Fix iconv truncation bug.
[htmlpurifier.git] / library / HTMLPurifier / Encoder.php
blob890c890ea05446b9498e5e0a01c68138a1dcdfdd
1 <?php
3 /**
4 * A UTF-8 specific character encoder that handles cleaning and transforming.
5 * @note All functions in this class should be static.
6 */
7 class HTMLPurifier_Encoder
10 /**
11 * Constructor throws fatal error if you attempt to instantiate class
13 private function __construct() {
14 trigger_error('Cannot instantiate encoder, call methods statically', E_USER_ERROR);
17 /**
18 * Error-handler that mutes errors, alternative to shut-up operator.
20 public static function muteErrorHandler() {}
22 /**
23 * iconv wrapper which mutes errors, but doesn't work around bugs.
25 public static function unsafeIconv($in, $out, $text) {
26 set_error_handler(array('HTMLPurifier_Encoder', 'muteErrorHandler'));
27 $r = iconv($in, $out, $text);
28 restore_error_handler();
29 return $r;
32 /**
33 * iconv wrapper which mutes errors and works around bugs.
35 public static function iconv($in, $out, $text, $max_chunk_size = 8000) {
36 $code = self::testIconvTruncateBug();
37 if ($code == self::ICONV_OK) {
38 return self::unsafeIconv($in, $out, $text);
39 } elseif ($code == self::ICONV_TRUNCATES) {
40 // we can only work around this if the input character set
41 // is utf-8
42 if ($in == 'utf-8') {
43 if ($max_chunk_size < 4) {
44 trigger_error('max_chunk_size is too small', E_USER_WARNING);
45 return false;
47 // split into 8000 byte chunks, but be careful to handle
48 // multibyte boundaries properly
49 if (($c = strlen($text)) <= $max_chunk_size) {
50 return self::unsafeIconv($in, $out, $text);
52 $r = '';
53 $i = 0;
54 while (true) {
55 if ($i + $max_chunk_size >= $c) {
56 $r .= self::unsafeIconv($in, $out, substr($text, $i));
57 break;
59 // wibble the boundary
60 if (0x80 != (0xC0 & ord($text[$i + $max_chunk_size]))) {
61 $chunk_size = $max_chunk_size;
62 } elseif (0x80 != (0xC0 & ord($text[$i + $max_chunk_size - 1]))) {
63 $chunk_size = $max_chunk_size - 1;
64 } elseif (0x80 != (0xC0 & ord($text[$i + $max_chunk_size - 2]))) {
65 $chunk_size = $max_chunk_size - 2;
66 } elseif (0x80 != (0xC0 & ord($text[$i + $max_chunk_size - 3]))) {
67 $chunk_size = $max_chunk_size - 3;
68 } else {
69 return false; // rather confusing UTF-8...
71 $chunk = substr($text, $i, $chunk_size); // substr doesn't mind overlong lengths
72 $r .= self::unsafeIconv($in, $out, $chunk);
73 $i += $chunk_size;
75 return $r;
76 } else {
77 return false;
79 } else {
80 return false;
84 /**
85 * Cleans a UTF-8 string for well-formedness and SGML validity
87 * It will parse according to UTF-8 and return a valid UTF8 string, with
88 * non-SGML codepoints excluded.
90 * @note Just for reference, the non-SGML code points are 0 to 31 and
91 * 127 to 159, inclusive. However, we allow code points 9, 10
92 * and 13, which are the tab, line feed and carriage return
93 * respectively. 128 and above the code points map to multibyte
94 * UTF-8 representations.
96 * @note Fallback code adapted from utf8ToUnicode by Henri Sivonen and
97 * hsivonen@iki.fi at <http://iki.fi/hsivonen/php-utf8/> under the
98 * LGPL license. Notes on what changed are inside, but in general,
99 * the original code transformed UTF-8 text into an array of integer
100 * Unicode codepoints. Understandably, transforming that back to
101 * a string would be somewhat expensive, so the function was modded to
102 * directly operate on the string. However, this discourages code
103 * reuse, and the logic enumerated here would be useful for any
104 * function that needs to be able to understand UTF-8 characters.
105 * As of right now, only smart lossless character encoding converters
106 * would need that, and I'm probably not going to implement them.
107 * Once again, PHP 6 should solve all our problems.
109 public static function cleanUTF8($str, $force_php = false) {
111 // UTF-8 validity is checked since PHP 4.3.5
112 // This is an optimization: if the string is already valid UTF-8, no
113 // need to do PHP stuff. 99% of the time, this will be the case.
