yt-dlp/yt_dlp/extractor/fc2.py
bashonly c6387abc1a
[cleanup] Misc (#10807)
Closes #10751, Closes #10769, Closes #10791
Authored by: bashonly, Codenade, pzhlkj6612, seproDev, coletdjnz, grqz, Grub4K

Co-authored-by: Codenade <amadeus.dorian04@gmail.com>
Co-authored-by: Mozi <29089388+pzhlkj6612@users.noreply.github.com>
Co-authored-by: sepro <4618135+seproDev@users.noreply.github.com>
Co-authored-by: coletdjnz <coletdjnz@protonmail.com>
Co-authored-by: N/Ame <173015200+grqz@users.noreply.github.com>
Co-authored-by: Simon Sawicki <contact@grub4k.xyz>
2024-09-27 22:46:22 +00:00

281 lines
10 KiB
Python

import re
import urllib.parse
from .common import InfoExtractor
from ..networking import Request
from ..utils import (
ExtractorError,
js_to_json,
traverse_obj,
update_url_query,
urlencode_postdata,
urljoin,
)
class FC2IE(InfoExtractor):
_VALID_URL = r'(?:https?://video\.fc2\.com/(?:[^/]+/)*content/|fc2:)(?P<id>[^/]+)'
IE_NAME = 'fc2'
_NETRC_MACHINE = 'fc2'
_TESTS = [{
'url': 'http://video.fc2.com/en/content/20121103kUan1KHs',
'md5': 'a6ebe8ebe0396518689d963774a54eb7',
'info_dict': {
'id': '20121103kUan1KHs',
'ext': 'flv',
'title': 'Boxing again with Puff',
},
}, {
'url': 'http://video.fc2.com/en/content/20150125cEva0hDn/',
'info_dict': {
'id': '20150125cEva0hDn',
'ext': 'mp4',
},
'params': {
'username': 'ytdl@yt-dl.org',
'password': '(snip)',
},
'skip': 'requires actual password',
}, {
'url': 'http://video.fc2.com/en/a/content/20130926eZpARwsF',
'only_matching': True,
}]
def _login(self):
username, password = self._get_login_info()
if username is None or password is None:
return False
# Log in
login_form_strs = {
'email': username,
'password': password,
'done': 'video',
'Submit': ' Login ',
}
login_data = urlencode_postdata(login_form_strs)
request = Request(
'https://secure.id.fc2.com/index.php?mode=login&switch_language=en', login_data)
login_results = self._download_webpage(request, None, note='Logging in', errnote='Unable to log in')
if 'mode=redirect&login=done' not in login_results:
self.report_warning('unable to log in: bad username or password')
return False
# this is also needed
login_redir = Request('http://id.fc2.com/?mode=redirect&login=done')
self._download_webpage(
login_redir, None, note='Login redirect', errnote='Login redirect failed')
return True
def _real_extract(self, url):
video_id = self._match_id(url)
self._login()
webpage = None
if not url.startswith('fc2:'):
webpage = self._download_webpage(url, video_id)
self.cookiejar.clear_session_cookies() # must clear
self._login()
title, thumbnail, description = None, None, None
if webpage is not None:
title = self._html_search_regex(
(r'<h2\s+class="videoCnt_title">([^<]+?)</h2>',
r'\s+href="[^"]+"\s*title="([^"]+?)"\s*rel="nofollow">\s*<img',
# there's two matches in the webpage
r'\s+href="[^"]+"\s*title="([^"]+?)"\s*rel="nofollow">\s*\1'),
webpage,
'title', fatal=False)
thumbnail = self._og_search_thumbnail(webpage)
description = self._og_search_description(webpage, default=None)
vidplaylist = self._download_json(
f'https://video.fc2.com/api/v3/videoplaylist/{video_id}?sh=1&fs=0', video_id,
note='Downloading info page')
vid_url = traverse_obj(vidplaylist, ('playlist', 'nq'))
if not vid_url:
raise ExtractorError('Unable to extract video URL')
vid_url = urljoin('https://video.fc2.com/', vid_url)
return {
'id': video_id,
'title': title,
'url': vid_url,
'ext': 'mp4',
'protocol': 'm3u8_native',
'description': description,
'thumbnail': thumbnail,
}
class FC2EmbedIE(InfoExtractor):
_VALID_URL = r'https?://video\.fc2\.com/flv2\.swf\?(?P<query>.+)'
IE_NAME = 'fc2:embed'
_TEST = {
'url': 'http://video.fc2.com/flv2.swf?t=201404182936758512407645&i=20130316kwishtfitaknmcgd76kjd864hso93htfjcnaogz629mcgfs6rbfk0hsycma7shkf85937cbchfygd74&i=201403223kCqB3Ez&d=2625&sj=11&lang=ja&rel=1&from=11&cmt=1&tk=TlRBM09EQTNNekU9&tl=プリズン・ブレイク%20S1-01%20マイケル%20【吹替】',
'md5': 'b8aae5334cb691bdb1193a88a6ab5d5a',
'info_dict': {
'id': '201403223kCqB3Ez',
'ext': 'flv',
'title': 'プリズン・ブレイク S1-01 マイケル 【吹替】',
'thumbnail': r're:^https?://.*\.jpg$',
},
}
def _real_extract(self, url):
mobj = self._match_valid_url(url)
query = urllib.parse.parse_qs(mobj.group('query'))
video_id = query['i'][-1]
title = query.get('tl', [f'FC2 video {video_id}'])[0]
sj = query.get('sj', [None])[0]
