1
0
mirror of https://github.com/l1ving/youtube-dl synced 2024-11-22 01:22:54 +08:00
youtube-dl/youtube_dl/extractor/vlive.py

87 lines
2.9 KiB
Python
Raw Normal View History

2015-08-20 12:56:11 +08:00
# coding: utf-8
from __future__ import unicode_literals
import hmac
from hashlib import sha1
from base64 import b64encode
from time import time
from .common import InfoExtractor
from ..utils import (
ExtractorError,
determine_ext
)
from ..compat import compat_urllib_parse
class VLiveIE(InfoExtractor):
IE_NAME = 'vlive'
# www.vlive.tv/video/ links redirect to m.vlive.tv/video/ for mobile devices
2015-08-20 12:56:11 +08:00
_VALID_URL = r'https?://(?:(www|m)\.)?vlive\.tv/video/(?P<id>[0-9]+)'
_TEST = {
'url': 'http://m.vlive.tv/video/1326',
'md5': 'cc7314812855ce56de70a06a27314983',
'info_dict': {
'id': '1326',
'ext': 'mp4',
'title': '[V] Girl\'s Day\'s Broadcast',
'creator': 'Girl\'s Day',
},
}
_SECRET = 'rFkwZet6pqk1vQt6SxxUkAHX7YL3lmqzUMrU4IDusTo4jEBdtOhNfT4BYYAdArwH'
def _real_extract(self, url):
video_id = self._match_id(url)
webpage = self._download_webpage(
'http://m.vlive.tv/video/%s' % video_id,
video_id, note='Download video page')
title = self._og_search_title(webpage)
thumbnail = self._og_search_thumbnail(webpage)
creator = self._html_search_regex(
2015-08-31 12:46:28 +08:00
r'<span[^>]+class="name">([^<>]+)</span>', webpage, 'creator')
2015-09-05 15:25:19 +08:00
2015-08-20 12:56:11 +08:00
url = 'http://global.apis.naver.com/globalV/globalV/vod/%s/playinfo?' % video_id
msgpad = '%.0f' % (time() * 1000)
md = b64encode(
hmac.new(self._SECRET.encode('ascii'),
(url[:255] + msgpad).encode('ascii'), sha1).digest()
)
url += '&' + compat_urllib_parse.urlencode({'msgpad': msgpad, 'md': md})
2015-08-20 12:56:11 +08:00
playinfo = self._download_json(url, video_id, 'Downloading video json')
if playinfo.get('message', '') != 'success':
2015-08-31 12:46:28 +08:00
raise ExtractorError(playinfo.get('message', 'JSON request unsuccessful'))
2015-08-20 12:56:11 +08:00
if not playinfo.get('result'):
raise ExtractorError('No videos found.')
formats = []
for vid in playinfo['result'].get('videos', {}).get('list', []):
formats.append({
'url': vid['source'],
'ext': 'mp4',
'abr': vid.get('bitrate', {}).get('audio'),
'vbr': vid.get('bitrate', {}).get('video'),
'format_id': vid['encodingOption']['name'],
'height': vid.get('height'),
'width': vid.get('width'),
})
self._sort_formats(formats)
subtitles = {}
for caption in playinfo['result'].get('captions', {}).get('list', []):
subtitles[caption['language']] = [
{'ext': determine_ext(caption['source'], default_ext='vtt'),
'url': caption['source']}]
return {
'id': video_id,
'title': title,
'creator': creator,
'thumbnail': thumbnail,
'formats': formats,
'subtitles': subtitles,
}