1
0
Fork 0
mirror of https://gitlab.com/dstftw/youtube-dl.git synced 2020-11-16 09:42:26 +00:00

[closertotruth] Add extractor

Removed print statement from code.

Replaced two regex searches with the corret ones.

Removed some unnecessary semicolumns

fixed title extraction

refactored everything to search_regex

processed comments on commit 5650b0d, fixed feedback from flake8

Improved regexes and returns info dict now.

Added support for closertotruth interview URL

Added support for episodes page
This commit is contained in:
Steven Gosseling 2016-02-26 13:31:52 +01:00 committed by Sergey M․
parent 90b6288cce
commit 41c1023300
No known key found for this signature in database
GPG key ID: 2C393E0F18A9236D

View file

@ -0,0 +1,69 @@
# coding: utf-8
from __future__ import unicode_literals
from .common import InfoExtractor
class CloserToTruthIE(InfoExtractor):
_VALID_URL = r'https?://(?:www\.)?closertotruth\.com/(episodes/|(series|interviews)/(?:[^#]+#video-)?(?P<id>\d+))'
_TESTS = [
{
'url': 'http://closertotruth.com/series/solutions-the-mind-body-problem#video-3688',
'md5': '5c548bde260a9247ddfdc07c7458ed29',
'info_dict': {
'id': '0_zof1ktre',
'ext': 'mov',
'title': 'Solutions to the Mind-Body Problem?',
'upload_date': '20140221',
'timestamp': 1392956007,
'uploader_id': 'CTTXML'
}
},
{
'url': 'http://closertotruth.com/interviews/1725',
'md5': 'b00598fd6a38372edb976408f72c5792',
'info_dict': {
'id': '0_19qv5rn1',
'ext': 'mov',
'title': 'AyaFr-002 - Francisco J. Ayala',
'upload_date': '20140307',
'timestamp': 1394236431,
'uploader_id': 'CTTXML'
}
},
{
'url': 'http://closertotruth.com/episodes/how-do-brains-work',
'md5': '4dd96aa0a5c296afa5c0bd24895c2f16',
'info_dict': {
'id': '0_iuxai6g6',
'ext': 'mov',
'title': 'How do Brains Work?',
'upload_date': '20140221',
'timestamp': 1392956024,
'uploader_id': 'CTTXML'
}
},
]
def _real_extract(self, url):
video_id = self._match_id(url)
webpage = self._download_webpage(url, video_id)
video_title = self._search_regex(r'<title>(.+) \|.+</title>', webpage, 'video title')
entry_id = self._search_regex(r'<a[^>]+id="(?:video-%s|embed-kaltura)"[^>]+data-kaltura="([^"]+)' % video_id, webpage, "video entry_id")
interviewee_name = self._search_regex(r'<div id="(?:node_interview_full_group_white_wrapper|node_interview_series_full_group_ajax_content)"(?:.|\n)*<h3>(.*)</h3>.+', webpage, "video interviewee_name", False)
if interviewee_name:
video_title = video_title + ' - ' + interviewee_name
p_id = self._search_regex(r'<script[^>]+src=["\'].+?partner_id/(\d+)', webpage, "kaltura partner_id")
return {
'_type': 'url_transparent',
'id': entry_id,
'url': 'kaltura:%s:%s' % (p_id, entry_id),
'ie_key': 'Kaltura',
'title': video_title
}