thumbnail = None
if sj:
# See thumbnailImagePath() in ServerConst.as of flv2.swf
thumbnail = 'http://video{}-thumbnail.fc2.com/up/pic/{}.jpg'.format(
sj, '/'.join((video_id[:6], video_id[6:8], video_id[-2], video_id[-1], video_id)))
return {
'_type': 'url_transparent',
'ie_key': FC2IE.ie_key(),
'url': f'fc2:{video_id}',
'title': title,
'thumbnail': thumbnail,
}
class FC2LiveIE(InfoExtractor):
_VALID_URL = r'https?://live\.fc2\.com/(?P<id>\d+)'
IE_NAME = 'fc2:live'
_TESTS = [{
'url': 'https://live.fc2.com/57892267/',
'info_dict': {
'id': '57892267',
'title': 'どこまで・・・',
'uploader': 'あつあげ',
'uploader_id': '57892267',
'thumbnail': r're:https?://.+fc2.+',
},
'skip': 'livestream',
}]
def _real_extract(self, url):
video_id = self._match_id(url)
webpage = self._download_webpage(f'https://live.fc2.com/{video_id}/', video_id)
self._set_cookie('live.fc2.com', 'js-player_size', '1')
member_api = self._download_json(
'https://live.fc2.com/api/memberApi.php', video_id, data=urlencode_postdata({
'channel': '1',
'profile': '1',
'user': '1',
'streamid': video_id,
}), note='Requesting member info')
control_server = self._download_json(
'https://live.fc2.com/api/getControlServer.php', video_id, note='Downloading ControlServer data',
data=urlencode_postdata({
'channel_id': video_id,
'mode': 'play',
'orz': '',
'channel_version': member_api['data']['channel_data']['version'],
'client_version': '2.1.0\n [1]',
'client_type': 'pc',
'client_app': 'browser_hls',
'ipv6': '',
}), headers={'X-Requested-With': 'XMLHttpRequest'})
self._set_cookie('live.fc2.com', 'l_ortkn', control_server['orz_raw'])
ws_url = update_url_query(control_server['url'], {'control_token': control_server['control_token']})
playlist_data = None
ws = self._request_webpage(Request(ws_url, headers={
'Origin': 'https://live.fc2.com',
}), video_id, note='Fetching HLS playlist info via WebSocket')
self.write_debug('Sending HLS server request')
while True:
recv = ws.recv()
if not recv:
continue
data = self._parse_json(recv, video_id, fatal=False)
if not data or not isinstance(data, dict):
continue
if data.get('name') == 'connect_complete':
break
ws.send(r'{"name":"get_hls_information","arguments":{},"id":1}')
while True:
recv = ws.recv()
if not recv:
continue
data = self._parse_json(recv, video_id, fatal=False)
if not data or not isinstance(data, dict):
continue
if data.get('name') == '_response_' and data.get('id') == 1:
self.write_debug('Goodbye')
playlist_data = data
break
self.write_debug('Server said: {}{}'.format(recv[:100], '...' if len(recv) > 100 else ''))
if not playlist_data:
raise ExtractorError('Unable to fetch HLS playlist info via WebSocket')
formats = []
for name, playlists in playlist_data['arguments'].items():
if not isinstance(playlists, list):
continue
for pl in playlists:
if pl.get('status') == 0 and 'master_playlist' in pl.get('url'):
formats.extend(self._extract_m3u8_formats(
pl['url'], video_id, ext='mp4', m3u8_id=name, live=True,
headers={
'Origin': 'https://live.fc2.com',
'Referer': url,
}))
for fmt in formats:
fmt.update({
'protocol': 'fc2_live',
'ws': ws,
})
title = self._html_search_meta(('og:title', 'twitter:title'), webpage, 'live title', fatal=False)
if not title:
title = self._html_extract_title(webpage, 'html title', fatal=False)
if title:
# remove service name in <title>
title = re.sub(r'\s+-\s+.+$', '', title)
uploader = None
if title:
match = self._search_regex(r'^(.+?)\s*\[(.+?)\]$', title, 'title and uploader', default=None, group=(1, 2))
if match and all(match):
title, uploader = match
live_info_view = self._search_regex(r'(?s)liveInfoView\s*:\s*({.+?}),\s*premiumStateView', webpage, 'user info', fatal=False) or None
if live_info_view:
# remove jQuery code from object literal
live_info_view = re.sub(r'\$\(.+?\)[^,]+,', '"",', live_info_view)
live_info_view = self._parse_json(js_to_json(live_info_view), video_id)
return {
'id': video_id,
'title': title or traverse_obj(live_info_view, 'title'),
'description': self._html_search_meta(
('og:description', 'twitter:description'),
webpage, 'live description', fatal=False) or traverse_obj(live_info_view, 'info'),
'formats': formats,
'uploader': uploader or traverse_obj(live_info_view, 'name'),
'uploader_id': video_id,
'thumbnail': traverse_obj(live_info_view, 'thumb'),
'is_live': True,
}