[p5sagit/p5-mst-13.2.git] / ext / Encode / Unicode / Unicode.pm

package Encode::Unicode;

use strict;
use warnings;

our $VERSION = do { my @r = (q$Revision: 1.38 $ =~ /\d+/g); sprintf "%d."."%02d" x $#r, @r };

use XSLoader;
XSLoader::load(__PACKAGE__,$VERSION);

#
# Object Generator 8 transcoders all at once!
#

require Encode;

for my $name (qw(UTF-16 UTF-16BE UTF-16LE
                 UTF-32 UTF-32BE UTF-32LE
                        UCS-2BE  UCS-2LE))
{
    my ($size, $endian, $ucs2, $mask);
    $name =~ /^(\w+)-(\d+)(\w*)$/o;
    if ($ucs2 = ($1 eq 'UCS')){
	$size = 2;
    }else{
	$size = $2/8;
    }
    $endian = ($3 eq 'BE') ? 'n' : ($3 eq 'LE') ? 'v' : '' ;
    $size == 4 and $endian = uc($endian);

    $Encode::Encoding{$name} = 	
	bless {
	       Name   =>   $name,
	       size   =>   $size,
	       endian => $endian,
	       ucs2   =>   $ucs2,
	      } => __PACKAGE__;

}

use base qw(Encode::Encoding);

#
# three implementations of (en|de)code exist.  The XS version is the
# fastest.  *_modern uses an array and *_classic sticks with substr.
# *_classic is  much slower but more memory conservative.
# *_xs is the default.

sub set_transcoder{
    no warnings qw(redefine);
    my $type = shift;
    if    ($type eq "xs"){
	*decode = \&decode_xs;
	*encode = \&encode_xs;
    }elsif($type eq "modern"){
	*decode = \&decode_modern;
	*encode = \&encode_modern;
    }elsif($type eq "classic"){
	*decode = \&decode_classic;
	*encode = \&encode_classic;
    }else{
	require Carp; 
	Carp::croak __PACKAGE__, "::set_transcoder(modern|classic|xs)";
    }
}

set_transcoder("xs");

#
# Aux. subs & constants
#

sub FBCHAR(){ 0xFFFd }
sub BOM_BE(){ 0xFeFF }
sub BOM16LE(){ 0xFFFe }
sub BOM32LE(){ 0xFFFe0000 }

sub valid_ucs2($){
    return 
	(0 <= $_[0] && $_[0] < 0xD800) 
	    || 	( 0xDFFF < $_[0] && $_[0] <= 0xFFFF);
}

sub issurrogate($){   0xD800 <= $_[0]  && $_[0] <= 0xDFFF }
sub isHiSurrogate($){ 0xD800 <= $_[0]  && $_[0] <  0xDC00 }
sub isLoSurrogate($){ 0xDC00 <= $_[0]  && $_[0] <= 0xDFFF }

sub ensurrogate($){
    use integer; # we have divisions
    my $uni = shift;
    my  $hi = ($uni - 0x10000) / 0x400 + 0xD800;
    my  $lo = ($uni - 0x10000) % 0x400 + 0xDC00;
    return ($hi, $lo);
}

sub desurrogate($$){
    my ($hi, $lo) = @_;
    return 0x10000 + ($hi - 0xD800)*0x400 + ($lo - 0xDC00);
}

sub Mask { {2 => 0xffff,  4 => 0xffffffff} }

#
# *_modern are much faster but guzzle more memory
#

sub decode_modern($$;$)
{
    my ($obj, $str, $chk ) = @_;
    my ($size, $endian, $ucs2) = @$obj{qw(size endian ucs2)};

