# coding: utf-8 import json import re from .common import InfoExtractor class TF1IE(InfoExtractor): """TF1 uses the wat.tv player.""" _VALID_URL = r'http://videos\.tf1\.fr/.*-(.*?)\.html' _TEST = { u'url': u'http://videos.tf1.fr/auto-moto/citroen-grand-c4-picasso-2013-presentation-officielle-8062060.html', u'file': u'10635995.mp4', u'md5': u'2e378cc28b9957607d5e88f274e637d8', u'info_dict': { u'title': u'Citroën Grand C4 Picasso 2013 : présentation officielle', u'description': u'Vidéo officielle du nouveau Citroën Grand C4 Picasso, lancé à l\'automne 2013.', }, u'skip': u'Sometimes wat serves the whole file with the --test option', } def _real_extract(self, url): mobj = re.match(self._VALID_URL, url) id = mobj.group(1) webpage = self._download_webpage(url, id) embed_url = self._html_search_regex(r'"(https://www.wat.tv/embedframe/.*?)"', webpage, 'embed url') embed_page = self._download_webpage(embed_url, id, u'Downloading embed player page') wat_id = self._search_regex(r'UVID=(.*?)&', embed_page, 'wat id') wat_info = self._download_webpage('http://www.wat.tv/interface/contentv3/%s' % wat_id, id, u'Downloading Wat info') wat_info = json.loads(wat_info)['media'] wat_url = wat_info['url'] return self.url_result(wat_url, 'Wat')