|
|
@ -2,7 +2,6 @@ import hashlib
|
|
|
|
import itertools
|
|
|
|
import itertools
|
|
|
|
import urllib.parse
|
|
|
|
import urllib.parse
|
|
|
|
|
|
|
|
|
|
|
|
from .brightcove import BrightcoveNewIE
|
|
|
|
|
|
|
|
from .common import InfoExtractor, SearchInfoExtractor
|
|
|
|
from .common import InfoExtractor, SearchInfoExtractor
|
|
|
|
from .youtube import YoutubeIE
|
|
|
|
from .youtube import YoutubeIE
|
|
|
|
from ..utils import (
|
|
|
|
from ..utils import (
|
|
|
@ -11,7 +10,6 @@ from ..utils import (
|
|
|
|
int_or_none,
|
|
|
|
int_or_none,
|
|
|
|
mimetype2ext,
|
|
|
|
mimetype2ext,
|
|
|
|
parse_iso8601,
|
|
|
|
parse_iso8601,
|
|
|
|
smuggle_url,
|
|
|
|
|
|
|
|
traverse_obj,
|
|
|
|
traverse_obj,
|
|
|
|
try_get,
|
|
|
|
try_get,
|
|
|
|
url_or_none,
|
|
|
|
url_or_none,
|
|
|
@ -337,121 +335,6 @@ class YahooSearchIE(SearchInfoExtractor):
|
|
|
|
break
|
|
|
|
break
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
class YahooGyaOPlayerIE(InfoExtractor):
|
|
|
|
|
|
|
|
IE_NAME = 'yahoo:gyao:player'
|
|
|
|
|
|
|
|
_VALID_URL = r'https?://(?:gyao\.yahoo\.co\.jp/(?:player|episode(?:/[^/]+)?)|streaming\.yahoo\.co\.jp/c/y)/(?P<id>\d+/v\d+/v\d+|[\da-f]{8}-[\da-f]{4}-[\da-f]{4}-[\da-f]{4}-[\da-f]{12})'
|
|
|
|
|
|
|
|
_TESTS = [{
|
|
|
|
|
|
|
|
'url': 'https://gyao.yahoo.co.jp/player/00998/v00818/v0000000000000008564/',
|
|
|
|
|
|
|
|
'info_dict': {
|
|
|
|
|
|
|
|
'id': '5993125228001',
|
|
|
|
|
|
|
|
'ext': 'mp4',
|
|
|
|
|
|
|
|
'title': 'フューリー 【字幕版】',
|
|
|
|
|
|
|
|
'description': 'md5:21e691c798a15330eda4db17a8fe45a5',
|
|
|
|
|
|
|
|
'uploader_id': '4235717419001',
|
|
|
|
|
|
|
|
'upload_date': '20190124',
|
|
|
|
|
|
|
|
'timestamp': 1548294365,
|
|
|
|
|
|
|
|
},
|
|
|
|
|
|
|
|
'params': {
|
|
|
|
|
|
|
|
# m3u8 download
|
|
|
|
|
|
|
|
'skip_download': True,
|
|
|
|
|
|
|
|
},
|
|
|
|
|
|
|
|
}, {
|
|
|
|
|
|
|
|
'url': 'https://streaming.yahoo.co.jp/c/y/01034/v00133/v0000000000000000706/',
|
|
|
|
|
|
|
|
'only_matching': True,
|
|
|
|
|
|
|
|
}, {
|
|
|
|
|
|
|
|
'url': 'https://gyao.yahoo.co.jp/episode/%E3%81%8D%E3%81%AE%E3%81%86%E4%BD%95%E9%A3%9F%E3%81%B9%E3%81%9F%EF%BC%9F%20%E7%AC%AC2%E8%A9%B1%202019%2F4%2F12%E6%94%BE%E9%80%81%E5%88%86/5cb02352-b725-409e-9f8d-88f947a9f682',
|
|
|
|
|
|
|
|
'only_matching': True,
|
|
|
|
|
|
|
|
}, {
|
|
|
|
|
|
|
|
'url': 'https://gyao.yahoo.co.jp/episode/5fa1226c-ef8d-4e93-af7a-fd92f4e30597',
|
|
|
|
|
|
|
|
'only_matching': True,
|
|
|
|
|
|
|
|
}]
|
|
|
|
|
|
|
|
_GEO_BYPASS = False
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
def _real_extract(self, url):
|
|
|
|
|
|
|
|
video_id = self._match_id(url).replace('/', ':')
|
|
|
|
|
|
|
|
headers = self.geo_verification_headers()
|
|
|
|
|
|
|
|
headers['Accept'] = 'application/json'
|
|
|
|
|
|
|
|
resp = self._download_json(
|
|
|
|
|
|
|
|
'https://gyao.yahoo.co.jp/apis/playback/graphql', video_id, query={
|
|
|
|
|
|
|
|
'appId': 'dj00aiZpPUNJeDh2cU1RazU3UCZzPWNvbnN1bWVyc2VjcmV0Jng9NTk-',
|
|
|
|
|
|
|
|
'query': '''{
|
|
|
|
|
|
|
|
content(parameter: {contentId: "%s", logicaAgent: PC_WEB}) {
|
|
|
|
|
|
|
|
video {
|
|
|
|
|
|
|
|
delivery {
|
|
|
|
|
|
|
|
id
|
|
|
|
|
|
|
|
}
|
|
|
|
|
|
|
|
title
|
|
|
|
|
|
|
|
}
|
|
|
|
|
|
|
|
}
|
|
|
|
|
|
|
|
}''' % video_id,
|
|
|
|
|
|
|
|
}, headers=headers)
|
|
|
|
|
|
|
|
content = resp['data']['content']
|
|
|
|
|
|
|
|
if not content:
|
|
|
|
|
|
|
|
msg = resp['errors'][0]['message']
|
|
