355c42cb30
Conflicts: include/config.php update.php
427 lines
18 KiB
PHP
427 lines
18 KiB
PHP
<?php
|
|
|
|
/**
|
|
* A UTF-8 specific character encoder that handles cleaning and transforming.
|
|
* @note All functions in this class should be static.
|
|
*/
|
|
class HTMLPurifier_Encoder
|
|
{
|
|
|
|
/**
|
|
* Constructor throws fatal error if you attempt to instantiate class
|
|
*/
|
|
private function __construct() {
|
|
trigger_error('Cannot instantiate encoder, call methods statically', E_USER_ERROR);
|
|
}
|
|
|
|
/**
|
|
* Error-handler that mutes errors, alternative to shut-up operator.
|
|
*/
|
|
public static function muteErrorHandler() {}
|
|
|
|
/**
|
|
* Cleans a UTF-8 string for well-formedness and SGML validity
|
|
*
|
|
* It will parse according to UTF-8 and return a valid UTF8 string, with
|
|
* non-SGML codepoints excluded.
|
|
*
|
|
* @note Just for reference, the non-SGML code points are 0 to 31 and
|
|
* 127 to 159, inclusive. However, we allow code points 9, 10
|
|
* and 13, which are the tab, line feed and carriage return
|
|
* respectively. 128 and above the code points map to multibyte
|
|
* UTF-8 representations.
|
|
*
|
|
* @note Fallback code adapted from utf8ToUnicode by Henri Sivonen and
|
|
* hsivonen@iki.fi at <http://iki.fi/hsivonen/php-utf8/> under the
|
|
* LGPL license. Notes on what changed are inside, but in general,
|
|
* the original code transformed UTF-8 text into an array of integer
|
|
* Unicode codepoints. Understandably, transforming that back to
|
|
* a string would be somewhat expensive, so the function was modded to
|
|
* directly operate on the string. However, this discourages code
|
|
* reuse, and the logic enumerated here would be useful for any
|
|
* function that needs to be able to understand UTF-8 characters.
|
|
* As of right now, only smart lossless character encoding converters
|
|
* would need that, and I'm probably not going to implement them.
|
|
* Once again, PHP 6 should solve all our problems.
|
|
*/
|
|
public static function cleanUTF8($str, $force_php = false) {
|
|
|
|
// UTF-8 validity is checked since PHP 4.3.5
|
|
// This is an optimization: if the string is already valid UTF-8, no
|
|
// need to do PHP stuff. 99% of the time, this will be the case.
|
|
// The regexp matches the XML char production, as well as well as excluding
|
|
// non-SGML codepoints U+007F to U+009F
|
|
if (preg_match('/^[\x{9}\x{A}\x{D}\x{20}-\x{7E}\x{A0}-\x{D7FF}\x{E000}-\x{FFFD}\x{10000}-\x{10FFFF}]*$/Du', $str)) {
|
|
return $str;
|
|
}
|
|
|
|
$mState = 0; // cached expected number of octets after the current octet
|
|
// until the beginning of the next UTF8 character sequence
|
|
$mUcs4 = 0; // cached Unicode character
|
|
$mBytes = 1; // cached expected number of octets in the current sequence
|
|
|
|
// original code involved an $out that was an array of Unicode
|
|
// codepoints. Instead of having to convert back into UTF-8, we've
|
|
// decided to directly append valid UTF-8 characters onto a string
|
|
// $out once they're done. $char accumulates raw bytes, while $mUcs4
|
|
// turns into the Unicode code point, so there's some redundancy.
|
|
|
|
$out = '';
|
|
$char = '';
|
|
|
|
$len = strlen($str);
|
|
for($i = 0; $i < $len; $i++) {
|
|
$in = ord($str{$i});
|
|
$char .= $str[$i]; // append byte to char
|
|
if (0 == $mState) {
|
|
// When mState is zero we expect either a US-ASCII character
|
|
// or a multi-octet sequence.
|
|
if (0 == (0x80 & ($in))) {
|
|
// US-ASCII, pass straight through.
