from __future__ import unicode_literals import re from .common import InfoExtractor from ..utils import ( compat_urllib_parse, determine_ext, ExtractorError, ) class AUEngineIE(InfoExtractor): _TEST = { 'url': 'http://auengine.com/embed.php?file=lfvlytY6&w=650&h=370', 'file': 'lfvlytY6.mp4', 'md5': '48972bdbcf1a3a2f5533e62425b41d4f', 'info_dict': { 'title': '[Commie]The Legend of the Legendary Heroes - 03 - Replication Eye (Alpha Stigma)[F9410F5A]' } } _VALID_URL = r'(?:http://)?(?:www\.)?auengine\.com/embed\.php\?.*?file=([^&]+).*?' def _real_extract(self, url): mobj = re.match(self._VALID_URL, url) video_id = mobj.group(1) webpage = self._download_webpage(url, video_id) title = self._html_search_regex(r'(?P<title>.+?)', webpage, 'title') title = title.strip() links = re.findall(r'\s(?:file|url):\s*["\']([^\'"]+)["\']', webpage) links = map(compat_urllib_parse.unquote, links) thumbnail = None video_url = None for link in links: if link.endswith('.png'): thumbnail = link elif '/videos/' in link: video_url = link if not video_url: raise ExtractorError(u'Could not find video URL') ext = '.' + determine_ext(video_url) if ext == title[-len(ext):]: title = title[:-len(ext)] return { 'id': video_id, 'url': video_url, 'title': title, 'thumbnail': thumbnail, }