1
0
mirror of https://codeberg.org/polarisfm/youtube-dl synced 2024-11-14 21:54:33 +01:00
youtube-dl/youtube_dl/extractor/cinchcast.py
2017-08-05 08:17:01 +07:00

59 lines
2.0 KiB
Python

# coding: utf-8
from __future__ import unicode_literals
from .common import InfoExtractor
from ..utils import (
unified_strdate,
xpath_text,
)
class CinchcastIE(InfoExtractor):
_VALID_URL = r'https?://player\.cinchcast\.com/.*?(?:assetId|show_id)=(?P<id>[0-9]+)'
_TESTS = [{
'url': 'http://player.cinchcast.com/?show_id=5258197&platformId=1&assetType=single',
'info_dict': {
'id': '5258197',
'ext': 'mp3',
'title': 'Train Your Brain to Up Your Game with Coach Mandy',
'upload_date': '20130816',
},
}, {
# Actual test is run in generic, look for undergroundwellness
'url': 'http://player.cinchcast.com/?platformId=1&#038;assetType=single&#038;assetId=7141703',
'only_matching': True,
}]
def _real_extract(self, url):
video_id = self._match_id(url)
doc = self._download_xml(
'http://www.blogtalkradio.com/playerasset/mrss?assetType=single&assetId=%s' % video_id,
video_id)
item = doc.find('.//item')
title = xpath_text(item, './title', fatal=True)
date_str = xpath_text(
item, './{http://developer.longtailvideo.com/trac/}date')
upload_date = unified_strdate(date_str, day_first=False)
# duration is present but wrong
formats = [{
'format_id': 'main',
'url': item.find('./{http://search.yahoo.com/mrss/}content').attrib['url'],
}]
backup_url = xpath_text(
item, './{http://developer.longtailvideo.com/trac/}backupContent')
if backup_url:
formats.append({
'preference': 2, # seems to be more reliable
'format_id': 'backup',
'url': backup_url,
})
self._sort_formats(formats)
return {
'id': video_id,
'title': title,
'upload_date': upload_date,
'formats': formats,
}