1 package Encode::Unicode;
6 our $VERSION = do { my @r = (q$Revision: 1.34 $ =~ /\d+/g); sprintf "%d."."%02d" x $#r, @r };
9 XSLoader::load(__PACKAGE__,$VERSION);
12 # Object Generator 8 transcoders all at once!
16 for my $name (qw(UTF-16 UTF-16BE UTF-16LE
17 UTF-32 UTF-32BE UTF-32LE
20 my ($size, $endian, $ucs2, $mask);
21 $name =~ /^(\w+)-(\d+)(\w*)$/o;
22 if ($ucs2 = ($1 eq 'UCS')){
27 $endian = ($3 eq 'BE') ? 'n' : ($3 eq 'LE') ? 'v' : '' ;
28 $size == 4 and $endian = uc($endian);
30 $Encode::Encoding{$name} =
40 sub name { shift->{'Name'} }
44 # Return the original if endian known
45 return $self if ($self->{endian});
47 return bless {%$self},ref($self);
50 sub needs_lines { 0 };
53 exists $INC{"PerlIO/encoding.pm"} or return 0;
59 # three implementations of (en|de)code exist. The XS version is the
60 # fastest. *_modern uses an array and *_classic sticks with substr.
61 # *_classic is much slower but more memory conservative.
62 # *_xs is the default.
65 no warnings qw(redefine);
68 *decode = \&decode_xs;
69 *encode = \&encode_xs;
70 }elsif($type eq "modern"){
71 *decode = \&decode_modern;
72 *encode = \&encode_modern;
73 }elsif($type eq "classic"){
74 *decode = \&decode_classic;
75 *encode = \&encode_classic;
78 Carp::croak __PACKAGE__, "::set_transcoder(modern|classic|xs)";
85 # Aux. subs & constants
88 sub FBCHAR(){ 0xFFFd }
89 sub BOM_BE(){ 0xFeFF }
90 sub BOM16LE(){ 0xFFFe }
91 sub BOM32LE(){ 0xFFFe0000 }
95 (0 <= $_[0] && $_[0] < 0xD800)
96 || ( 0xDFFF < $_[0] && $_[0] <= 0xFFFF);
99 sub issurrogate($){ 0xD800 <= $_[0] && $_[0] <= 0xDFFF }
100 sub isHiSurrogate($){ 0xD800 <= $_[0] && $_[0] < 0xDC00 }
101 sub isLoSurrogate($){ 0xDC00 <= $_[0] && $_[0] <= 0xDFFF }
104 use integer; # we have divisions
106 my $hi = ($uni - 0x10000) / 0x400 + 0xD800;
107 my $lo = ($uni - 0x10000) % 0x400 + 0xDC00;
113 return 0x10000 + ($hi - 0xD800)*0x400 + ($lo - 0xDC00);
116 sub Mask { {2 => 0xffff, 4 => 0xffffffff} }
119 # *_modern are much faster but guzzle more memory
122 sub decode_modern($$;$)
124 my ($obj, $str, $chk ) = @_;
125 my ($size, $endian, $ucs2) = @$obj{qw(size endian ucs2)};
127 # warn "$size, $endian, $ucs2";
128 $endian ||= BOMB($size, substr($str, 0, $size, ''))
129 or poisoned2death($obj, "Where's the BOM?");
130 my $mask = Mask->{$size};
132 my @ord = unpack("$endian*", $str);
133 undef $str; # to conserve memory
135 my $ord = shift @ord;
136 unless ($size == 4 or valid_ucs2($ord &= $mask)){
139 poisoned2death($obj, "no surrogates allowed", $ord);
140 shift @ord; # skip the next one as well
143 unless (isHiSurrogate($ord)){
144 poisoned2death($obj, "Malformed HI surrogate", $ord);
147 unless (isLoSurrogate($lo &= $mask)){
148 poisoned2death($obj, "Malformed LO surrogate", $ord, $lo);
150 $ord = desurrogate($ord, $lo);
155 utf8::upgrade($utf8);
159 sub encode_modern($$;$)
161 my ($obj, $utf8, $chk) = @_;
162 my ($size, $endian, $ucs2) = @$obj{qw(size endian ucs2)};
165 $endian = ($size == 4) ? 'N' : 'n';
168 my @ord = unpack("U*", $utf8);
169 undef $utf8; # to conserve memory
171 unless ($size == 4 or valid_ucs2($ord)) {
172 unless(issurrogate($ord)){
175 poisoned2death($obj, "code point too high", $ord);
180 push @str, ensurrogate($ord);
182 }else{ # not supposed to happen
189 return pack("$endian*", @str);
