* @return string a clean, shiny, normalized UTF-8 string
*/
function cleanUp( $string ) {
- if( UtfNormal::quickIsNFCVerify( $string ) )
+ if( NORMALIZE_ICU ) {
+ # We exclude a few chars that ICU would not.
+ $string = preg_replace(
+ '/[\x00-\x08\x0b\x0c\x0e-\x1f]/',
+ UTF8_REPLACEMENT,
+ $string );
+ $string = str_replace( UTF8_FFFE, UTF8_REPLACEMENT, $string );
+ $string = str_replace( UTF8_FFFF, UTF8_REPLACEMENT, $string );
+
+ # UnicodeString constructor fails if the string ends with a
+ # head byte. Add a junk char at the end, we'll strip it off.
+ return rtrim( utf8_normalize( $string . "\x01", UNORM_NFC ), "\x01" );
+ } elseif( UtfNormal::quickIsNFCVerify( $string ) ) {
+ # Side effect -- $string has had UTF-8 errors cleaned up.
return $string;
- else
+ } else {
return UtfNormal::NFC( $string );
+ }
}
/**
* Returns true if the string is _definitely_ in NFC.
* Returns false if not or uncertain.
* @param string $string a UTF-8 string, altered on output to be valid UTF-8 safe for XML.
- * @return bool
*/
function quickIsNFCVerify( &$string ) {
+ # Screen out some characters that eg won't be allowed in XML
+ $string = preg_replace( '/[\x00-\x08\x0b\x0c\x0e-\x1f]/', UTF8_REPLACEMENT, $string );
+
# ASCII is always valid NFC!
+ # If we're only ever given plain ASCII, we can avoid the overhead
+ # of initializing the decomposition tables by skipping out early.
if( !preg_match( '/[\x80-\xff]/', $string ) ) return true;
- UtfNormal::loadData();
- global $utfCheckNFC, $utfCombiningClass;
- $len = strlen( $string );
- $out = '';
- $state = UTF8_HEAD;
- $looksNormal = true;
+ static $checkit = null, $tailBytes = null, $utfCheckOrCombining = null;
+ if( !isset( $checkit ) ) {
+ # Load/build some scary lookup tables...
+ UtfNormal::loadData();
+ global $utfCheckNFC, $utfCombiningClass;
+
+ $utfCheckOrCombining = array_merge( $utfCheckNFC, $utfCombiningClass );
+
+ # Head bytes for sequences which we should do further validity checks
+ $checkit = array_flip( array_map( 'chr',
+ array( 0xc0, 0xc1, 0xe0, 0xed, 0xef,
+ 0xf0, 0xf1, 0xf2, 0xf3, 0xf4, 0xf5, 0xf6, 0xf7,
+ 0xf8, 0xf9, 0xfa, 0xfb, 0xfc, 0xfd, 0xfe, 0xff ) ) );
+
+ # Each UTF-8 head byte is followed by a certain
+ # number of tail bytes.
+ $tailBytes = array();
+ for( $n = 0; $n < 256; $n++ ) {
+ if( $n < 0xc0 ) {
+ $remaining = 0;
+ } elseif( $n < 0xe0 ) {
+ $remaining = 1;
+ } elseif( $n < 0xf0 ) {
+ $remaining = 2;
+ } elseif( $n < 0xf8 ) {
+ $remaining = 3;
+ } elseif( $n < 0xfc ) {
+ $remaining = 4;
+ } elseif( $n < 0xfe ) {
+ $remaining = 5;
+ } else {
+ $remaining = 0;
+ }
+ $tailBytes[chr($n)] = $remaining;
+ }
+ }
- $rep = false;
- $head = 0;
- for( $i = 0; $i < $len; $i++ ) {
- $c = $string{$i};
- $n = ord( $c );
- if( $state == UTF8_TAIL ) {
- if( $n >= 0x80 && $n < 0xc0 ) {
- $sequence .= $c;
- if( --$remaining == 0 ) {
- if( $head < 0xc2 || $head == 0xed || $head == 0xe0 || $head > 0xee ) {
- if( ( $sequence >= UTF8_SURROGATE_FIRST
- && $sequence <= UTF8_SURROGATE_LAST)
- || ($head == 0xc0 && $sequence <= UTF8_OVERLONG_A)
- || ($head == 0xc1 && $sequence <= UTF8_OVERLONG_A)
- || ($head == 0xe0 && $sequence <= UTF8_OVERLONG_B)
- || ($head == 0xef &&
- ($sequence >= UTF8_FDD0 && $sequence <= UTF8_FDEF)
- || ($sequence == UTF8_FFFE)
+ # Chop the text into pure-ASCII and non-ASCII areas;
+ # large ASCII parts can be handled much more quickly.
