969 lines
		
	
	
		
			No EOL
		
	
	
		
			38 KiB
		
	
	
	
		
			PHP
		
	
	
	
	
	
			
		
		
	
	
			969 lines
		
	
	
		
			No EOL
		
	
	
		
			38 KiB
		
	
	
	
		
			PHP
		
	
	
	
	
	
<?php
 | 
						|
// {{{ license
 | 
						|
 | 
						|
/* vim: set expandtab tabstop=4 shiftwidth=4 softtabstop=4 foldmethod=marker: */
 | 
						|
//
 | 
						|
// +----------------------------------------------------------------------+
 | 
						|
// | This library is free software; you can redistribute it and/or modify |
 | 
						|
// | it under the terms of the GNU Lesser General Public License as       |
 | 
						|
// | published by the Free Software Foundation; either version 2.1 of the |
 | 
						|
// | License, or (at your option) any later version.                      |
 | 
						|
// |                                                                      |
 | 
						|
// | This library is distributed in the hope that it will be useful, but  |
 | 
						|
// | WITHOUT ANY WARRANTY; without even the implied warranty of           |
 | 
						|
// | MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the GNU    |
 | 
						|
// | Lesser General Public License for more details.                      |
 | 
						|
// |                                                                      |
 | 
						|
// | You should have received a copy of the GNU Lesser General Public     |
 | 
						|
// | License along with this library; if not, write to the Free Software  |
 | 
						|
// | Foundation, Inc., 59 Temple Place, Suite 330, Boston, MA  02111-1307 |
 | 
						|
// | USA.                                                                 |
 | 
						|
// +----------------------------------------------------------------------+
 | 
						|
//
 | 
						|
 | 
						|
// }}}
 | 
						|
 | 
						|
/**
 | 
						|
 * Encode/decode Internationalized Domain Names.
 | 
						|
 *
 | 
						|
 * The class allows to convert internationalized domain names
 | 
						|
 * (see RFC 3490 for details) as they can be used with various registries worldwide
 | 
						|
 * to be translated between their original (localized) form and their encoded form
 | 
						|
 * as it will be used in the DNS (Domain Name System).
 | 
						|
 *
 | 
						|
 * The class provides two public methods, encode() and decode(), which do exactly
 | 
						|
 * what you would expect them to do. You are allowed to use complete domain names,
 | 
						|
 * simple strings and complete email addresses as well. That means, that you might
 | 
						|
 * use any of the following notations:
 | 
						|
 *
 | 
						|
 * - www.nörgler.com
 | 
						|
 * - xn--nrgler-wxa
 | 
						|
 * - xn--brse-5qa.xn--knrz-1ra.info
 | 
						|
 *
 | 
						|
 * Unicode input might be given as either UTF-8 string, UCS-4 string or UCS-4
 | 
						|
 * array. Unicode output is available in the same formats.
 | 
						|
 * You can select your preferred format via {@link set_paramter()}.
 | 
						|
 *
 | 
						|
 * ACE input and output is always expected to be ASCII.
 | 
						|
 *
 | 
						|
 * @author  Matthias Sommerfeld <mso@phlylabs.de>
 | 
						|
 * @copyright 2004-2007 phlyLabs Berlin, http://phlylabs.de
 | 
						|
 * @version 0.5.1
 | 
						|
 *
 | 
						|
 */
 | 
						|
class idna_convert
 | 
						|
{
 | 
						|
    /**
 | 
						|
     * Holds all relevant mapping tables, loaded from a seperate file on construct
 | 
						|
     * See RFC3454 for details
 | 
						|
     *
 | 
						|
     * @var array
 | 
						|
     * @access private
 | 
						|
     */
 | 
						|
    var $NP = array();
 | 
						|
 | 
						|
    // Internal settings, do not mess with them
 | 
						|
    var $_punycode_prefix = 'xn--';
 | 
						|
    var $_invalid_ucs =     0x80000000;
 | 
						|
    var $_max_ucs =         0x10FFFF;
 | 
						|
    var $_base =            36;
 | 
						|
    var $_tmin =            1;
 | 
						|
    var $_tmax =            26;
 | 
						|
    var $_skew =            38;
 | 
						|
    var $_damp =            700;
 | 
						|
    var $_initial_bias =    72;
 | 
						|
    var $_initial_n =       0x80;
 | 
						|
    var $_sbase =           0xAC00;
 | 
						|
    var $_lbase =           0x1100;
 | 
						|
    var $_vbase =           0x1161;
 | 
						|
    var $_tbase =           0x11A7;
 | 
						|
    var $_lcount =          19;
 | 
						|
    var $_vcount =          21;
 | 
						|
    var $_tcount =          28;
 | 
						|
    var $_ncount =          588;   // _vcount * _tcount
 | 
						|
    var $_scount =          11172; // _lcount * _tcount * _vcount
 | 
						|
    var $_error =           false;
 | 
						|
 | 
						|
    // See {@link set_paramter()} for details of how to change the following
 | 
						|
    // settings from within your script / application
 | 
						|
    var $_api_encoding   =  'utf8'; // Default input charset is UTF-8
 | 
						|
    var $_allow_overlong =  false;  // Overlong UTF-8 encodings are forbidden
 | 
						|
    var $_strict_mode    =  false;  // Behave strict or not
 | 
						|
 | 
						|
    // The constructor
 | 
						|
    function idna_convert($options = false)
 | 
						|
    {
 | 
						|
        $this->slast = $this->_sbase + $this->_lcount * $this->_vcount * $this->_tcount;
 | 
						|
        if (function_exists('file_get_contents')) {
 | 
						|
            $this->NP = unserialize(file_get_contents(dirname(__FILE__).'/npdata.ser'));
 | 
						|
        } else {
 | 
						|
            $this->NP = unserialize(join('', file(dirname(__FILE__).'/npdata.ser')));
 | 
						|
        }
 | 
						|
        // If parameters are given, pass these to the respective method
 | 
						|
        if (is_array($options)) {
 | 
						|
            return $this->set_parameter($options);
 | 
						|
        }
 | 
						|
        return true;
 | 
						|
    }
 | 
						|
 | 
						|
    /**
 | 
						|
     * Sets a new option value. Available options and values:
 | 
						|
     * [encoding - Use either UTF-8, UCS4 as array or UCS4 as string as input ('utf8' for UTF-8,
 | 
						|
     *         'ucs4_string' and 'ucs4_array' respectively for UCS4); The output is always UTF-8]
 | 
						|
     * [overlong - Unicode does not allow unnecessarily long encodings of chars,
 | 
