import json import netrc import re import socket from .common import InfoExtractor from ..utils import ( compat_http_client, compat_str, compat_urllib_error, compat_urllib_parse, compat_urllib_request, ExtractorError, ) class FacebookIE(InfoExtractor): """Information Extractor for Facebook""" _VALID_URL = r'^(?:https?://)?(?:\w+\.)?facebook\.com/(?:video/video|photo)\.php\?(?:.*?)v=(?P\d+)(?:.*)' _LOGIN_URL = 'https://login.facebook.com/login.php?m&next=http%3A%2F%2Fm.facebook.com%2Fhome.php&' _NETRC_MACHINE = 'facebook' IE_NAME = u'facebook' _TEST = { u'url': u'https://www.facebook.com/photo.php?v=120708114770723', u'file': u'120708114770723.mp4', u'md5': u'48975a41ccc4b7a581abd68651c1a5a8', u'info_dict': { u"duration": 279, u"title": u"PEOPLE ARE AWESOME 2013" } } def report_login(self): """Report attempt to log in.""" self.to_screen(u'Logging in') def _real_initialize(self): if self._downloader is None: return useremail = None password = None downloader_params = self._downloader.params # Attempt to use provided username and password or .netrc data if downloader_params.get('username', None) is not None: useremail = downloader_params['username'] password = downloader_params['password'] elif downloader_params.get('usenetrc', False): try: info = netrc.netrc().authenticators(self._NETRC_MACHINE) if info is not None: useremail = info[0] password = info[2] else: raise netrc.NetrcParseError('No authenticators for %s' % self._NETRC_MACHINE) except (IOError, netrc.NetrcParseError) as err: self._downloader.report_warning(u'parsing .netrc: %s' % compat_str(err)) return if useremail is None: return # Log in login_form = { 'email': useremail, 'pass': password, 'login': 'Log+In' } request = compat_urllib_request.Request(self._LOGIN_URL, compat_urllib_parse.urlencode(login_form)) try: self.report_login() login_results = compat_urllib_request.urlopen(request).read() if re.search(r'', login_results) is not None: self._downloader.report_warning(u'unable to log in: bad username/password, or exceded login rate limit (~3/min). Check credentials or wait.') return except (compat_urllib_error.URLError, compat_http_client.HTTPException, socket.error) as err: self._downloader.report_warning(u'unable to log in: %s' % compat_str(err)) return def _real_extract(self, url): mobj = re.match(self._VALID_URL, url) if mobj is None: raise ExtractorError(u'Invalid URL: %s' % url) video_id = mobj.group('ID') url = 'https://www.facebook.com/video/video.php?v=%s' % video_id webpage = self._download_webpage(url, video_id) BEFORE = '{swf.addParam(param[0], param[1]);});\n' AFTER = '.forEach(function(variable) {swf.addVariable(variable[0], variable[1]);});' m = re.search(re.escape(BEFORE) + '(.*?)' + re.escape(AFTER), webpage) if not m: raise ExtractorError(u'Cannot parse data') data = dict(json.loads(m.group(1))) params_raw = compat_urllib_parse.unquote(data['params']) params = json.loads(params_raw) video_data = params['video_data'][0] video_url = video_data.get('hd_src') if not video_url: video_url = video_data['sd_src'] if not video_url: raise ExtractorError(u'Cannot find video URL') video_duration = int(video_data['video_duration']) thumbnail = video_data['thumbnail_src'] video_title = self._html_search_regex( r'

([^<]*)

', webpage, u'title') info = { 'id': video_id, 'title': video_title, 'url': video_url, 'ext': 'mp4', 'duration': video_duration, 'thumbnail': thumbnail, } return [info]