yt-dlp/yt_dlp/extractor/zaiko.py
c-basalt 345b4c0aed
[extractor/zaiko] Add extractor (#7254)
Closes #7196
Authored by: c-basalt
2023-06-12 18:12:09 +00:00

92 lines
4 KiB
Python

from .common import InfoExtractor
from ..utils import (
ExtractorError,
extract_attributes,
int_or_none,
str_or_none,
traverse_obj,
unescapeHTML,
url_or_none,
)
class ZaikoIE(InfoExtractor):
_VALID_URL = r'https?://(?:[\w-]+\.)?zaiko\.io/event/(?P<id>\d+)/stream(?:/\d+)+'
_TESTS = [{
'url': 'https://zaiko.io/event/324868/stream/20571/20571',
'info_dict': {
'id': '324868',
'ext': 'mp4',
'title': 'ZAIKO STREAMING TEST',
'alt_title': '[VOD] ZAIKO STREAMING TEST_20210603(Do Not Delete)',
'uploader_id': '454',
'uploader': 'ZAIKO ZERO',
'release_timestamp': 1583809200,
'thumbnail': r're:https://[a-z0-9]+.cloudfront.net/[a-z0-9_]+/[a-z0-9_]+',
'release_date': '20200310',
'categories': ['Tech House'],
'live_status': 'was_live',
},
'params': {'skip_download': 'm3u8'},
}]
def _parse_vue_element_attr(self, name, string, video_id):
page_elem = self._search_regex(rf'(<{name}[^>]+>)', string, name)
attrs = {}
for key, value in extract_attributes(page_elem).items():
if key.startswith(':'):
attrs[key[1:]] = self._parse_json(
value, video_id, transform_source=unescapeHTML, fatal=False)
return attrs
def _real_extract(self, url):
video_id = self._match_id(url)
webpage, urlh = self._download_webpage_handle(url, video_id)
final_url = urlh.geturl()
if 'zaiko.io/login' in final_url:
self.raise_login_required()
elif '/_buy/' in final_url:
raise ExtractorError('Your account does not have tickets to this event', expected=True)
stream_meta = self._parse_vue_element_attr('stream-page', webpage, video_id)
player_page = self._download_webpage(
stream_meta['stream-access']['video_source'], video_id,
'Downloading player page', headers={'referer': 'https://zaiko.io/'})
player_meta = self._parse_vue_element_attr('player', player_page, video_id)
status = traverse_obj(player_meta, ('initial_event_info', 'status', {str}))
live_status, msg, expected = {
'vod': ('was_live', 'No VOD stream URL was found', False),
'archiving': ('post_live', 'Event VOD is still being processed', True),
'deleting': ('post_live', 'This event has ended', True),
'deleted': ('post_live', 'This event has ended', True),
'error': ('post_live', 'This event has ended', True),
'disconnected': ('post_live', 'Stream has been disconnected', True),
'live_to_disconnected': ('post_live', 'Stream has been disconnected', True),
'live': ('is_live', 'No livestream URL found was found', False),
'waiting': ('is_upcoming', 'Live event has not yet started', True),
'cancelled': ('not_live', 'Event has been cancelled', True),
}.get(status) or ('not_live', f'Unknown event status "{status}"', False)
stream_url = traverse_obj(player_meta, ('initial_event_info', 'endpoint', {url_or_none}))
formats = self._extract_m3u8_formats(
stream_url, video_id, live=True, fatal=False) if stream_url else []
if not formats:
self.raise_no_formats(msg, expected=expected)
return {
'id': video_id,
'formats': formats,
'live_status': live_status,
**traverse_obj(stream_meta, {
'title': ('event', 'name', {str}),
'uploader': ('profile', 'name', {str}),
'uploader_id': ('profile', 'id', {str_or_none}),
'release_timestamp': ('stream', 'start', 'timestamp', {int_or_none}),
'categories': ('event', 'genres', ..., {lambda x: x or None}),
}),
**traverse_obj(player_meta, ('initial_event_info', {
'alt_title': ('title', {str}),
'thumbnail': ('poster_url', {url_or_none}),
})),
}