						|
     *             to allow this, set this parameter to true, else to false;
 | 
						|
     *             default is false.]
 | 
						|
     * [strict - true: strict mode, good for registration purposes - Causes errors
 | 
						|
     *           on failures; false: loose mode, ideal for "wildlife" applications
 | 
						|
     *           by silently ignoring errors and returning the original input instead
 | 
						|
     *
 | 
						|
     * @param    mixed     Parameter to set (string: single parameter; array of Parameter => Value pairs)
 | 
						|
     * @param    string    Value to use (if parameter 1 is a string)
 | 
						|
     * @return   boolean   true on success, false otherwise
 | 
						|
     * @access   public
 | 
						|
     */
 | 
						|
    function set_parameter($option, $value = false)
 | 
						|
    {
 | 
						|
        if (!is_array($option)) {
 | 
						|
            $option = array($option => $value);
 | 
						|
        }
 | 
						|
        foreach ($option as $k => $v) {
 | 
						|
            switch ($k) {
 | 
						|
            case 'encoding':
 | 
						|
                switch ($v) {
 | 
						|
                case 'utf8':
 | 
						|
                case 'ucs4_string':
 | 
						|
                case 'ucs4_array':
 | 
						|
                    $this->_api_encoding = $v;
 | 
						|
                    break;
 | 
						|
                default:
 | 
						|
                    $this->_error('Set Parameter: Unknown parameter '.$v.' for option '.$k);
 | 
						|
                    return false;
 | 
						|
                }
 | 
						|
                break;
 | 
						|
            case 'overlong':
 | 
						|
                $this->_allow_overlong = ($v) ? true : false;
 | 
						|
                break;
 | 
						|
            case 'strict':
 | 
						|
                $this->_strict_mode = ($v) ? true : false;
 | 
						|
                break;
 | 
						|
            default:
 | 
						|
                $this->_error('Set Parameter: Unknown option '.$k);
 | 
						|
                return false;
 | 
						|
            }
 | 
						|
        }
 | 
						|
        return true;
 | 
						|
    }
 | 
						|
 | 
						|
    /**
 | 
						|
     * Decode a given ACE domain name
 | 
						|
     * @param    string   Domain name (ACE string)
 | 
						|
     * [@param    string   Desired output encoding, see {@link set_parameter}]
 | 
						|
     * @return   string   Decoded Domain name (UTF-8 or UCS-4)
 | 
						|
     * @access   public
 | 
						|
     */
 | 
						|
    function decode($input, $one_time_encoding = false)
 | 
						|
    {
 | 
						|
        // Optionally set
 | 
						|
        if ($one_time_encoding) {
 | 
						|
            switch ($one_time_encoding) {
 | 
						|
            case 'utf8':
 | 
						|
            case 'ucs4_string':
 | 
						|
            case 'ucs4_array':
 | 
						|
                break;
 | 
						|
            default:
 | 
						|
                $this->_error('Unknown encoding '.$one_time_encoding);
 | 
						|
                return false;
 | 
						|
            }
 | 
						|
        }
 | 
						|
        // Make sure to drop any newline characters around
 | 
						|
        $input = trim($input);
 | 
						|
 | 
						|
        // Negotiate input and try to determine, whether it is a plain string,
 | 
						|
        // an email address or something like a complete URL
 | 
						|
        if (strpos($input, '@')) { // Maybe it is an email address
 | 
						|
            // No no in strict mode
 | 
						|
            if ($this->_strict_mode) {
 | 
						|
                $this->_error('Only simple domain name parts can be handled in strict mode');
 | 
						|
                return false;
 | 
						|
            }
 | 
						|
            list ($email_pref, $input) = explode('@', $input, 2);
 | 
						|
            $arr = explode('.', $input);
 | 
						|
            foreach ($arr as $k => $v) {
 | 
						|
                if (preg_match('!^'.preg_quote($this->_punycode_prefix, '!').'!', $v)) {
 | 
						|
                    $conv = $this->_decode($v);
 | 
						|
                    if ($conv) $arr[$k] = $conv;
 | 
						|
                }
 | 
						|
            }
 | 
						|
            $input = join('.', $arr);
 | 
						|
            $arr = explode('.', $email_pref);
 | 
						|
            foreach ($arr as $k => $v) {
 | 
						|
                if (preg_match('!^'.preg_quote($this->_punycode_prefix, '!').'!', $v)) {
 | 
						|
                    $conv = $this->_decode($v);
 | 
						|
                    if ($conv) $arr[$k] = $conv;
 | 
						|
                }
 | 
						|
            }
 | 
						|
            $email_pref = join('.', $arr);
 | 
						|
            $return = $email_pref . '@' . $input;
 | 
						|
        } elseif (preg_match('![:\./]!', $input)) { // Or a complete domain name (with or without paths / parameters)
 | 
						|
            // No no in strict mode
 | 
						|
            if ($this->_strict_mode) {
 | 
						|
                $this->_error('Only simple domain name parts can be handled in strict mode');
 | 
						|
                return false;
 | 
						|
            }
 | 
						|
            $parsed = parse_url($input);
 | 
						|
            if (isset($parsed['host'])) {
 | 
						|
                $arr = explode('.', $parsed['host']);
 | 
						|
                foreach ($arr as $k => $v) {
 | 
						|
                    $conv = $this->_decode($v);
 | 
						|
                    if ($conv) $arr[$k] = $conv;
 | 
						|
                }
 | 
						|
                $parsed['host'] = join('.', $arr);
 | 
						|
                $return =
 | 
						|
                        (empty($parsed['scheme']) ? '' : $parsed['scheme'].(strtolower($parsed['scheme']) == 'mailto' ? ':' : '://'))
 | 
						|
                        .(empty($parsed['user']) ? '' : $parsed['user'].(empty($parsed['pass']) ? '' : ':'.$parsed['pass']).'@')
 | 
						|
                        .$parsed['host']
 | 
						|
                        .(empty($parsed['port']) ? '' : ':'.$parsed['port'])
 | 
						|
                        .(empty($parsed['path']) ? '' : $parsed['path'])
 | 
						|
                        .(empty($parsed['query']) ? '' : '?'.$parsed['query'])