|
|
if (($in <= 31 || $in == 127) &&
|
|
!($in == 9 || $in == 13 || $in == 10) // save \r\t\n
|
|
) {
|
|
// control characters, remove
|
|
} else {
|
|
$out .= $char;
|
|
}
|
|
// reset
|
|
$char = '';
|
|
$mBytes = 1;
|
|
} elseif (0xC0 == (0xE0 & ($in))) {
|
|
// First octet of 2 octet sequence
|
|
$mUcs4 = ($in);
|
|
$mUcs4 = ($mUcs4 & 0x1F) << 6;
|
|
$mState = 1;
|
|
$mBytes = 2;
|
|
} elseif (0xE0 == (0xF0 & ($in))) {
|
|
// First octet of 3 octet sequence
|
|
$mUcs4 = ($in);
|
|
$mUcs4 = ($mUcs4 & 0x0F) << 12;
|
|
$mState = 2;
|
|
$mBytes = 3;
|
|
} elseif (0xF0 == (0xF8 & ($in))) {
|
|
// First octet of 4 octet sequence
|
|
$mUcs4 = ($in);
|
|
$mUcs4 = ($mUcs4 & 0x07) << 18;
|
|
$mState = 3;
|
|
$mBytes = 4;
|
|
} elseif (0xF8 == (0xFC & ($in))) {
|
|
// First octet of 5 octet sequence.
|
|
//
|
|
// This is illegal because the encoded codepoint must be
|
|
// either:
|
|
// (a) not the shortest form or
|
|
// (b) outside the Unicode range of 0-0x10FFFF.
|
|
// Rather than trying to resynchronize, we will carry on
|
|
// until the end of the sequence and let the later error
|
|
// handling code catch it.
|
|
$mUcs4 = ($in);
|
|
$mUcs4 = ($mUcs4 & 0x03) << 24;
|
|
$mState = 4;
|
|
$mBytes = 5;
|
|
} elseif (0xFC == (0xFE & ($in))) {
|
|
// First octet of 6 octet sequence, see comments for 5
|
|
// octet sequence.
|
|
$mUcs4 = ($in);
|
|
$mUcs4 = ($mUcs4 & 1) << 30;
|
|
$mState = 5;
|
|
$mBytes = 6;
|
|
} else {
|
|
// Current octet is neither in the US-ASCII range nor a
|
|
// legal first octet of a multi-octet sequence.
|
|
$mState = 0;
|
|
$mUcs4 = 0;
|
|
$mBytes = 1;
|
|
$char = '';
|
|
}
|
|
} else {
|
|
// When mState is non-zero, we expect a continuation of the
|
|
// multi-octet sequence
|
|
if (0x80 == (0xC0 & ($in))) {
|
|
// Legal continuation.
|
|
$shift = ($mState - 1) * 6;
|
|
$tmp = $in;
|
|
$tmp = ($tmp & 0x0000003F) << $shift;
|
|
$mUcs4 |= $tmp;
|
|
|
|
if (0 == --$mState) {
|
|
// End of the multi-octet sequence. mUcs4 now contains
|
|
// the final Unicode codepoint to be output
|
|
|
|
// Check for illegal sequences and codepoints.
|
|
|
|
// From Unicode 3.1, non-shortest form is illegal
|
|
if (((2 == $mBytes) && ($mUcs4 < 0x0080)) ||
|
|
((3 == $mBytes) && ($mUcs4 < 0x0800)) ||
|
|
((4 == $mBytes) && ($mUcs4 < 0x10000)) ||
|
|
(4 < $mBytes) ||
|
|
// From Unicode 3.2, surrogate characters = illegal
|
|
(($mUcs4 & 0xFFFFF800) == 0xD800) ||
|
|
// Codepoints outside the Unicode range are illegal
|
|
($mUcs4 > 0x10FFFF)
|
|
) {
|
|
|
|
} elseif (0xFEFF != $mUcs4 && // omit BOM
|
|
// check for valid Char unicode codepoints
|
|
(
|
|
0x9 == $mUcs4 ||
|
|
0xA == $mUcs4 ||
|
|
0xD == $mUcs4 ||
|
|
(0x20 <= $mUcs4 && 0x7E >= $mUcs4) ||
|
|
// 7F-9F is not strictly prohibited by XML,
|
|
// but it is non-SGML, and thus we don't allow it
|
|
(0xA0 <= $mUcs4 && 0xD7FF >= $mUcs4) ||
|
|
(0x10000 <= $mUcs4 && 0x10FFFF >= $mUcs4)
|
|
)
|
|
) {
|
|
$out .= $char;
|
|
}
|
|
// initialize UTF8 cache (reset)
|
|
$mState = 0;
|
|
$mUcs4 = 0;
|
|
$mBytes = 1;
|
|
$char = '';
|
|
}
|
|
} else {
|
|
// ((0xC0 & (*in) != 0x80) && (mState != 0))
|
|
// Incomplete multi-octet sequence.
