1
0
mirror of https://github.com/l1ving/youtube-dl synced 2024-11-24 02:22:56 +08:00
youtube-dl/youtube_dl/extractor/vessel.py

134 lines
4.7 KiB
Python
Raw Normal View History

# coding: utf-8
from __future__ import unicode_literals
import json
from .common import InfoExtractor
from ..compat import compat_urllib_request
from ..utils import (
ExtractorError,
parse_iso8601,
)
class VesselIE(InfoExtractor):
_VALID_URL = r'https?://(?:www\.)?vessel\.com/videos/(?P<id>[0-9a-zA-Z]+)'
_API_URL_TEMPLATE = 'https://www.vessel.com/api/view/items/%s'
_LOGIN_URL = 'https://www.vessel.com/api/account/login'
_NETRC_MACHINE = 'vessel'
_TEST = {
'url': 'https://www.vessel.com/videos/HDN7G5UMs',
'md5': '455cdf8beb71c6dd797fd2f3818d05c4',
'info_dict': {
'id': 'HDN7G5UMs',
'ext': 'mp4',
'title': 'Nvidia GeForce GTX Titan X - The Best Video Card on the Market?',
'thumbnail': 're:^https?://.*\.jpg$',
'upload_date': '20150317',
'description': 'Did Nvidia pull out all the stops on the Titan X, or does its performance leave something to be desired?',
'timestamp': int,
},
}
@staticmethod
def make_json_request(url, data):
payload = json.dumps(data).encode('utf-8')
req = compat_urllib_request.Request(url, payload)
req.add_header('Content-Type', 'application/json; charset=utf-8')
return req
@staticmethod
def find_assets(data, asset_type, asset_id=None):
for asset in data.get('assets', []):
if not asset.get('type') == asset_type:
continue
elif asset_id is not None and not asset.get('id') == asset_id:
continue
else:
yield asset
def _check_access_rights(self, data):
access_info = data.get('__view', {})
2015-03-27 02:51:40 +08:00
if not access_info.get('allow_access', True):
err_code = access_info.get('error_code') or ''
if err_code == 'ITEM_PAID_ONLY':
raise ExtractorError(
'This video requires subscription.', expected=True)
else:
raise ExtractorError(
'Access to this content is restricted. (%s said: %s)' % (self.IE_NAME, err_code), expected=True)
def _login(self):
(username, password) = self._get_login_info()
if username is None:
return
self.report_login()
data = {
'client_id': 'web',
'type': 'password',
'user_key': username,
'password': password,
}
login_request = VesselIE.make_json_request(self._LOGIN_URL, data)
self._download_webpage(login_request, None, False, 'Wrong login info')
def _real_initialize(self):
self._login()
def _real_extract(self, url):
video_id = self._match_id(url)
webpage = self._download_webpage(url, video_id)
data = self._parse_json(self._search_regex(
r'App\.bootstrapData\((.*?)\);', webpage, 'data'), video_id)
asset_id = data['model']['data']['id']
req = VesselIE.make_json_request(
self._API_URL_TEMPLATE % asset_id, {'client': 'web'})
data = self._download_json(req, video_id)
video_asset_id = data.get('main_video_asset')
self._check_access_rights(data)
try:
video_asset = next(
VesselIE.find_assets(data, 'video', asset_id=video_asset_id))
except StopIteration:
raise ExtractorError('No video assets found')
formats = []
for f in video_asset.get('sources', []):
if f['name'] == 'hls-index':
formats.extend(self._extract_m3u8_formats(
f['location'], video_id, ext='mp4', m3u8_id='m3u8'))
else:
formats.append({
'format_id': f['name'],
'tbr': f.get('bitrate'),
'height': f.get('height'),
'width': f.get('width'),
'url': f['location'],
})
self._sort_formats(formats)
thumbnails = []
for im_asset in VesselIE.find_assets(data, 'image'):
thumbnails.append({
'url': im_asset['location'],
'width': im_asset.get('width', 0),
'height': im_asset.get('height', 0),
})
return {
'id': video_id,
'title': data['title'],
'formats': formats,
'thumbnails': thumbnails,
'description': data.get('short_description'),
'duration': data.get('duration'),
'comment_count': data.get('comment_count'),
'like_count': data.get('like_count'),
'view_count': data.get('view_count'),
'timestamp': parse_iso8601(data.get('released_at')),
}