 | 
						|
                        .(empty($parsed['fragment']) ? '' : '#'.$parsed['fragment']);
 | 
						|
            } else { // parse_url seems to have failed, try without it
 | 
						|
                $arr = explode('.', $input);
 | 
						|
                foreach ($arr as $k => $v) {
 | 
						|
                    $conv = $this->_decode($v);
 | 
						|
                    $arr[$k] = ($conv) ? $conv : $v;
 | 
						|
                }
 | 
						|
                $return = join('.', $arr);
 | 
						|
            }
 | 
						|
        } else { // Otherwise we consider it being a pure domain name string
 | 
						|
            $return = $this->_decode($input);
 | 
						|
            if (!$return) $return = $input;
 | 
						|
        }
 | 
						|
        // The output is UTF-8 by default, other output formats need conversion here
 | 
						|
        // If one time encoding is given, use this, else the objects property
 | 
						|
        switch (($one_time_encoding) ? $one_time_encoding : $this->_api_encoding) {
 | 
						|
        case 'utf8':
 | 
						|
            return $return;
 | 
						|
            break;
 | 
						|
        case 'ucs4_string':
 | 
						|
           return $this->_ucs4_to_ucs4_string($this->_utf8_to_ucs4($return));
 | 
						|
           break;
 | 
						|
        case 'ucs4_array':
 | 
						|
            return $this->_utf8_to_ucs4($return);
 | 
						|
            break;
 | 
						|
        default:
 | 
						|
            $this->_error('Unsupported output format');
 | 
						|
            return false;
 | 
						|
        }
 | 
						|
    }
 | 
						|
 | 
						|
    /**
 | 
						|
     * Encode a given UTF-8 domain name
 | 
						|
     * @param    string   Domain name (UTF-8 or UCS-4)
 | 
						|
     * [@param    string   Desired input encoding, see {@link set_parameter}]
 | 
						|
     * @return   string   Encoded Domain name (ACE string)
 | 
						|
     * @access   public
 | 
						|
     */
 | 
						|
    function encode($decoded, $one_time_encoding = false)
 | 
						|
    {
 | 
						|
        // Forcing conversion of input to UCS4 array
 | 
						|
        // If one time encoding is given, use this, else the objects property
 | 
						|
        switch ($one_time_encoding ? $one_time_encoding : $this->_api_encoding) {
 | 
						|
        case 'utf8':
 | 
						|
            $decoded = $this->_utf8_to_ucs4($decoded);
 | 
						|
            break;
 | 
						|
        case 'ucs4_string':
 | 
						|
           $decoded = $this->_ucs4_string_to_ucs4($decoded);
 | 
						|
        case 'ucs4_array':
 | 
						|
           break;
 | 
						|
        default:
 | 
						|
            $this->_error('Unsupported input format: '.($one_time_encoding ? $one_time_encoding : $this->_api_encoding));
 | 
						|
            return false;
 | 
						|
        }
 | 
						|
 | 
						|
        // No input, no output, what else did you expect?
 | 
						|
        if (empty($decoded)) return '';
 | 
						|
 | 
						|
        // Anchors for iteration
 | 
						|
        $last_begin = 0;
 | 
						|
        // Output string
 | 
						|
        $output = '';
 | 
						|
        foreach ($decoded as $k => $v) {
 | 
						|
            // Make sure to use just the plain dot
 | 
						|
            switch($v) {
 | 
						|
            case 0x3002:
 | 
						|
            case 0xFF0E:
 | 
						|
            case 0xFF61:
 | 
						|
                $decoded[$k] = 0x2E;
 | 
						|
                // Right, no break here, the above are converted to dots anyway
 | 
						|
            // Stumbling across an anchoring character
 | 
						|
            case 0x2E:
 | 
						|
            case 0x2F:
 | 
						|
            case 0x3A:
 | 
						|
            case 0x3F:
 | 
						|
            case 0x40:
 | 
						|
                // Neither email addresses nor URLs allowed in strict mode
 | 
						|
                if ($this->_strict_mode) {
 | 
						|
                   $this->_error('Neither email addresses nor URLs are allowed in strict mode.');
 | 
						|
                   return false;
 | 
						|
                } else {
 | 
						|
                    // Skip first char
 | 
						|
                    if ($k) {
 | 
						|
                        $encoded = '';
 | 
						|
                        $encoded = $this->_encode(array_slice($decoded, $last_begin, (($k)-$last_begin)));
 | 
						|
                        if ($encoded) {
 | 
						|
                            $output .= $encoded;
 | 
						|
                        } else {
 | 
						|
                            $output .= $this->_ucs4_to_utf8(array_slice($decoded, $last_begin, (($k)-$last_begin)));
 | 
						|
                        }
 | 
						|
                        $output .= chr($decoded[$k]);
 | 
						|
                    }
 | 
						|
                    $last_begin = $k + 1;
 | 
						|
                }
 | 
						|
            }
 | 
						|
        }
 | 
						|
        // Catch the rest of the string
 | 
						|
        if ($last_begin) {
 | 
						|
            $inp_len = sizeof($decoded);
 | 
						|
            $encoded = '';
 | 
						|
            $encoded = $this->_encode(array_slice($decoded, $last_begin, (($inp_len)-$last_begin)));
 | 
						|
            if ($encoded) {
 | 
						|
                $output .= $encoded;
 | 
						|
            } else {
 | 
						|
                $output .= $this->_ucs4_to_utf8(array_slice($decoded, $last_begin, (($inp_len)-$last_begin)));
 | 
						|
            }
 | 
						|
            return $output;
 | 
						|
        } else {
 | 
						|
            if ($output = $this->_encode($decoded)) {
 | 
						|
                return $output;
 | 
						|
            } else {
 | 
						|
                return $this->_ucs4_to_utf8($decoded);
 | 
						|
            }
 | 
						|
        }
 | 
						|
    }
 | 
						|
 | 
						|
    /**
 | 
						|
     * Use this method to get the last error ocurred
 | 
						|
     * @param    void
 | 
						|
     * @return   string   The last error, that occured
 | 
						|
     * @access   public
 | 
						|
     */
 | 
						|
    function get_last_error()
 | 
						|
    {
 | 
						|
        return $this->_error;
 | 
						|
    }
 | 
						|
 | 
