mirror of
https://github.com/yt-dlp/yt-dlp.git
synced 2024-11-24 11:51:11 +01:00
98 lines
3.6 KiB
Python
98 lines
3.6 KiB
Python
# coding: utf-8
|
|
from __future__ import unicode_literals
|
|
|
|
from .common import InfoExtractor
|
|
from ..utils import (
|
|
int_or_none,
|
|
js_to_json,
|
|
ExtractorError,
|
|
urlencode_postdata,
|
|
extract_attributes,
|
|
smuggle_url,
|
|
)
|
|
|
|
|
|
class TouTvIE(InfoExtractor):
|
|
_NETRC_MACHINE = 'toutv'
|
|
IE_NAME = 'tou.tv'
|
|
_VALID_URL = r'https?://ici\.tou\.tv/(?P<id>[a-zA-Z0-9_-]+/S[0-9]+E[0-9]+)'
|
|
_access_token = None
|
|
_claims = None
|
|
|
|
_TEST = {
|
|
'url': 'http://ici.tou.tv/garfield-tout-court/S2015E17',
|
|
'info_dict': {
|
|
'id': '122017',
|
|
'ext': 'mp4',
|
|
'title': 'Saison 2015 Épisode 17',
|
|
'description': 'La photo de famille 2',
|
|
'upload_date': '20100717',
|
|
},
|
|
'params': {
|
|
# m3u8 download
|
|
'skip_download': True,
|
|
},
|
|
'skip': '404 Not Found',
|
|
}
|
|
|
|
def _real_initialize(self):
|
|
email, password = self._get_login_info()
|
|
if email is None:
|
|
return
|
|
state = 'http://ici.tou.tv//'
|
|
webpage = self._download_webpage(state, None, 'Downloading homepage')
|
|
toutvlogin = self._parse_json(self._search_regex(
|
|
r'(?s)toutvlogin\s*=\s*({.+?});', webpage, 'toutvlogin'), None, js_to_json)
|
|
authorize_url = toutvlogin['host'] + '/auth/oauth/v2/authorize'
|
|
login_webpage = self._download_webpage(
|
|
authorize_url, None, 'Downloading login page', query={
|
|
'client_id': toutvlogin['clientId'],
|
|
'redirect_uri': 'https://ici.tou.tv/login/loginCallback',
|
|
'response_type': 'token',
|
|
'scope': 'media-drmt openid profile email id.write media-validation.read.privileged',
|
|
'state': state,
|
|
})
|
|
login_form = self._search_regex(
|
|
r'(?s)(<form[^>]+id="Form-login".+?</form>)', login_webpage, 'login form')
|
|
form_data = self._hidden_inputs(login_form)
|
|
form_data.update({
|
|
'login-email': email,
|
|
'login-password': password,
|
|
})
|
|
post_url = extract_attributes(login_form).get('action') or authorize_url
|
|
_, urlh = self._download_webpage_handle(
|
|
post_url, None, 'Logging in', data=urlencode_postdata(form_data))
|
|
self._access_token = self._search_regex(
|
|
r'access_token=([\da-f]{8}-[\da-f]{4}-[\da-f]{4}-[\da-f]{4}-[\da-f]{12})',
|
|
urlh.geturl(), 'access token')
|
|
self._claims = self._download_json(
|
|
'https://services.radio-canada.ca/media/validation/v2/getClaims',
|
|
None, 'Extracting Claims', query={
|
|
'token': self._access_token,
|
|
'access_token': self._access_token,
|
|
})['claims']
|
|
|
|
def _real_extract(self, url):
|
|
path = self._match_id(url)
|
|
metadata = self._download_json('http://ici.tou.tv/presentation/%s' % path, path)
|
|
if metadata.get('IsDrm'):
|
|
raise ExtractorError('This video is DRM protected.', expected=True)
|
|
video_id = metadata['IdMedia']
|
|
details = metadata['Details']
|
|
title = details['OriginalTitle']
|
|
video_url = 'radiocanada:%s:%s' % (metadata.get('AppCode', 'toutv'), video_id)
|
|
if self._access_token and self._claims:
|
|
video_url = smuggle_url(video_url, {
|
|
'access_token': self._access_token,
|
|
'claims': self._claims,
|
|
})
|
|
|
|
return {
|
|
'_type': 'url_transparent',
|
|
'url': video_url,
|
|
'id': video_id,
|
|
'title': title,
|
|
'thumbnail': details.get('ImageUrl'),
|
|
'duration': int_or_none(details.get('LengthInSeconds')),
|
|
}
|