|
|
|
@ -8,8 +8,7 @@ from ..utils import (
|
|
|
|
|
ExtractorError,
|
|
|
|
|
parse_duration,
|
|
|
|
|
try_get,
|
|
|
|
|
unified_strdate,
|
|
|
|
|
)
|
|
|
|
|
unified_strdate)
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
def get_thumbnail(data):
|
|
|
|
@ -35,10 +34,8 @@ class KanBaseIE(InfoExtractor):
|
|
|
|
|
self._search_regex(
|
|
|
|
|
r'<script[^>]+id="kan_app_search_data"[^>]*>([^<]+)</script>',
|
|
|
|
|
webpage,
|
|
|
|
|
'data',
|
|
|
|
|
),
|
|
|
|
|
video_id,
|
|
|
|
|
)
|
|
|
|
|
'data'),
|
|
|
|
|
video_id)
|
|
|
|
|
title = data.get('title') or self._og_search_title(webpage)
|
|
|
|
|
m3u8_url = try_get(data, lambda x: x['content']['src'], compat_str)
|
|
|
|
|
formats = self._extract_m3u8_formats(m3u8_url, video_id, ext='mp4')
|
|
|
|
@ -106,8 +103,7 @@ class KanPlaylistIE(KanBaseIE):
|
|
|
|
|
video_url = 'https://www.kan.org.il/Item/?itemId=%s' % video_id
|
|
|
|
|
entries.append(self.extract_item(
|
|
|
|
|
video_id,
|
|
|
|
|
self.download_webpage(video_url, video_id))
|
|
|
|
|
)
|
|
|
|
|
self.download_webpage(video_url, video_id)))
|
|
|
|
|
if not entries:
|
|
|
|
|
raise ExtractorError('Unable to extract playlist entries')
|
|
|
|
|
|
|
|
|
|