maintainer = "kraoc"; $this->name = "Twitter Bridge Tweaked"; $this->uri = "https://twitter.com/"; $this->description = "(same as Twitter Bridge Extended, but with cleaned title & content)"; $this->update = "2016-08-09"; $this->parameters["By keyword or hashtag"] = '[ { "name" : "Keyword or #hashtag", "identifier" : "q" } ]'; $this->parameters["By username"] = '[ { "name" : "username", "identifier" : "u" } ]'; } private function containsTLD($string) { preg_match( "/(AC($|\/)|\.AD($|\/)|\.AE($|\/)|\.AERO($|\/)|\.AF($|\/)|\.AG($|\/)|\.AI($|\/)|\.AL($|\/)|\.AM($|\/)|\.AN($|\/)|\.AO($|\/)|\.AQ($|\/)|\.AR($|\/)|\.ARPA($|\/)|\.AS($|\/)|\.ASIA($|\/)|\.AT($|\/)|\.AU($|\/)|\.AW($|\/)|\.AX($|\/)|\.AZ($|\/)|\.BA($|\/)|\.BB($|\/)|\.BD($|\/)|\.BE($|\/)|\.BF($|\/)|\.BG($|\/)|\.BH($|\/)|\.BI($|\/)|\.BIZ($|\/)|\.BJ($|\/)|\.BM($|\/)|\.BN($|\/)|\.BO($|\/)|\.BR($|\/)|\.BS($|\/)|\.BT($|\/)|\.BV($|\/)|\.BW($|\/)|\.BY($|\/)|\.BZ($|\/)|\.CA($|\/)|\.CAT($|\/)|\.CC($|\/)|\.CD($|\/)|\.CF($|\/)|\.CG($|\/)|\.CH($|\/)|\.CI($|\/)|\.CK($|\/)|\.CL($|\/)|\.CM($|\/)|\.CN($|\/)|\.CO($|\/)|\.COM($|\/)|\.COOP($|\/)|\.CR($|\/)|\.CU($|\/)|\.CV($|\/)|\.CX($|\/)|\.CY($|\/)|\.CZ($|\/)|\.DE($|\/)|\.DJ($|\/)|\.DK($|\/)|\.DM($|\/)|\.DO($|\/)|\.DZ($|\/)|\.EC($|\/)|\.EDU($|\/)|\.EE($|\/)|\.EG($|\/)|\.ER($|\/)|\.ES($|\/)|\.ET($|\/)|\.EU($|\/)|\.FI($|\/)|\.FJ($|\/)|\.FK($|\/)|\.FM($|\/)|\.FO($|\/)|\.FR($|\/)|\.GA($|\/)|\.GB($|\/)|\.GD($|\/)|\.GE($|\/)|\.GF($|\/)|\.GG($|\/)|\.GH($|\/)|\.GI($|\/)|\.GL($|\/)|\.GM($|\/)|\.GN($|\/)|\.GOV($|\/)|\.GP($|\/)|\.GQ($|\/)|\.GR($|\/)|\.GS($|\/)|\.GT($|\/)|\.GU($|\/)|\.GW($|\/)|\.GY($|\/)|\.HK($|\/)|\.HM($|\/)|\.HN($|\/)|\.HR($|\/)|\.HT($|\/)|\.HU($|\/)|\.ID($|\/)|\.IE($|\/)|\.IL($|\/)|\.IM($|\/)|\.IN($|\/)|\.INFO($|\/)|\.INT($|\/)|\.IO($|\/)|\.IQ($|\/)|\.IR($|\/)|\.IS($|\/)|\.IT($|\/)|\.JE($|\/)|\.JM($|\/)|\.JO($|\/)|\.JOBS($|\/)|\.JP($|\/)|\.KE($|\/)|\.KG($|\/)|\.KH($|\/)|\.KI($|\/)|\.KM($|\/)|\.KN($|\/)|\.KP($|\/)|\.KR($|\/)|\.KW($|\/)|\.KY($|\/)|\.KZ($|\/)|\.LA($|\/)|\.LB($|\/)|\.LC($|\/)|\.LI($|\/)|\.LK($|\/)|\.LR($|\/)|\.LS($|\/)|\.LT($|\/)|\.LU($|\/)|\.LV($|\/)|\.LY($|\/)|\.MA($|\/)|\.MC($|\/)|\.MD($|\/)|\.ME($|\/)|\.MG($|\/)|\.MH($|\/)|\.MIL($|\/)|\.MK($|\/)|\.ML($|\/)|\.MM($|\/)|\.MN($|\/)|\.MO($|\/)|\.MOBI($|\/)|\.MP($|\/)|\.MQ($|\/)|\.MR($|\/)|\.MS($|\/)|\.MT($|\/)|\.MU($|\/)|\.MUSEUM($|\/)|\.MV($|\/)|\.MW($|\/)|\.MX($|\/)|\.MY($|\/)|\.MZ($|\/)|\.NA($|\/)|\.NAME($|\/)|\.NC($|\/)|\.NE($|\/)|\.NET($|\/)|\.NF($|\/)|\.NG($|\/)|\.NI($|\/)|\.NL($|\/)|\.NO($|\/)|\.NP($|\/)|\.NR($|\/)|\.NU($|\/)|\.NZ($|\/)|\.OM($|\/)|\.ORG($|\/)|\.PA($|\/)|\.PE($|\/)|\.PF($|\/)|\.PG($|\/)|\.PH($|\/)|\.PK($|\/)|\.PL($|\/)|\.PM($|\/)|\.PN($|\/)|\.PR($|\/)|\.PRO($|\/)|\.PS($|\/)