1
0
mirror of https://github.com/l1ving/youtube-dl synced 2024-11-20 21:42:58 +08:00

[vtm] Add extractor (closes #9974)

Implementation of the approach described in #9974.
This commit is contained in:
mrBliss 2017-01-31 13:59:18 +01:00 committed by Sergey M․
parent 3e5856d860
commit e7a51a4c02
No known key found for this signature in database
GPG Key ID: 2C393E0F18A9236D
2 changed files with 137 additions and 0 deletions

View File

@ -1174,6 +1174,7 @@ from .voxmedia import VoxMediaIE
from .vporn import VpornIE
from .vrt import VRTIE
from .vrak import VrakIE
from .vtm import VTMIE
from .vube import VubeIE
from .vuclip import VuClipIE
from .vvvvid import VVVVIDIE

136
youtube_dl/extractor/vtm.py Normal file
View File

@ -0,0 +1,136 @@
from __future__ import unicode_literals
import re
from .generic import GenericIE
from .common import InfoExtractor
from ..utils import (
urlencode_postdata,
compat_urllib_parse_urlencode,
ExtractorError,
remove_end,
)
class VTMIE(InfoExtractor):
"""Download full episodes that require an account from vtm.be or q2.be.
The generic extractor can be used to download clips that do no require an
account.
"""
_VALID_URL = r'https?://(?:www\.)?(?P<site_id>vtm|q2)\.be/video[/?].+?'
_NETRC_MACHINE = 'vtm'
_APIKEY = '3_HZ0FtkMW_gOyKlqQzW5_0FHRC7Nd5XpXJZcDdXY4pk5eES2ZWmejRW5egwVm4ug-'
_TESTS = [
{
'url': 'http://vtm.be/video/volledige-afleveringen/id/vtm_20170219_VM0678361_vtmwatch',
'info_dict': {
'id': 'vtm_20170219_VM0678361_vtmwatch',
'ext': 'mp4',
'title': 'Allemaal Chris afl. 6',
'description': 'md5:4be86427521e7b07e0adb0c9c554ddb2',
},
'skip_download': True,
},
{
'url': 'http://vtm.be/video/volledige-afleveringen/id/257107153551000',
'only_matching': True,
},
{
'url': 'http://vtm.be/video?aid=163157',
'only_matching': True,
},
{
'url': 'http://www.q2.be/video/volledige-afleveringen/id/2be_20170301_VM0684442_q2',
'only_matching': True,
},
{
'url': 'http://vtm.be/video?aid=168332',
'info_dict': {
'id': 'video?aid=168332',
'ext': 'mp4',
'title': 'Videozone',
},
},
]
def _real_initialize(self):
self._logged_in = False
def _login(self):
(username, password) = self._get_login_info()
if username is None or password is None:
self.raise_login_required()
auth_data = {
'APIKey': self._APIKEY,
'sdk': 'js_6.1',
'format': 'json',
'loginID': username,
'password': password,
}
auth_info = self._download_json(
'https://accounts.eu1.gigya.com/accounts.login', None,
note='Logging in', errnote='Unable to log in',
data=urlencode_postdata(auth_data), fatal=True)
error_message = auth_info.get('errorDetails')
if error_message:
raise ExtractorError(
'Unable to login: %s' % error_message, expected=True)
self._uid = auth_info['UID']
self._uid_signature = auth_info['UIDSignature']
self._signature_timestamp = auth_info['signatureTimestamp']
self._logged_in = True
def _real_extract(self, url):
mobj = re.match(self._VALID_URL, url)
site_id = mobj.group('site_id')
webpage = self._download_webpage(url, None, "Downloading webpage")
# The URL sometimes contains the video id, but not always, e.g., test
# case 3. Fortunately, all webpages of videos requiring authentication
# contain the video id.
video_id = self._search_regex(
r'\\"vodId\\":\\"(.+?)\\"', webpage, 'video_id', default=None)
# It was most likely a video not requiring authentication.
if not video_id:
return self.url_result(url, 'Generic')
if not self._logged_in:
self._login()
title = self._html_search_regex(
r'\\"title\\":\\"(.+?)\\"', webpage, 'title', default=None)
description = self._html_search_regex(
r'<div[^>]+class="field-item\s+even">\s*<p>(.+?)</p>',
webpage, 'description', default=None)
data_url = 'http://vod.medialaan.io/api/1.0/item/%s/video' % video_id
m3u8_data = {
'app_id': 'vtm_watch' if site_id == 'vtm' else 'q2',
'user_network': 'vtm-sso',
'UID': self._uid,
'UIDSignature': self._uid_signature,
'signatureTimestamp': self._signature_timestamp,
}
data = self._download_json(data_url, video_id, query=m3u8_data)
formats = self._extract_m3u8_formats(
data['response']['uri'], video_id, entry_protocol='m3u8_native',
ext='mp4', m3u8_id='hls')
self._sort_formats(formats)
return {
'id': video_id,
'title': title,
'description': description,
'formats': formats,
}