mirror of
https://codeberg.org/polarisfm/youtube-dl
synced 2024-11-26 18:34:32 +01:00
Add new extractor
This commit is contained in:
parent
6722ebd437
commit
47f2d01a5a
@ -274,6 +274,7 @@ from .karrierevideos import KarriereVideosIE
|
|||||||
from .keezmovies import KeezMoviesIE
|
from .keezmovies import KeezMoviesIE
|
||||||
from .khanacademy import KhanAcademyIE
|
from .khanacademy import KhanAcademyIE
|
||||||
from .kickstarter import KickStarterIE
|
from .kickstarter import KickStarterIE
|
||||||
|
from .kika import KikaIE
|
||||||
from .keek import KeekIE
|
from .keek import KeekIE
|
||||||
from .kontrtube import KontrTubeIE
|
from .kontrtube import KontrTubeIE
|
||||||
from .krasview import KrasViewIE
|
from .krasview import KrasViewIE
|
||||||
|
115
youtube_dl/extractor/kika.py
Normal file
115
youtube_dl/extractor/kika.py
Normal file
@ -0,0 +1,115 @@
|
|||||||
|
# coding: utf-8
|
||||||
|
from __future__ import unicode_literals
|
||||||
|
|
||||||
|
from .common import InfoExtractor
|
||||||
|
from ..utils import ExtractorError
|
||||||
|
|
||||||
|
|
||||||
|
class KikaIE(InfoExtractor):
|
||||||
|
_VALID_URL = r'https?://(?:www\.)?kika\.de/(?:[a-z-]+/)*(?:video|sendung)(?P<id>\d+).*'
|
||||||
|
|
||||||
|
_TESTS = [
|
||||||
|
{
|
||||||
|
'url': 'http://www.kika.de/baumhaus/videos/video9572.html',
|
||||||
|
'md5': '94fc748cf5d64916571d275a07ffe2d5',
|
||||||
|
'info_dict': {
|
||||||
|
'id': '9572',
|
||||||
|
'ext': 'mp4',
|
||||||
|
'title': 'Baumhaus vom 29. Oktober 2014',
|
||||||
|
'description': None
|
||||||
|
}
|
||||||
|
},
|
||||||
|
{
|
||||||
|
'url': 'http://www.kika.de/sendungen/einzelsendungen/weihnachtsprogramm/videos/video8182.html',
|
||||||
|
'md5': '5fe9c4dd7d71e3b238f04b8fdd588357',
|
||||||
|
'info_dict': {
|
||||||
|
'id': '8182',
|
||||||
|
'ext': 'mp4',
|
||||||
|
'title': 'Beutolomäus und der geheime Weihnachtswunsch',
|
||||||
|
'description': 'md5:b69d32d7b2c55cbe86945ab309d39bbd'
|
||||||
|
}
|
||||||
|
},
|
||||||
|
{
|
||||||
|
'url': 'http://www.kika.de/videos/allevideos/video9572_zc-32ca94ad_zs-3f535991.html',
|
||||||
|
'md5': '94fc748cf5d64916571d275a07ffe2d5',
|
||||||
|
'info_dict': {
|
||||||
|
'id': '9572',
|
||||||
|
'ext': 'mp4',
|
||||||
|
'title': 'Baumhaus vom 29. Oktober 2014',
|
||||||
|
'description': None
|
||||||
|
}
|
||||||
|
},
|
||||||
|
{
|
||||||
|
'url': 'http://www.kika.de/sendungen/einzelsendungen/weihnachtsprogramm/videos/sendung81244_zc-81d703f8_zs-f82d5e31.html',
|
||||||
|
'md5': '5fe9c4dd7d71e3b238f04b8fdd588357',
|
||||||
|
'info_dict': {
|
||||||
|
'id': '8182',
|
||||||
|
'ext': 'mp4',
|
||||||
|
'title': 'Beutolomäus und der geheime Weihnachtswunsch',
|
||||||
|
'description': 'md5:b69d32d7b2c55cbe86945ab309d39bbd'
|
||||||
|
}
|
||||||
|
}
|
||||||
|
]
|
||||||
|
|
||||||
|
