mirror of
https://github.com/yt-dlp/yt-dlp.git
synced 2024-12-04 17:30:49 +01:00
86e5f3ed2e
Using https://github.com/asottile/pyupgrade 1. `__future__` imports and `coding: utf-8` were removed 2. Files were rewritten with `pyupgrade --py36-plus --keep-percent-format` 3. f-strings were cherry-picked from `pyupgrade --py36-plus` Extractors are left untouched (except removing header) to avoid unnecessary merge conflicts
93 lines
3.6 KiB
Python
93 lines
3.6 KiB
Python
from .common import InfoExtractor
|
|
from ..utils import try_get
|
|
|
|
|
|
class ThetaStreamIE(InfoExtractor):
|
|
_VALID_URL = r'https?://(?:www\.)?theta\.tv/(?!video/)(?P<id>[a-z0-9-]+)'
|
|
_TESTS = [{
|
|
'url': 'https://www.theta.tv/davirus',
|
|
'skip': 'The live may have ended',
|
|
'info_dict': {
|
|
'id': 'DaVirus',
|
|
'ext': 'mp4',
|
|
'title': 'I choose you - My Community is King -👀 - YO HABLO ESPANOL - CODE DAVIRUS',
|
|
'thumbnail': r're:https://live-thumbnails-prod-theta-tv\.imgix\.net/thumbnail/.+\.jpg',
|
|
}
|
|
}, {
|
|
'url': 'https://www.theta.tv/mst3k',
|
|
'note': 'This channel is live 24/7',
|
|
'info_dict': {
|
|
'id': 'MST3K',
|
|
'ext': 'mp4',
|
|
'title': 'Mystery Science Theatre 3000 24/7 Powered by the THETA Network.',
|
|
'thumbnail': r're:https://user-prod-theta-tv\.imgix\.net/.+\.jpg',
|
|
}
|
|
}, {
|
|
'url': 'https://www.theta.tv/contv-anime',
|
|
'info_dict': {
|
|
'id': 'ConTVAnime',
|
|
'ext': 'mp4',
|
|
'title': 'CONTV ANIME 24/7. Powered by THETA Network.',
|
|
'thumbnail': r're:https://user-prod-theta-tv\.imgix\.net/.+\.jpg',
|
|
}
|
|
}]
|
|
|
|
def _real_extract(self, url):
|
|
channel_id = self._match_id(url)
|
|
info = self._download_json(f'https://api.theta.tv/v1/channel?alias={channel_id}', channel_id)['body']
|
|
|
|
m3u8_playlist = next(
|
|
data['url'] for data in info['live_stream']['video_urls']
|
|
if data.get('type') != 'embed' and data.get('resolution') in ('master', 'source'))
|
|
|
|
formats = self._extract_m3u8_formats(m3u8_playlist, channel_id, 'mp4', m3u8_id='hls', live=True)
|
|
self._sort_formats(formats)
|
|
|
|
channel = try_get(info, lambda x: x['user']['username']) # using this field instead of channel_id due to capitalization
|
|
|
|
return {
|
|
'id': channel,
|
|
'title': try_get(info, lambda x: x['live_stream']['title']),
|
|
'channel': channel,
|
|
'view_count': try_get(info, lambda x: x['live_stream']['view_count']),
|
|
'is_live': True,
|
|
'formats': formats,
|
|
'thumbnail': try_get(info, lambda x: x['live_stream']['thumbnail_url']),
|
|
}
|
|
|
|
|
|
class ThetaVideoIE(InfoExtractor):
|
|
_VALID_URL = r'https?://(?:www\.)?theta\.tv/video/(?P<id>vid[a-z0-9]+)'
|
|
_TEST = {
|
|
'url': 'https://www.theta.tv/video/vidiq6aaet3kzf799p0',
|
|
'md5': '633d8c29eb276bb38a111dbd591c677f',
|
|
'info_dict': {
|
|
'id': 'vidiq6aaet3kzf799p0',
|
|
'ext': 'mp4',
|
|
'title': 'Theta EdgeCast Tutorial',
|
|
'uploader': 'Pixiekittie',
|
|
'description': 'md5:e316253f5bdced8b5a46bb50ae60a09f',
|
|
'thumbnail': r're:https://user-prod-theta-tv\.imgix\.net/.+/vod_thumb/.+.jpg',
|
|
}
|
|
}
|
|
|
|
def _real_extract(self, url):
|
|
video_id = self._match_id(url)
|
|
info = self._download_json(f'https://api.theta.tv/v1/video/{video_id}/raw', video_id)['body']
|
|
|
|
m3u8_playlist = try_get(info, lambda x: x['video_urls'][0]['url'])
|
|
|
|
formats = self._extract_m3u8_formats(m3u8_playlist, video_id, 'mp4', m3u8_id='hls')
|
|
self._sort_formats(formats)
|
|
|
|
return {
|
|
'id': video_id,
|
|
'title': info.get('title'),
|
|
'uploader': try_get(info, lambda x: x['user']['username']),
|
|
'description': info.get('description'),
|
|
'view_count': info.get('view_count'),
|
|
'like_count': info.get('like_count'),
|
|
'formats': formats,
|
|
'thumbnail': info.get('thumbnail_url'),
|
|
}
|