0
0
Fork 0
mirror of https://github.com/yt-dlp/yt-dlp.git synced 2024-11-23 02:25:11 +00:00
yt-dlp/yt_dlp/extractor/tennistv.py
pukkandan 52efa4b312
[extractor] Add _perform_login function (#2943)
* Adds new functions `_initialize_pre_login` and `_perform_login` as part of the extractor API
* Adds `ie.supports_login` to the public API
2022-03-18 13:53:33 -07:00

113 lines
3.9 KiB
Python

# coding: utf-8
from __future__ import unicode_literals
import json
from .common import InfoExtractor
from ..utils import (
ExtractorError,
unified_timestamp,
)
class TennisTVIE(InfoExtractor):
_VALID_URL = r'https?://(?:www\.)?tennistv\.com/videos/(?P<id>[-a-z0-9]+)'
_TEST = {
'url': 'https://www.tennistv.com/videos/indian-wells-2018-verdasco-fritz',
'info_dict': {
'id': 'indian-wells-2018-verdasco-fritz',
'ext': 'mp4',
'title': 'Fernando Verdasco v Taylor Fritz',
'description': 're:^After his stunning victory.{174}$',
'thumbnail': 'https://atp-prod.akamaized.net/api/images/v1/images/112831/landscape/1242/0',
'timestamp': 1521017381,
'upload_date': '20180314',
},
'params': {
'skip_download': True,
},
'skip': 'Requires email and password of a subscribed account',
}
_NETRC_MACHINE = 'tennistv'
_session_token = None
def _perform_login(self, username, password):
login_form = {
'Email': username,
'Password': password,
}
login_json = json.dumps(login_form).encode('utf-8')
headers = {
'content-type': 'application/json',
'Referer': 'https://www.tennistv.com/login',
'Origin': 'https://www.tennistv.com',
}
login_result = self._download_json(
'https://www.tennistv.com/api/users/v1/login', None,
note='Logging in',
errnote='Login failed (wrong password?)',
headers=headers,
data=login_json)
if login_result['error']['errorCode']:
raise ExtractorError('Login failed, %s said: %r' % (self.IE_NAME, login_result['error']['errorMessage']))
if login_result['entitlement'] != 'SUBSCRIBED':
self.report_warning('%s may not be subscribed to %s.' % (username, self.IE_NAME))
self._session_token = login_result['sessionToken']
def _real_initialize(self):
if not self._session_token:
raise self.raise_login_required('Login info is needed for this website', method='password')
def _real_extract(self, url):
video_id = self._match_id(url)
webpage = self._download_webpage(url, video_id)
internal_id = self._search_regex(r'video=([\w-]+)', webpage, 'internal video id')
headers = {
'Origin': 'https://www.tennistv.com',
'authorization': 'ATP %s' % self._session_token,
'content-type': 'application/json',
'Referer': url,
}
check_data = {
'videoID': internal_id,
'VideoUrlType': 'HLS',
}
check_json = json.dumps(check_data).encode('utf-8')
check_result = self._download_json(
'https://www.tennistv.com/api/users/v1/entitlementchecknondiva',
video_id, note='Checking video authorization', headers=headers, data=check_json)
formats = self._extract_m3u8_formats(check_result['contentUrl'], video_id, ext='mp4')
self._sort_formats(formats)
vdata = self._download_json(
'https://www.tennistv.com/api/en/v2/none/common/video/%s' % video_id,
video_id, headers=headers)
timestamp = unified_timestamp(vdata['timestamp'])
thumbnail = vdata['video']['thumbnailUrl']
description = vdata['displayText']['description']
title = vdata['video']['title']
series = vdata['tour']
venue = vdata['displayText']['venue']
round_str = vdata['seo']['round']
return {
'id': video_id,
'title': title,
'description': description,
'formats': formats,
'thumbnail': thumbnail,
'timestamp': timestamp,
'series': series,
'season': venue,
'episode': round_str,
}