mirror of
https://github.com/yt-dlp/yt-dlp.git
synced 2024-12-01 10:47:30 +01:00
[adn] fix formats extraction
This commit is contained in:
parent
d16c0121b9
commit
20e2c9de04
@ -15,6 +15,7 @@
|
|||||||
intlist_to_bytes,
|
intlist_to_bytes,
|
||||||
srt_subtitles_timecode,
|
srt_subtitles_timecode,
|
||||||
strip_or_none,
|
strip_or_none,
|
||||||
|
urljoin,
|
||||||
)
|
)
|
||||||
|
|
||||||
|
|
||||||
@ -31,25 +32,28 @@ class ADNIE(InfoExtractor):
|
|||||||
'description': 'md5:2f7b5aa76edbc1a7a92cedcda8a528d5',
|
'description': 'md5:2f7b5aa76edbc1a7a92cedcda8a528d5',
|
||||||
}
|
}
|
||||||
}
|
}
|
||||||
|
_BASE_URL = 'http://animedigitalnetwork.fr'
|
||||||
|
|
||||||
def _get_subtitles(self, sub_path, video_id):
|
def _get_subtitles(self, sub_path, video_id):
|
||||||
if not sub_path:
|
if not sub_path:
|
||||||
return None
|
return None
|
||||||
|
|
||||||
enc_subtitles = self._download_webpage(
|
enc_subtitles = self._download_webpage(
|
||||||
'http://animedigitalnetwork.fr/' + sub_path,
|
urljoin(self._BASE_URL, sub_path),
|
||||||
video_id, fatal=False)
|
video_id, fatal=False, headers={
|
||||||
|
'User-Agent': 'Mozilla/5.0 (X11; Linux x86_64; rv:53.0) Gecko/20100101 Firefox/53.0',
|
||||||
|
})
|
||||||
if not enc_subtitles:
|
if not enc_subtitles:
|
||||||
return None
|
return None
|
||||||
|
|
||||||
# http://animedigitalnetwork.fr/components/com_vodvideo/videojs/adn-vjs.min.js
|
# http://animedigitalnetwork.fr/components/com_vodvideo/videojs/adn-vjs.min.js
|
||||||
dec_subtitles = intlist_to_bytes(aes_cbc_decrypt(
|
dec_subtitles = intlist_to_bytes(aes_cbc_decrypt(
|
||||||
bytes_to_intlist(base64.b64decode(enc_subtitles[24:])),
|
bytes_to_intlist(base64.b64decode(enc_subtitles[24:])),
|
||||||
bytes_to_intlist(b'\nd\xaf\xd2J\xd0\xfc\xe1\xfc\xdf\xb61\xe8\xe1\xf0\xcc'),
|
bytes_to_intlist(b'\x1b\xe0\x29\x61\x38\x94\x24\x00\x12\xbd\xc5\x80\xac\xce\xbe\xb0'),
|
||||||
bytes_to_intlist(base64.b64decode(enc_subtitles[:24]))
|
bytes_to_intlist(base64.b64decode(enc_subtitles[:24]))
|
||||||
))
|
))
|
||||||
subtitles_json = self._parse_json(
|
subtitles_json = self._parse_json(
|
||||||
dec_subtitles[:-compat_ord(dec_subtitles[-1])],
|
dec_subtitles[:-compat_ord(dec_subtitles[-1])].decode(),
|
||||||
None, fatal=False)
|
None, fatal=False)
|
||||||
if not subtitles_json:
|
if not subtitles_json:
|
||||||
return None
|
return None
|
||||||
@ -103,9 +107,16 @@ def _real_extract(self, url):
|
|||||||
metas = options.get('metas') or {}
|
metas = options.get('metas') or {}
|
||||||
title = metas.get('title') or video_info['title']
|
title = metas.get('title') or video_info['title']
|
||||||
links = player_config.get('links') or {}
|
links = player_config.get('links') or {}
|
||||||
|
if not links:
|
||||||
|
links_url = player_config['linksurl']
|
||||||
|
links_data = self._download_json(urljoin(
|
||||||
|
self._BASE_URL, links_url), video_id)
|
||||||
|
links = links_data.get('links') or {}
|
||||||
|
|
||||||
formats = []
|
formats = []
|
||||||
for format_id, qualities in links.items():
|
for format_id, qualities in links.items():
|
||||||
|
if not isinstance(qualities, dict):
|
||||||
|
continue
|
||||||
for load_balancer_url in qualities.values():
|
for load_balancer_url in qualities.values():
|
||||||
load_balancer_data = self._download_json(
|
load_balancer_data = self._download_json(
|
||||||
load_balancer_url, video_id, fatal=False) or {}
|
load_balancer_url, video_id, fatal=False) or {}
|
||||||
|
Loading…
Reference in New Issue
Block a user