|\.PT($|\/)|\.PW($|\/)|\.PY($|\/)|\.QA($|\/)|\.RE($|\/)|\.RO($|\/)|\.RS($|\/)|\.RU($|\/)|\.RW($|\/)|\.SA($|\/)|\.SB($|\/)|\.SC($|\/)|\.SD($|\/)|\.SE($|\/)|\.SG($|\/)|\.SH($|\/)|\.SI($|\/)|\.SJ($|\/)|\.SK($|\/)|\.SL($|\/)|\.SM($|\/)|\.SN($|\/)|\.SO($|\/)|\.SR($|\/)|\.ST($|\/)|\.SU($|\/)|\.SV($|\/)|\.SY($|\/)|\.SZ($|\/)|\.TC($|\/)|\.TD($|\/)|\.TEL($|\/)|\.TF($|\/)|\.TG($|\/)|\.TH($|\/)|\.TJ($|\/)|\.TK($|\/)|\.TL($|\/)|\.TM($|\/)|\.TN($|\/)|\.TO($|\/)|\.TP($|\/)|\.TR($|\/)|\.TRAVEL($|\/)|\.TT($|\/)|\.TV($|\/)|\.TW($|\/)|\.TZ($|\/)|\.UA($|\/)|\.UG($|\/)|\.UK($|\/)|\.US($|\/)|\.UY($|\/)|\.UZ($|\/)|\.VA($|\/)|\.VC($|\/)|\.VE($|\/)|\.VG($|\/)|\.VI($|\/)|\.VN($|\/)|\.VU($|\/)|\.WF($|\/)|\.WS($|\/)|\.XN--0ZWM56D($|\/)|\.XN--11B5BS3A9AJ6G($|\/)|\.XN--80AKHBYKNJ4F($|\/)|\.XN--9T4B11YI5A($|\/)|\.XN--DEBA0AD($|\/)|\.XN--G6W251D($|\/)|\.XN--HGBK6AJ7F53BBA($|\/)|\.XN--HLCJ6AYA9ESC7A($|\/)|\.XN--JXALPDLP($|\/)|\.XN--KGBECHTV($|\/)|\.XN--ZCKZAH($|\/)|\.YE($|\/)|\.YT($|\/)|\.YU($|\/)|\.ZA($|\/)|\.ZM($|\/)|\.ZW)/i", $string, $M ); $has_tld = (count($M) > 0) ? true : false; return $has_tld; } private function cleaner($url) { $U = explode(' ', $url); $W =array(); foreach ($U as $k => $u) { if (stristr($u,".")) { //only preg_match if there is a dot if ($this->containsTLD($u) === true) { unset($U[$k]); return $this->cleaner( implode(' ', $U) ); } } } return implode(' ', $U); } // (c) Kraoc / urlclean // https://github.com/kraoc/Leed-market/blob/master/urlclean/urlclean.plugin.disabled.php private function resolve_url($link) { // fallback to crawl to real url (slowest method and unsecure to privacy) if (function_exists('curl_init') && !ini_get('safe_mode')) { curl_setopt($ch, CURLOPT_USERAGENT, $ua); curl_setopt($ch, CURLOPT_URL, $link); curl_setopt($ch, CURLOPT_HEADER, true); curl_setopt($ch, CURLOPT_FOLLOWLOCATION, true); curl_setopt($ch, CURLOPT_RETURNTRANSFER, true); // >>> anonimization curl_setopt($ch, CURLOPT_COOKIESESSION, true); curl_setopt($ch, CURLOPT_REFERER, ''); // <<< anonimization $ch = curl_init(); $ua = 'Mozilla/5.0 (X11; Linux x86_64) AppleWebKit/537.16 (KHTML, like Gecko) Chrome/24.0.1304.0 Safari/537.16'; $a = curl_exec($ch); $link = curl_getinfo($ch, CURLINFO_EFFECTIVE_URL); } $link = preg_replace("/[&#?]xtor=(.)+/", "", $link); // remove: xtor $link = preg_replace("/utm_([^&#]|(&))+&*/", "", $link); // remove: utm_ // cleanup end of url $link = preg_replace("/\?&/", "", $link); if (isset($link[strlen($link) -1])){ if ($link[strlen($link) -1] == '?') $link = substr($link, 0, strlen($link) -1); } return $link; } public function collectData(array $param){ $html = ''; if (isset($param['q'])) { /* keyword search mode */ $html = $this->file_get_html('https://twitter.com/search?q='.urlencode($param['q']).'