|
|
// used to result in complete fail, but we'll reset
|
|
$mState = 0;
|
|
$mUcs4 = 0;
|
|
$mBytes = 1;
|
|
$char ='';
|
|
}
|
|
}
|
|
}
|
|
return $out;
|
|
}
|
|
|
|
/**
|
|
* Translates a Unicode codepoint into its corresponding UTF-8 character.
|
|
* @note Based on Feyd's function at
|
|
* <http://forums.devnetwork.net/viewtopic.php?p=191404#191404>,
|
|
* which is in public domain.
|
|
* @note While we're going to do code point parsing anyway, a good
|
|
* optimization would be to refuse to translate code points that
|
|
* are non-SGML characters. However, this could lead to duplication.
|
|
* @note This is very similar to the unichr function in
|
|
* maintenance/generate-entity-file.php (although this is superior,
|
|
* due to its sanity checks).
|
|
*/
|
|
|
|
// +----------+----------+----------+----------+
|
|
// | 33222222 | 22221111 | 111111 | |
|
|
// | 10987654 | 32109876 | 54321098 | 76543210 | bit
|
|
// +----------+----------+----------+----------+
|
|
// | | | | 0xxxxxxx | 1 byte 0x00000000..0x0000007F
|
|
// | | | 110yyyyy | 10xxxxxx | 2 byte 0x00000080..0x000007FF
|
|
// | | 1110zzzz | 10yyyyyy | 10xxxxxx | 3 byte 0x00000800..0x0000FFFF
|
|
// | 11110www | 10wwzzzz | 10yyyyyy | 10xxxxxx | 4 byte 0x00010000..0x0010FFFF
|
|
// +----------+----------+----------+----------+
|
|
// | 00000000 | 00011111 | 11111111 | 11111111 | Theoretical upper limit of legal scalars: 2097151 (0x001FFFFF)
|
|
// | 00000000 | 00010000 | 11111111 | 11111111 | Defined upper limit of legal scalar codes
|
|
// +----------+----------+----------+----------+
|
|
|
|
public static function unichr($code) {
|
|
if($code > 1114111 or $code < 0 or
|
|
($code >= 55296 and $code <= 57343) ) {
|
|
// bits are set outside the "valid" range as defined
|
|
// by UNICODE 4.1.0
|
|
return '';
|
|
}
|
|
|
|
$x = $y = $z = $w = 0;
|
|
if ($code < 128) {
|
|
// regular ASCII character
|
|
$x = $code;
|
|
} else {
|
|
// set up bits for UTF-8
|
|
$x = ($code & 63) | 128;
|
|
if ($code < 2048) {
|
|
$y = (($code & 2047) >> 6) | 192;
|
|
} else {
|
|
$y = (($code & 4032) >> 6) | 128;
|
|
if($code < 65536) {
|
|
$z = (($code >> 12) & 15) | 224;
|
|
} else {
|
|
$z = (($code >> 12) & 63) | 128;
|
|
$w = (($code >> 18) & 7) | 240;
|
|
}
|
|
}
|
|
}
|
|
// set up the actual character
|
|
$ret = '';
|
|
if($w) $ret .= chr($w);
|
|
if($z) $ret .= chr($z);
|
|
if($y) $ret .= chr($y);
|
|
$ret .= chr($x);
|
|
|
|
return $ret;
|
|
}
|
|
|
|
/**
|
|
* Converts a string to UTF-8 based on configuration.
