from __future__ import unicode_literals import re from .common import InfoExtractor class IconosquareIE(InfoExtractor): _VALID_URL = r'https?://(www\.)?(?:iconosquare\.com|statigr\.am)/p/(?P<id>[^/]+)' _TEST = { 'url': 'http://statigr.am/p/522207370455279102_24101272', 'md5': '6eb93b882a3ded7c378ee1d6884b1814', 'info_dict': { 'id': '522207370455279102_24101272', 'ext': 'mp4', 'uploader_id': 'aguynamedpatrick', 'title': 'Instagram photo by @aguynamedpatrick (Patrick Janelle)', 'description': 'md5:644406a9ec27457ed7aa7a9ebcd4ce3d', }, } def _real_extract(self, url): mobj = re.match(self._VALID_URL, url) video_id = mobj.group('id') webpage = self._download_webpage(url, video_id) html_title = self._html_search_regex( r'<title>(.+?)</title>', webpage, 'title') title = re.sub(r'(?: *\(Videos?\))? \| (?:Iconosquare|Statigram)$', '', html_title) uploader_id = self._html_search_regex( r'@([^ ]+)', title, 'uploader name', fatal=False) return { 'id': video_id, 'url': self._og_search_video_url(webpage), 'title': title, 'description': self._og_search_description(webpage), 'thumbnail': self._og_search_thumbnail(webpage), 'uploader_id': uploader_id }