mirror of
https://github.com/ytdl-org/youtube-dl.git
synced 2024-12-15 01:36:31 +00:00
75 lines
2.7 KiB
Python
75 lines
2.7 KiB
Python
# coding: utf-8
|
|
from __future__ import unicode_literals
|
|
|
|
from .adobepass import AdobePassIE
|
|
from ..utils import (
|
|
NO_DEFAULT,
|
|
smuggle_url,
|
|
update_url_query,
|
|
)
|
|
|
|
|
|
class USANetworkIE(AdobePassIE):
|
|
_VALID_URL = r'https?://(?:www\.)?usanetwork\.com/(?:[^/]+/videos|movies)/(?P<id>[^/?#]+)'
|
|
_TEST = {
|
|
'url': 'http://www.usanetwork.com/mrrobot/videos/hpe-cybersecurity',
|
|
'md5': '33c0d2ba381571b414024440d08d57fd',
|
|
'info_dict': {
|
|
'id': '3086229',
|
|
'ext': 'mp4',
|
|
'title': 'HPE Cybersecurity',
|
|
'description': 'The more we digitize our world, the more vulnerable we are.',
|
|
'upload_date': '20160818',
|
|
'timestamp': 1471535460,
|
|
'uploader': 'NBCU-USA',
|
|
},
|
|
}
|
|
|
|
def _real_extract(self, url):
|
|
display_id = self._match_id(url)
|
|
webpage = self._download_webpage(url, display_id)
|
|
|
|
def _x(name, default=NO_DEFAULT):
|
|
return self._search_regex(
|
|
r'data-%s\s*=\s*(["\'])(?P<value>(?:(?!\1).)+)\1' % name,
|
|
webpage, name, default=default, group='value')
|
|
|
|
video_id = _x('mpx-guid')
|
|
title = _x('episode-title')
|
|
mpx_account_id = _x('mpx-account-id', '2304992029')
|
|
|
|
query = {
|
|
'mbr': 'true',
|
|
}
|
|
if _x('is-full-episode', None) == '1':
|
|
query['manifest'] = 'm3u'
|
|
|
|
if _x('is-entitlement', None) == '1':
|
|
adobe_pass = {}
|
|
drupal_settings = self._search_regex(
|
|
r'jQuery\.extend\(Drupal\.settings\s*,\s*({.+?})\);',
|
|
webpage, 'drupal settings', fatal=False)
|
|
if drupal_settings:
|
|
drupal_settings = self._parse_json(drupal_settings, video_id, fatal=False)
|
|
if drupal_settings:
|
|
adobe_pass = drupal_settings.get('adobePass', {})
|
|
resource = self._get_mvpd_resource(
|
|
adobe_pass.get('adobePassResourceId', 'usa'),
|
|
title, video_id, _x('episode-rating', 'TV-14'))
|
|
query['auth'] = self._extract_mvpd_auth(
|
|
url, video_id, adobe_pass.get('adobePassRequestorId', 'usa'), resource)
|
|
|
|
info = self._search_json_ld(webpage, video_id, default={})
|
|
info.update({
|
|
'_type': 'url_transparent',
|
|
'url': smuggle_url(update_url_query(
|
|
'http://link.theplatform.com/s/HNK2IC/media/guid/%s/%s' % (mpx_account_id, video_id),
|
|
query), {'force_smil_url': True}),
|
|
'id': video_id,
|
|
'title': title,
|
|
'series': _x('show-title', None),
|
|
'episode': title,
|
|
'ie_key': 'ThePlatform',
|
|
})
|
|
return info
|