						|
    /**
 | 
						|
     * The actual decoding algorithm
 | 
						|
     * @access   private
 | 
						|
     */
 | 
						|
    function _decode($encoded)
 | 
						|
    {
 | 
						|
        // We do need to find the Punycode prefix
 | 
						|
        if (!preg_match('!^'.preg_quote($this->_punycode_prefix, '!').'!', $encoded)) {
 | 
						|
            $this->_error('This is not a punycode string');
 | 
						|
            return false;
 | 
						|
        }
 | 
						|
        $encode_test = preg_replace('!^'.preg_quote($this->_punycode_prefix, '!').'!', '', $encoded);
 | 
						|
        // If nothing left after removing the prefix, it is hopeless
 | 
						|
        if (!$encode_test) {
 | 
						|
            $this->_error('The given encoded string was empty');
 | 
						|
            return false;
 | 
						|
        }
 | 
						|
        // Find last occurence of the delimiter
 | 
						|
        $delim_pos = strrpos($encoded, '-');
 | 
						|
        if ($delim_pos > strlen($this->_punycode_prefix)) {
 | 
						|
            for ($k = strlen($this->_punycode_prefix); $k < $delim_pos; ++$k) {
 | 
						|
                $decoded[] = ord($encoded{$k});
 | 
						|
            }
 | 
						|
        } else {
 | 
						|
            $decoded = array();
 | 
						|
        }
 | 
						|
        $deco_len = count($decoded);
 | 
						|
        $enco_len = strlen($encoded);
 | 
						|
 | 
						|
        // Wandering through the strings; init
 | 
						|
        $is_first = true;
 | 
						|
        $bias     = $this->_initial_bias;
 | 
						|
        $idx      = 0;
 | 
						|
        $char     = $this->_initial_n;
 | 
						|
 | 
						|
        for ($enco_idx = ($delim_pos) ? ($delim_pos + 1) : 0; $enco_idx < $enco_len; ++$deco_len) {
 | 
						|
            for ($old_idx = $idx, $w = 1, $k = $this->_base; 1 ; $k += $this->_base) {
 | 
						|
                $digit = $this->_decode_digit($encoded{$enco_idx++});
 | 
						|
                $idx += $digit * $w;
 | 
						|
                $t = ($k <= $bias) ? $this->_tmin :
 | 
						|
                        (($k >= $bias + $this->_tmax) ? $this->_tmax : ($k - $bias));
 | 
						|
                if ($digit < $t) break;
 | 
						|
                $w = (int) ($w * ($this->_base - $t));
 | 
						|
            }
 | 
						|
            $bias = $this->_adapt($idx - $old_idx, $deco_len + 1, $is_first);
 | 
						|
            $is_first = false;
 | 
						|
            $char += (int) ($idx / ($deco_len + 1));
 | 
						|
            $idx %= ($deco_len + 1);
 | 
						|
            if ($deco_len > 0) {
 | 
						|
                // Make room for the decoded char
 | 
						|
                for ($i = $deco_len; $i > $idx; $i--) {
 | 
						|
                    $decoded[$i] = $decoded[($i - 1)];
 | 
						|
                }
 | 
						|
            }
 | 
						|
            $decoded[$idx++] = $char;
 | 
						|
        }
 | 
						|
        return $this->_ucs4_to_utf8($decoded);
 | 
						|
    }
 | 
						|
 | 
						|
    /**
 | 
						|
     * The actual encoding algorithm
 | 
						|
     * @access   private
 | 
						|
     */
 | 
						|
    function _encode($decoded)
 | 
						|
    {
 | 
						|
        // We cannot encode a domain name containing the Punycode prefix
 | 
						|
        $extract = strlen($this->_punycode_prefix);
 | 
						|
        $check_pref = $this->_utf8_to_ucs4($this->_punycode_prefix);
 | 
						|
        $check_deco = array_slice($decoded, 0, $extract);
 | 
						|
 | 
						|
        if ($check_pref == $check_deco) {
 | 
						|
            $this->_error('This is already a punycode string');
 | 
						|
            return false;
 | 
						|
        }
 | 
						|
        // We will not try to encode strings consisting of basic code points only
 | 
						|
        $encodable = false;
 | 
						|
        foreach ($decoded as $k => $v) {
 | 
						|
            if ($v > 0x7a) {
 | 
						|
                $encodable = true;
 | 
						|
                break;
 | 
						|
            }
 | 
						|
        }
 | 
						|
        if (!$encodable) {
 | 
						|
            $this->_error('The given string does not contain encodable chars');
 | 
						|
            return false;
 | 
						|
        }
 | 
						|
 | 
						|
        // Do NAMEPREP
 | 
						|
        $decoded = $this->_nameprep($decoded);
 | 
						|
        if (!$decoded || !is_array($decoded)) return false; // NAMEPREP failed
 | 
						|
 | 
						|
        $deco_len  = count($decoded);
 | 
						|
        if (!$deco_len) return false; // Empty array
 | 
						|
 | 
						|
        $codecount = 0; // How many chars have been consumed
 | 
						|
 | 
						|
        $encoded = '';
 | 
						|
        // Copy all basic code points to output
 | 
						|
        for ($i = 0; $i < $deco_len; ++$i) {
 | 
						|
            $test = $decoded[$i];
 | 
						|
            // Will match [-0-9a-zA-Z]
 | 
						|
            if ((0x2F < $test && $test < 0x40) || (0x40 < $test && $test < 0x5B)
 | 
						|
                    || (0x60 < $test && $test <= 0x7B) || (0x2D == $test)) {
 | 
						|
                $encoded .= chr($decoded[$i]);
 | 
						|
                $codecount++;
 | 
						|
            }
 | 
						|
        }
 | 
						|
        if ($codecount == $deco_len) return $encoded; // All codepoints were basic ones
 | 
						|
 | 
						|
        // Start with the prefix; copy it to output
 | 
						|
        $encoded = $this->_punycode_prefix.$encoded;
 | 
						|
 | 
						|
        // If we have basic code points in output, add an hyphen to the end
 | 
						|
        if ($codecount) $encoded .= '-';
 | 
						|
 | 
						|
        // Now find and encode all non-basic code points
 | 
						|
        $is_first  = true;
 | 
						|
        $cur_code  = $this->_initial_n;
 | 
						|
        $bias      = $this->_initial_bias;
 | 
						|
        $delta     = 0;
 | 
						|
        while ($codecount < $deco_len) {
 | 
						|
            // Find the smallest code point >= the current code point and
 | 
						|
            // remember the last ouccrence of it in the input
 | 
						|
            for ($i = 0, $next_code = $this->_max_ucs; $i < $deco_len; $i++) {
 | 
						|