193 # *_classic are slower but more memory conservative
196 sub decode_classic($$;$)
198 my ($obj, $str, $chk ) = @_;
199 my ($size, $endian, $ucs2) = @$obj{qw(size endian ucs2)};
201 # warn "$size, $endian, $ucs2";
202 $endian ||= BOMB($size, substr($str, 0, $size, ''))
203 or poisoned2death($obj, "Where's the BOM?");
204 my $mask = Mask->{$size};
206 my @ord = unpack("$endian*", $str);
207 while (length($str)){
208 my $ord = unpack($endian, substr($str, 0, $size, ''));
209 unless ($size == 4 or valid_ucs2($ord &= $mask)){
212 poisoned2death($obj, "no surrogates allowed", $ord);
213 substr($str,0,$size,''); # skip the next one as well
216 unless (isHiSurrogate($ord)){
217 poisoned2death($obj, "Malformed HI surrogate", $ord);
219 my $lo = unpack($endian ,substr($str,0,$size,''));
220 unless (isLoSurrogate($lo &= $mask)){
221 poisoned2death($obj, "Malformed LO surrogate", $ord, $lo);
223 $ord = desurrogate($ord, $lo);
228 utf8::upgrade($utf8);
232 sub encode_classic($$;$)
234 my ($obj, $utf8, $chk) = @_;
235 my ($size, $endian, $ucs2) = @$obj{qw(size endian ucs2)};
236 # warn join ", ", $size, $ucs2, $endian, $mask;
239 $endian = ($size == 4) ? 'N' : 'n';
240 $str .= pack($endian, BOM_BE);
242 while (length($utf8)){
243 my $ord = ord(substr($utf8,0,1,''));
244 unless ($size == 4 or valid_ucs2($ord)) {
245 unless(issurrogate($ord)){
248 poisoned2death($obj, "code point too high", $ord);
249 $str .= pack($endian, FBCHAR);
251 $str .= pack($endian.2, ensurrogate($ord));
253 }else{ # not supposed to happen
254 $str .= pack($endian, FBCHAR);
257 $str .= pack($endian, $ord);
264 my ($size, $bom) = @_;
265 my $N = $size == 2 ? 'n' : 'N';
266 my $ord = unpack($N, $bom);
267 return ($ord eq BOM_BE) ? $N :
268 ($ord eq BOM16LE) ? 'v' : ($ord eq BOM32LE) ? 'V' : undef;
274 my $pair = join(", ", map {sprintf "\\x%x", $_} @_);
276 Carp::croak $obj->name, ":", $msg, "<$pair>.", caller;
284 Encode::Unicode -- Various Unicode Transformation Formats
290 use Encode qw/encode decode/;
291 $ucs2 = encode("UCS-2BE", $utf8);
292 $utf8 = decode("UCS-2BE", $ucs2);
296 This module implements all Character Encoding Schemes of Unicode that
297 are officially documented by Unicode Consortium (except, of course,
298 for UTF-8, which is a native format in perl).
302 =item L<http://www.unicode.org/glossary/> says:
304 I<Character Encoding Scheme> A character encoding form plus byte
305 serialization. There are seven character encoding schemes in Unicode:
306 UTF-8, UTF-16, UTF-16BE, UTF-16LE, UTF-32, UTF-32BE and UTF-32LE.
308 =item Quick Reference
310 Decodes from ord(N) Encodes chr(N) to...
311 octet/char BOM S.P d800-dfff ord > 0xffff \x{1abcd} ==
312 ---------------+-----------------+------------------------------
313 UCS-2BE 2 N N is bogus Not Available
314 UCS-2LE 2 N N bogus Not Available
315 UTF-16 2/4 Y Y is S.P S.P BE/LE
316 UTF-16BE 2/4 N Y S.P S.P 0xd82a,0xdfcd
317 UTF-16LE 2 N Y S.P S.P 0x2ad8,0xcddf
318 UTF-32 4 Y - is bogus As is BE/LE
319 UTF-32BE 4 N - bogus As is 0x0001abcd
320 UTF-32LE 4 N - bogus As is 0xcdab0100
321 UTF-8 1-4 - - bogus >= 4 octets \xf0\x9a\af\8d
322 ---------------+-----------------+------------------------------
326 =head1 Size, Endianness, and BOM
328 You can categorize these CES by 3 criteria: size of each character,
329 endianness, and Byte Order Mark.