&f=tweets') or $this->returnError('No results for this query.', 404); } elseif (isset($param['u'])) { /* user timeline mode */ $html = $this->file_get_html('https://twitter.com/'.urlencode($param['u']).'/with_replies') or $this->returnError('Requested username can\'t be found.', 404); } else { $this->returnError('You must specify a keyword (?q=...) or a Twitter username (?u=...).', 400); } foreach($html->find('div.js-stream-tweet') as $tweet) { $item = new \Item(); // extract username and sanitize $item->username = $tweet->getAttribute('data-screen-name'); // extract fullname (pseudonym) $item->fullname = $tweet->getAttribute('data-name'); // get avatar link $item->avatar = $tweet->find('img', 0)->src; // get TweetID $item->id = $tweet->getAttribute('data-tweet-id'); // get tweet link $item->uri = 'https://twitter.com'.$tweet->find('a.js-permalink', 0)->getAttribute('href'); // extract tweet timestamp $item->timestamp = $tweet->find('span.js-short-timestamp', 0)->getAttribute('data-time'); // extract plaintext $item->content_simple = str_replace('href="/', 'href="https://twitter.com/', html_entity_decode(strip_tags($tweet->find('p.js-tweet-text', 0)->innertext, ''))); // processing content links foreach($tweet->find('a') as $link) { if($link->hasAttribute('data-expanded-url') ) { $link->href = $link->getAttribute('data-expanded-url'); } $link->removeAttribute('data-expanded-url'); $link->removeAttribute('data-query-source'); $link->removeAttribute('rel'); $link->removeAttribute('class'); $link->removeAttribute('target'); $link->removeAttribute('title'); } // get tweet text $item->content = 'avatar'.$item->username.' '.$item->fullname.'
'.str_replace('href="/', 'href="https://twitter.com/', $tweet->find('p.js-tweet-text', 0)->innertext).'
'; // generate the title // $item->title = $item->fullname . ' (@'. $item->username . ') | ' . $item->content_simple; $item->title = $item->content_simple; $item->title = preg_replace('|https?://www\.[a-z\.0-9]+|i', '', $item->title); // remove http(s) links $item->title = preg_replace('|www\.[a-z\.0-9]+|i', '', $item->title); // remove www. links $item->title = $this->cleaner($item->title); // remove all remaining links $item->title = trim($item->title); // remove extra spaces at beginning and end // convert all content links to real ones $regex = "/(http|https|ftp|ftps)\:\/\/[a-zA-Z0-9\-\.]+\.[a-zA-Z]{2,3}(\/\S*)?/"; $item->content = preg_replace_callback($regex, function($url) { // do stuff with $url[0] here return $this->resolve_url($url[0]); }, $item->content); // put out $this->items[] = $item; } } public function getCacheDuration(){ return 300; // 5 minutes } }