yt-dlp/yt_dlp/extractor/pialive.py
2024-08-21 11:43:45 +08:00

56 lines
2.1 KiB
Python

from .common import InfoExtractor
from .piaulizaportal import PIAULIZAPortalAPIIE
from ..networking import Request
from ..utils import multipart_encode, smuggle_url
class PiaLiveIE(InfoExtractor):
PLAYER_ROOT_URL = 'https://player.pia-live.jp'
PIA_LIVE_API_URL = 'https://api.pia-live.jp'
_VALID_URL = r'https?://player\.pia-live\.jp/stream/(?P<id>[\w-]+)'
def _real_extract(self, url):
video_key = self._match_id(url)
webpage = self._download_webpage(url, video_key)
program_code = self._search_regex(r"const programCode = '(.*?)';", webpage, 'program code')
prod_configure = self._download_webpage(
self.PLAYER_ROOT_URL + self._search_regex(r'<script [^>]*\bsrc="(/statics/js/s_prod[^"]+)"', webpage, 'prod configure page'),
program_code,
headers={'Referer': self.PLAYER_ROOT_URL},
note='Fetching prod configure page', errnote='Unable to fetch prod configure page',
)
api_key = self._search_regex(r"const APIKEY = '(.*?)';", prod_configure, 'api key')
payload, content_type = multipart_encode({
'play_url': video_key,
'api_key': api_key,
})
player_tag_list = self._download_json(
Request(
url=f'{self.PIA_LIVE_API_URL}/perf/player-tag-list/{program_code}',
method='POST',
data=payload,
headers={
'Content-Type': content_type,
},
),
program_code,
)
player_data_url = self._search_regex(PIAULIZAPortalAPIIE.SCRIPT_TAG_REGEX_PATTERN,
player_tag_list['data']['movie_one_tag'], 'player data url')
return {
**self.url_result(
smuggle_url(
player_data_url,
{'video_id': program_code, 'referer': self.PLAYER_ROOT_URL, 'info_dict': {
'id': program_code,
'title': self._html_extract_title(webpage),
}},
),
ie=PIAULIZAPortalAPIIE.ie_key(),
),
}