You can not select more than 25 topics
Topics must start with a letter or number, can include dashes ('-') and can be up to 35 characters long.
|
|
from __future__ import unicode_literals import os.path
from .common import InfoExtractor from ..utils import ( compat_urllib_parse_urlparse,
ExtractorError, )
class MySpassIE(InfoExtractor): _VALID_URL = r'http://www\.myspass\.de/.*' _TEST = { 'url': 'http://www.myspass.de/myspass/shows/tvshows/absolute-mehrheit/Absolute-Mehrheit-vom-17022013-Die-Highlights-Teil-2--/11741/', 'md5': '0b49f4844a068f8b33f4b7c88405862b', 'info_dict': { 'id': '11741', 'ext': 'mp4', "description": "Wer kann in die Fu\u00dfstapfen von Wolfgang Kubicki treten und die Mehrheit der Zuschauer hinter sich versammeln? Wird vielleicht sogar die Absolute Mehrheit geknackt und der Jackpot von 200.000 Euro mit nach Hause genommen?", "title": "Absolute Mehrheit vom 17.02.2013 - Die Highlights, Teil 2", }, }
def _real_extract(self, url): META_DATA_URL_TEMPLATE = 'http://www.myspass.de/myspass/includes/apps/video/getvideometadataxml.php?id=%s'
# video id is the last path element of the URL # usually there is a trailing slash, so also try the second but last url_path = compat_urllib_parse_urlparse(url).path url_parent_path, video_id = os.path.split(url_path) if not video_id: _, video_id = os.path.split(url_parent_path)
# get metadata metadata_url = META_DATA_URL_TEMPLATE % video_id metadata = self._download_xml(metadata_url, video_id)
# extract values from metadata url_flv_el = metadata.find('url_flv') if url_flv_el is None: raise ExtractorError('Unable to extract download url') video_url = url_flv_el.text title_el = metadata.find('title') if title_el is None: raise ExtractorError('Unable to extract title') title = title_el.text format_id_el = metadata.find('format_id') if format_id_el is None: format = 'mp4' else: format = format_id_el.text description_el = metadata.find('description') if description_el is not None: description = description_el.text else: description = None imagePreview_el = metadata.find('imagePreview') if imagePreview_el is not None: thumbnail = imagePreview_el.text else: thumbnail = None
return { 'id': video_id, 'url': video_url, 'title': title, 'format': format, 'thumbnail': thumbnail, 'description': description, }
|