mirror of
https://github.com/WordPress/WordPress.git
synced 2024-12-23 17:48:01 +01:00
6e181bb941
git-svn-id: http://svn.automattic.com/wordpress/trunk@7140 1a063a9b-81f0-0310-95a4-ce76da25c4cd
100 lines
2.6 KiB
PHP
100 lines
2.6 KiB
PHP
<?php
|
|
/**
|
|
* WordPress implementation for PHP functions missing from older PHP versions.
|
|
*
|
|
* @package PHP
|
|
* @access private
|
|
*/
|
|
|
|
// Added in PHP 5.0
|
|
|
|
if (!function_exists('http_build_query')) {
|
|
function http_build_query($data, $prefix=null, $sep=null) {
|
|
return _http_build_query($data, $prefix, $sep);
|
|
}
|
|
}
|
|
|
|
// from php.net (modified by Mark Jaquith to behave like the native PHP5 function)
|
|
function _http_build_query($data, $prefix=null, $sep=null, $key='', $urlencode=true) {
|
|
$ret = array();
|
|
|
|
foreach ( (array) $data as $k => $v ) {
|
|
if ( $urlencode)
|
|
$k = urlencode($k);
|
|
if ( is_int($k) && $prefix != null )
|
|
$k = $prefix.$k;
|
|
if ( !empty($key) )
|
|
$k = $key . '%5B' . $k . '%5D';
|
|
if ( $v === NULL )
|
|
continue;
|
|
elseif ( $v === FALSE )
|
|
$v = '0';
|
|
|
|
if ( is_array($v) || is_object($v) )
|
|
array_push($ret,_http_build_query($v, '', $sep, $k, $urlencode));
|
|
elseif ( $urlencode )
|
|
array_push($ret, $k.'='.urlencode($v));
|
|
else
|
|
array_push($ret, $k.'='.$v);
|
|
}
|
|
|
|
if ( NULL === $sep )
|
|
$sep = ini_get('arg_separator.output');
|
|
|
|
return implode($sep, $ret);
|
|
}
|
|
|
|
if ( !function_exists('_') ) {
|
|
function _($string) {
|
|
return $string;
|
|
}
|
|
}
|
|
|
|
if (!function_exists('stripos')) {
|
|
function stripos($haystack, $needle, $offset = 0) {
|
|
return strpos(strtolower($haystack), strtolower($needle), $offset);
|
|
}
|
|
}
|
|
|
|
if ( ! function_exists('hash_hmac') ):
|
|
function hash_hmac($algo, $data, $key, $raw_output = false) {
|
|
$packs = array('md5' => 'H32', 'sha1' => 'H40');
|
|
|
|
if ( !isset($packs[$algo]) )
|
|
return false;
|
|
|
|
$pack = $packs[$algo];
|
|
|
|
if (strlen($key) > 64)
|
|
$key = pack($pack, $algo($key));
|
|
else if (strlen($key) < 64)
|
|
$key = str_pad($key, 64, chr(0));
|
|
|
|
$ipad = (substr($key, 0, 64) ^ str_repeat(chr(0x36), 64));
|
|
$opad = (substr($key, 0, 64) ^ str_repeat(chr(0x5C), 64));
|
|
|
|
return $algo($opad . pack($pack, $algo($ipad . $data)));
|
|
}
|
|
endif;
|
|
|
|
if ( ! function_exists('mb_strcut') ):
|
|
function mb_strcut( $str, $start, $length=null, $encoding=null ) {
|
|
return _mb_strcut($str, $start, $length, $encoding);
|
|
}
|
|
endif;
|
|
|
|
function _mb_strcut( $str, $start, $length=null, $encoding=null ) {
|
|
// the solution below, works only for utf-8, so in case of a different
|
|
// charset, just use built-in substr
|
|
$charset = get_option( 'blog_charset' );
|
|
if ( !in_array( $charset, array('utf8', 'utf-8', 'UTF8', 'UTF-8') ) ) {
|
|
return is_null( $length )? substr( $str, $start ) : substr( $str, $start, $length);
|
|
}
|
|
// use the regex unicode support to separate the UTF-8 characters into an array
|
|
preg_match_all( '/./us', $str, $match );
|
|
$chars = is_null( $length )? array_slice( $match[0], $start ) : array_slice( $match[0], $start, $length );
|
|
return implode( '', $chars );
|
|
}
|
|
|
|
?>
|