114 // The regexp matches the XML char production, as well as well as excluding
115 // non-SGML codepoints U+007F to U+009F
116 if (preg_match('/^[\x{9}\x{A}\x{D}\x{20}-\x{7E}\x{A0}-\x{D7FF}\x{E000}-\x{FFFD}\x{10000}-\x{10FFFF}]*$/Du', $str)) {
117 return $str;
120 $mState = 0; // cached expected number of octets after the current octet
121 // until the beginning of the next UTF8 character sequence
122 $mUcs4 = 0; // cached Unicode character
123 $mBytes = 1; // cached expected number of octets in the current sequence
125 // original code involved an $out that was an array of Unicode
126 // codepoints. Instead of having to convert back into UTF-8, we've
127 // decided to directly append valid UTF-8 characters onto a string
128 // $out once they're done. $char accumulates raw bytes, while $mUcs4
129 // turns into the Unicode code point, so there's some redundancy.
131 $out = '';
132 $char = '';
134 $len = strlen($str);
135 for($i = 0; $i < $len; $i++) {
136 $in = ord($str{$i});
137 $char .= $str[$i]; // append byte to char
138 if (0 == $mState) {
139 // When mState is zero we expect either a US-ASCII character
140 // or a multi-octet sequence.
141 if (0 == (0x80 & ($in))) {
142 // US-ASCII, pass straight through.
143 if (($in <= 31 || $in == 127) &&
144 !($in == 9 || $in == 13 || $in == 10) // save \r\t\n
146 // control characters, remove
147 } else {
148 $out .= $char;
150 // reset
151 $char = '';
152 $mBytes = 1;
153 } elseif (0xC0 == (0xE0 & ($in))) {
154 // First octet of 2 octet sequence
155 $mUcs4 = ($in);
156 $mUcs4 = ($mUcs4 & 0x1F) << 6;
157 $mState = 1;
158 $mBytes = 2;
159 } elseif (0xE0 == (0xF0 & ($in))) {
160 // First octet of 3 octet sequence
161 $mUcs4 = ($in);
162 $mUcs4 = ($mUcs4 & 0x0F) << 12;
163 $mState = 2;
164 $mBytes = 3;
165 } elseif (0xF0 == (0xF8 & ($in))) {
166 // First octet of 4 octet sequence
167 $mUcs4 = ($in);
168 $mUcs4 = ($mUcs4 & 0x07) << 18;
169 $mState = 3;
170 $mBytes = 4;
171 } elseif (0xF8 == (0xFC & ($in))) {
172 // First octet of 5 octet sequence.
174 // This is illegal because the encoded codepoint must be
175 // either:
176 // (a) not the shortest form or
177 // (b) outside the Unicode range of 0-0x10FFFF.
178 // Rather than trying to resynchronize, we will carry on
179 // until the end of the sequence and let the later error
180 // handling code catch it.
181 $mUcs4 = ($in);
182 $mUcs4 = ($mUcs4 & 0x03) << 24;
183 $mState = 4;
184 $mBytes = 5;
185 } elseif (0xFC == (0xFE & ($in))) {
186 // First octet of 6 octet sequence, see comments for 5
187 // octet sequence.
188 $mUcs4 = ($in);
189 $mUcs4 = ($mUcs4 & 1) << 30;
190 $mState = 5;
191 $mBytes = 6;
192 } else {
193 // Current octet is neither in the US-ASCII range nor a
194 // legal first octet of a multi-octet sequence.
195 $mState = 0;
196 $mUcs4 = 0;
197 $mBytes = 1;
198 $char = '';
200 } else {
201 // When mState is non-zero, we expect a continuation of the
202 // multi-octet sequence
203 if (0x80 == (0xC0 & ($in))) {
204 // Legal continuation.
205 $shift = ($mState - 1) * 6;
206 $tmp = $in;
207 $tmp = ($tmp & 0x0000003F) << $shift;
208 $mUcs4 |= $tmp;
210 if (0 == --$mState) {
211 // End of the multi-octet sequence. mUcs4 now contains
212 // the final Unicode codepoint to be output
214 // Check for illegal sequences and codepoints.
