1
0
mirror of https://codeberg.org/polarisfm/youtube-dl synced 2024-11-30 12:17:54 +01:00
youtube-dl/youtube_dl/extractor/medici.py

71 lines
2.2 KiB
Python
Raw Normal View History

2017-04-08 17:38:37 +02:00
# coding: utf-8
from __future__ import unicode_literals
from .common import InfoExtractor
from ..utils import (
unified_strdate,
update_url_query,
urlencode_postdata,
)
class MediciIE(InfoExtractor):
_VALID_URL = r'https?://(?:www\.)?medici\.tv/#!/(?P<id>[^?#&]+)'
_TEST = {
'url': 'http://www.medici.tv/#!/daniel-harding-frans-helmerson-verbier-festival-music-camp',
'md5': '004c21bb0a57248085b6ff3fec72719d',
'info_dict': {
'id': '3059',
'ext': 'flv',
'title': 'Daniel Harding conducts the Verbier Festival Music Camp \u2013 With Frans Helmerson',
'description': 'md5:322a1e952bafb725174fd8c1a8212f58',
'thumbnail': r're:^https?://.*\.jpg$',
'upload_date': '20170408',
},
}
def _real_extract(self, url):
video_id = self._match_id(url)
# Sets csrftoken cookie
self._download_webpage(url, video_id)
MEDICI_URL = 'http://www.medici.tv/'
data = self._download_json(
MEDICI_URL, video_id,
data=urlencode_postdata({
'json': 'true',
'page': '/%s' % video_id,
'timezone_offset': -420,
}), headers={
'X-CSRFToken': self._get_cookies(url)['csrftoken'].value,
'X-Requested-With': 'XMLHttpRequest',
'Referer': MEDICI_URL,
'Content-Type': 'application/x-www-form-urlencoded',
})
video = data['video']['videos']['video1']
title = video.get('nom') or data['title']
video_id = video.get('id') or video_id
formats = self._extract_f4m_formats(
update_url_query(video['url_akamai'], {
'hdcore': '3.1.0',
'plugin=aasp': '3.1.0.43.124',
}), video_id, f4m_id='hds')
description = data.get('meta_description')
thumbnail = video.get('url_thumbnail') or data.get('main_image')
upload_date = unified_strdate(data['video'].get('date'))
return {
'id': video_id,
'title': title,
'description': description,
'thumbnail': thumbnail,
'upload_date': upload_date,
'formats': formats,
}