youtube-dl/youtube_dl/extractor/bild.py
2014-10-18 22:15:47 +02:00

47 lines
1.5 KiB
Python

from __future__ import unicode_literals
import re
from .common import InfoExtractor
class BildIE(InfoExtractor):
IE_NAME = 'bild'
_TEST = {
'url': 'http://www.bild.de/video/clip/apple-ipad-air/das-koennen-die-neuen-ipads-38184146.bild.html',
'info_dict': {
'id': '38184146',
'title': 'BILD hat sie getestet',
'thumbnail': 'http://bilder.bild.de/fotos/stand-das-koennen-die-neuen-ipads-38184138/Bild/1.bild.jpg',
'duration': 196,
}
}
#http://www.bild.de/video/clip/apple-ipad-air/das-koennen-die-neuen-ipads-38184146.bild.html
_VALID_URL = r'http?://(?:www\.)?bild\.de/(?:[^/]+/)+(?P<display_id>[^/]+)-(?P<id>\d+)(?:,auto=true)?\.bild\.html'
def _real_extract(self, url):
m = re.match(self._VALID_URL, url)
video_id = m.group('id')
#webpage = self._download_webpage(url, video_id)
xml_url = url.split(".bild.html")[0]+",view=xml.bild.xml"
doc = self._download_xml(xml_url, video_id)
video_url = doc.attrib['src']
title = doc.attrib['ueberschrift']
description = doc.attrib['text']
thumbnail = doc.attrib['img']
duration = int(doc.attrib['duration'])/1000
return {
'id': video_id,
'title': title,
'description': description,
'url': video_url,
'thumbnail': thumbnail,
'duration': duration,
}