Home > php教程 > php手册 > body text

PHP 采集程序 常用函数

WBOY
Release: 2016-06-13 12:25:45
Original
780 people have browsed it

当前的脚本网址
function get_php_url(){
if(!empty($_SERVER["REQUEST_URI"])){
$scriptName = $_SERVER["REQUEST_URI"];
$nowurl = $scriptName;
}else{
$scriptName = $_SERVER["PHP_SELF"];
if(empty($_SERVER["QUERY_STRING"])) $nowurl = $scriptName;
else $nowurl = $scriptName."?".$_SERVER["QUERY_STRING"];
}
return $nowurl;
}
//把全角数字转为半角数字
function GetAlabNum($fnum){
$nums = array("0","1","2","3","4","5","6","7","8","9");
$fnums = "0123456789";
for($i=0;$i$fnum = ereg_replace("[^0-9\.]|^0{1,}","",$fnum);
if($fnum=="") $fnum=0;
return $fnum;
}
//去除HTML标记
function Text2Html($txt){
$txt = str_replace(" "," ",$txt);
$txt = str_replace("$txt = str_replace(">",">",$txt);
$txt = preg_replace("/[\r\n]{1,}/isU","
\r\n",$txt);
return $txt;
}
//清除HTML标记
function ClearHtml($str){
$str = str_replace('$str = str_replace('>','>',$str);
return $str;
}
//相对路径转化成绝对路径
function relative_to_absolute($content, $feed_url) {
preg_match('/(http|https|ftp):\/\//', $feed_url, $protocol);
$server_url = preg_replace("/(http|https|ftp|news):\/\//", "", $feed_url);
$server_url = preg_replace("/\/.*/", "", $server_url);
if ($server_url == '') {
return $content;
}
if (isset($protocol[0])) {
$new_content = preg_replace('/href="\//', 'href="'.$protocol[0].$server_url.'/', $content);
$new_content = preg_replace('/src="\//', 'src="'.$protocol[0].$server_url.'/', $new_content);
} else {
$new_content = $content;
}
return $new_content;
}
//取得所有链接
function get_all_url($code){
preg_match_all('/"\' ]+)["|\']?\s*[^>]*>([^>]+)/i',$code,$arr);
return array('name'=>$arr[2],'url'=>$arr[1]);
}
//获取指定标记中的内容
function get_tag_data($str, $start, $end){
if ( $start == '' || $end == '' ){
return;
}
$str = explode($start, $str);
$str = explode($end, $str[1]);
return $str[0];
}
//HTML表格的每行转为CSV格式数组
function get_tr_array($table) {
$table = preg_replace("'

]*?>'si",'"',$table);
$table = str_replace("",'",',$table);
$table = str_replace("","{tr}",$table);
//去掉 HTML 标记
$table = preg_replace("']*?>'si","",$table);
//去掉空白字符
$table = preg_replace("'([\r\n])[\s]+'","",$table);
$table = str_replace(" ","",$table);
$table = str_replace(" ","",$table);
$table = explode(",{tr}",$table);
array_pop($table);
return $table;
}
//将HTML表格的每行每列转为数组,采集表格数据
function get_td_array($table) {
$table = preg_replace("']*?>'si","",$table);
$table = preg_replace("']*?>'si","",$table);
$table = preg_replace("'","{tr}",$table);
$table = str_replace("","{td}",$table);
//去掉 HTML 标记
$table = preg_replace("']*?>'si","",$table);
//去掉空白字符
$table = preg_replace("'([\r\n])[\s]+'","",$table);
$table = str_replace(" ","",$table);
$table = str_replace(" ","",$table);
$table = explode('{tr}', $table);
array_pop($table);
foreach ($table as $key=>$tr) {
$td = explode('{td}', $tr);
array_pop($td);
$td_array[] = $td;
}
return $td_array;
}
//返回字符串中的所有单词 $distinct=true 去除重复
function split_en_str($str,$distinct=true) {
preg_match_all('/([a-zA-Z]+)/',$str,$match);
if ($distinct == true) {
$match[1] = array_unique($match[1]);
}
sort($match[1]);
return $match[1];
}
]*?>'si","",$table);
$table = str_replace("
source:php.cn
Statement of this Website
The content of this article is voluntarily contributed by netizens, and the copyright belongs to the original author. This site does not assume corresponding legal responsibility. If you find any content suspected of plagiarism or infringement, please contact admin@php.cn
Popular Recommendations
Popular Tutorials
More>
Latest Downloads
More>
Web Effects
Website Source Code
Website Materials
Front End Template