|
|
|
|
|
|
if msg == 'not in japan':
|
|
|
|
|
|
|
|
self.raise_geo_restricted(countries=['JP'])
|
|
|
|
|
|
|
|
raise ExtractorError(msg)
|
|
|
|
|
|
|
|
video = content['video']
|
|
|
|
|
|
|
|
return {
|
|
|
|
|
|
|
|
'_type': 'url_transparent',
|
|
|
|
|
|
|
|
'id': video_id,
|
|
|
|
|
|
|
|
'title': video['title'],
|
|
|
|
|
|
|
|
'url': smuggle_url(
|
|
|
|
|
|
|
|
'http://players.brightcove.net/4235717419001/SyG5P0gjb_default/index.html?videoId=' + video['delivery']['id'],
|
|
|
|
|
|
|
|
{'geo_countries': ['JP']}),
|
|
|
|
|
|
|
|
'ie_key': BrightcoveNewIE.ie_key(),
|
|
|
|
|
|
|
|
}
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
class YahooGyaOIE(InfoExtractor):
|
|
|
|
|
|
|
|
IE_NAME = 'yahoo:gyao'
|
|
|
|
|
|
|
|
_VALID_URL = r'https?://(?:gyao\.yahoo\.co\.jp/(?:p|title(?:/[^/]+)?)|streaming\.yahoo\.co\.jp/p/y)/(?P<id>\d+/v\d+|[\da-f]{8}-[\da-f]{4}-[\da-f]{4}-[\da-f]{4}-[\da-f]{12})'
|
|
|
|
|
|
|
|
_TESTS = [{
|
|
|
|
|
|
|
|
'url': 'https://gyao.yahoo.co.jp/title/%E3%82%BF%E3%82%A4%E3%83%A0%E3%83%9C%E3%82%AB%E3%83%B3%E3%82%B7%E3%83%AA%E3%83%BC%E3%82%BA%20%E3%83%A4%E3%83%83%E3%82%BF%E3%83%BC%E3%83%9E%E3%83%B3/5f60ceb3-6e5e-40ef-ba40-d68b598d067f',
|
|
|
|
|
|
|
|
'info_dict': {
|
|
|
|
|
|
|
|
'id': '5f60ceb3-6e5e-40ef-ba40-d68b598d067f',
|
|
|
|
|
|
|
|
},
|
|
|
|
|
|
|
|
'playlist_mincount': 80,
|
|
|
|
|
|
|
|
}, {
|
|
|
|
|
|
|
|
'url': 'https://gyao.yahoo.co.jp/p/00449/v03102/',
|
|
|
|
|
|
|
|
'only_matching': True,
|
|
|
|
|
|
|
|
}, {
|
|
|
|
|
|
|
|
'url': 'https://streaming.yahoo.co.jp/p/y/01034/v00133/',
|
|
|
|
|
|
|
|
'only_matching': True,
|
|
|
|
|
|
|
|
}, {
|
|
|
|
|
|
|
|
'url': 'https://gyao.yahoo.co.jp/title/%E3%81%97%E3%82%83%E3%81%B9%E3%81%8F%E3%82%8A007/5b025a49-b2e5-4dc7-945c-09c6634afacf',
|
|
|
|
|
|
|
|
'only_matching': True,
|
|
|
|
|
|
|
|
}, {
|
|
|
|
|
|
|
|
'url': 'https://gyao.yahoo.co.jp/title/5b025a49-b2e5-4dc7-945c-09c6634afacf',
|
|
|
|
|
|
|
|
'only_matching': True,
|
|
|
|
|
|
|
|
}]
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
def _entries(self, program_id):
|
|
|
|
|
|
|
|
page = 1
|
|
|
|
|
|
|
|
while True:
|
|
|
|
|
|
|
|
playlist = self._download_json(
|
|
|
|
|
|
|
|
f'https://gyao.yahoo.co.jp/api/programs/{program_id}/videos?page={page}&serviceId=gy', program_id,
|
|
|
|
|
|
|
|
note=f'Downloading JSON metadata page {page}')
|
|
|
|
|
|
|
|
if not playlist:
|
|
|
|
|
|
|
|
break
|
|
|
|
|
|
|
|
for video in playlist['videos']:
|
|
|
|
|
|
|
|
video_id = video.get('id')
|
|
|
|
|
|
|
|
if not video_id:
|
|
|
|
|
|
|
|
continue
|
|
|
|
|
|
|
|
if video.get('streamingAvailability') == 'notYet':
|
|
|
|
|
|
|
|
continue
|
|
|
|
|
|
|
|
yield self.url_result(
|
|
|
|
|
|
|
|
'https://gyao.yahoo.co.jp/player/%s/' % video_id.replace(':', '/'),
|
|
|
|
|
|
|
|
YahooGyaOPlayerIE.ie_key(), video_id)
|
|
|
|
|
|
|
|
if playlist.get('ended'):
|
|
|
|
|
|
|
|
break
|
|
|
|
|
|
|
|
page += 1
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
def _real_extract(self, url):
|
|
|
|
|
|
|
|
program_id = self._match_id(url).replace('/', ':')
|
|
|
|
|
|
|
|
return self.playlist_result(self._entries(program_id), program_id)
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
class YahooJapanNewsIE(InfoExtractor):
|
|
|
|
class YahooJapanNewsIE(InfoExtractor):
|
|
|
|
IE_NAME = 'yahoo:japannews'
|
|
|
|
IE_NAME = 'yahoo:japannews'
|
|
|
|
IE_DESC = 'Yahoo! Japan News'
|
|
|
|
IE_DESC = 'Yahoo! Japan News'
|
|
|
|