mirror of
https://github.com/yt-dlp/yt-dlp.git
synced 2024-11-09 12:42:28 +01:00
58 lines
2.1 KiB
Python
58 lines
2.1 KiB
Python
|
import random
|
||
|
|
||
|
from .common import InfoExtractor
|
||
|
from ..utils import ExtractorError, traverse_obj
|
||
|
|
||
|
|
||
|
class CamsodaIE(InfoExtractor):
|
||
|
_VALID_URL = r'https?://www\.camsoda\.com/(?P<id>[\w-]+)'
|
||
|
_TESTS = [{
|
||
|
'url': 'https://www.camsoda.com/lizzhopf',
|
||
|
'info_dict': {
|
||
|
'id': 'lizzhopf',
|
||
|
'ext': 'mp4',
|
||
|
'title': 'lizzhopf (lizzhopf) Nude on Cam. Free Live Sex Chat Room - CamSoda',
|
||
|
'description': str,
|
||
|
'is_live': True,
|
||
|
'age_limit': 18,
|
||
|
},
|
||
|
'skip': 'Room is offline',
|
||
|
}]
|
||
|
|
||
|
def _real_extract(self, url):
|
||
|
video_id = self._match_id(url)
|
||
|
webpage = self._download_webpage(url, video_id, headers=self.geo_verification_headers())
|
||
|
|
||
|
data = self._download_json(
|
||
|
f'https://camsoda.com/api/v1/video/vtoken/{video_id}', video_id,
|
||
|
query={'username': f'guest_{random.randrange(10000, 99999)}'},
|
||
|
headers=self.geo_verification_headers())
|
||
|
if not data:
|
||
|
raise ExtractorError('Unable to find configuration for stream.')
|
||
|
elif data.get('private_servers'):
|
||
|
raise ExtractorError('Model is in private show.', expected=True)
|
||
|
elif not data.get('stream_name'):
|
||
|
raise ExtractorError('Model is offline.', expected=True)
|
||
|
|
||
|
stream_name = traverse_obj(data, 'stream_name', expected_type=str)
|
||
|
token = traverse_obj(data, 'token', expected_type=str)
|
||
|
|
||
|
formats = []
|
||
|
for server in traverse_obj(data, ('edge_servers', ...)):
|
||
|
formats = self._extract_m3u8_formats(
|
||
|
f'https://{server}/{stream_name}_v1/index.m3u8?token={token}',
|
||
|
video_id, ext='mp4', m3u8_id='hls', fatal=False, live=True)
|
||
|
if formats:
|
||
|
break
|
||
|
if not formats:
|
||
|
self.raise_no_formats('No active streams found', expected=True)
|
||
|
|
||
|
return {
|
||
|
'id': video_id,
|
||
|
'title': self._html_extract_title(webpage),
|
||
|
'description': self._html_search_meta('description', webpage, default=None),
|
||
|
'is_live': True,
|
||
|
'formats': formats,
|
||
|
'age_limit': 18,
|
||
|
}
|