216 // From Unicode 3.1, non-shortest form is illegal
217 if (((2 == $mBytes) && ($mUcs4 < 0x0080)) ||
218 ((3 == $mBytes) && ($mUcs4 < 0x0800)) ||
219 ((4 == $mBytes) && ($mUcs4 < 0x10000)) ||
220 (4 < $mBytes) ||
221 // From Unicode 3.2, surrogate characters = illegal
222 (($mUcs4 & 0xFFFFF800) == 0xD800) ||
223 // Codepoints outside the Unicode range are illegal
224 ($mUcs4 > 0x10FFFF)
227 } elseif (0xFEFF != $mUcs4 && // omit BOM
228 // check for valid Char unicode codepoints
230 0x9 == $mUcs4 ||
231 0xA == $mUcs4 ||
232 0xD == $mUcs4 ||
233 (0x20 <= $mUcs4 && 0x7E >= $mUcs4) ||
234 // 7F-9F is not strictly prohibited by XML,
235 // but it is non-SGML, and thus we don't allow it
236 (0xA0 <= $mUcs4 && 0xD7FF >= $mUcs4) ||
237 (0x10000 <= $mUcs4 && 0x10FFFF >= $mUcs4)
240 $out .= $char;
242 // initialize UTF8 cache (reset)
243 $mState = 0;
244 $mUcs4 = 0;
245 $mBytes = 1;
246 $char = '';
248 } else {
249 // ((0xC0 & (*in) != 0x80) && (mState != 0))
250 // Incomplete multi-octet sequence.
251 // used to result in complete fail, but we'll reset
252 $mState = 0;
253 $mUcs4 = 0;
254 $mBytes = 1;
255 $char ='';
259 return $out;
263 * Translates a Unicode codepoint into its corresponding UTF-8 character.
264 * @note Based on Feyd's function at
265 * <http://forums.devnetwork.net/viewtopic.php?p=191404#191404>,
266 * which is in public domain.
267 * @note While we're going to do code point parsing anyway, a good
268 * optimization would be to refuse to translate code points that
269 * are non-SGML characters. However, this could lead to duplication.
270 * @note This is very similar to the unichr function in
271 * maintenance/generate-entity-file.php (although this is superior,
272 * due to its sanity checks).
275 // +----------+----------+----------+----------+
276 // | 33222222 | 22221111 | 111111 | |
277 // | 10987654 | 32109876 | 54321098 | 76543210 | bit
278 // +----------+----------+----------+----------+
279 // | | | | 0xxxxxxx | 1 byte 0x00000000..0x0000007F
280 // | | | 110yyyyy | 10xxxxxx | 2 byte 0x00000080..0x000007FF
281 // | | 1110zzzz | 10yyyyyy | 10xxxxxx | 3 byte 0x00000800..0x0000FFFF
282 // | 11110www | 10wwzzzz | 10yyyyyy | 10xxxxxx | 4 byte 0x00010000..0x0010FFFF
283 // +----------+----------+----------+----------+
284 // | 00000000 | 00011111 | 11111111 | 11111111 | Theoretical upper limit of legal scalars: 2097151 (0x001FFFFF)
285 // | 00000000 | 00010000 | 11111111 | 11111111 | Defined upper limit of legal scalar codes
286 // +----------+----------+----------+----------+
288 public static function unichr($code) {
289 if($code > 1114111 or $code < 0 or
290 ($code >= 55296 and $code <= 57343) ) {
291 // bits are set outside the "valid" range as defined
292 // by UNICODE 4.1.0
293 return '';
296 $x = $y = $z = $w = 0;
297 if ($code < 128) {
298 // regular ASCII character
299 $x = $code;
300 } else {
301 // set up bits for UTF-8
302 $x = ($code & 63) | 128;
303 if ($code < 2048) {
304 $y = (($code & 2047) >> 6) | 192;
305 } else {
306 $y = (($code & 4032) >> 6) | 128;
307 if($code < 65536) {
308 $z = (($code >> 12) & 15) | 224;
309 } else {
310 $z = (($code >> 12) & 63) | 128;
311 $w = (($code >> 18) & 7) | 240;
315 // set up the actual character
316 $ret = '';
317 if($w) $ret .= chr($w);
318 if($z) $ret .= chr($z);
319 if($y) $ret .= chr($y);
320 $ret .= chr($x);
322 return $ret;
325 public static function iconvAvailable() {
326 static $iconv = null;
327 if ($iconv === null) {
328 $iconv = function_exists('iconv') && self::testIconvTruncateBug() != self::ICONV_UNUSABLE;
330 return $iconv;
334 * Converts a string to UTF-8 based on configuration.
