1 package Encode::Unicode;
5 no warnings 'redefine';
7 our $VERSION = do { my @r = ( q$Revision: 2.9 $ =~ /\d+/g ); sprintf "%d." . "%02d" x $#r, @r };
10 XSLoader::load( __PACKAGE__, $VERSION );
13 # Object Generator 8 transcoders all at once!
18 our %BOM_Unknown = map { $_ => 1 } qw(UTF-16 UTF-32);
21 qw(UTF-16 UTF-16BE UTF-16LE
22 UTF-32 UTF-32BE UTF-32LE
26 my ( $size, $endian, $ucs2, $mask );
27 $name =~ /^(\w+)-(\d+)(\w*)$/o;
28 if ( $ucs2 = ( $1 eq 'UCS' ) ) {
34 $endian = ( $3 eq 'BE' ) ? 'n' : ( $3 eq 'LE' ) ? 'v' : '';
35 $size == 4 and $endian = uc($endian);
37 $Encode::Encoding{$name} = bless {
45 use parent qw(Encode::Encoding);
49 $BOM_Unknown{ $self->name } or return $self;
50 my $clone = bless {%$self} => ref($self);
51 $clone->{renewed}++; # so the caller knows it is renewed.
55 # There used to be a perl implementation of (en|de)code but with
56 # XS version is ripe, perl version is zapped for optimal speed
58 *decode = \&decode_xs;
59 *encode = \&encode_xs;
66 Encode::Unicode -- Various Unicode Transformation Formats
72 use Encode qw/encode decode/;
73 $ucs2 = encode("UCS-2BE", $utf8);
74 $utf8 = decode("UCS-2BE", $ucs2);
78 This module implements all Character Encoding Schemes of Unicode that
79 are officially documented by Unicode Consortium (except, of course,
80 for UTF-8, which is a native format in perl).
84 =item L<http://www.unicode.org/glossary/> says:
86 I<Character Encoding Scheme> A character encoding form plus byte
87 serialization. There are Seven character encoding schemes in Unicode:
88 UTF-8, UTF-16, UTF-16BE, UTF-16LE, UTF-32 (UCS-4), UTF-32BE (UCS-4BE) and
89 UTF-32LE (UCS-4LE), and UTF-7.
91 Since UTF-7 is a 7-bit (re)encoded version of UTF-16BE, It is not part of
92 Unicode's Character Encoding Scheme. It is separately implemented in
93 Encode::Unicode::UTF7. For details see L<Encode::Unicode::UTF7>.
97 Decodes from ord(N) Encodes chr(N) to...
98 octet/char BOM S.P d800-dfff ord > 0xffff \x{1abcd} ==
99 ---------------+-----------------+------------------------------
100 UCS-2BE 2 N N is bogus Not Available
101 UCS-2LE 2 N N bogus Not Available
102 UTF-16 2/4 Y Y is S.P S.P BE/LE
103 UTF-16BE 2/4 N Y S.P S.P 0xd82a,0xdfcd
104 UTF-16LE 2/4 N Y S.P S.P 0x2ad8,0xcddf
105 UTF-32 4 Y - is bogus As is BE/LE
106 UTF-32BE 4 N - bogus As is 0x0001abcd
107 UTF-32LE 4 N - bogus As is 0xcdab0100
108 UTF-8 1-4 - - bogus >= 4 octets \xf0\x9a\af\8d
109 ---------------+-----------------+------------------------------
113 =head1 Size, Endianness, and BOM
115 You can categorize these CES by 3 criteria: size of each character,
116 endianness, and Byte Order Mark.
120 UCS-2 is a fixed-length encoding with each character taking 16 bits.
121 It B<does not> support I<surrogate pairs>. When a surrogate pair
122 is encountered during decode(), its place is filled with \x{FFFD}
123 if I<CHECK> is 0, or the routine croaks if I<CHECK> is 1. When a
124 character whose ord value is larger than 0xFFFF is encountered,
125 its place is filled with \x{FFFD} if I<CHECK> is 0, or the routine
126 croaks if I<CHECK> is 1.
128 UTF-16 is almost the same as UCS-2 but it supports I<surrogate pairs>.
129 When it encounters a high surrogate (0xD800-0xDBFF), it fetches the
130 following low surrogate (0xDC00-0xDFFF) and C<desurrogate>s them to
131 form a character. Bogus surrogates result in death. When \x{10000}
132 or above is encountered during encode(), it C<ensurrogate>s them and
133 pushes the surrogate pair to the output stream.
135 UTF-32 (UCS-4) is a fixed-length encoding with each character taking 32 bits.
136 Since it is 32-bit, there is no need for I<surrogate pairs>.
140 The first (and now failed) goal of Unicode was to map all character
141 repertoires into a fixed-length integer so that programmers are happy.