+ # Don't chop up Unicode areas for punctuation, though,
+ # that wastes energy.
+ preg_match_all(
+ '/([\x00-\x7f]+|[\x80-\xff][\x00-\x40\x5b-\x5f\x7b-\xff]*)/',
+ $string, $matches );
+
+ $looksNormal = true;
+ $base = 0;
+ $replace = array();
+ foreach( $matches[1] as $str ) {
+ $chunk = strlen( $str );
+
+ if( $str{0} < "\x80" ) {
+ # ASCII chunk: guaranteed to be valid UTF-8
+ # and in normal form C, so skip over it.
+ $base += $chunk;
+ continue;
+ }
+
+ # We'll have to examine the chunk byte by byte to ensure
+ # that it consists of valid UTF-8 sequences, and to see
+ # if any of them might not be normalized.
+ #
+ # Since PHP is not the fastest language on earth, some of
+ # this code is a little ugly with inner loop optimizations.
+
+ $head = '';
+ $len = $chunk + 1; # Counting down is faster. I'm *so* sorry.
+
+ for( $i = -1; --$len; ) {
+ if( $remaining = $tailBytes[$c = $str{++$i}] ) {
+ # UTF-8 head byte!
+ $sequence = $head = $c;
+ do {
+ # Look for the defined number of tail bytes...
+ if( --$len && ( $c = $str{++$i} ) >= "\x80" && $c < "\xc0" ) {
+ # Legal tail bytes are nice.
+ $sequence .= $c;
+ } else {
+ if( 0 == $len ) {
+ # Premature end of string!
+ # Drop a replacement character into output to
+ # represent the invalid UTF-8 sequence.
+ $replace[] = array( UTF8_REPLACEMENT,
+ $base + $i + 1 - strlen( $sequence ),
+ strlen( $sequence ) );
+ break 2;
+ } else {
+ # Illegal tail byte; abandon the sequence.
+ $replace[] = array( UTF8_REPLACEMENT,
+ $base + $i - strlen( $sequence ),
+ strlen( $sequence ) );
+ # Back up and reprocess this byte; it may itself
+ # be a legal ASCII or UTF-8 sequence head.
+ --$i;
+ ++$len;
+ continue 2;
+ }
+ }
+ } while( --$remaining );
+
+ if( isset( $checkit[$head] ) ) {
+ # Do some more detailed validity checks, for
+ # invalid characters and illegal sequences.
+ if( $head == "\xed" ) {
+ # 0xed is relatively frequent in Korean, which
+ # abuts the surrogate area, so we're doing
+ # this check separately to speed things up.
+
+ if( $sequence >= UTF8_SURROGATE_FIRST ) {
+ # Surrogates are legal only in UTF-16 code.
+ # They are totally forbidden here in UTF-8
+ # utopia.
+ $replace[] = array( UTF8_REPLACEMENT,
+ $base + $i + 1 - strlen( $sequence ),
+ strlen( $sequence ) );
+ $head = '';
+ continue;
+ }
+ } else {
+ # Slower, but rarer checks...
+ $n = ord( $head );
+ if(
+ # "Overlong sequences" are those that are syntactically
+ # correct but use more UTF-8 bytes than are necessary to
+ # encode a character. Naïve string comparisons can be
+ # tricked into failing to see a match for an ASCII
+ # character, for instance, which can be a security hole
+ # if blacklist checks are being used.
+ ($n < 0xc2 && $sequence <= UTF8_OVERLONG_A)
+ || ($n == 0xe0 && $sequence <= UTF8_OVERLONG_B)
+ || ($n == 0xf0 && $sequence <= UTF8_OVERLONG_C)
+
+ # U+FFFE and U+FFFF are explicitly forbidden in Unicode.