                if ($decoded[$i] >= $cur_code && $decoded[$i] <= $next_code) {
 | 
						|
                    $next_code = $decoded[$i];
 | 
						|
                }
 | 
						|
            }
 | 
						|
 | 
						|
            $delta += ($next_code - $cur_code) * ($codecount + 1);
 | 
						|
            $cur_code = $next_code;
 | 
						|
 | 
						|
            // Scan input again and encode all characters whose code point is $cur_code
 | 
						|
            for ($i = 0; $i < $deco_len; $i++) {
 | 
						|
                if ($decoded[$i] < $cur_code) {
 | 
						|
                    $delta++;
 | 
						|
                } elseif ($decoded[$i] == $cur_code) {
 | 
						|
                    for ($q = $delta, $k = $this->_base; 1; $k += $this->_base) {
 | 
						|
                        $t = ($k <= $bias) ? $this->_tmin :
 | 
						|
                                (($k >= $bias + $this->_tmax) ? $this->_tmax : $k - $bias);
 | 
						|
                        if ($q < $t) break;
 | 
						|
                        $encoded .= $this->_encode_digit(intval($t + (($q - $t) % ($this->_base - $t)))); //v0.4.5 Changed from ceil() to intval()
 | 
						|
                        $q = (int) (($q - $t) / ($this->_base - $t));
 | 
						|
                    }
 | 
						|
                    $encoded .= $this->_encode_digit($q);
 | 
						|
                    $bias = $this->_adapt($delta, $codecount+1, $is_first);
 | 
						|
                    $codecount++;
 | 
						|
                    $delta = 0;
 | 
						|
                    $is_first = false;
 | 
						|
                }
 | 
						|
            }
 | 
						|
            $delta++;
 | 
						|
            $cur_code++;
 | 
						|
        }
 | 
						|
        return $encoded;
 | 
						|
    }
 | 
						|
 | 
						|
    /**
 | 
						|
     * Adapt the bias according to the current code point and position
 | 
						|
     * @access   private
 | 
						|
     */
 | 
						|
    function _adapt($delta, $npoints, $is_first)
 | 
						|
    {
 | 
						|
        $delta = intval($is_first ? ($delta / $this->_damp) : ($delta / 2));
 | 
						|
        $delta += intval($delta / $npoints);
 | 
						|
        for ($k = 0; $delta > (($this->_base - $this->_tmin) * $this->_tmax) / 2; $k += $this->_base) {
 | 
						|
            $delta = intval($delta / ($this->_base - $this->_tmin));
 | 
						|
        }
 | 
						|
        return intval($k + ($this->_base - $this->_tmin + 1) * $delta / ($delta + $this->_skew));
 | 
						|
    }
 | 
						|
 | 
						|
    /**
 | 
						|
     * Encoding a certain digit
 | 
						|
     * @access   private
 | 
						|
     */
 | 
						|
    function _encode_digit($d)
 | 
						|
    {
 | 
						|
        return chr($d + 22 + 75 * ($d < 26));
 | 
						|
    }
 | 
						|
 | 
						|
    /**
 | 
						|
     * Decode a certain digit
 | 
						|
     * @access   private
 | 
						|
     */
 | 
						|
    function _decode_digit($cp)
 | 
						|
    {
 | 
						|
        $cp = ord($cp);
 | 
						|
        return ($cp - 48 < 10) ? $cp - 22 : (($cp - 65 < 26) ? $cp - 65 : (($cp - 97 < 26) ? $cp - 97 : $this->_base));
 | 
						|
    }
 | 
						|
 | 
						|
    /**
 | 
						|
     * Internal error handling method
 | 
						|
     * @access   private
 | 
						|
     */
 | 
						|
    function _error($error = '')
 | 
						|
    {
 | 
						|
        $this->_error = $error;
 | 
						|
    }
 | 
						|
 | 
						|
    /**
 | 
						|
     * Do Nameprep according to RFC3491 and RFC3454
 | 
						|
     * @param    array    Unicode Characters
 | 
						|
     * @return   string   Unicode Characters, Nameprep'd
 | 
						|
     * @access   private
 | 
						|
     */
 | 
						|
    function _nameprep($input)
 | 
						|
    {
 | 
						|
        $output = array();
 | 
						|
        $error = false;
 | 
						|
        //
 | 
						|
        // Mapping
 | 
						|
        // Walking through the input array, performing the required steps on each of
 | 
						|
        // the input chars and putting the result into the output array
 | 
						|
        // While mapping required chars we apply the cannonical ordering
 | 
						|
        foreach ($input as $v) {
 | 
						|
            // Map to nothing == skip that code point
 | 
						|
            if (in_array($v, $this->NP['map_nothing'])) continue;
 | 
						|
 | 
						|
            // Try to find prohibited input
 | 
						|
            if (in_array($v, $this->NP['prohibit']) || in_array($v, $this->NP['general_prohibited'])) {
 | 
						|
                $this->_error('NAMEPREP: Prohibited input U+'.sprintf('%08X', $v));
 | 
						|
                return false;
 | 
						|
            }
 | 
						|
            foreach ($this->NP['prohibit_ranges'] as $range) {
 | 
						|
                if ($range[0] <= $v && $v <= $range[1]) {
 | 
						|
                    $this->_error('NAMEPREP: Prohibited input U+'.sprintf('%08X', $v));
 | 
						|
                    return false;
 | 
						|
                }
 | 
						|
            }
 | 
						|
            //
 | 
						|
            // Hangul syllable decomposition
 | 
						|
            if (0xAC00 <= $v && $v <= 0xD7AF) {
 | 
						|
                foreach ($this->_hangul_decompose($v) as $out) {
 | 
						|
                    $output[] = (int) $out;
 | 
						|
                }
 | 
						|
            // There's a decomposition mapping for that code point
 | 
						|
            } elseif (isset($this->NP['replacemaps'][$v])) {
 | 
						|
                foreach ($this->_apply_cannonical_ordering($this->NP['replacemaps'][$v]) as $out) {
 | 
						|
                    $output[] = (int) $out;
 | 
						|
                }
 | 
						|
            } else {
 | 
						|
                $output[] = (int) $v;
 | 
						|
            }
 | 
						|
        }
 | 
						|
        // Before applying any Combining, try to rearrange any Hangul syllables
 | 
						|
        $output = $this->_hangul_compose($output);
 | 
						|
        //
 | 
						|
        // Combine code points
 | 
						|
        //
 | 
						|
        $last_class   = 0;
 | 
						|
        $last_starter = 0;
 | 
						|
        $out_len      = count($output);
 | 
						|
        for ($i = 0; $i < $out_len; ++$i) {
 | 
						|
