0
0
Fork 0
mirror of https://github.com/yt-dlp/yt-dlp.git synced 2024-11-22 02:15:12 +00:00
yt-dlp/yt_dlp/extractor/airtv.py
HobbyistDev e318b5b87a
[extractor/airtv] Add extractor (#5533)
Authored by: HobbyistDev
Closes #5132
2022-12-10 13:59:13 +05:30

96 lines
4 KiB
Python

from .common import InfoExtractor
from .youtube import YoutubeIE
from ..utils import (
determine_ext,
int_or_none,
mimetype2ext,
parse_iso8601,
traverse_obj
)
class AirTVIE(InfoExtractor):
_VALID_URL = r'https?://www\.air\.tv/watch\?v=(?P<id>\w+)'
_TESTS = [{
# without youtube_id
'url': 'https://www.air.tv/watch?v=W87jcWleSn2hXZN47zJZsQ',
'info_dict': {
'id': 'W87jcWleSn2hXZN47zJZsQ',
'ext': 'mp4',
'release_date': '20221003',
'release_timestamp': 1664792603,
'channel_id': 'vgfManQlRQKgoFQ8i8peFQ',
'title': 'md5:c12d49ed367c3dadaa67659aff43494c',
'upload_date': '20221003',
'duration': 151,
'view_count': int,
'thumbnail': 'https://cdn-sp-gcs.air.tv/videos/W/8/W87jcWleSn2hXZN47zJZsQ/b13fc56464f47d9d62a36d110b9b5a72-4096x2160_9.jpg',
'timestamp': 1664792603,
}
}, {
# with youtube_id
'url': 'https://www.air.tv/watch?v=sv57EC8tRXG6h8dNXFUU1Q',
'info_dict': {
'id': '2ZTqmpee-bQ',
'ext': 'mp4',
'comment_count': int,
'tags': 'count:11',
'channel_follower_count': int,
'like_count': int,
'uploader': 'Newsflare',
'thumbnail': 'https://i.ytimg.com/vi_webp/2ZTqmpee-bQ/maxresdefault.webp',
'availability': 'public',
'title': 'Geese Chase Alligator Across Golf Course',
'uploader_id': 'NewsflareBreaking',
'channel_url': 'https://www.youtube.com/channel/UCzSSoloGEz10HALUAbYhngQ',
'description': 'md5:99b21d9cea59330149efbd9706e208f5',
'age_limit': 0,
'channel_id': 'UCzSSoloGEz10HALUAbYhngQ',
'uploader_url': 'http://www.youtube.com/user/NewsflareBreaking',
'view_count': int,
'categories': ['News & Politics'],
'live_status': 'not_live',
'playable_in_embed': True,
'channel': 'Newsflare',
'duration': 37,
'upload_date': '20180511',
}
}]
def _get_formats_and_subtitle(self, json_data, video_id):
formats, subtitles = [], {}
for source in traverse_obj(json_data, 'sources', 'sources_desktop', ...):
ext = determine_ext(source.get('src'), mimetype2ext(source.get('type')))
if ext == 'm3u8':
fmts, subs = self._extract_m3u8_formats_and_subtitles(source.get('src'), video_id)
formats.extend(fmts)
self._merge_subtitles(subs, target=subtitles)
else:
formats.append({'url': source.get('src'), 'ext': ext})
return formats, subtitles
def _real_extract(self, url):
display_id = self._match_id(url)
webpage = self._download_webpage(url, display_id)
nextjs_json = self._search_nextjs_data(webpage, display_id)['props']['pageProps']['initialState']['videos'][display_id]
if nextjs_json.get('youtube_id'):
return self.url_result(
f'https://www.youtube.com/watch?v={nextjs_json.get("youtube_id")}', YoutubeIE)
formats, subtitles = self._get_formats_and_subtitle(nextjs_json, display_id)
return {
'id': display_id,
'title': nextjs_json.get('title') or self._html_search_meta('og:title', webpage),
'formats': formats,
'subtitles': subtitles,
'description': nextjs_json.get('description') or None,
'duration': int_or_none(nextjs_json.get('duration')),
'thumbnails': [
{'url': thumbnail}
for thumbnail in traverse_obj(nextjs_json, ('default_thumbnails', ...))],
'channel_id': traverse_obj(nextjs_json, 'channel', 'channel_slug'),
'timestamp': parse_iso8601(nextjs_json.get('created')),
'release_timestamp': parse_iso8601(nextjs_json.get('published')),
'view_count': int_or_none(nextjs_json.get('views')),
}