336 public static function convertToUTF8($str, $config, $context) {
337 $encoding = $config->get('Core.Encoding');
338 if ($encoding === 'utf-8') return $str;
339 static $iconv = null;
340 if ($iconv === null) $iconv = self::iconvAvailable();
341 if ($iconv && !$config->get('Test.ForceNoIconv')) {
342 // unaffected by bugs, since UTF-8 support all characters
343 $str = self::unsafeIconv($encoding, 'utf-8//IGNORE', $str);
344 if ($str === false) {
345 // $encoding is not a valid encoding
346 trigger_error('Invalid encoding ' . $encoding, E_USER_ERROR);
347 return '';
349 // If the string is bjorked by Shift_JIS or a similar encoding
350 // that doesn't support all of ASCII, convert the naughty
351 // characters to their true byte-wise ASCII/UTF-8 equivalents.
352 $str = strtr($str, self::testEncodingSupportsASCII($encoding));
353 return $str;
354 } elseif ($encoding === 'iso-8859-1') {
355 $str = utf8_encode($str);
356 return $str;
358 trigger_error('Encoding not supported, please install iconv', E_USER_ERROR);
362 * Converts a string from UTF-8 based on configuration.
363 * @note Currently, this is a lossy conversion, with unexpressable
364 * characters being omitted.
366 public static function convertFromUTF8($str, $config, $context) {
367 $encoding = $config->get('Core.Encoding');
368 if ($encoding === 'utf-8') return $str;
369 static $iconv = null;
370 if ($iconv === null) $iconv = self::iconvAvailable();
371 if ($escape = $config->get('Core.EscapeNonASCIICharacters')) {
372 $str = self::convertToASCIIDumbLossless($str);
374 if ($iconv && !$config->get('Test.ForceNoIconv')) {
375 // Undo our previous fix in convertToUTF8, otherwise iconv will barf
376 $ascii_fix = self::testEncodingSupportsASCII($encoding);
377 if (!$escape && !empty($ascii_fix)) {
378 $clear_fix = array();
379 foreach ($ascii_fix as $utf8 => $native) $clear_fix[$utf8] = '';
380 $str = strtr($str, $clear_fix);
382 $str = strtr($str, array_flip($ascii_fix));
383 // Normal stuff
384 $str = self::iconv('utf-8', $encoding . '//IGNORE', $str);
385 return $str;
386 } elseif ($encoding === 'iso-8859-1') {
387 $str = utf8_decode($str);
388 return $str;
390 trigger_error('Encoding not supported', E_USER_ERROR);
391 // You might be tempted to assume that the ASCII representation
392 // might be OK, however, this is *not* universally true over all
393 // encodings. So we take the conservative route here, rather
394 // than forcibly turn on %Core.EscapeNonASCIICharacters
398 * Lossless (character-wise) conversion of HTML to ASCII
399 * @param $str UTF-8 string to be converted to ASCII
400 * @returns ASCII encoded string with non-ASCII character entity-ized
401 * @warning Adapted from MediaWiki, claiming fair use: this is a common
402 * algorithm. If you disagree with this license fudgery,
403 * implement it yourself.
404 * @note Uses decimal numeric entities since they are best supported.
405 * @note This is a DUMB function: it has no concept of keeping
406 * character entities that the projected character encoding
407 * can allow. We could possibly implement a smart version
408 * but that would require it to also know which Unicode
409 * codepoints the charset supported (not an easy task).
