You cannot select more than 25 topics Topics must start with a letter or number, can include dashes ('-') and can be up to 35 characters long.
yt-dlp/yt_dlp/extractor/mixch.py

56 lines
2.0 KiB
Python

from __future__ import unicode_literals
from .common import InfoExtractor
from ..utils import (
ExtractorError,
traverse_obj,
)
class MixchIE(InfoExtractor):
IE_NAME = 'mixch'
_VALID_URL = r'https?://(?:www\.)?mixch\.tv/u/(?P<id>\d+)'
TESTS = [{
'url': 'https://mixch.tv/u/16236849/live',
'skip': 'don\'t know if this live persists',
'info_dict': {
'id': '16236849',
'title': '24配信シェア⭕投票🙏💦',
'comment_count': 13145,
'view_count': 28348,
'timestamp': 1636189377,
'uploader': '🦥伊咲👶🏻#フレアワ',
'uploader_id': '16236849',
}
}, {
'url': 'https://mixch.tv/u/16137876/live',
'only_matching': True,
}]
def _real_extract(self, url):
video_id = self._match_id(url)
webpage = self._download_webpage(f'https://mixch.tv/u/{video_id}/live', video_id)
initial_js_state = self._parse_json(self._search_regex(
r'(?m)^\s*window\.__INITIAL_JS_STATE__\s*=\s*(\{.+?\});\s*$', webpage, 'initial JS state'), video_id)
if not initial_js_state.get('liveInfo'):
raise ExtractorError('Livestream has ended.', expected=True)
return {
'id': video_id,
'title': traverse_obj(initial_js_state, ('liveInfo', 'title')),
'comment_count': traverse_obj(initial_js_state, ('liveInfo', 'comments')),
'view_count': traverse_obj(initial_js_state, ('liveInfo', 'visitor')),
'timestamp': traverse_obj(initial_js_state, ('liveInfo', 'created')),
'uploader': traverse_obj(initial_js_state, ('broadcasterInfo', 'name')),
'uploader_id': video_id,
'formats': [{
'format_id': 'hls',
'url': traverse_obj(initial_js_state, ('liveInfo', 'hls')) or 'https://d1hd0ww6piyb43.cloudfront.net/hls/torte_%s.m3u8' % video_id,
'ext': 'mp4',
'protocol': 'm3u8',
}],
'is_live': True,
}