|
|
|
@ -1,143 +0,0 @@
|
|
|
|
|
from .common import InfoExtractor
|
|
|
|
|
from ..compat import compat_str
|
|
|
|
|
from ..utils import (
|
|
|
|
|
format_field,
|
|
|
|
|
int_or_none,
|
|
|
|
|
str_or_none,
|
|
|
|
|
)
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
class LineLiveBaseIE(InfoExtractor):
|
|
|
|
|
_API_BASE_URL = 'https://live-api.line-apps.com/web/v4.0/channel/'
|
|
|
|
|
|
|
|
|
|
def _parse_broadcast_item(self, item):
|
|
|
|
|
broadcast_id = compat_str(item['id'])
|
|
|
|
|
title = item['title']
|
|
|
|
|
is_live = item.get('isBroadcastingNow')
|
|
|
|
|
|
|
|
|
|
thumbnails = []
|
|
|
|
|
for thumbnail_id, thumbnail_url in (item.get('thumbnailURLs') or {}).items():
|
|
|
|
|
if not thumbnail_url:
|
|
|
|
|
continue
|
|
|
|
|
thumbnails.append({
|
|
|
|
|
'id': thumbnail_id,
|
|
|
|
|
'url': thumbnail_url,
|
|
|
|
|
})
|
|
|
|
|
|
|
|
|
|
channel = item.get('channel') or {}
|
|
|
|
|
channel_id = str_or_none(channel.get('id'))
|
|
|
|
|
|
|
|
|
|
return {
|
|
|
|
|
'id': broadcast_id,
|
|
|
|
|
'title': title,
|
|
|
|
|
'thumbnails': thumbnails,
|
|
|
|
|
'timestamp': int_or_none(item.get('createdAt')),
|
|
|
|
|
'channel': channel.get('name'),
|
|
|
|
|
'channel_id': channel_id,
|
|
|
|
|
'channel_url': format_field(channel_id, None, 'https://live.line.me/channels/%s'),
|
|
|
|
|
'duration': int_or_none(item.get('archiveDuration')),
|
|
|
|
|
'view_count': int_or_none(item.get('viewerCount')),
|
|
|
|
|
'comment_count': int_or_none(item.get('chatCount')),
|
|
|
|
|
'is_live': is_live,
|
|
|
|
|
}
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
class LineLiveIE(LineLiveBaseIE):
|
|
|
|
|
_VALID_URL = r'https?://live\.line\.me/channels/(?P<channel_id>\d+)/broadcast/(?P<id>\d+)'
|
|
|
|
|
_TESTS = [{
|
|
|
|
|
'url': 'https://live.line.me/channels/5833718/broadcast/18373277',
|
|
|
|
|
'md5': '2c15843b8cb3acd55009ddcb2db91f7c',
|
|
|
|
|
'info_dict': {
|
|
|
|
|
'id': '18373277',
|
|
|
|
|
'title': '2021/12/05 (15分犬)定例譲渡会🐶',
|
|
|
|
|
'ext': 'mp4',
|
|
|
|
|
'timestamp': 1638674925,
|
|
|
|
|
'upload_date': '20211205',
|
|
|
|
|
'thumbnail': 'md5:e1f5817e60f4a72b7e43377cf308d7ef',
|
|
|
|
|
'channel_url': 'https://live.line.me/channels/5833718',
|
|
|
|
|
'channel': 'Yahooニュース掲載🗞プロフ見てね🐕🐕',
|
|
|
|
|
'channel_id': '5833718',
|
|
|
|
|
'duration': 937,
|
|
|
|
|
'view_count': int,
|
|
|
|
|
'comment_count': int,
|
|
|
|
|
'is_live': False,
|
|
|
|
|
}
|
|
|
|
|
}, {
|
|
|
|
|
# archiveStatus == 'DELETED'
|
|
|
|
|
'url': 'https://live.line.me/channels/4778159/broadcast/16378488',
|
|
|
|
|
'only_matching': True,
|
|
|
|
|
}]
|
|
|
|
|
|
|
|
|
|
def _real_extract(self, url):
|
|
|
|
|
channel_id, broadcast_id = self._match_valid_url(url).groups()
|
|
|
|
|
