UtfNormal.php

Go to the documentation of this file.

<?php
define( 'NORMALIZE_ICU', function_exists( 'utf8_normalize' ) );
define( 'NORMALIZE_INTL', function_exists( 'normalizer_normalize' ) );

class UtfNormal {
    const UNORM_NONE = 1;
    const UNORM_NFD  = 2;
    const UNORM_NFKD = 3;
    const UNORM_NFC  = 4;
    const UNORM_NFKC = 5;
    const UNORM_FCD  = 6;
    const UNORM_DEFAULT = self::UNORM_NFC;

    static $utfCombiningClass = null;
    static $utfCanonicalComp = null;
    static $utfCanonicalDecomp = null;

    # Load compatibility decompositions on demand if they are needed.
    static $utfCompatibilityDecomp = null;

    static $utfCheckNFC;

    static function cleanUp( $string ) {
        if( NORMALIZE_ICU ) {
            $string = self::replaceForNativeNormalize( $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", self::UNORM_NFC ), "\x01" );
        } elseif( NORMALIZE_INTL ) {
            $string = self::replaceForNativeNormalize( $string );
            $norm = normalizer_normalize( $string, Normalizer::FORM_C );
            if( $norm === null || $norm === false ) {
                # normalizer_normalize will either return false or null
                # (depending on which doc you read) if invalid utf8 string.
                # quickIsNFCVerify cleans up invalid sequences.

                if( UtfNormal::quickIsNFCVerify( $string ) ) {
                    # if that's true, the string is actually already normal.
                    return $string;
                } else {
                    # Now we are valid but non-normal
                    return normalizer_normalize( $string, Normalizer::FORM_C );
                }
            } else {
                return $norm;
            }
        } elseif( UtfNormal::quickIsNFCVerify( $string ) ) {
            # Side effect -- $string has had UTF-8 errors cleaned up.
            return $string;
        } else {
            return UtfNormal::NFC( $string );
        }
    }

    static function toNFC( $string ) {
        if( NORMALIZE_INTL )
            return normalizer_normalize( $string, Normalizer::FORM_C );
        elseif( NORMALIZE_ICU )
            return utf8_normalize( $string, self::UNORM_NFC );
        elseif( UtfNormal::quickIsNFC( $string ) )
            return $string;
        else
            return UtfNormal::NFC( $string );
    }

    static function toNFD( $string ) {
        if( NORMALIZE_INTL )
            return normalizer_normalize( $string, Normalizer::FORM_D );
        elseif( NORMALIZE_ICU )
            return utf8_normalize( $string, self::UNORM_NFD );
        elseif( preg_match( '/[\x80-\xff]/', $string ) )
            return UtfNormal::NFD( $string );
        else
            return $string;
    }

    static function toNFKC( $string ) {
        if( NORMALIZE_INTL )
            return normalizer_normalize( $string, Normalizer::FORM_KC );
        elseif( NORMALIZE_ICU )
            return utf8_normalize( $string, self::UNORM_NFKC );
        elseif( preg_match( '/[\x80-\xff]/', $string ) )
            return UtfNormal::NFKC( $string );
        else
            return $string;
    }

    static function toNFKD( $string ) {
        if( NORMALIZE_INTL )
            return normalizer_normalize( $string, Normalizer::FORM_KD );
        elseif( NORMALIZE_ICU )
            return utf8_normalize( $string, self::UNORM_NFKD );
        elseif( preg_match( '/[\x80-\xff]/', $string ) )
            return UtfNormal::NFKD( $string );
        else
            return $string;
    }

    static function loadData() {
        if( !isset( self::$utfCombiningClass ) ) {
            require_once __DIR__ . '/UtfNormalData.inc';
        }
    }

    static function quickIsNFC( $string ) {
        # ASCII is always valid NFC!
        # If it's pure ASCII, let it through.
        if( !preg_match( '/[\x80-\xff]/', $string ) ) return true;

        UtfNormal::loadData();
        $len = strlen( $string );
        for( $i = 0; $i < $len; $i++ ) {
            $c = $string[$i];
            $n = ord( $c );
            if( $n < 0x80 ) {
                continue;
            } elseif( $n >= 0xf0 ) {
                $c = substr( $string, $i, 4 );
                $i += 3;
            } elseif( $n >= 0xe0 ) {
                $c = substr( $string, $i, 3 );
                $i += 2;
            } elseif( $n >= 0xc0 ) {
                $c = substr( $string, $i, 2 );
                $i++;
            }
            if( isset( self::$utfCheckNFC[$c] ) ) {
                # If it's NO or MAYBE, bail and do the slow check.
                return false;
            }
            if( isset( self::$utfCombiningClass[$c] ) ) {
                # Combining character? We might have to do sorting, at least.
                return false;
            }
        }
        return true;
    }

    static 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;

        static $checkit = null, $tailBytes = null, $utfCheckOrCombining = null;
        if( !isset( $checkit ) ) {
            # Load/build some scary lookup tables...
            UtfNormal::loadData();

