mirror of
https://github.com/ytdl-org/youtube-dl.git
synced 2024-11-23 02:32:16 +00:00
57 lines
2.6 KiB
Python
57 lines
2.6 KiB
Python
# coding: utf-8
|
|
from __future__ import unicode_literals
|
|
|
|
import re
|
|
|
|
from .common import InfoExtractor
|
|
|
|
|
|
class SeekerIE(InfoExtractor):
|
|
_VALID_URL = r'https?://(?:www\.)?seeker\.com/(?P<display_id>.*)-(?P<article_id>\d+)\.html'
|
|
_TESTS = [{
|
|
# player.loadRevision3Item
|
|
'url': 'http://www.seeker.com/should-trump-be-required-to-release-his-tax-returns-1833805621.html',
|
|
'md5': '30c1dc4030cc715cf05b423d0947ac18',
|
|
'info_dict': {
|
|
'id': '76243',
|
|
'ext': 'webm',
|
|
'title': 'Should Trump Be Required To Release His Tax Returns?',
|
|
'description': 'Donald Trump has been secretive about his "big," "beautiful" tax returns. So what can we learn if he decides to release them?',
|
|
'uploader': 'Seeker Daily',
|
|
'uploader_id': 'seekerdaily',
|
|
}
|
|
}, {
|
|
'url': 'http://www.seeker.com/changes-expected-at-zoos-following-recent-gorilla-lion-shootings-1834116536.html',
|
|
'playlist': [
|
|
{
|
|
'md5': '83bcd157cab89ad7318dd7b8c9cf1306',
|
|
'info_dict': {
|
|
'id': '67558',
|
|
'ext': 'mp4',
|
|
'title': 'The Pros & Cons Of Zoos',
|
|
'description': 'Zoos are often depicted as a terrible place for animals to live, but is there any truth to this?',
|
|
'uploader': 'DNews',
|
|
'uploader_id': 'dnews',
|
|
},
|
|
}
|
|
],
|
|
'info_dict': {
|
|
'id': '1834116536',
|
|
'title': 'After Gorilla Killing, Changes Ahead for Zoos',
|
|
'description': 'The largest association of zoos and others are hoping to learn from recent incidents that led to the shooting deaths of a gorilla and two lions.',
|
|
},
|
|
}]
|
|
|
|
def _real_extract(self, url):
|
|
display_id, article_id = re.match(self._VALID_URL, url).groups()
|
|
webpage = self._download_webpage(url, display_id)
|
|
mobj = re.search(r"player\.loadRevision3Item\('([^']+)'\s*,\s*(\d+)\);", webpage)
|
|
if mobj:
|
|
playlist_type, playlist_id = mobj.groups()
|
|
return self.url_result(
|
|
'revision3:%s:%s' % (playlist_type, playlist_id), 'Revision3Embed', playlist_id)
|
|
else:
|
|
entries = [self.url_result('revision3:video_id:%s' % video_id, 'Revision3Embed', video_id) for video_id in re.findall(
|
|
r'<iframe[^>]+src=[\'"](?:https?:)?//api\.seekernetwork\.com/player/embed\?videoId=(\d+)', webpage)]
|
|
return self.playlist_result(
|
|
entries, article_id, self._og_search_title(webpage), self._og_search_description(webpage))
|