1
0
Fork 0
mirror of https://github.com/ytdl-org/youtube-dl.git synced 2024-11-18 00:05:46 +00:00
youtube-dl/youtube_dl/extractor/closertotruth.py

93 lines
3 KiB
Python
Raw Normal View History

# coding: utf-8
from __future__ import unicode_literals
import re
from .common import InfoExtractor
class CloserToTruthIE(InfoExtractor):
_VALID_URL = r'https?://(?:www\.)?closertotruth\.com/(?:[^/]+/)*(?P<id>[^/?#&]+)'
_TESTS = [{
'url': 'http://closertotruth.com/series/solutions-the-mind-body-problem#video-3688',
'info_dict': {
'id': '0_zof1ktre',
'display_id': 'solutions-the-mind-body-problem',
'ext': 'mov',
'title': 'Solutions to the Mind-Body Problem?',
'upload_date': '20140221',
'timestamp': 1392956007,
'uploader_id': 'CTTXML'
},
'params': {
'skip_download': True,
},
}, {
'url': 'http://closertotruth.com/episodes/how-do-brains-work',
'info_dict': {
'id': '0_iuxai6g6',
'display_id': 'how-do-brains-work',
'ext': 'mov',
'title': 'How do Brains Work?',
'upload_date': '20140221',
'timestamp': 1392956024,
'uploader_id': 'CTTXML'
},
'params': {
'skip_download': True,
},
}, {
'url': 'http://closertotruth.com/interviews/1725',
'info_dict': {
'id': '1725',
'title': 'AyaFr-002',
},
'playlist_mincount': 2,
}]
def _real_extract(self, url):
display_id = self._match_id(url)
webpage = self._download_webpage(url, display_id)
partner_id = self._search_regex(
r'<script[^>]+src=["\'].*?\b(?:partner_id|p)/(\d+)',
webpage, 'kaltura partner_id')
title = self._search_regex(
r'<title>(.+?)\s*\|\s*.+?</title>', webpage, 'video title')
select = self._search_regex(
r'(?s)<select[^>]+id="select-version"[^>]*>(.+?)</select>',
webpage, 'select version', default=None)
if select:
entry_ids = set()
entries = []
for mobj in re.finditer(
r'<option[^>]+value=(["\'])(?P<id>[0-9a-z_]+)(?:#.+?)?\1[^>]*>(?P<title>[^<]+)',
webpage):
entry_id = mobj.group('id')
if entry_id in entry_ids:
continue
entry_ids.add(entry_id)
entries.append({
'_type': 'url_transparent',
'url': 'kaltura:%s:%s' % (partner_id, entry_id),
'ie_key': 'Kaltura',
'title': mobj.group('title'),
})
if entries:
return self.playlist_result(entries, display_id, title)
entry_id = self._search_regex(
r'<a[^>]+id=(["\'])embed-kaltura\1[^>]+data-kaltura=(["\'])(?P<id>[0-9a-z_]+)\2',
webpage, 'kaltura entry_id', group='id')
return {
'_type': 'url_transparent',
'display_id': display_id,
'url': 'kaltura:%s:%s' % (partner_id, entry_id),
'ie_key': 'Kaltura',
'title': title
}