            $utfCheckOrCombining = array_merge( self::$utfCheckNFC, self::$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;
            }
        }

        # 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.
        $matches = array();
        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; ) {
                $remaining = $tailBytes[$c = $str[++$i]];
                if( $remaining ) {
                    # 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) )

                                # 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( $utfCheckOrCombining[$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 {
                    # Miscellaneous freaks.
                    $replace[] = array( UTF8_REPLACEMENT, $base + $i, 1 );
                    $head = '';
                }
            }
            $base += $chunk;
        }
        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;
        }
        return $looksNormal;
    }

    # These take a string and run the normalization on them, without
    # checking for validity or any optimization etc. Input must be
    # VALID UTF-8!

    static function NFC( $string ) {
        return UtfNormal::fastCompose( UtfNormal::NFD( $string ) );
    }

    static function NFD( $string ) {
        UtfNormal::loadData();

        return UtfNormal::fastCombiningSort(
            UtfNormal::fastDecompose( $string, self::$utfCanonicalDecomp ) );
    }

    static function NFKC( $string ) {
        return UtfNormal::fastCompose( UtfNormal::NFKD( $string ) );
    }

    static function NFKD( $string ) {
        if( !isset( self::$utfCompatibilityDecomp ) ) {
            require_once 'UtfNormalDataK.inc';
        }
        return self::fastCombiningSort(
            self::fastDecompose( $string, self::$utfCompatibilityDecomp ) );
    }


    static function fastDecompose( $string, $map ) {
        UtfNormal::loadData();
        $len = strlen( $string );
        $out = '';
        for( $i = 0; $i < $len; $i++ ) {
            $c = $string[$i];
            $n = ord( $c );
            if( $n < 0x80 ) {
                # ASCII chars never decompose
                # THEY ARE IMMORTAL
                $out .= $c;
                continue;
            } elseif( $n >= 0xf0 ) {
                $c = substr( $string, $i, 4 );
                $i += 3;
            } elseif( $n >= 0xe0 ) {
                $c = substr( $string, $i, 3 );
                $i += 2;
            } elseif( $n >= 0xc0 ) {
                $c = substr( $string, $i, 2 );
                $i++;
            }
            if( isset( $map[$c] ) ) {
                $out .= $map[$c];
                continue;
            } else {
                if( $c >= UTF8_HANGUL_FIRST && $c <= UTF8_HANGUL_LAST ) {
                    # Decompose a hangul syllable into jamo;
                    # hardcoded for three-byte UTF-8 sequence.
                    # A lookup table would be slightly faster,
                    # but adds a lot of memory & disk needs.
                    #
                    $index = ( (ord( $c[0] ) & 0x0f) << 12
                             | (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);
                    $t = $index % UNICODE_HANGUL_TCOUNT;
                    $out .= "\xe1\x84" . chr( 0x80 + $l ) . "\xe1\x85" . chr( 0xa1 + $v );
                    if( $t >= 25 ) {
                        $out .= "\xe1\x87" . chr( 0x80 + $t - 25 );
                    } elseif( $t ) {
                        $out .= "\xe1\x86" . chr( 0xa7 + $t );
                    }
                    continue;
                }
            }
            $out .= $c;
        }
        return $out;
    }

    static function fastCombiningSort( $string ) {
        UtfNormal::loadData();
        $len = strlen( $string );
        $out = '';
        $combiners = array();
        $lastClass = -1;
        for( $i = 0; $i < $len; $i++ ) {
            $c = $string[$i];
            $n = ord( $c );
            if( $n >= 0x80 ) {
                if( $n >= 0xf0 ) {
                    $c = substr( $string, $i, 4 );
                    $i += 3;
                } elseif( $n >= 0xe0 ) {
                    $c = substr( $string, $i, 3 );
                    $i += 2;
                } elseif( $n >= 0xc0 ) {
                    $c = substr( $string, $i, 2 );
                    $i++;
                }
                if( isset( self::$utfCombiningClass[$c] ) ) {
                    $lastClass = self::$utfCombiningClass[$c];
                    if( isset( $combiners[$lastClass] ) ) {
                        $combiners[$lastClass] .= $c;
                    } else {
                        $combiners[$lastClass] = $c;
                    }
                    continue;
                }
            }
            if( $lastClass ) {
                ksort( $combiners );
                $out .= implode( '', $combiners );
                $combiners = array();
            }
            $out .= $c;
            $lastClass = 0;
        }
        if( $lastClass ) {
            ksort( $combiners );
            $out .= implode( '', $combiners );
        }
        return $out;
    }

