具体如下:
<?php
class crawler{
private $_depth=5;
private $_urls=array();
function extract_links($url)
{
if(!$this->_started){
$this->_started=1;
$curr_depth=0;
}else{
$curr_depth++;
}
if($curr_depth<$this->_depth)
{
$data=file_get_contents($url);
if(preg_match_all('/((?:http|https)://(?:www.)*(?:[a-zA-Z0-9_-]{1,15}.+[a-zA-Z0-9_]{1,}){1,}(?:[a-zA-Z0-9_/.-?&:%,!;]*))/',$data,$urls12))
{
foreach($urls12[0] as $k=>$v){
$check=get_headers($v,1);
if(strstr($v,$url) && $check[0]=='HTTP/1.1 200 OK' && !array_search($v,$this->_urls) && $curr_depth<$this->_depth){
$this->_urls[]=$v;
$this->extract_links($v);
}
}
}
}
return $this->_urls;
}
}
?>
更多信息请查看IT技术专栏