mirror of
https://github.com/yt-dlp/yt-dlp.git
synced 2024-11-14 20:33:19 +00:00
65 lines
2.3 KiB
Python
65 lines
2.3 KiB
Python
# coding: utf-8
|
|
from __future__ import unicode_literals
|
|
|
|
from .common import InfoExtractor
|
|
from ..utils import int_or_none
|
|
|
|
|
|
class MGTVIE(InfoExtractor):
|
|
_VALID_URL = r'https?://www\.mgtv\.com/v/(?:[^/]+/)*(?P<id>\d+)\.html'
|
|
IE_DESC = '芒果TV'
|
|
|
|
_TEST = {
|
|
'url': 'http://www.mgtv.com/v/1/290525/f/3116640.html',
|
|
'md5': '1bdadcf760a0b90946ca68ee9a2db41a',
|
|
'info_dict': {
|
|
'id': '3116640',
|
|
'ext': 'mp4',
|
|
'title': '我是歌手第四季双年巅峰会:韩红李玟“双王”领军对抗',
|
|
'description': '我是歌手第四季双年巅峰会',
|
|
'duration': 7461,
|
|
'thumbnail': 're:^https?://.*\.jpg$',
|
|
},
|
|
}
|
|
|
|
def _real_extract(self, url):
|
|
video_id = self._match_id(url)
|
|
api_data = self._download_json(
|
|
'http://v.api.mgtv.com/player/video', video_id,
|
|
query={'video_id': video_id})['data']
|
|
info = api_data['info']
|
|
|
|
formats = []
|
|
for idx, stream in enumerate(api_data['stream']):
|
|
stream_url = stream.get('url')
|
|
if not stream_url:
|
|
continue
|
|
tbr = int_or_none(self._search_regex(
|
|
r'(\d+)\.mp4', stream_url, 'tbr', default=None))
|
|
|
|
def extract_format(stream_url, format_id, idx, query={}):
|
|
format_info = self._download_json(
|
|
stream_url, video_id,
|
|
note='Download video info for format %s' % format_id or '#%d' % idx, query=query)
|
|
return {
|
|
'format_id': format_id,
|
|
'url': format_info['info'],
|
|
'ext': 'mp4',
|
|
'tbr': tbr,
|
|
}
|
|
|
|
formats.append(extract_format(
|
|
stream_url, 'hls-%d' % tbr if tbr else None, idx * 2))
|
|
formats.append(extract_format(stream_url.replace(
|
|
'/playlist.m3u8', ''), 'http-%d' % tbr if tbr else None, idx * 2 + 1, {'pno': 1031}))
|
|
self._sort_formats(formats)
|
|
|
|
return {
|
|
'id': video_id,
|
|
'title': info['title'].strip(),
|
|
'formats': formats,
|
|
'description': info.get('desc'),
|
|
'duration': int_or_none(info.get('duration')),
|
|
'thumbnail': info.get('thumb'),
|
|
}
|