1
1
mirror of https://github.com/ytdl-org/youtube-dl synced 2024-12-25 06:25:51 +01:00
youtube-dl/youtube_dl/extractor/telewebion.py
2017-01-06 00:58:56 +08:00

56 lines
1.8 KiB
Python

# coding: utf-8
from __future__ import unicode_literals
from .common import InfoExtractor
class TelewebionIE(InfoExtractor):
_VALID_URL = r'https?://(?:www\.)?telewebion\.com/#!/episode/(?P<id>\d+)'
_TEST = {
'url': 'http://www.telewebion.com/#!/episode/1263668/',
'info_dict': {
'id': '1263668',
'ext': 'mp4',
'title': 'قرعه\u200cکشی لیگ قهرمانان اروپا',
'thumbnail': r're:^https?://.*\.jpg',
'view_count': int,
},
'params': {
# m3u8 download
'skip_download': True,
},
}
def _real_extract(self, url):
video_id = self._match_id(url)
secure_token = self._download_webpage(
'http://m.s2.telewebion.com/op/op?action=getSecurityToken', video_id)
episode_details = self._download_json(
'http://m.s2.telewebion.com/op/op', video_id,
query={'action': 'getEpisodeDetails', 'episode_id': video_id})
m3u8_url = 'http://m.s1.telewebion.com/smil/%s.m3u8?filepath=%s&m3u8=1&secure_token=%s' % (
video_id, episode_details['file_path'], secure_token)
formats = self._extract_m3u8_formats(
m3u8_url, video_id, ext='mp4', m3u8_id='hls')
picture_paths = [
episode_details.get('picture_path'),
episode_details.get('large_picture_path'),
]
thumbnails = [{
'url': picture_path,
'preference': idx,
} for idx, picture_path in enumerate(picture_paths) if picture_path is not None]
return {
'id': video_id,
'title': episode_details['title'],
'formats': formats,
'thumbnails': thumbnails,
'view_count': episode_details.get('view_count'),
}