410 * @note Sort of with cleanUTF8() but it assumes that $str is
411 * well-formed UTF-8
413 public static function convertToASCIIDumbLossless($str) {
414 $bytesleft = 0;
415 $result = '';
416 $working = 0;
417 $len = strlen($str);
418 for( $i = 0; $i < $len; $i++ ) {
419 $bytevalue = ord( $str[$i] );
420 if( $bytevalue <= 0x7F ) { //0xxx xxxx
421 $result .= chr( $bytevalue );
422 $bytesleft = 0;
423 } elseif( $bytevalue <= 0xBF ) { //10xx xxxx
424 $working = $working << 6;
425 $working += ($bytevalue & 0x3F);
426 $bytesleft--;
427 if( $bytesleft <= 0 ) {
428 $result .= "&#" . $working . ";";
430 } elseif( $bytevalue <= 0xDF ) { //110x xxxx
431 $working = $bytevalue & 0x1F;
432 $bytesleft = 1;
433 } elseif( $bytevalue <= 0xEF ) { //1110 xxxx
434 $working = $bytevalue & 0x0F;
435 $bytesleft = 2;
436 } else { //1111 0xxx
437 $working = $bytevalue & 0x07;
438 $bytesleft = 3;
441 return $result;
444 /** No bugs detected in iconv. */
445 const ICONV_OK = 0;
447 /** Iconv truncates output if converting from UTF-8 to another
448 * character set with //IGNORE, and a non-encodable character is found */
449 const ICONV_TRUNCATES = 1;
451 /** Iconv does not support //IGNORE, making it unusable for
452 * transcoding purposes */
453 const ICONV_UNUSABLE = 2;
456 * glibc iconv has a known bug where it doesn't handle the magic
457 * //IGNORE stanza correctly. In particular, rather than ignore
458 * characters, it will return an EILSEQ after consuming some number
459 * of characters, and expect you to restart iconv as if it were
460 * an E2BIG. Old versions of PHP did not respect the errno, and
461 * returned the fragment, so as a result you would see iconv
462 * mysteriously truncating output. We can work around this by
463 * manually chopping our input into segments of about 8000
464 * characters, as long as PHP ignores the error code. If PHP starts
465 * paying attention to the error code, iconv becomes unusable.
467 * @returns Error code indicating severity of bug.
469 public static function testIconvTruncateBug() {
470 static $code = null;
471 if ($code === null) {
472 // better not use iconv, otherwise infinite loop!
473 $r = self::unsafeIconv('utf-8', 'ascii//IGNORE', "\xCE\xB1" . str_repeat('a', 9000));
474 if ($r === false) {
475 $code = self::ICONV_UNUSABLE;
476 } elseif (($c = strlen($r)) < 9000) {
477 $code = self::ICONV_TRUNCATES;
478 } elseif ($c > 9000) {
479 trigger_error('Your copy of iconv is extremely buggy. Please notify HTML Purifier maintainers: include your iconv version as per phpversion()', E_USER_ERROR);
480 } else {
481 $code = self::ICONV_OK;
484 return $code;
488 * This expensive function tests whether or not a given character
489 * encoding supports ASCII. 7/8-bit encodings like Shift_JIS will
490 * fail this test, and require special processing. Variable width
491 * encodings shouldn't ever fail.
493 * @param string $encoding Encoding name to test, as per iconv format
494 * @param bool $bypass Whether or not to bypass the precompiled arrays.
495 * @return Array of UTF-8 characters to their corresponding ASCII,
496 * which can be used to "undo" any overzealous iconv action.
498 public static function testEncodingSupportsASCII($encoding, $bypass = false) {
499 // All calls to iconv here are unsafe, proof by case analysis:
500 // If ICONV_OK, no difference.
501 // If ICONV_TRUNCATE, all calls involve one character inputs,
502 // so bug is not triggered.
503 // If ICONV_UNUSABLE, this call is irrelevant
504 static $encodings = array();
505 if (!$bypass) {
506 if (isset($encodings[$encoding])) return $encodings[$encoding];
507 $lenc = strtolower($encoding);
508 switch ($lenc) {
509 case 'shift_jis':
510 return array("\xC2\xA5" => '\\', "\xE2\x80\xBE" => '~');
511 case 'johab':
512 return array("\xE2\x82\xA9" => '\\');
514 if (strpos($lenc, 'iso-8859-') === 0) return array();
516 $ret = array();
517 if (self::unsafeIconv('UTF-8', $encoding, 'a') === false) return false;
518 for ($i = 0x20; $i <= 0x7E; $i++) { // all printable ASCII chars
519 $c = chr($i); // UTF-8 char
520 $r = self::unsafeIconv('UTF-8', "$encoding//IGNORE", $c); // initial conversion
521 if (
522 $r === '' ||
523 // This line is needed for iconv implementations that do not
524 // omit characters that do not exist in the target character set
525 ($r === $c && self::unsafeIconv($encoding, 'UTF-8//IGNORE', $r) !== $c)
527 // Reverse engineer: what's the UTF-8 equiv of this byte
528 // sequence? This assumes that there's no variable width
529 // encoding that doesn't support ASCII.
530 $ret[self::unsafeIconv($encoding, 'UTF-8//IGNORE', $c)] = $c;
533 $encodings[$encoding] = $ret;
534 return $ret;
540 // vim: et sw=4 sts=4