package Unicode::Normalize; BEGIN { unless ("A" eq pack('U', 0x41)) { die "Unicode::Normalize cannot stringify a Unicode code point\n"; } } use 5.006; use strict; use warnings; use Carp; no warnings 'utf8'; our $VERSION = '0.30'; our $PACKAGE = __PACKAGE__; require Exporter; require DynaLoader; our @ISA = qw(Exporter DynaLoader); our @EXPORT = qw( NFC NFD NFKC NFKD ); our @EXPORT_OK = qw( normalize decompose reorder compose checkNFD checkNFKD checkNFC checkNFKC check getCanon getCompat getComposite getCombinClass isExclusion isSingleton isNonStDecomp isComp2nd isComp_Ex isNFD_NO isNFC_NO isNFC_MAYBE isNFKD_NO isNFKC_NO isNFKC_MAYBE FCD checkFCD FCC checkFCC composeContiguous splitOnLastStarter ); our %EXPORT_TAGS = ( all => [ @EXPORT, @EXPORT_OK ], normalize => [ @EXPORT, qw/normalize decompose reorder compose/ ], check => [ qw/checkNFD checkNFKD checkNFC checkNFKC check/ ], fast => [ qw/FCD checkFCD FCC checkFCC composeContiguous/ ], ); ###### bootstrap Unicode::Normalize $VERSION; ###### sub pack_U { return pack('U*', @_); } sub unpack_U { return unpack('U*', pack('U*').shift); } ## ## normalization forms ## use constant COMPAT => 1; sub NFD ($) { reorder(decompose($_[0])) } sub NFKD ($) { reorder(decompose($_[0], COMPAT)) } sub NFC ($) { compose(reorder(decompose($_[0]))) } sub NFKC ($) { compose(reorder(decompose($_[0], COMPAT))) } sub FCD ($) { my $str = shift; return checkFCD($str) ? $str : NFD($str); } sub FCC ($) { composeContiguous(reorder(decompose($_[0]))) } our %formNorm = ( NFC => \&NFC, C => \&NFC, NFD => \&NFD, D => \&NFD, NFKC => \&NFKC, KC => \&NFKC, NFKD => \&NFKD, KD => \&NFKD, FCD => \&FCD, FCC => \&FCC, ); sub normalize($$) { my $form = shift; my $str = shift; return exists $formNorm{$form} ? $formNorm{$form}->($str) : croak $PACKAGE."::normalize: invalid form name: $form"; } ## ## quick check ## our %formCheck = ( NFC => \&checkNFC, C => \&checkNFC, NFD => \&checkNFD, D => \&checkNFD, NFKC => \&checkNFKC, KC => \&checkNFKC, NFKD => \&checkNFKD, KD => \&checkNFKD, FCD => \&checkFCD, FCC => \&checkFCC, ); sub check($$) { my $form = shift; my $str = shift; return exists $formCheck{$form} ? $formCheck{$form}->($str) : croak $PACKAGE."::check: invalid form name: $form"; } 1; __END__ =head1 NAME Unicode::Normalize - Unicode Normalization Forms =head1 SYNOPSIS (1) using function names exported by default: use Unicode::Normalize; $NFD_string = NFD($string); # Normalization Form D $NFC_string = NFC($string); # Normalization Form C $NFKD_string = NFKD($string); # Normalization Form KD $NFKC_string = NFKC($string); # Normalization Form KC (2) using function names exported on request: use Unicode::Normalize 'normalize'; $NFD_string = normalize('D', $string); # Normalization Form D $NFC_string = normalize('C', $string); # Normalization Form C $NFKD_string = normalize('KD', $string); # Normalization Form KD $NFKC_string = normalize('KC', $string); # Normalization Form KC =head1 DESCRIPTION Parameters: C<$string> is used as a string under character semantics (see F). C<$codepoint> should be an unsigned integer representing a Unicode code point. Note: Between XS edition and pure Perl edition, interpretation of C<$codepoint> as a decimal number has incompatibility. XS converts C<$codepoint> to an unsigned integer, but pure Perl does not. Do not use a floating point nor a negative sign in C<$codepoint>. =head2 Normalization Forms =over 4 =item C<$NFD_string = NFD($string)> returns the Normalization Form D (formed by canonical decomposition). =item C<$NFC_string = NFC($string)> returns the Normalization Form C (formed by canonical decomposition followed by canonical composition). =item C<$NFKD_string = NFKD($string)> returns the Normalization Form KD (formed by compatibility decomposition). =item C<$NFKC_string = NFKC($string)> returns the Normalization Form KC (formed by compatibility decomposition followed by B composition). =item C<$FCD_string = FCD($string)> If the given string is in FCD ("Fast C or D" form; cf. UTN #5), returns it without modification; otherwise returns an FCD string. Note: FCD is not always unique, then plural forms may be equivalent each other. C will return one of these equivalent forms. =item C<$FCC_string = FCC($string)> returns the FCC form ("Fast C Contiguous"; cf. UTN #5). Note: FCC is unique, as well as four normalization forms (NF*). =item C<$normalized_string = normalize($form_name, $string)> As C<$form_name>, one of the following names must be given. 