mirror of
https://github.com/yt-dlp/yt-dlp.git
synced 2024-12-29 06:55:25 +00:00
[ufctv] add support for more domains and remove compatibility code(closes #23332)
This commit is contained in:
parent
7d53fa475a
commit
4067a23270
@ -17,19 +17,12 @@ from ..utils import (
|
||||
class ImgGamingBaseIE(InfoExtractor):
|
||||
_API_BASE = 'https://dce-frontoffice.imggaming.com/api/v2/'
|
||||
_API_KEY = '857a1e5d-e35e-4fdf-805b-a87b6f8364bf'
|
||||
_DOMAIN = None
|
||||
_HEADERS = None
|
||||
_LOGIN_REQUIRED = True
|
||||
_LOGIN_SUFFIX = ''
|
||||
_MANIFEST_HEADERS = {'Accept-Encoding': 'identity'}
|
||||
_REALM = None
|
||||
_TOKEN = None
|
||||
_VALID_URL_TEMPL = r'https?://%s/(?P<type>live|playlist|video)/(?P<id>\d+)(?:\?.*?\bplaylistId=(?P<playlist_id>\d+))?'
|
||||
_VALID_URL_TEMPL = r'https?://(?P<domain>(?:(?:app|www)\.)?%s)/(?P<type>live|playlist|video)/(?P<id>\d+)(?:\?.*?\bplaylistId=(?P<playlist_id>\d+))?'
|
||||
|
||||
def _real_initialize(self):
|
||||
if not self._LOGIN_REQUIRED:
|
||||
return
|
||||
|
||||
self._HEADERS = {
|
||||
'Realm': 'dce.' + self._REALM,
|
||||
'x-api-key': self._API_KEY,
|
||||
@ -42,7 +35,7 @@ class ImgGamingBaseIE(InfoExtractor):
|
||||
p_headers = self._HEADERS.copy()
|
||||
p_headers['Content-Type'] = 'application/json'
|
||||
self._HEADERS['Authorization'] = 'Bearer ' + self._download_json(
|
||||
self._API_BASE + 'login' + self._LOGIN_SUFFIX,
|
||||
self._API_BASE + 'login',
|
||||
None, 'Logging in', data=json.dumps({
|
||||
'id': email,
|
||||
'secret': password,
|
||||
@ -52,9 +45,6 @@ class ImgGamingBaseIE(InfoExtractor):
|
||||
return self._download_json(
|
||||
self._API_BASE + path + media_id, media_id, headers=self._HEADERS)
|
||||
|
||||
def _extract_media_id(self, url, display_id):
|
||||
return display_id
|
||||
|
||||
def _extract_dve_api_url(self, media_id, media_type):
|
||||
stream_path = 'stream'
|
||||
if media_type == 'video':
|
||||
@ -72,8 +62,7 @@ class ImgGamingBaseIE(InfoExtractor):
|
||||
raise
|
||||
|
||||
def _real_extract(self, url):
|
||||
media_type, display_id, playlist_id = re.match(self._VALID_URL, url).groups()
|
||||
media_id = self._extract_media_id(url, display_id)
|
||||
domain, media_type, media_id, playlist_id = re.match(self._VALID_URL, url).groups()
|
||||
|
||||
if playlist_id:
|
||||
if self._downloader.params.get('noplaylist'):
|
||||
@ -90,7 +79,7 @@ class ImgGamingBaseIE(InfoExtractor):
|
||||
if not video_id:
|
||||
continue
|
||||
entries.append(self.url_result(
|
||||
'https://%s/video/%s' % (self._DOMAIN, video_id),
|
||||
'https://%s/video/%s' % (domain, video_id),
|
||||
self.ie_key(), video_id))
|
||||
return self.playlist_result(
|
||||
entries, media_id, playlist.get('title'),
|
||||
@ -133,7 +122,6 @@ class ImgGamingBaseIE(InfoExtractor):
|
||||
|
||||
return {
|
||||
'id': media_id,
|
||||
'display_id': display_id,
|
||||
'title': title,
|
||||
'formats': formats,
|
||||
'thumbnail': video_data.get('thumbnailUrl'),
|
||||
|
@ -5,14 +5,12 @@ from .imggaming import ImgGamingBaseIE
|
||||
|
||||
|
||||
class UFCTVIE(ImgGamingBaseIE):
|
||||
_VALID_URL = ImgGamingBaseIE._VALID_URL_TEMPL % r'(?:www\.)?ufc\.tv'
|
||||
_VALID_URL = ImgGamingBaseIE._VALID_URL_TEMPL % r'(?:ufc\.tv|(?:ufc)?fightpass\.com)'
|
||||
_NETRC_MACHINE = 'ufctv'
|
||||
_DOMAIN = 'ufc.tv'
|
||||
_REALM = 'ufc'
|
||||
|
||||
|
||||
class UFCArabiaIE(ImgGamingBaseIE):
|
||||
_VALID_URL = ImgGamingBaseIE._VALID_URL_TEMPL % r'app\.ufcarabia\.com'
|
||||
_VALID_URL = ImgGamingBaseIE._VALID_URL_TEMPL % r'ufcarabia\.(?:ae|com)'
|
||||
_NETRC_MACHINE = 'ufcarabia'
|
||||
_DOMAIN = 'app.ufcarabia.com'
|
||||
_REALM = 'admufc'
|
||||
|
Loading…
Reference in New Issue
Block a user