_id, note=note)
- jsonDumpIn = json.dumps(raw_data,indent = 1)
return raw_data['data']
'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['stream']))]
+ } for i in range(max(len(v.get('segs')) for v in data1['stream']))]
for stream in data1['stream']:
fm = stream.get('stream_type')
video_urls = video_urls_dict[fm]