333 UCS-2 is a fixed-length encoding with each character taking 16 bits.
334 It B<does not> support I<surrogate pairs>. When a surrogate pair
335 is encountered during decode(), its place is filled with \x{FFFD}
336 if I<CHECK> is 0, or the routine croaks if I<CHECK> is 1. When a
337 character whose ord value is larger than 0xFFFF is encountered,
338 its place is filled with \x{FFFD} if I<CHECK> is 0, or the routine
339 croaks if I<CHECK> is 1.
341 UTF-16 is almost the same as UCS-2 but it supports I<surrogate pairs>.
342 When it encounters a high surrogate (0xD800-0xDBFF), it fetches the
343 following low surrogate (0xDC00-0xDFFF) and C<desurrogate>s them to
344 form a character. Bogus surrogates result in death. When \x{10000}
345 or above is encountered during encode(), it C<ensurrogate>s them and
346 pushes the surrogate pair to the output stream.
348 UTF-32 is a fixed-length encoding with each character taking 32 bits.
349 Since it is 32-bit, there is no need for I<surrogate pairs>.
353 The first (and now failed) goal of Unicode was to map all character
354 repertoires into a fixed-length integer so that programmers are happy.
355 Since each character is either a I<short> or I<long> in C, you have to
356 pay attention to the endianness of each platform when you pass data
359 Anything marked as BE is Big Endian (or network byte order) and LE is
360 Little Endian (aka VAX byte order). For anything not marked either
361 BE or LE, a character called Byte Order Mark (BOM) indicating the
362 endianness is prepended to the string.
366 =item BOM as integer when fetched in network byte order
369 -------------------------
372 -------------------------
376 This modules handles the BOM as follows.
382 When BE or LE is explicitly stated as the name of encoding, BOM is
383 simply treated as a normal character (ZERO WIDTH NO-BREAK SPACE).
387 When BE or LE is omitted during decode(), it checks if BOM is at the
388 beginning of the string; if one is found, the endianness is set to
389 what the BOM says. If no BOM is found, the routine dies.
393 When BE or LE is omitted during encode(), it returns a BE-encoded
394 string with BOM prepended. So when you want to encode a whole text
395 file, make sure you encode() the whole text at once, not line by line
396 or each line, not file, will have a BOM prepended.
400 C<UCS-2> is an exception. Unlike others, this is an alias of UCS-2BE.
401 UCS-2 is already registered by IANA and others that way.
405 =head1 Surrogate Pairs
407 To say the least, surrogate pairs were the biggest mistake of the
408 Unicode Consortium. But according to the late Douglas Adams in I<The
409 Hitchhiker's Guide to the Galaxy> Trilogy, C<In the beginning the
410 Universe was created. This has made a lot of people very angry and
411 been widely regarded as a bad move>. Their mistake was not of this
412 magnitude so let's forgive them.
414 (I don't dare make any comparison with Unicode Consortium and the
415 Vogons here ;) Or, comparing Encode to Babel Fish is completely
416 appropriate -- if you can only stick this into your ear :)
418 Surrogate pairs were born when the Unicode Consortium finally
419 admitted that 16 bits were not big enough to hold all the world's
420 character repertoires. But they already made UCS-2 16-bit. What
423 Back then, the range 0xD800-0xDFFF was not allocated. Let's split
424 that range in half and use the first half to represent the C<upper
425 half of a character> and the second half to represent the C<lower
426 half of a character>. That way, you can represent 1024 * 1024 =
427 1048576 more characters. Now we can store character ranges up to
428 \x{10ffff} even with 16-bit encodings. This pair of half-character is
429 now called a I<surrogate pair> and UTF-16 is the name of the encoding
432 Here is a formula to ensurrogate a Unicode character \x{10000} and
435 $hi = ($uni - 0x10000) / 0x400 + 0xD800;
436 $lo = ($uni - 0x10000) % 0x400 + 0xDC00;
440 $uni = 0x10000 + ($hi - 0xD800) * 0x400 + ($lo - 0xDC00);
442 Note this move has made \x{D800}-\x{DFFF} into a forbidden zone but
443 perl does not prohibit the use of characters within this range. To perl,
444 every one of \x{0000_0000} up to \x{ffff_ffff} (*) is I<a character>.
446 (*) or \x{ffff_ffff_ffff_ffff} if your perl is compiled with 64-bit
451 L<Encode>, L<http://www.unicode.org/glossary/>,
453 RFC 2781 L<http://rfc.net/rfc2781.html>,
455 L<http://www.unicode.org/unicode/faq/utf_bom.html>
457 Ch. 15, pp. 403 of C<Programming Perl (3rd Edition)>
458 by Larry Wall, Tom Christiansen, Jon Orwant;
459 O'Reilly & Associates; ISBN 0-596-00027-8