            $class = $this->_get_combining_class($output[$i]);
 | 
						|
            if ((!$last_class || $last_class > $class) && $class) {
 | 
						|
                // Try to match
 | 
						|
                $seq_len = $i - $last_starter;
 | 
						|
                $out = $this->_combine(array_slice($output, $last_starter, $seq_len));
 | 
						|
                // On match: Replace the last starter with the composed character and remove
 | 
						|
                // the now redundant non-starter(s)
 | 
						|
                if ($out) {
 | 
						|
                    $output[$last_starter] = $out;
 | 
						|
                    if (count($out) != $seq_len) {
 | 
						|
                        for ($j = $i+1; $j < $out_len; ++$j) {
 | 
						|
                            $output[$j-1] = $output[$j];
 | 
						|
                        }
 | 
						|
                        unset($output[$out_len]);
 | 
						|
                    }
 | 
						|
                    // Rewind the for loop by one, since there can be more possible compositions
 | 
						|
                    $i--;
 | 
						|
                    $out_len--;
 | 
						|
                    $last_class = ($i == $last_starter) ? 0 : $this->_get_combining_class($output[$i-1]);
 | 
						|
                    continue;
 | 
						|
                }
 | 
						|
            }
 | 
						|
            // The current class is 0
 | 
						|
            if (!$class) $last_starter = $i;
 | 
						|
            $last_class = $class;
 | 
						|
        }
 | 
						|
        return $output;
 | 
						|
    }
 | 
						|
 | 
						|
    /**
 | 
						|
     * Decomposes a Hangul syllable
 | 
						|
     * (see http://www.unicode.org/unicode/reports/tr15/#Hangul
 | 
						|
     * @param    integer  32bit UCS4 code point
 | 
						|
     * @return   array    Either Hangul Syllable decomposed or original 32bit value as one value array
 | 
						|
     * @access   private
 | 
						|
     */
 | 
						|
    function _hangul_decompose($char)
 | 
						|
    {
 | 
						|
        $sindex = (int) $char - $this->_sbase;
 | 
						|
        if ($sindex < 0 || $sindex >= $this->_scount) {
 | 
						|
            return array($char);
 | 
						|
        }
 | 
						|
        $result = array();
 | 
						|
        $result[] = (int) $this->_lbase + $sindex / $this->_ncount;
 | 
						|
        $result[] = (int) $this->_vbase + ($sindex % $this->_ncount) / $this->_tcount;
 | 
						|
        $T = intval($this->_tbase + $sindex % $this->_tcount);
 | 
						|
        if ($T != $this->_tbase) $result[] = $T;
 | 
						|
        return $result;
 | 
						|
    }
 | 
						|
    /**
 | 
						|
     * Ccomposes a Hangul syllable
 | 
						|
     * (see http://www.unicode.org/unicode/reports/tr15/#Hangul
 | 
						|
     * @param    array    Decomposed UCS4 sequence
 | 
						|
     * @return   array    UCS4 sequence with syllables composed
 | 
						|
     * @access   private
 | 
						|
     */
 | 
						|
    function _hangul_compose($input)
 | 
						|
    {
 | 
						|
        $inp_len = count($input);
 | 
						|
        if (!$inp_len) return array();
 | 
						|
        $result = array();
 | 
						|
        $last = (int) $input[0];
 | 
						|
        $result[] = $last; // copy first char from input to output
 | 
						|
 | 
						|
        for ($i = 1; $i < $inp_len; ++$i) {
 | 
						|
            $char = (int) $input[$i];
 | 
						|
            $sindex = $last - $this->_sbase;
 | 
						|
            $lindex = $last - $this->_lbase;
 | 
						|
            $vindex = $char - $this->_vbase;
 | 
						|
            $tindex = $char - $this->_tbase;
 | 
						|
            // Find out, whether two current characters are LV and T
 | 
						|
            if (0 <= $sindex && $sindex < $this->_scount && ($sindex % $this->_tcount == 0)
 | 
						|
                    && 0 <= $tindex && $tindex <= $this->_tcount) {
 | 
						|
                // create syllable of form LVT
 | 
						|
                $last += $tindex;
 | 
						|
                $result[(count($result) - 1)] = $last; // reset last
 | 
						|
                continue; // discard char
 | 
						|
            }
 | 
						|
            // Find out, whether two current characters form L and V
 | 
						|
            if (0 <= $lindex && $lindex < $this->_lcount && 0 <= $vindex && $vindex < $this->_vcount) {
 | 
						|
                // create syllable of form LV
 | 
						|
                $last = (int) $this->_sbase + ($lindex * $this->_vcount + $vindex) * $this->_tcount;
 | 
						|
                $result[(count($result) - 1)] = $last; // reset last
 | 
						|
                continue; // discard char
 | 
						|
            }
 | 
						|
            // if neither case was true, just add the character
 | 
						|
            $last = $char;
 | 
						|
            $result[] = $char;
 | 
						|
        }
 | 
						|
        return $result;
 | 
						|
    }
 | 
						|
 | 
						|
    /**
 | 
						|
     * Returns the combining class of a certain wide char
 | 
						|
     * @param    integer    Wide char to check (32bit integer)
 | 
						|
     * @return   integer    Combining class if found, else 0
 | 
						|
     * @access   private
 | 
						|
     */
 | 
						|
    function _get_combining_class($char)
 | 
						|
    {
 | 
						|
        return isset($this->NP['norm_combcls'][$char]) ? $this->NP['norm_combcls'][$char] : 0;
 | 
						|
    }
 | 
						|
 | 
						|
    /**
 | 
						|
     * Apllies the cannonical ordering of a decomposed UCS4 sequence
 | 
						|
     * @param    array      Decomposed UCS4 sequence
 | 
						|
     * @return   array      Ordered USC4 sequence
 | 
						|
     * @access   private
 | 
						|
     */
 | 
						|
    function _apply_cannonical_ordering($input)
 | 
						|
    {
 | 
						|
        $swap = true;
 | 
						|
        $size = count($input);
 | 
						|
        while ($swap) {
 | 
						|
            $swap = false;
 | 
						|
            $last = $this->_get_combining_class(intval($input[0]));
 | 
						|
            for ($i = 0; $i < $size-1; ++$i) {
 | 
						|
                $next = $this->_get_combining_class(intval($input[$i+1]));
 | 
						|
                if ($next != 0 && $last > $next) {
 | 
						|