|
|
*/
|
|
public static function convertToUTF8($str, $config, $context) {
|
|
$encoding = $config->get('Core.Encoding');
|
|
if ($encoding === 'utf-8') return $str;
|
|
static $iconv = null;
|
|
if ($iconv === null) $iconv = function_exists('iconv');
|
|
set_error_handler(array('HTMLPurifier_Encoder', 'muteErrorHandler'));
|
|
if ($iconv && !$config->get('Test.ForceNoIconv')) {
|
|
$str = iconv($encoding, 'utf-8//IGNORE', $str);
|
|
if ($str === false) {
|
|
// $encoding is not a valid encoding
|
|
restore_error_handler();
|
|
trigger_error('Invalid encoding ' . $encoding, E_USER_ERROR);
|
|
return '';
|
|
}
|
|
// If the string is bjorked by Shift_JIS or a similar encoding
|
|
// that doesn't support all of ASCII, convert the naughty
|
|
// characters to their true byte-wise ASCII/UTF-8 equivalents.
|
|
$str = strtr($str, HTMLPurifier_Encoder::testEncodingSupportsASCII($encoding));
|
|
restore_error_handler();
|
|
return $str;
|
|
} elseif ($encoding === 'iso-8859-1') {
|
|
$str = utf8_encode($str);
|
|
restore_error_handler();
|
|
return $str;
|
|
}
|
|
trigger_error('Encoding not supported, please install iconv', E_USER_ERROR);
|
|
}
|
|
|
|
/**
|
|
* Converts a string from UTF-8 based on configuration.
|
|
* @note Currently, this is a lossy conversion, with unexpressable
|
|
* characters being omitted.
|
|
*/
|
|
public static function convertFromUTF8($str, $config, $context) {
|
|
$encoding = $config->get('Core.Encoding');
|
|
if ($encoding === 'utf-8') return $str;
|
|
static $iconv = null;
|
|
if ($iconv === null) $iconv = function_exists('iconv');
|
|
if ($escape = $config->get('Core.EscapeNonASCIICharacters')) {
|
|
$str = HTMLPurifier_Encoder::convertToASCIIDumbLossless($str);
|
|
}
|
|
set_error_handler(array('HTMLPurifier_Encoder', 'muteErrorHandler'));
|
|
if ($iconv && !$config->get('Test.ForceNoIconv')) {
|
|
// Undo our previous fix in convertToUTF8, otherwise iconv will barf
|
|
$ascii_fix = HTMLPurifier_Encoder::testEncodingSupportsASCII($encoding);
|
|
if (!$escape && !empty($ascii_fix)) {
|
|
$clear_fix = array();
|
|
foreach ($ascii_fix as $utf8 => $native) $clear_fix[$utf8] = '';
|
|
$str = strtr($str, $clear_fix);
|
|
}
|
|
$str = strtr($str, array_flip($ascii_fix));
|
|
// Normal stuff
|
|
$str = iconv('utf-8', $encoding . '//IGNORE', $str);
|
|
restore_error_handler();
|
|
return $str;
|
|
} elseif ($encoding === 'iso-8859-1') {
|
|
$str = utf8_decode($str);
|
|
restore_error_handler();
|
|
return $str;
|
|
}
|
|
trigger_error('Encoding not supported', E_USER_ERROR);
|
|
}
|
|
|
|
/**
|
|
* Lossless (character-wise) conversion of HTML to ASCII
|
|
* @param $str UTF-8 string to be converted to ASCII
|
|
* @returns ASCII encoded string with non-ASCII character entity-ized
|
|
* @warning Adapted from MediaWiki, claiming fair use: this is a common
|
|
* algorithm. If you disagree with this license fudgery,
|
|
* implement it yourself.
|
|
* @note Uses decimal numeric entities since they are best supported.
|
|
* @note This is a DUMB function: it has no concept of keeping
|
|
* character entities that the projected character encoding
|
|
* can allow. We could possibly implement a smart version
|
|
* but that would require it to also know which Unicode
|
|
* codepoints the charset supported (not an easy task).