    static function fastCompose( $string ) {
        UtfNormal::loadData();
        $len = strlen( $string );
        $out = '';
        $lastClass = -1;
        $lastHangul = 0;
        $startChar = '';
        $combining = '';
        $x1 = ord(substr(UTF8_HANGUL_VBASE, 0, 1));
        $x2 = ord(substr(UTF8_HANGUL_TEND, 0, 1));
        for( $i = 0; $i < $len; $i++ ) {
            $c = $string[$i];
            $n = ord( $c );
            if( $n < 0x80 ) {
                # No combining characters here...
                $out .= $startChar;
                $out .= $combining;
                $startChar = $c;
                $combining = '';
                $lastClass = 0;
                continue;
            } elseif( $n >= 0xf0 ) {
                $c = substr( $string, $i, 4 );
                $i += 3;
            } elseif( $n >= 0xe0 ) {
                $c = substr( $string, $i, 3 );
                $i += 2;
            } elseif( $n >= 0xc0 ) {
                $c = substr( $string, $i, 2 );
                $i++;
            }
            $pair = $startChar . $c;
            if( $n > 0x80 ) {
                if( isset( self::$utfCombiningClass[$c] ) ) {
                    # A combining char; see what we can do with it
                    $class = self::$utfCombiningClass[$c];
                    if( !empty( $startChar ) &&
                        $lastClass < $class &&
                        $class > 0 &&
                        isset( self::$utfCanonicalComp[$pair] ) ) {
                        $startChar = self::$utfCanonicalComp[$pair];
                        $class = 0;
                    } else {
                        $combining .= $c;
                    }
                    $lastClass = $class;
                    $lastHangul = 0;
                    continue;
                }
            }
            # New start char
            if( $lastClass == 0 ) {
                if( isset( self::$utfCanonicalComp[$pair] ) ) {
                    $startChar = self::$utfCanonicalComp[$pair];
                    $lastHangul = 0;
                    continue;
                }
                if( $n >= $x1 && $n <= $x2 ) {
                    # WARNING: Hangul code is painfully slow.
                    # I apologize for this ugly, ugly code; however
                    # performance is even more teh suck if we call
                    # out to nice clean functions. Lookup tables are
                    # marginally faster, but require a lot of space.
                    #
                    if( $c >= UTF8_HANGUL_VBASE &&
                        $c <= UTF8_HANGUL_VEND &&
                        $startChar >= UTF8_HANGUL_LBASE &&
                        $startChar <= UTF8_HANGUL_LEND ) {
                        #
                        #$lIndex = utf8ToCodepoint( $startChar ) - UNICODE_HANGUL_LBASE;
                        #$vIndex = utf8ToCodepoint( $c ) - UNICODE_HANGUL_VBASE;
                        $lIndex = ord( $startChar[2] ) - 0x80;
                        $vIndex = ord( $c[2]         ) - 0xa1;

                        $hangulPoint = UNICODE_HANGUL_FIRST +
                            UNICODE_HANGUL_TCOUNT *
                            (UNICODE_HANGUL_VCOUNT * $lIndex + $vIndex);

                        # Hardcode the limited-range UTF-8 conversion:
                        $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 &&
                              $startChar <= UTF8_HANGUL_LAST &&
                              !$lastHangul ) {
                        # $tIndex = utf8ToCodepoint( $c ) - UNICODE_HANGUL_TBASE;
                        $tIndex = ord( $c[2] ) - 0xa7;
                        if( $tIndex < 0 ) $tIndex = ord( $c[2] ) - 0x80 + (0x11c0 - 0x11a7);

                        # Increment the code point by $tIndex, without
                        # the function overhead of decoding and recoding UTF-8
                        #
                        $tail = ord( $startChar[2] ) + $tIndex;
                        if( $tail > 0xbf ) {
                            $tail -= 0x40;
                            $mid = ord( $startChar[1] ) + 1;
                            if( $mid > 0xbf ) {
                                $startChar[0] = chr( ord( $startChar[0] ) + 1 );
                                $mid -= 0x40;
                            }
                            $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;
                    }
                }
            }
            $out .= $startChar;
            $out .= $combining;
            $startChar = $c;
            $combining = '';
            $lastClass = 0;
            $lastHangul = 0;
        }
        $out .= $startChar . $combining;
        return $out;
    }

    static function placebo( $string ) {
        $len = strlen( $string );
        $out = '';
        for( $i = 0; $i < $len; $i++ ) {
            $out .= $string[$i];
        }
        return $out;
    }
    private static function replaceForNativeNormalize( $string ) {
        $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 );
        return $string;
    }
}