[Letv] Add new extractor. Single video only

This commit is contained in:
Yen Chi Hsuan 2015-02-24 23:55:57 +08:00
parent 4f3b21e1c7
commit 7f09a662a0
2 changed files with 121 additions and 0 deletions

View file

@ -237,6 +237,7 @@ from .krasview import KrasViewIE
from .ku6 import Ku6IE
from .la7 import LA7IE
from .laola1tv import Laola1TvIE
from .letv import LetvIE
from .lifenews import LifeNewsIE
from .liveleak import LiveLeakIE
from .livestream import (

View file

@ -0,0 +1,120 @@
# coding: utf-8
from __future__ import unicode_literals
import os.path
import time
import datetime
from .common import InfoExtractor
from ..compat import (compat_urlparse, compat_urllib_parse)
from ..utils import (ExtractorError, parse_iso8601)
class LetvIE(InfoExtractor):
_VALID_URL = r'http://www.letv.com/ptv/vplay/(?P<id>\d+).html'
_TESTS = [{
'url': 'http://www.letv.com/ptv/vplay/22005890.html',
'md5': 'cab23bd68d5a8db9be31c9a222c1e8df',
'info_dict': {
'id': '22005890',
'ext': 'mp4',
'title': '第87届奥斯卡颁奖礼完美落幕 《鸟人》成最大赢家',
'timestamp': 1424747397,
'upload_date': '20150224',
}
}, {
'url': 'http://www.letv.com/ptv/vplay/1118082.html',
'info_dict': {
'id': '1118082',
'ext': 'mp4',
}
}]
@staticmethod
def urshift(val, n):
return val >> n if val >= 0 else (val + 0x100000000) >> n
# ror() and calcTimeKey() are reversed from a embedded swf file in KLetvPlayer.swf
def ror(self, param1, param2):
_loc3_ = 0
while _loc3_ < param2:
param1 = self.urshift(param1, 1) + ((param1 & 1) << 31)
_loc3_ += 1
return param1
def calcTimeKey(self, param1):
_loc2_ = 773625421
_loc3_ = self.ror(param1, _loc2_ % 13)
_loc3_ = _loc3_ ^ _loc2_
_loc3_ = self.ror(_loc3_, _loc2_ % 17)
return _loc3_
def _real_extract(self, url):
media_id = self._match_id(url)
page = self._download_webpage(url, media_id)
params = {
'id': media_id,
'platid': 1,
'splatid': 101,
'format': 1,
'tkey': self.calcTimeKey(int(time.time())),
'domain': 'www.letv.com'
}
play_json = self._download_json(
'http://api.letv.com/mms/out/video/playJson?' + compat_urllib_parse.urlencode(params),
media_id, 'playJson data')
# Check for errors
playstatus = play_json['playstatus']
if playstatus['status'] == 0:
flag = playstatus['flag']
if flag == 1:
msg = 'Country %s auth error' % playstatus['country']
else:
msg = 'Generic error. flag = %d' % flag
raise ExtractorError(msg, expected=True)
playurl = play_json['playurl']
formats = ['350', '1000', '1300', '720p', '1080p']
dispatch = playurl['dispatch']
urls = []
for format_id in formats:
if format_id in dispatch:
media_url = playurl['domain'][0] + dispatch[format_id][0]
# Mimic what flvxz.com do
url_parts = list(compat_urlparse.urlparse(media_url))
qs = dict(compat_urlparse.parse_qs(url_parts[4]))
qs.update({
'platid': '14',
'splatid': '1401',
'tss': 'no',
'retry': 1
})
url_parts[4] = compat_urllib_parse.urlencode(qs)
media_url = compat_urlparse.urlunparse(url_parts)
url_info_dict = {
'url': media_url,
'ext': os.path.splitext(dispatch[format_id][1])[1][1:]
}
if format_id[-1:] == 'p':
url_info_dict['height'] = format_id[:-1]
urls.append(url_info_dict)
publish_time = parse_iso8601(self._html_search_regex(
r'发布时间&nbsp;([^<>]+) ', page, 'publish time', fatal=False),
delimiter=' ', timezone=datetime.timedelta(hours=8))
return {
'id': media_id,
'formats': urls,
'title': playurl['title'],
'thumbnail': playurl['pic'],
'timestamp': publish_time,
}