from .common import InfoExtractor from math import ceil import re class ToypicsIE(InfoExtractor): _VALID_URL = r'(?:http://)?videos\.toypics\.net/.*' _TEST = { 'url': 'http://videos.toypics.net/view/514/chancebulged,-2-1/', #'md5': '8a8b546956bbd0e769dbe28f6e80abb3', == $head -c10K 12929646011616163504.mp4 |md5sum //no idea why it fails 'info_dict': { 'id': '514', 'ext': 'mp4', 'title': 'Chance-Bulge\'d, 2', 'age_limit': 18 } } PAGINATED=8 def _real_extract(self, url): mobj = re.match(r'(http://)?videos\.toypics\.net/(?P[^/?]+)$', url) if not mobj: return self.extract_one(url) return [self.extract_one(u) for u in self.process_paginated(url, r'public/">Public Videos \((?P[0-9]+)\)', r'

\n\s*' )] def process_paginated(self, profile_url, re_total, re_video_page): profile_page = self._download_webpage(profile_url, 'profile' , 'getting profile page: '+profile_url) videos_count = self._html_search_regex(re_total, profile_page, 'videos count') lst = [] for n in xrange(1,int(ceil(float(videos_count)/self.PAGINATED)) +1): lpage_url = profile_url +'/public/%d'%n lpage = self._download_webpage(lpage_url, 'page %d'%n) lst.extend(re.findall(re_video_page, lpage)) return lst def extract_one(self,url): mobj = re.match(r'(http://)?videos\.toypics\.net/view/(?P[0-9]+)/.*', url) video_id = mobj.group('videoid') page = self._download_webpage(url, video_id, 'getting page: '+url) video_url = self._html_search_regex( r'src:\s+"(http://static[0-9]+\.toypics\.net/flvideo/[^"]+)"', page, 'video URL') title = self._html_search_regex( r'Toypics - ([^<]+)', page, 'title') username = self._html_search_regex( r'toypics.net/([^/"]+)" class="user-name">', page, 'username') return { 'id': video_id, 'url': video_url, 'ext': video_url[-3:], 'title': title, 'uploader': username, 'age_limit': 18 }