'C' or 'NFC' for Normalization Form C (UAX #15) 'D' or 'NFD' for Normalization Form D (UAX #15) 'KC' or 'NFKC' for Normalization Form KC (UAX #15) 'KD' or 'NFKD' for Normalization Form KD (UAX #15) 'FCD' for "Fast C or D" Form (UTN #5) 'FCC' for "Fast C Contiguous" (UTN #5) =back =head2 Decomposition and Composition =over 4 =item C<$decomposed_string = decompose($string)> =item C<$decomposed_string = decompose($string, $useCompatMapping)> Decomposes the specified string and returns the result. If the second parameter (a boolean) is omitted or false, decomposes it using the Canonical Decomposition Mapping. If true, decomposes it using the Compatibility Decomposition Mapping. The string returned is not always in NFD/NFKD. Reordering may be required. $NFD_string = reorder(decompose($string)); # eq. to NFD() $NFKD_string = reorder(decompose($string, TRUE)); # eq. to NFKD() =item C<$reordered_string = reorder($string)> Reorders the combining characters and the like in the canonical ordering and returns the result. E.g., when you have a list of NFD/NFKD strings, you can get the concatenated NFD/NFKD string from them, saying $concat_NFD = reorder(join '', @NFD_strings); $concat_NFKD = reorder(join '', @NFKD_strings); =item C<$composed_string = compose($string)> Returns the string where composable pairs are composed. E.g., when you have a NFD/NFKD string, you can get its NFC/NFKC string, saying $NFC_string = compose($NFD_string); $NFKC_string = compose($NFKD_string); =back =head2 Quick Check (see Annex 8, UAX #15; and F) The following functions check whether the string is in that normalization form. The result returned will be: YES The string is in that normalization form. NO The string is not in that normalization form. MAYBE Dubious. Maybe yes, maybe no. =over 4 =item C<$result = checkNFD($string)> returns C (C<1>) or C (C). =item C<$result = checkNFC($string)> returns C (C<1>), C (C), or C (C). =item C<$result = checkNFKD($string)> returns C (C<1>) or C (C). =item C<$result = checkNFKC($string)> returns C (C<1>), C (C), or C (C). =item C<$result = checkFCD($string)> returns C (C<1>) or C (C). =item C<$result = checkFCC($string)> returns C (C<1>), C (C), or C (C). If a string is not in FCD, it must not be in FCC. So C should return C. =item C<$result = check($form_name, $string)> returns C (C<1>), C (C), or C (C). C<$form_name> is alike to that for C. =back B In the cases of NFD, NFKD, and FCD, the answer must be either C or C. The answer C may be returned in the cases of NFC, NFKC, and FCC. A C string should contain at least one combining character or the like. For example, C has the MAYBE_NFC/MAYBE_NFKC property. Both C and C will return C. C<"A\N{COMBINING ACUTE ACCENT}"> is not in NFC (its NFC is C<"\N{LATIN CAPITAL LETTER A WITH ACUTE}">), while C<"B\N{COMBINING ACUTE ACCENT}"> is in NFC. If you want to check exactly, compare the string with its NFC/NFKC/FCC; i.e., $string eq NFC($string) # thorough than checkNFC($string) $string eq NFKC($string) # thorough than checkNFKC($string) $string eq FCC($string) # thorough than checkFCC($string) =head2 Character Data These functions are interface of character data used internally. If you want only to get Unicode normalization forms, you don't need call them yourself. =over 4 =item C<$canonical_decomposed = getCanon($codepoint)> If the character of the specified codepoint is canonically decomposable (including Hangul Syllables), returns the B string canonically equivalent to it. If it is not decomposable, returns C. =item C<$compatibility_decomposed = getCompat($codepoint)> If the character of the specified codepoint is compatibility decomposable (including Hangul Syllables), returns the B string compatibility equivalent to it. If it is not decomposable, returns C. =item C<$codepoint_composite = getComposite($codepoint_here, $codepoint_next)> If two characters here and next (as codepoints) are composable (including Hangul Jamo/Syllables and Composition Exclusions), returns the codepoint of the composite. If they are not composable, returns C. =item C<$combining_class = getCombinClass($codepoint)> Returns the combining class of the character as an integer. =item C<$is_exclusion = isExclusion($codepoint)> Returns a boolean whether the character of the specified codepoint is a composition exclusion. =item C<$is_singleton = isSingleton($codepoint)> Returns a boolean whether the character of the specified codepoint is a singleton. =item C<$is_non_starter_decomposition = isNonStDecomp($codepoint)> Returns a boolean whether the canonical decomposition of the character of the specified codepoint is a Non-Starter Decomposition. =item C<$may_be_composed_with_prev_char = isComp2nd($codepoint)> Returns a boolean whether the character of the specified codepoint may be composed with the previous one in a certain composition (including Hangul Compositions, but excluding Composition Exclusions and Non-Starter Decompositions). =back =head2 EXPORT C, C, C, C: by default. C and other some functions: on request. =head1 AUTHOR SADAHIRO Tomoyuki http://homepage1.nifty.com/nomenclator/perl/ Copyright(C) 2001-2004, SADAHIRO Tomoyuki. Japan. All rights reserved. This module is free software; you can redistribute it and/or modify it under the same terms as Perl itself. =head1 SEE ALSO =over 4 =item http://www.unicode.org/reports/tr15/ Unicode Normalization Forms - UAX #15 =item http://www.unicode.org/Public/UNIDATA/DerivedNormalizationProps.txt Derived Normalization Properties =item http://www.unicode.org/notes/tn5/ Canonical Equivalence in Applications - UTN #5 =back =cut