# coding: utf-8 from __future__ import unicode_literals from .common import InfoExtractor from ..compat import ( compat_str, compat_urlparse, ) from ..utils import ( determine_ext, ) class GolemIE(InfoExtractor): _VALID_URL = r'^https?://video\.golem\.de/.+?/(?P.+?)/' _TEST = { 'url': 'http://video.golem.de/handy/14095/iphone-6-und-6-plus-test.html', 'md5': 'c1a2c0a3c863319651c7c992c5ee29bf', 'info_dict': { 'id': '14095', 'format_id': 'high', 'ext': 'mp4', 'title': 'iPhone 6 und 6 Plus - Test', 'duration': 300.44, 'filesize': 65309548, } } _PREFIX = 'http://video.golem.de' def _real_extract(self, url): video_id = self._match_id(url) config = self._download_xml( 'https://video.golem.de/xml/{0}.xml'.format(video_id), video_id) info = { 'id': video_id, 'title': config.findtext('./title', 'golem'), 'duration': self._float(config.findtext('./playtime'), 'duration'), } formats = [] for e in config: url = e.findtext('./url') if not url: continue formats.append({ 'format_id': compat_str(e.tag), 'url': compat_urlparse.urljoin(self._PREFIX, url), 'height': self._int(e.get('height'), 'height'), 'width': self._int(e.get('width'), 'width'), 'filesize': self._int(e.findtext('filesize'), 'filesize'), 'ext': determine_ext(e.findtext('./filename')), }) self._sort_formats(formats) info['formats'] = formats thumbnails = [] for e in config.findall('.//teaser'): url = e.findtext('./url') if not url: continue thumbnails.append({ 'url': compat_urlparse.urljoin(self._PREFIX, url), 'width': self._int(e.get('width'), 'thumbnail width'), 'height': self._int(e.get('height'), 'thumbnail height'), }) info['thumbnails'] = thumbnails return info