mirror of
https://github.com/ytdl-org/youtube-dl.git
synced 2024-11-25 19:52:11 +00:00
83 lines
2.6 KiB
Python
83 lines
2.6 KiB
Python
# coding: utf-8
|
|
from __future__ import unicode_literals
|
|
|
|
from .common import InfoExtractor
|
|
from ..compat import compat_str
|
|
from ..utils import (
|
|
ExtractorError,
|
|
int_or_none,
|
|
try_get,
|
|
unified_timestamp,
|
|
)
|
|
|
|
|
|
class FreshLiveIE(InfoExtractor):
|
|
_VALID_URL = r'https?://freshlive\.tv/[^/]+/(?P<id>\d+)'
|
|
_TEST = {
|
|
'url': 'https://freshlive.tv/satotv/74712',
|
|
'md5': '9f0cf5516979c4454ce982df3d97f352',
|
|
'info_dict': {
|
|
'id': '74712',
|
|
'ext': 'mp4',
|
|
'title': 'テスト',
|
|
'description': 'テスト',
|
|
'thumbnail': r're:^https?://.*\.jpg$',
|
|
'duration': 1511,
|
|
'timestamp': 1483619655,
|
|
'upload_date': '20170105',
|
|
'uploader': 'サトTV',
|
|
'uploader_id': 'satotv',
|
|
'view_count': int,
|
|
'comment_count': int,
|
|
'is_live': False,
|
|
}
|
|
}
|
|
|
|
def _real_extract(self, url):
|
|
video_id = self._match_id(url)
|
|
|
|
webpage = self._download_webpage(url, video_id)
|
|
|
|
options = self._parse_json(
|
|
self._search_regex(
|
|
r'window\.__CONTEXT__\s*=\s*({.+?});\s*</script>',
|
|
webpage, 'initial context'),
|
|
video_id)
|
|
|
|
info = options['context']['dispatcher']['stores']['ProgramStore']['programs'][video_id]
|
|
|
|
title = info['title']
|
|
|
|
if info.get('status') == 'upcoming':
|
|
raise ExtractorError('Stream %s is upcoming' % video_id, expected=True)
|
|
|
|
stream_url = info.get('liveStreamUrl') or info['archiveStreamUrl']
|
|
|
|
is_live = info.get('liveStreamUrl') is not None
|
|
|
|
formats = self._extract_m3u8_formats(
|
|
stream_url, video_id, 'mp4',
|
|
'm3u8_native', m3u8_id='hls')
|
|
|
|
if is_live:
|
|
title = self._live_title(title)
|
|
|
|
return {
|
|
'id': video_id,
|
|
'formats': formats,
|
|
'title': title,
|
|
'description': info.get('description'),
|
|
'thumbnail': info.get('thumbnailUrl'),
|
|
'duration': int_or_none(info.get('airTime')),
|
|
'timestamp': unified_timestamp(info.get('createdAt')),
|
|
'uploader': try_get(
|
|
info, lambda x: x['channel']['title'], compat_str),
|
|
'uploader_id': try_get(
|
|
info, lambda x: x['channel']['code'], compat_str),
|
|
'uploader_url': try_get(
|
|
info, lambda x: x['channel']['permalink'], compat_str),
|
|
'view_count': int_or_none(info.get('viewCount')),
|
|
'comment_count': int_or_none(info.get('commentCount')),
|
|
'tags': info.get('tags', []),
|
|
'is_live': is_live,
|
|
}
|