def _real_extract(self, url):
|
||||||
|
# broadcast_id may be the same as the video_id
|
||||||
|
broadcast_id = self._match_id(url)
|
||||||
|
webpage = self._download_webpage(url, broadcast_id)
|
||||||
|
|
||||||
|
xml_re = r'sectionArticle[ "](?:(?!sectionA[ "])(?:.|\n))*?dataURL:\'(?:/[a-z-]+?)*?/video(\d+)-avCustom\.xml'
|
||||||
|
video_id = self._search_regex(xml_re, webpage, "xml_url", default=None)
|
||||||
|
if not video_id:
|
||||||
|
# Video is not available online
|
||||||
|
err_msg = 'Video %s is not available online' % broadcast_id
|
||||||
|
raise ExtractorError(err_msg, expected=True)
|
||||||
|
|
||||||
|
xml_url = 'http://www.kika.de/video%s-avCustom.xml' % (video_id)
|
||||||
|
xml_tree = self._download_xml(xml_url, video_id)
|
||||||
|
|
||||||
|
title = xml_tree.find('title').text
|
||||||
|
webpage_url = xml_tree.find('htmlUrl').text
|
||||||
|
|
||||||
|
# Try to get the description, not available for all videos
|
||||||
|
try:
|
||||||
|
broadcast_elem = xml_tree.find('broadcast')
|
||||||
|
description = broadcast_elem.find('broadcastDescription').text
|
||||||
|
except AttributeError:
|
||||||
|
# No description available
|
||||||
|
description = None
|
||||||
|
|
||||||
|
# duration string format is mm:ss (even if it is >= 1 hour, e.g. 78:42)
|
||||||
|
tmp = xml_tree.find('duration').text.split(':')
|
||||||
|
duration = int(tmp[0]) * 60 + int(tmp[1])
|
||||||
|
|
||||||
|
formats_list = []
|
||||||
|
for elem in xml_tree.find('assets'):
|
||||||
|
format_dict = {}
|
||||||
|
format_dict['url'] = elem.find('progressiveDownloadUrl').text
|
||||||
|
format_dict['ext'] = elem.find('mediaType').text.lower()
|
||||||
|
format_dict['format'] = elem.find('profileName').text
|
||||||
|
width = int(elem.find('frameWidth').text)
|
||||||
|
height = int(elem.find('frameHeight').text)
|
||||||
|
format_dict['width'] = width
|
||||||
|
format_dict['height'] = height
|
||||||
|
format_dict['resolution'] = '%dx%d' % (width, height)
|
||||||
|
format_dict['abr'] = int(elem.find('bitrateAudio').text)
|
||||||
|
format_dict['vbr'] = int(elem.find('bitrateVideo').text)
|
||||||
|
format_dict['tbr'] = format_dict['abr'] + format_dict['vbr']
|
||||||
|
format_dict['filesize'] = int(elem.find('fileSize').text)
|
||||||
|
|
||||||
|
# append resolution and dict for sorting by resolution
|
||||||
|
formats_list.append((width * height, format_dict))
|
||||||
|
|
||||||
|
# Sort by resolution (=quality)
|
||||||
|
formats_list.sort()
|
||||||
|
|
||||||
|
out_list = [x[1] for x in formats_list]
|
||||||
|
|
||||||
|
return {
|
||||||
|
'id': video_id,
|
||||||
|
'title': title,
|
||||||
|
'description': description,
|
||||||
|
'formats': out_list,
|
||||||
|
'duration': duration,
|
||||||
|
'webpage_url': webpage_url
|
||||||
|
}
|
Loading…
Reference in New Issue
Block a user