+ || ($n == 0xef &&
+ ($sequence == UTF8_FFFE)
|| ($sequence == UTF8_FFFF) )
- || ($head == 0xf0 && $sequence <= UTF8_OVERLONG_C)
- || ($sequence > UTF8_MAX) ) {
- $out .= UTF8_REPLACEMENT;
- $state = UTF8_HEAD;
+
+ # Unicode has been limited to 21 bits; longer
+ # sequences are not allowed.
+ || ($n >= 0xf0 && $sequence > UTF8_MAX) ) {
+
+ $replace[] = array( UTF8_REPLACEMENT,
+ $base + $i + 1 - strlen( $sequence ),
+ strlen( $sequence ) );
+ $head = '';
continue;
}
}
- if( isset( $utfCheckNFC[$sequence] ) ||
- isset( $utfCombiningClass[$sequence] ) ) {
- # If it's NO or MAYBE, we'll have to do the slow check.
- $looksNormal = false;
- }
- $out .= $sequence;
- $state = UTF8_HEAD;
- $head = 0;
}
- continue;
- }
- # Not a valid tail byte! DIscard the char we've been building.
- #printf ("Invalid '%x' in tail with %d remaining bytes\n", $n, $remaining );
- $state = UTF8_HEAD;
- $out .= UTF8_REPLACEMENT;
- }
- if( $n < 0x20 ) {
- if( $n < 0x09 ) {
- $out .= UTF8_REPLACEMENT;
- } elseif( $n == 0x0a ) {
- $out .= $c;
- } elseif( $n < 0x0d ) {
- $out .= UTF8_REPLACEMENT;
- } elseif( $n == 0x0d ) {
- # Strip \r silently
+
+ if( isset( $utfCheckOrCombining[$sequence] ) ) {
+ # If it's NO or MAYBE, we'll have to rip
+ # the string apart and put it back together.
+ # That's going to be mighty slow.
+ $looksNormal = false;
+ }
+
+ # The sequence is legal!
+ $head = '';
+ } elseif( $c < "\x80" ) {
+ # ASCII byte.
+ $head = '';
+ } elseif( $c < "\xc0" ) {
+ # Illegal tail bytes
+ if( $head == '' ) {
+ # Out of the blue!
+ $replace[] = array( UTF8_REPLACEMENT, $base + $i, 1 );
+ } else {
+ # Don't add if we're continuing a broken sequence;
+ # we already put a replacement character when we looked
+ # at the broken sequence.
+ $replace[] = array( '', $base + $i, 1 );
+ }
} else {
- $out .= UTF8_REPLACEMENT;
+ # Miscellaneous freaks.
+ $replace[] = array( UTF8_REPLACEMENT, $base + $i, 1 );
+ $head = '';
}
- } elseif( $n < 0x80 ) {
- # Friendly ASCII chars.
- $out .= $c;
- } elseif( $n < 0xc0 ) {
- # illegal tail bytes or head byte of overlong sequence
- if( $head == 0 ) $out .= UTF8_REPLACEMENT;
- } elseif( $n < 0xe0 ) {
- $state = UTF8_TAIL;
- $remaining = 1;
- $sequence = $c;
- $head = $n;
- } elseif( $n < 0xf0 ) {
- $state = UTF8_TAIL;
- $remaining = 2;
- $sequence = $c;
- $head = $n;
- } elseif( $n < 0xf8 ) {
- $state = UTF8_TAIL;
- $remaining = 3;
- $sequence = $c;
- $head = $n;
- } elseif( $n < 0xfc ) {
- $state = UTF8_TAIL;
- $remaining = 4;
- $sequence = $c;
- $head = $n;
- } elseif( $n < 0xfe ) {
- $state = UTF8_TAIL;
- $remaining = 5;
- $sequence = $c;
- $head = $n;
- } else {
- $out .= UTF8_REPLACEMENT;
}
+ $base += $chunk;
}
- if( $state == UTF8_TAIL ) {
- $out .= UTF8_REPLACEMENT;
+ if( count( $replace ) ) {
+ # There were illegal UTF-8 sequences we need to fix up.
