mirror of
https://github.com/yt-dlp/yt-dlp.git
synced 2024-11-24 02:35:12 +00:00
55 lines
2.2 KiB
Python
55 lines
2.2 KiB
Python
|
from .common import InfoExtractor
|
||
|
from ..utils import js_to_json
|
||
|
|
||
|
|
||
|
class PeerTVIE(InfoExtractor):
|
||
|
IE_NAME = 'peer.tv'
|
||
|
_VALID_URL = r'https?://(?:www\.)?peer\.tv/(?:de|it|en)/(?P<id>\d+)'
|
||
|
_TESTS = [{
|
||
|
'url': 'https://www.peer.tv/de/841',
|
||
|
'info_dict': {
|
||
|
'id': '841',
|
||
|
'ext': 'mp4',
|
||
|
'title': 'Die Brunnenburg',
|
||
|
'description': 'md5:4395f6142b090338340ab88a3aae24ed',
|
||
|
},
|
||
|
}, {
|
||
|
'url': 'https://www.peer.tv/it/404',
|
||
|
'info_dict': {
|
||
|
'id': '404',
|
||
|
'ext': 'mp4',
|
||
|
'title': 'Cascate di ghiaccio in Val Gardena',
|
||
|
'description': 'md5:e8e5907f236171842674e8090e3577b8',
|
||
|
},
|
||
|
}]
|
||
|
|
||
|
def _real_extract(self, url):
|
||
|
video_id = self._match_id(url)
|
||
|
webpage = self._download_webpage(url, video_id)
|
||
|
|
||
|
video_key = self._html_search_regex(r'player\.peer\.tv/js/([a-zA-Z0-9]+)', webpage, 'video key')
|
||
|
|
||
|
js = self._download_webpage(f'https://player.peer.tv/js/{video_key}/', video_id,
|
||
|
headers={'Referer': 'https://www.peer.tv/'}, note='Downloading session id')
|
||
|
|
||
|
session_id = self._search_regex(r'["\']session_id["\']:\s*["\']([a-zA-Z0-9]+)["\']', js, 'session id')
|
||
|
|
||
|
player_webpage = self._download_webpage(
|
||
|
f'https://player.peer.tv/jsc/{video_key}/{session_id}?jsr=aHR0cHM6Ly93d3cucGVlci50di9kZS84NDE=&cs=UTF-8&mq=2&ua=0&webm=p&mp4=p&hls=1',
|
||
|
video_id, note='Downloading player webpage')
|
||
|
|
||
|
m3u8_url = self._search_regex(r'["\']playlist_url["\']:\s*(["\'][^"\']+["\'])', player_webpage, 'm3u8 url')
|
||
|
m3u8_url = self._parse_json(m3u8_url, video_id, transform_source=js_to_json)
|
||
|
|
||
|
formats = self._extract_m3u8_formats(m3u8_url, video_id, m3u8_id='hls')
|
||
|
|
||
|
self._sort_formats(formats)
|
||
|
|
||
|
return {
|
||
|
'id': video_id,
|
||
|
'title': self._html_search_regex(r'<h1>(.+?)</h1>', webpage, 'title').replace('\xa0', ' '),
|
||
|
'formats': formats,
|
||
|
'description': self._html_search_meta(('og:description', 'description'), webpage),
|
||
|
'thumbnail': self._html_search_meta(('og:image', 'image'), webpage)
|
||
|
}
|