                    // Move item leftward until it fits
 | 
						|
                    for ($j = $i + 1; $j > 0; --$j) {
 | 
						|
                        if ($this->_get_combining_class(intval($input[$j-1])) <= $next) break;
 | 
						|
                        $t = intval($input[$j]);
 | 
						|
                        $input[$j] = intval($input[$j-1]);
 | 
						|
                        $input[$j-1] = $t;
 | 
						|
                        $swap = true;
 | 
						|
                    }
 | 
						|
                    // Reentering the loop looking at the old character again
 | 
						|
                    $next = $last;
 | 
						|
                }
 | 
						|
                $last = $next;
 | 
						|
            }
 | 
						|
        }
 | 
						|
        return $input;
 | 
						|
    }
 | 
						|
 | 
						|
    /**
 | 
						|
     * Do composition of a sequence of starter and non-starter
 | 
						|
     * @param    array      UCS4 Decomposed sequence
 | 
						|
     * @return   array      Ordered USC4 sequence
 | 
						|
     * @access   private
 | 
						|
     */
 | 
						|
    function _combine($input)
 | 
						|
    {
 | 
						|
        $inp_len = count($input);
 | 
						|
        foreach ($this->NP['replacemaps'] as $np_src => $np_target) {
 | 
						|
            if ($np_target[0] != $input[0]) continue;
 | 
						|
            if (count($np_target) != $inp_len) continue;
 | 
						|
            $hit = false;
 | 
						|
            foreach ($input as $k2 => $v2) {
 | 
						|
                if ($v2 == $np_target[$k2]) {
 | 
						|
                    $hit = true;
 | 
						|
                } else {
 | 
						|
                    $hit = false;
 | 
						|
                    break;
 | 
						|
                }
 | 
						|
            }
 | 
						|
            if ($hit) return $np_src;
 | 
						|
        }
 | 
						|
        return false;
 | 
						|
    }
 | 
						|
 | 
						|
    /**
 | 
						|
     * This converts an UTF-8 encoded string to its UCS-4 representation
 | 
						|
     * By talking about UCS-4 "strings" we mean arrays of 32bit integers representing
 | 
						|
     * each of the "chars". This is due to PHP not being able to handle strings with
 | 
						|
     * bit depth different from 8. This apllies to the reverse method _ucs4_to_utf8(), too.
 | 
						|
     * The following UTF-8 encodings are supported:
 | 
						|
     * bytes bits  representation
 | 
						|
     * 1        7  0xxxxxxx
 | 
						|
     * 2       11  110xxxxx 10xxxxxx
 | 
						|
     * 3       16  1110xxxx 10xxxxxx 10xxxxxx
 | 
						|
     * 4       21  11110xxx 10xxxxxx 10xxxxxx 10xxxxxx
 | 
						|
     * 5       26  111110xx 10xxxxxx 10xxxxxx 10xxxxxx 10xxxxxx
 | 
						|
     * 6       31  1111110x 10xxxxxx 10xxxxxx 10xxxxxx 10xxxxxx 10xxxxxx
 | 
						|
     * Each x represents a bit that can be used to store character data.
 | 
						|
     * The five and six byte sequences are part of Annex D of ISO/IEC 10646-1:2000
 | 
						|
     * @access   private
 | 
						|
     */
 | 
						|
    function _utf8_to_ucs4($input)
 | 
						|
    {
 | 
						|
        $output = array();
 | 
						|
        $out_len = 0;
 | 
						|
        $inp_len = strlen($input);
 | 
						|
        $mode = 'next';
 | 
						|
        $test = 'none';
 | 
						|
        for ($k = 0; $k < $inp_len; ++$k) {
 | 
						|
            $v = ord($input{$k}); // Extract byte from input string
 | 
						|
 | 
						|
            if ($v < 128) { // We found an ASCII char - put into stirng as is
 | 
						|
                $output[$out_len] = $v;
 | 
						|
                ++$out_len;
 | 
						|
                if ('add' == $mode) {
 | 
						|
                    $this->_error('Conversion from UTF-8 to UCS-4 failed: malformed input at byte '.$k);
 | 
						|
                    return false;
 | 
						|
                }
 | 
						|
                continue;
 | 
						|
            }
 | 
						|
            if ('next' == $mode) { // Try to find the next start byte; determine the width of the Unicode char
 | 
						|
                $start_byte = $v;
 | 
						|
                $mode = 'add';
 | 
						|
                $test = 'range';
 | 
						|
                if ($v >> 5 == 6) { // &110xxxxx 10xxxxx
 | 
						|
                    $next_byte = 0; // Tells, how many times subsequent bitmasks must rotate 6bits to the left
 | 
						|
                    $v = ($v - 192) << 6;
 | 
						|
                } elseif ($v >> 4 == 14) { // &1110xxxx 10xxxxxx 10xxxxxx
 | 
						|
                    $next_byte = 1;
 | 
						|
                    $v = ($v - 224) << 12;
 | 
						|
                } elseif ($v >> 3 == 30) { // &11110xxx 10xxxxxx 10xxxxxx 10xxxxxx
 | 
						|
                    $next_byte = 2;
 | 
						|
                    $v = ($v - 240) << 18;
 | 
						|
                } elseif ($v >> 2 == 62) { // &111110xx 10xxxxxx 10xxxxxx 10xxxxxx 10xxxxxx
 | 
						|
                    $next_byte = 3;
 | 
						|
                    $v = ($v - 248) << 24;
 | 
						|
                } elseif ($v >> 1 == 126) { // &1111110x 10xxxxxx 10xxxxxx 10xxxxxx 10xxxxxx 10xxxxxx
 | 
						|
                    $next_byte = 4;
 | 
						|
                    $v = ($v - 252) << 30;
 | 
						|
                } else {
 | 
						|
                    $this->_error('This might be UTF-8, but I don\'t understand it at byte '.$k);
 | 
						|
                    return false;
 | 
						|
                }
 | 
						|
                if ('add' == $mode) {
 | 
						|
                    $output[$out_len] = (int) $v;
 | 
						|
                    ++$out_len;
 | 
						|
                    continue;
 | 
						|
                }
 | 
						|
            }
 | 
						|
            if ('add' == $mode) {
 | 
						|
                if (!$this->_allow_overlong && $test == 'range') {
 | 
						|
                    $test = 'none';
 | 
						|
                    if (($v < 0xA0 && $start_byte == 0xE0) || ($v < 0x90 && $start_byte == 0xF0) || ($v > 0x8F && $start_byte == 0xF4)) {
 | 
						|
                        $this->_error('Bogus UTF-8 character detected (out of legal range) at byte '.$k);
 | 
						|
                        return false;
 | 
						|
                    }
 | 
						|
