# encoding: utf-8 from __future__ import unicode_literals import re from .common import InfoExtractor from ..utils import ( parse_iso8601, float_or_none, int_or_none, ) class XboxClipsIE(InfoExtractor): _VALID_URL = r'https?://(?:www\.)?xboxclips\.com/video\.php\?.*vid=(?P[\w-]{36})' _TEST = { 'url': 'https://xboxclips.com/video.php?uid=2533274823424419&gamertag=Iabdulelah&vid=074a69a9-5faf-46aa-b93b-9909c1720325', 'md5': 'fbe1ec805e920aeb8eced3c3e657df5d', 'info_dict': { 'id': '074a69a9-5faf-46aa-b93b-9909c1720325', 'ext': 'mp4', 'title': 'Iabdulelah playing Upload Studio', 'filesize_approx': 28101836.8, 'timestamp': 1407388500, 'upload_date': '20140807', 'duration': 56, } } def _real_extract(self, url): mobj = re.match(self._VALID_URL, url) video_id = mobj.group('id') webpage = self._download_webpage(url, video_id) video_url = self._html_search_regex( r'>Link: ', webpage, 'video URL') title = self._html_search_regex( r'XboxClips \| ([^<]+)', webpage, 'title') timestamp = parse_iso8601(self._html_search_regex( r'>Recorded: ([^<]+)<', webpage, 'upload date', fatal=False)) filesize = float_or_none(self._html_search_regex( r'>Size: ([\d\.]+)MB<', webpage, 'file size', fatal=False), invscale=1024 * 1024) duration = int_or_none(self._html_search_regex( r'>Duration: (\d+) Seconds<', webpage, 'duration', fatal=False)) view_count = int_or_none(self._html_search_regex( r'>Views: (\d+)<', webpage, 'view count', fatal=False)) return { 'id': video_id, 'url': video_url, 'title': title, 'timestamp': timestamp, 'filesize_approx': filesize, 'duration': duration, 'view_count': view_count, }