X-Git-Url: https://git.rapsys.eu/youtubedl/blobdiff_plain/d1a5eac72c0b1116e8efbbf50766d8d00678f5c8..75535d182355c0cc00271ceae302e8e4aba9c04e:/youtube_dl/extractor/dreisat.py?ds=sidebyside diff --git a/youtube_dl/extractor/dreisat.py b/youtube_dl/extractor/dreisat.py index 0b11d1f..69ca754 100644 --- a/youtube_dl/extractor/dreisat.py +++ b/youtube_dl/extractor/dreisat.py @@ -1,39 +1,37 @@ -# coding: utf-8 +from __future__ import unicode_literals import re from .common import InfoExtractor -from ..utils import ( - unified_strdate, -) +from ..utils import unified_strdate class DreiSatIE(InfoExtractor): IE_NAME = '3sat' _VALID_URL = r'(?:http://)?(?:www\.)?3sat\.de/mediathek/(?:index\.php)?\?(?:(?:mode|display)=[^&]+&)*obj=(?P[0-9]+)$' _TEST = { - u"url": u"http://www.3sat.de/mediathek/index.php?obj=36983", - u'file': u'36983.mp4', - u'md5': u'9dcfe344732808dbfcc901537973c922', - u'info_dict': { - u"title": u"Kaffeeland Schweiz", - u"description": u"Über 80 Kaffeeröstereien liefern in der Schweiz das Getränk, in das das Land so vernarrt ist: Mehr als 1000 Tassen trinkt ein Schweizer pro Jahr. SCHWEIZWEIT nimmt die Kaffeekultur unter die...", - u"uploader": u"3sat", - u"upload_date": u"20130622" + 'url': 'http://www.3sat.de/mediathek/index.php?obj=36983', + 'md5': '9dcfe344732808dbfcc901537973c922', + 'info_dict': { + 'id': '36983', + 'ext': 'mp4', + 'title': 'Kaffeeland Schweiz', + 'description': 'md5:cc4424b18b75ae9948b13929a0814033', + 'uploader': '3sat', + 'upload_date': '20130622' } } - def _real_extract(self, url): mobj = re.match(self._VALID_URL, url) video_id = mobj.group('id') details_url = 'http://www.3sat.de/mediathek/xmlservice/web/beitragsDetails?ak=web&id=%s' % video_id - details_doc = self._download_xml(details_url, video_id, note=u'Downloading video details') + details_doc = self._download_xml(details_url, video_id, 'Downloading video details') thumbnail_els = details_doc.findall('.//teaserimage') thumbnails = [{ - 'width': te.attrib['key'].partition('x')[0], - 'height': te.attrib['key'].partition('x')[2], + 'width': int(te.attrib['key'].partition('x')[0]), + 'height': int(te.attrib['key'].partition('x')[2]), 'url': te.text, } for te in thumbnail_els]