youtube-dl/youtube_dl/extractor/vlive.py

172 lines
6.1 KiB
Python
Raw Normal View History

2015-08-20 13:56:11 +09:00
# coding: utf-8
from __future__ import unicode_literals
2015-08-20 13:56:11 +09:00
2016-04-26 23:30:24 +09:00
import re
2016-04-29 20:29:00 +09:00
2015-08-20 13:56:11 +09:00
from .common import InfoExtractor
from ..utils import (
dict_get,
2016-04-29 20:29:00 +09:00
ExtractorError,
float_or_none,
int_or_none,
remove_start,
urlencode_postdata,
2015-08-20 13:56:11 +09:00
)
from ..compat import compat_urllib_parse_urlencode
2015-08-20 13:56:11 +09:00
class VLiveIE(InfoExtractor):
IE_NAME = 'vlive'
_VALID_URL = r'https?://(?:(?:www|m)\.)?vlive\.tv/video/(?P<id>[0-9]+)'
2016-11-16 00:11:47 +09:00
_TESTS = [{
'url': 'http://www.vlive.tv/video/1326',
2015-08-20 13:56:11 +09:00
'md5': 'cc7314812855ce56de70a06a27314983',
'info_dict': {
'id': '1326',
'ext': 'mp4',
'title': "[V LIVE] Girl's Day's Broadcast",
'creator': "Girl's Day",
'view_count': int,
2015-08-20 13:56:11 +09:00
},
2016-11-16 00:11:47 +09:00
}, {
'url': 'http://www.vlive.tv/video/16937',
'info_dict': {
'id': '16937',
'ext': 'mp4',
'title': '[V LIVE] 첸백시 걍방',
'creator': 'EXO',
'view_count': int,
'subtitles': 'mincount:12',
},
'params': {
'skip_download': True,
},
}]
2015-08-20 13:56:11 +09:00
def _real_extract(self, url):
video_id = self._match_id(url)
webpage = self._download_webpage(
'http://www.vlive.tv/video/%s' % video_id, video_id)
2015-08-20 13:56:11 +09:00
VIDEO_PARAMS_RE = r'\bvlive\.video\.init\(([^)]+)'
VIDEO_PARAMS_FIELD = 'video params'
2016-12-06 06:16:12 +09:00
params = self._parse_json(self._search_regex(
VIDEO_PARAMS_RE, webpage, VIDEO_PARAMS_FIELD, default=''), video_id,
transform_source=lambda s: '[' + s + ']', fatal=False)
if not params or len(params) < 7:
params = self._search_regex(
VIDEO_PARAMS_RE, webpage, VIDEO_PARAMS_FIELD)
params = [p.strip(r'"') for p in re.split(r'\s*,\s*', params)]
status, long_video_id, key = params[2], params[5], params[6]
status = remove_start(status, 'PRODUCT_')
2016-04-26 23:30:24 +09:00
if status == 'LIVE_ON_AIR' or status == 'BIG_EVENT_ON_AIR':
2016-12-06 06:16:12 +09:00
return self._live(video_id, webpage)
2016-04-26 23:30:24 +09:00
elif status == 'VOD_ON_AIR' or status == 'BIG_EVENT_INTRO':
if long_video_id and key:
return self._replay(video_id, webpage, long_video_id, key)
else:
status = 'COMING_SOON'
if status == 'LIVE_END':
raise ExtractorError('Uploading for replay. Please wait...',
expected=True)
elif status == 'COMING_SOON':
raise ExtractorError('Coming soon!', expected=True)
2016-04-26 23:30:24 +09:00
elif status == 'CANCELED':
raise ExtractorError('We are sorry, '
'but the live broadcast has been canceled.',
expected=True)
else:
raise ExtractorError('Unknown status %s' % status)
def _get_common_fields(self, webpage):
2015-08-20 13:56:11 +09:00
title = self._og_search_title(webpage)
2016-04-26 23:30:24 +09:00
creator = self._html_search_regex(
r'<div[^>]+class="info_area"[^>]*>\s*<a\s+[^>]*>([^<]+)',
webpage, 'creator', fatal=False)
thumbnail = self._og_search_thumbnail(webpage)
return {
'title': title,
'creator': creator,
'thumbnail': thumbnail,
}
2015-09-05 16:25:19 +09:00
2016-12-06 06:16:12 +09:00
def _live(self, video_id, webpage):
init_page = self._download_webpage(
'http://www.vlive.tv/video/init/view',
video_id, note='Downloading live webpage',
data=urlencode_postdata({'videoSeq': video_id}),
headers={
2016-12-06 06:16:12 +09:00
'Referer': 'http://www.vlive.tv/video/%s' % video_id,
'Content-Type': 'application/x-www-form-urlencoded'
})
live_params = self._search_regex(
r'"liveStreamInfo"\s*:\s*(".*"),',
init_page, 'live stream info')
2016-12-06 06:16:12 +09:00
live_params = self._parse_json(live_params, video_id)
live_params = self._parse_json(live_params, video_id)
2016-04-26 23:30:24 +09:00
formats = []
for vid in live_params.get('resolutions', []):
formats.extend(self._extract_m3u8_formats(
vid['cdnUrl'], video_id, 'mp4',
m3u8_id=vid.get('name'),
fatal=False, live=True))
self._sort_formats(formats)
info = self._get_common_fields(webpage)
info.update({
'title': self._live_title(info['title']),
'id': video_id,
'formats': formats,
'is_live': True,
})
return info
2016-04-26 23:30:24 +09:00
def _replay(self, video_id, webpage, long_video_id, key):
playinfo = self._download_json(
'http://global.apis.naver.com/rmcnmv/rmcnmv/vod_play_videoInfo.json?%s'
% compat_urllib_parse_urlencode({
'videoId': long_video_id,
'key': key,
'ptc': 'http',
'doct': 'json', # document type (xml or json)
'cpt': 'vtt', # captions type (vtt or ttml)
}), video_id)
2015-08-20 13:56:11 +09:00
formats = [{
'url': vid['source'],
'format_id': vid.get('encodingOption', {}).get('name'),
'abr': float_or_none(vid.get('bitrate', {}).get('audio')),
'vbr': float_or_none(vid.get('bitrate', {}).get('video')),
'width': int_or_none(vid.get('encodingOption', {}).get('width')),
'height': int_or_none(vid.get('encodingOption', {}).get('height')),
'filesize': int_or_none(vid.get('size')),
} for vid in playinfo.get('videos', {}).get('list', []) if vid.get('source')]
2015-08-20 13:56:11 +09:00
self._sort_formats(formats)
view_count = int_or_none(playinfo.get('meta', {}).get('count'))
2015-08-20 13:56:11 +09:00
subtitles = {}
for caption in playinfo.get('captions', {}).get('list', []):
lang = dict_get(caption, ('locale', 'language', 'country', 'label'))
if lang and caption.get('source'):
subtitles[lang] = [{
'ext': 'vtt',
'url': caption['source']}]
2015-08-20 13:56:11 +09:00
info = self._get_common_fields(webpage)
info.update({
'id': video_id,
'formats': formats,
'view_count': view_count,
'subtitles': subtitles,
})
return info