+ $out = '';
+ $last = 0;
+ foreach( $replace as $rep ) {
+ list( $replacement, $start, $length ) = $rep;
+ if( $last < $start ) {
+ $out .= substr( $string, $last, $start - $last );
+ }
+ $out .= $replacement;
+ $last = $start + $length;
+ }
+ if( $last < strlen( $string ) ) {
+ $out .= substr( $string, $last );
+ }
+ $string = $out;
}
- $string = $out;
return $looksNormal;
}
* (depending on which decomposition map is passed to us).
* Input is assumed to be *valid* UTF-8. Invalid code will break.
* @access private
- * @param string &$string Valid UTF-8 string
- * @param array &$map hash of expanded decomposition map
+ * @param string $string Valid UTF-8 string
+ * @param array $map hash of expanded decomposition map
* @return string a UTF-8 string decomposed, not yet normalized (needs sorting)
*/
- function fastDecompose( &$string, &$map ) {
+ function fastDecompose( $string, &$map ) {
UtfNormal::loadData();
$len = strlen( $string );
$out = '';
}
if( isset( $map[$c] ) ) {
$out .= $map[$c];
+ continue;
} else {
if( $c >= UTF8_HANGUL_FIRST && $c <= UTF8_HANGUL_LAST ) {
# Decompose a hangul syllable into jamo;
| (ord( $c{1} ) & 0x3f) << 6
| (ord( $c{2} ) & 0x3f) )
- UNICODE_HANGUL_FIRST;
- $l = IntVal( $index / UNICODE_HANGUL_NCOUNT );
- $v = IntVal( ($index % UNICODE_HANGUL_NCOUNT) / UNICODE_HANGUL_TCOUNT);
+ $l = intval( $index / UNICODE_HANGUL_NCOUNT );
+ $v = intval( ($index % UNICODE_HANGUL_NCOUNT) / UNICODE_HANGUL_TCOUNT);
$t = $index % UNICODE_HANGUL_TCOUNT;
$out .= "\xe1\x84" . chr( 0x80 + $l ) . "\xe1\x85" . chr( 0xa1 + $v );
if( $t >= 25 ) {
} elseif( $t ) {
$out .= "\xe1\x86" . chr( 0xa7 + $t );
}
- } else {
- $out .= $c;
+ continue;
}
}
+ $out .= $c;
}
return $out;
}
$len = strlen( $string );
$out = '';
$lastClass = -1;
+ $lastHangul = 0;
$startChar = '';
$combining = '';
$x1 = ord(substr(UTF8_HANGUL_VBASE,0,1));
$combining .= $c;
}
$lastClass = $class;
+ $lastHangul = 0;
continue;
}
}
if( $lastClass == 0 ) {
if( isset( $utfCanonicalComp[$pair] ) ) {
$startChar = $utfCanonicalComp[$pair];
+ $lastHangul = 0;
continue;
}
if( $n >= $x1 && $n <= $x2 ) {
$startChar = chr( $hangulPoint >> 12 & 0x0f | 0xe0 ) .
chr( $hangulPoint >> 6 & 0x3f | 0x80 ) .
chr( $hangulPoint & 0x3f | 0x80 );
+ $lastHangul = 0;
continue;
} elseif( $c >= UTF8_HANGUL_TBASE &&
$c <= UTF8_HANGUL_TEND &&
$startChar >= UTF8_HANGUL_FIRST &&
- $startChar <= UTF8_HANGUL_LAST ) {
+ $startChar <= UTF8_HANGUL_LAST &&
+ !$lastHangul ) {
# $tIndex = utf8ToCodepoint( $c ) - UNICODE_HANGUL_TBASE;
$tIndex = ord( $c{2} ) - 0xa7;
if( $tIndex < 0 ) $tIndex = ord( $c{2} ) - 0x80 + (0x11c0 - 0x11a7);
$startChar{1} = chr( $mid );
}
$startChar{2} = chr( $tail );
+
+ # If there's another jamo char after this, *don't* try to merge it.
+ $lastHangul = 1;
continue;
}
}
$startChar = $c;
$combining = '';
$lastClass = 0;
+ $lastHangul = 0;
}
$out .= $startChar . $combining;
return $out;