mirror of
https://github.com/yt-dlp/yt-dlp.git
synced 2024-11-23 19:33:59 +01:00
[extractor/mocha] Add extractor (#4213)
Closes https://github.com/yt-dlp/yt-dlp/issues/3752 Authored by: HobbyistDev
This commit is contained in:
parent
17a23f0930
commit
65ea4cba29
|
@ -948,6 +948,7 @@
|
|||
)
|
||||
from .mlssoccer import MLSSoccerIE
|
||||
from .mnet import MnetIE
|
||||
from .mocha import MochaVideoIE
|
||||
from .moevideo import MoeVideoIE
|
||||
from .mofosex import (
|
||||
MofosexIE,
|
||||
|
|
66
yt_dlp/extractor/mocha.py
Normal file
66
yt_dlp/extractor/mocha.py
Normal file
|
@ -0,0 +1,66 @@
|
|||
from .common import InfoExtractor
|
||||
from ..utils import int_or_none, traverse_obj
|
||||
|
||||
|
||||
class MochaVideoIE(InfoExtractor):
|
||||
_VALID_URL = r'https?://video.mocha.com.vn/(?P<video_slug>[\w-]+)'
|
||||
_TESTS = [{
|
||||
'url': 'http://video.mocha.com.vn/chuyen-meo-gia-su-tu-thong-diep-cuoc-song-v18694039',
|
||||
'info_dict': {
|
||||
'id': '18694039',
|
||||
'title': 'Chuyện mèo giả sư tử | Thông điệp cuộc sống',
|
||||
'ext': 'mp4',
|
||||
'view_count': int,
|
||||
'like_count': int,
|
||||
'dislike_count': int,
|
||||
'display_id': 'chuyen-meo-gia-su-tu-thong-diep-cuoc-song',
|
||||
'thumbnail': 'http://mcvideomd1fr.keeng.net/playnow/images/20220505/ad0a055d-2f69-42ca-b888-4790041fe6bc_640x480.jpg',
|
||||
'description': '',
|
||||
'duration': 70,
|
||||
'timestamp': 1652254203,
|
||||
'upload_date': '20220511',
|
||||
'comment_count': int,
|
||||
'categories': ['Kids']
|
||||
}
|
||||
}]
|
||||
|
||||
def _real_extract(self, url):
|
||||
video_slug = self._match_valid_url(url).group('video_slug')
|
||||
json_data = self._download_json(
|
||||
'http://apivideo.mocha.com.vn:8081/onMediaBackendBiz/mochavideo/getVideoDetail',
|
||||
video_slug, query={'url': url, 'token': ''})['data']['videoDetail']
|
||||
video_id = str(json_data['id'])
|
||||
video_urls = (json_data.get('list_resolution') or []) + [json_data.get('original_path')]
|
||||
|
||||
formats, subtitles = [], {}
|
||||
for video in video_urls:
|
||||
if isinstance(video, str):
|
||||
formats.extend([{'url': video, 'ext': 'mp4'}])
|
||||
else:
|
||||
fmts, subs = self._extract_m3u8_formats_and_subtitles(
|
||||
video.get('video_path'), video_id, ext='mp4')
|
||||
formats.extend(fmts)
|
||||
self._merge_subtitles(subs, target=subtitles)
|
||||
|
||||
self._sort_formats(formats)
|
||||
|
||||
return {
|
||||
'id': video_id,
|
||||
'display_id': json_data.get('slug') or video_slug,
|
||||
'title': json_data.get('name'),
|
||||
'formats': formats,
|
||||
'subtitles': subtitles,
|
||||
'description': json_data.get('description'),
|
||||
'duration': json_data.get('durationS'),
|
||||
'view_count': json_data.get('total_view'),
|
||||
'like_count': json_data.get('total_like'),
|
||||
'dislike_count': json_data.get('total_unlike'),
|
||||
'thumbnail': json_data.get('image_path_thumb'),
|
||||
'timestamp': int_or_none(json_data.get('publish_time'), scale=1000),
|
||||
'is_live': json_data.get('isLive'),
|
||||
'channel': traverse_obj(json_data, ('channels', '0', 'name')),
|
||||
'channel_id': traverse_obj(json_data, ('channels', '0', 'id')),
|
||||
'channel_follower_count': traverse_obj(json_data, ('channels', '0', 'numfollow')),
|
||||
'categories': traverse_obj(json_data, ('categories', ..., 'categoryname')),
|
||||
'comment_count': json_data.get('total_comment'),
|
||||
}
|
Loading…
Reference in New Issue
Block a user