broadcast = self._download_json(
|
|
|
|
|
self._API_BASE_URL + '%s/broadcast/%s' % (channel_id, broadcast_id),
|
|
|
|
|
broadcast_id)
|
|
|
|
|
item = broadcast['item']
|
|
|
|
|
info = self._parse_broadcast_item(item)
|
|
|
|
|
protocol = 'm3u8' if info['is_live'] else 'm3u8_native'
|
|
|
|
|
formats = []
|
|
|
|
|
for k, v in (broadcast.get(('live' if info['is_live'] else 'archived') + 'HLSURLs') or {}).items():
|
|
|
|
|
if not v:
|
|
|
|
|
continue
|
|
|
|
|
if k == 'abr':
|
|
|
|
|
formats.extend(self._extract_m3u8_formats(
|
|
|
|
|
v, broadcast_id, 'mp4', protocol,
|
|
|
|
|
m3u8_id='hls', fatal=False))
|
|
|
|
|
continue
|
|
|
|
|
f = {
|
|
|
|
|
'ext': 'mp4',
|
|
|
|
|
'format_id': 'hls-' + k,
|
|
|
|
|
'protocol': protocol,
|
|
|
|
|
'url': v,
|
|
|
|
|
}
|
|
|
|
|
if not k.isdigit():
|
|
|
|
|
f['vcodec'] = 'none'
|
|
|
|
|
formats.append(f)
|
|
|
|
|
if not formats:
|
|
|
|
|
archive_status = item.get('archiveStatus')
|
|
|
|
|
if archive_status != 'ARCHIVED':
|
|
|
|
|
self.raise_no_formats('this video has been ' + archive_status.lower(), expected=True)
|
|
|
|
|
info['formats'] = formats
|
|
|
|
|
return info
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
class LineLiveChannelIE(LineLiveBaseIE):
|
|
|
|
|
_VALID_URL = r'https?://live\.line\.me/channels/(?P<id>\d+)(?!/broadcast/\d+)(?:[/?&#]|$)'
|
|
|
|
|
_TEST = {
|
|
|
|
|
'url': 'https://live.line.me/channels/5893542',
|
|
|
|
|
'info_dict': {
|
|
|
|
|
'id': '5893542',
|
|
|
|
|
'title': 'いくらちゃんだよぉ🦒',
|
|
|
|
|
'description': 'md5:4d418087973ad081ceb1b3481f0b1816',
|
|
|
|
|
},
|
|
|
|
|
'playlist_mincount': 29
|
|
|
|
|
}
|
|
|
|
|
|
|
|
|
|
def _archived_broadcasts_entries(self, archived_broadcasts, channel_id):
|
|
|
|
|
while True:
|
|
|
|
|
for row in (archived_broadcasts.get('rows') or []):
|
|
|
|
|
share_url = str_or_none(row.get('shareURL'))
|
|
|
|
|
if not share_url:
|
|
|
|
|
continue
|
|
|
|
|
info = self._parse_broadcast_item(row)
|
|
|
|
|
info.update({
|
|
|
|
|
'_type': 'url',
|
|
|
|
|
'url': share_url,
|
|
|
|
|
'ie_key': LineLiveIE.ie_key(),
|
|
|
|
|
})
|
|
|
|
|
yield info
|
|
|
|
|
if not archived_broadcasts.get('hasNextPage'):
|
|
|
|
|
return
|
|
|
|
|
archived_broadcasts = self._download_json(
|
|
|
|
|
self._API_BASE_URL + channel_id + '/archived_broadcasts',
|
|
|
|
|
channel_id, query={
|
|
|
|
|
'lastId': info['id'],
|
|
|
|
|
})
|
|
|
|
|
|
|
|
|
|
def _real_extract(self, url):
|
|
|
|
|
channel_id = self._match_id(url)
|
|
|
|
|
channel = self._download_json(self._API_BASE_URL + channel_id, channel_id)
|
|
|
|
|
return self.playlist_result(
|
|
|
|
|
self._archived_broadcasts_entries(channel.get('archivedBroadcasts') or {}, channel_id),
|
|
|
|
|
channel_id, channel.get('title'), channel.get('information'))
|