                }
 | 
						|
                if ($v >> 6 == 2) { // Bit mask must be 10xxxxxx
 | 
						|
                    $v = ($v - 128) << ($next_byte * 6);
 | 
						|
                    $output[($out_len - 1)] += $v;
 | 
						|
                    --$next_byte;
 | 
						|
                } else {
 | 
						|
                    $this->_error('Conversion from UTF-8 to UCS-4 failed: malformed input at byte '.$k);
 | 
						|
                    return false;
 | 
						|
                }
 | 
						|
                if ($next_byte < 0) {
 | 
						|
                    $mode = 'next';
 | 
						|
                }
 | 
						|
            }
 | 
						|
        } // for
 | 
						|
        return $output;
 | 
						|
    }
 | 
						|
 | 
						|
    /**
 | 
						|
     * Convert UCS-4 string into UTF-8 string
 | 
						|
     * See _utf8_to_ucs4() for details
 | 
						|
     * @access   private
 | 
						|
     */
 | 
						|
    function _ucs4_to_utf8($input)
 | 
						|
    {
 | 
						|
        $output = '';
 | 
						|
        $k = 0;
 | 
						|
        foreach ($input as $v) {
 | 
						|
            ++$k;
 | 
						|
            // $v = ord($v);
 | 
						|
            if ($v < 128) { // 7bit are transferred literally
 | 
						|
                $output .= chr($v);
 | 
						|
            } elseif ($v < (1 << 11)) { // 2 bytes
 | 
						|
                $output .= chr(192 + ($v >> 6)) . chr(128 + ($v & 63));
 | 
						|
            } elseif ($v < (1 << 16)) { // 3 bytes
 | 
						|
                $output .= chr(224 + ($v >> 12)) . chr(128 + (($v >> 6) & 63)) . chr(128 + ($v & 63));
 | 
						|
            } elseif ($v < (1 << 21)) { // 4 bytes
 | 
						|
                $output .= chr(240 + ($v >> 18)) . chr(128 + (($v >> 12) & 63))
 | 
						|
                         . chr(128 + (($v >> 6) & 63)) . chr(128 + ($v & 63));
 | 
						|
            } elseif ($v < (1 << 26)) { // 5 bytes
 | 
						|
                $output .= chr(248 + ($v >> 24)) . chr(128 + (($v >> 18) & 63))
 | 
						|
                         . chr(128 + (($v >> 12) & 63)) . chr(128 + (($v >> 6) & 63))
 | 
						|
                         . chr(128 + ($v & 63));
 | 
						|
            } elseif ($v < (1 << 31)) { // 6 bytes
 | 
						|
                $output .= chr(252 + ($v >> 30)) . chr(128 + (($v >> 24) & 63))
 | 
						|
                         . chr(128 + (($v >> 18) & 63)) . chr(128 + (($v >> 12) & 63))
 | 
						|
                         . chr(128 + (($v >> 6) & 63)) . chr(128 + ($v & 63));
 | 
						|
            } else {
 | 
						|
                $this->_error('Conversion from UCS-4 to UTF-8 failed: malformed input at byte '.$k);
 | 
						|
                return false;
 | 
						|
            }
 | 
						|
        }
 | 
						|
        return $output;
 | 
						|
    }
 | 
						|
 | 
						|
    /**
 | 
						|
      * Convert UCS-4 array into UCS-4 string
 | 
						|
      *
 | 
						|
      * @access   private
 | 
						|
      */
 | 
						|
    function _ucs4_to_ucs4_string($input)
 | 
						|
    {
 | 
						|
        $output = '';
 | 
						|
        // Take array values and split output to 4 bytes per value
 | 
						|
        // The bit mask is 255, which reads &11111111
 | 
						|
        foreach ($input as $v) {
 | 
						|
            $output .= chr(($v >> 24) & 255).chr(($v >> 16) & 255).chr(($v >> 8) & 255).chr($v & 255);
 | 
						|
        }
 | 
						|
        return $output;
 | 
						|
    }
 | 
						|
 | 
						|
    /**
 | 
						|
      * Convert UCS-4 strin into UCS-4 garray
 | 
						|
      *
 | 
						|
      * @access   private
 | 
						|
      */
 | 
						|
    function _ucs4_string_to_ucs4($input)
 | 
						|
    {
 | 
						|
        $output = array();
 | 
						|
        $inp_len = strlen($input);
 | 
						|
        // Input length must be dividable by 4
 | 
						|
        if ($inp_len % 4) {
 | 
						|
            $this->_error('Input UCS4 string is broken');
 | 
						|
            return false;
 | 
						|
        }
 | 
						|
        // Empty input - return empty output
 | 
						|
        if (!$inp_len) return $output;
 | 
						|
        for ($i = 0, $out_len = -1; $i < $inp_len; ++$i) {
 | 
						|
            // Increment output position every 4 input bytes
 | 
						|
            if (!($i % 4)) {
 | 
						|
                $out_len++;
 | 
						|
                $output[$out_len] = 0;
 | 
						|
            }
 | 
						|
            $output[$out_len] += ord($input{$i}) << (8 * (3 - ($i % 4) ) );
 | 
						|
        }
 | 
						|
        return $output;
 | 
						|
    }
 | 
						|
}
 | 
						|
 | 
						|
/**
 | 
						|
* Adapter class for aligning the API of idna_convert with that of Net_IDNA
 | 
						|
* @author  Matthias Sommerfeld <mso@phlylabs.de>
 | 
						|
*/
 | 
						|
class Net_IDNA_php4 extends idna_convert
 | 
						|
{
 | 
						|
    /**
 | 
						|
     * Sets a new option value. Available options and values:
 | 
						|
     * [encoding - Use either UTF-8, UCS4 as array or UCS4 as string as input ('utf8' for UTF-8,
 | 
						|
     *         'ucs4_string' and 'ucs4_array' respectively for UCS4); The output is always UTF-8]
 | 
						|
     * [overlong - Unicode does not allow unnecessarily long encodings of chars,
 | 
						|
     *             to allow this, set this parameter to true, else to false;
 | 
						|
     *             default is false.]
 | 
						|
     * [strict - true: strict mode, good for registration purposes - Causes errors
 | 
						|
     *           on failures; false: loose mode, ideal for "wildlife" applications
 | 
						|
     *           by silently ignoring errors and returning the original input instead
 | 
						|
     *
 | 
						|
     * @param    mixed     Parameter to set (string: single parameter; array of Parameter => Value pairs)
 | 
						|
     * @param    string    Value to use (if parameter 1 is a string)
 | 
						|
     * @return   boolean   true on success, false otherwise
 | 
						|
     * @access   public
 | 
						|
     */
 | 
						|
    function setParams($option, $param = false)
 | 
						|
    {
 | 
						|
        return $this->IC->set_parameters($option, $param);
 | 
						|
    }
 | 
						|
}
 | 
						|
 | 
						|
?>
 |