142 Since each character is either a I<short> or I<long> in C, you have to
143 pay attention to the endianness of each platform when you pass data
146 Anything marked as BE is Big Endian (or network byte order) and LE is
147 Little Endian (aka VAX byte order). For anything not marked either
148 BE or LE, a character called Byte Order Mark (BOM) indicating the
149 endianness is prepended to the string.
151 CAVEAT: Though BOM in utf8 (\xEF\xBB\xBF) is valid, it is meaningless
152 and as of this writing Encode suite just leave it as is (\x{FeFF}).
156 =item BOM as integer when fetched in network byte order
159 -------------------------
162 -------------------------
166 This modules handles the BOM as follows.
172 When BE or LE is explicitly stated as the name of encoding, BOM is
173 simply treated as a normal character (ZERO WIDTH NO-BREAK SPACE).
177 When BE or LE is omitted during decode(), it checks if BOM is at the
178 beginning of the string; if one is found, the endianness is set to
179 what the BOM says. If no BOM is found, the routine dies.
183 When BE or LE is omitted during encode(), it returns a BE-encoded
184 string with BOM prepended. So when you want to encode a whole text
185 file, make sure you encode() the whole text at once, not line by line
186 or each line, not file, will have a BOM prepended.
190 C<UCS-2> is an exception. Unlike others, this is an alias of UCS-2BE.
191 UCS-2 is already registered by IANA and others that way.
195 =head1 Surrogate Pairs
197 To say the least, surrogate pairs were the biggest mistake of the
198 Unicode Consortium. But according to the late Douglas Adams in I<The
199 Hitchhiker's Guide to the Galaxy> Trilogy, C<In the beginning the
200 Universe was created. This has made a lot of people very angry and
201 been widely regarded as a bad move>. Their mistake was not of this
202 magnitude so let's forgive them.
204 (I don't dare make any comparison with Unicode Consortium and the
205 Vogons here ;) Or, comparing Encode to Babel Fish is completely
206 appropriate -- if you can only stick this into your ear :)
208 Surrogate pairs were born when the Unicode Consortium finally
209 admitted that 16 bits were not big enough to hold all the world's
210 character repertoires. But they already made UCS-2 16-bit. What
213 Back then, the range 0xD800-0xDFFF was not allocated. Let's split
214 that range in half and use the first half to represent the C<upper
215 half of a character> and the second half to represent the C<lower
216 half of a character>. That way, you can represent 1024 * 1024 =
217 1048576 more characters. Now we can store character ranges up to
218 \x{10ffff} even with 16-bit encodings. This pair of half-character is
219 now called a I<surrogate pair> and UTF-16 is the name of the encoding
222 Here is a formula to ensurrogate a Unicode character \x{10000} and
225 $hi = ($uni - 0x10000) / 0x400 + 0xD800;
226 $lo = ($uni - 0x10000) % 0x400 + 0xDC00;
230 $uni = 0x10000 + ($hi - 0xD800) * 0x400 + ($lo - 0xDC00);
232 Note this move has made \x{D800}-\x{DFFF} into a forbidden zone but
233 perl does not prohibit the use of characters within this range. To perl,
234 every one of \x{0000_0000} up to \x{ffff_ffff} (*) is I<a character>.
236 (*) or \x{ffff_ffff_ffff_ffff} if your perl is compiled with 64-bit
239 =head1 Error Checking
241 Unlike most encodings which accept various ways to handle errors,
242 Unicode encodings simply croaks.
244 % perl -MEncode -e'$_ = "\xfe\xff\xd8\xd9\xda\xdb\0\n"' \
245 -e'Encode::from_to($_, "utf16","shift_jis", 0); print'
246 UTF-16:Malformed LO surrogate d8d9 at /path/to/Encode.pm line 184.
247 % perl -MEncode -e'$a = "BOM missing"' \
248 -e' Encode::from_to($a, "utf16", "shift_jis", 0); print'
249 UTF-16:Unrecognised BOM 424f at /path/to/Encode.pm line 184.
251 Unlike other encodings where mappings are not one-to-one against
252 Unicode, UTFs are supposed to map 100% against one another. So Encode
253 is more strict on UTFs.
255 Consider that "division by zero" of Encode :)
259 L<Encode>, L<Encode::Unicode::UTF7>, L<http://www.unicode.org/glossary/>,
260 L<http://www.unicode.org/unicode/faq/utf_bom.html>,
262 RFC 2781 L<http://www.ietf.org/rfc/rfc2781.txt>,
264 The whole Unicode standard L<http://www.unicode.org/unicode/uni2book/u2.html>
266 Ch. 15, pp. 403 of C<Programming Perl (3rd Edition)>
267 by Larry Wall, Tom Christiansen, Jon Orwant;
268 O'Reilly & Associates; ISBN 0-596-00027-8