|
|
* @note Sort of with cleanUTF8() but it assumes that $str is
|
|
* well-formed UTF-8
|
|
*/
|
|
public static function convertToASCIIDumbLossless($str) {
|
|
$bytesleft = 0;
|
|
$result = '';
|
|
$working = 0;
|
|
$len = strlen($str);
|
|
for( $i = 0; $i < $len; $i++ ) {
|
|
$bytevalue = ord( $str[$i] );
|
|
if( $bytevalue <= 0x7F ) { //0xxx xxxx
|
|
$result .= chr( $bytevalue );
|
|
$bytesleft = 0;
|
|
} elseif( $bytevalue <= 0xBF ) { //10xx xxxx
|
|
$working = $working << 6;
|
|
$working += ($bytevalue & 0x3F);
|
|
$bytesleft--;
|
|
if( $bytesleft <= 0 ) {
|
|
$result .= "&#" . $working . ";";
|
|
}
|
|
} elseif( $bytevalue <= 0xDF ) { //110x xxxx
|
|
$working = $bytevalue & 0x1F;
|
|
$bytesleft = 1;
|
|
} elseif( $bytevalue <= 0xEF ) { //1110 xxxx
|
|
$working = $bytevalue & 0x0F;
|
|
$bytesleft = 2;
|
|
} else { //1111 0xxx
|
|
$working = $bytevalue & 0x07;
|
|
$bytesleft = 3;
|
|
}
|
|
}
|
|
return $result;
|
|
}
|
|
|
|
/**
|
|
* This expensive function tests whether or not a given character
|
|
* encoding supports ASCII. 7/8-bit encodings like Shift_JIS will
|
|
* fail this test, and require special processing. Variable width
|
|
* encodings shouldn't ever fail.
|
|
*
|
|
* @param string $encoding Encoding name to test, as per iconv format
|
|
* @param bool $bypass Whether or not to bypass the precompiled arrays.
|
|
* @return Array of UTF-8 characters to their corresponding ASCII,
|
|
* which can be used to "undo" any overzealous iconv action.
|
|
*/
|
|
public static function testEncodingSupportsASCII($encoding, $bypass = false) {
|
|
static $encodings = array();
|
|
if (!$bypass) {
|
|
if (isset($encodings[$encoding])) return $encodings[$encoding];
|
|
$lenc = strtolower($encoding);
|
|
switch ($lenc) {
|
|
case 'shift_jis':
|
|
return array("\xC2\xA5" => '\\', "\xE2\x80\xBE" => '~');
|
|
case 'johab':
|
|
return array("\xE2\x82\xA9" => '\\');
|
|
}
|
|
if (strpos($lenc, 'iso-8859-') === 0) return array();
|
|
}
|
|
$ret = array();
|
|
set_error_handler(array('HTMLPurifier_Encoder', 'muteErrorHandler'));
|
|
if (iconv('UTF-8', $encoding, 'a') === false) return false;
|
|
for ($i = 0x20; $i <= 0x7E; $i++) { // all printable ASCII chars
|
|
$c = chr($i); // UTF-8 char
|
|
$r = iconv('UTF-8', "$encoding//IGNORE", $c); // initial conversion
|
|
if (
|
|
$r === '' ||
|
|
// This line is needed for iconv implementations that do not
|
|
// omit characters that do not exist in the target character set
|
|
($r === $c && iconv($encoding, 'UTF-8//IGNORE', $r) !== $c)
|
|
) {
|
|
// Reverse engineer: what's the UTF-8 equiv of this byte
|
|
// sequence? This assumes that there's no variable width
|
|
// encoding that doesn't support ASCII.
|
|
$ret[iconv($encoding, 'UTF-8//IGNORE', $c)] = $c;
|
|
}
|
|
}
|
|
restore_error_handler();
|
|
$encodings[$encoding] = $ret;
|
|
return $ret;
|
|
}
|
|
|
|
|
|
}
|
|
|
|
// vim: et sw=4 sts=4
|