    # warn "$size, $endian, $ucs2";
    $endian ||= BOMB($size, substr($str, 0, $size, ''))
	or poisoned2death($obj, "Where's the BOM?");
    my  $mask = Mask->{$size};
    my $utf8   = '';
    my @ord = unpack("$endian*", $str);
    undef $str; # to conserve memory
    while (@ord){
	my $ord = shift @ord;
	unless ($size == 4 or valid_ucs2($ord &= $mask)){
	    if ($ucs2){
		$chk and 
		    poisoned2death($obj, "no surrogates allowed", $ord);
		shift @ord; # skip the next one as well
		$ord = FBCHAR;
	    }else{
		unless (isHiSurrogate($ord)){
		    poisoned2death($obj, "Malformed HI surrogate", $ord);
		}
		my $lo = shift @ord;
		unless (isLoSurrogate($lo &= $mask)){
		    poisoned2death($obj, "Malformed LO surrogate", $ord, $lo);
		}
		$ord = desurrogate($ord, $lo);
	    }
	}
	$utf8 .= chr($ord);
    }
    utf8::upgrade($utf8);
    return $utf8;
}

sub encode_modern($$;$)
{
    my ($obj, $utf8, $chk) = @_;
    my ($size, $endian, $ucs2) = @$obj{qw(size endian ucs2)};
    my @str = ();
    unless ($endian){
	$endian = ($size == 4) ? 'N' : 'n';
	push @str, BOM_BE;
    }
    my @ord = unpack("U*", $utf8);
    undef $utf8; # to conserve memory
    for my $ord (@ord){
	unless ($size == 4 or valid_ucs2($ord)) {
	    unless(issurrogate($ord)){
		if ($ucs2){
		    $chk and 
			poisoned2death($obj, "code point too high", $ord);

		    push @str, FBCHAR;
		}else{
		 
		    push @str, ensurrogate($ord);
		}
	    }else{  # not supposed to happen
		push @str, FBCHAR;
	    }
	}else{
	    push @str, $ord;
	}
    }
    return pack("$endian*", @str);
}

#
# *_classic are slower but more memory conservative
#

sub decode_classic($$;$)
{
    my ($obj, $str, $chk ) = @_;
    my ($size, $endian, $ucs2) = @$obj{qw(size endian ucs2)};

    # warn "$size, $endian, $ucs2";
    $endian ||= BOMB($size, substr($str, 0, $size, ''))
	or poisoned2death($obj, "Where's the BOM?");
    my  $mask = Mask->{$size};
    my $utf8   = '';
    my @ord = unpack("$endian*", $str);
    while (length($str)){
	 my $ord = unpack($endian, substr($str, 0, $size, ''));
	unless ($size == 4 or valid_ucs2($ord &= $mask)){
	    if ($ucs2){
		$chk and 
		    poisoned2death($obj, "no surrogates allowed", $ord);
		substr($str,0,$size,''); # skip the next one as well
		$ord = FBCHAR;
	    }else{
		unless (isHiSurrogate($ord)){
		    poisoned2death($obj, "Malformed HI surrogate", $ord);
		}
		my $lo = unpack($endian ,substr($str,0,$size,''));
		unless (isLoSurrogate($lo &= $mask)){
		    poisoned2death($obj, "Malformed LO surrogate", $ord, $lo);
		}
		$ord = desurrogate($ord, $lo);
	    }
	}
	$utf8 .= chr($ord);
    }
    utf8::upgrade($utf8);
    return $utf8;
}

sub encode_classic($$;$)
{
    my ($obj, $utf8, $chk) = @_;
    my ($size, $endian, $ucs2) = @$obj{qw(size endian ucs2)};
    # warn join ", ", $size, $ucs2, $endian, $mask;
    my $str   = '';
    unless ($endian){
	$endian = ($size == 4) ? 'N' : 'n';
	$str .= pack($endian, BOM_BE);
    }
    while (length($utf8)){
	my $ord  = ord(substr($utf8,0,1,''));
	unless ($size == 4 or valid_ucs2($ord)) {
	    unless(issurrogate($ord)){
		if ($ucs2){
		    $chk and 
			poisoned2death($obj, "code point too high", $ord);
		    $str .= pack($endian, FBCHAR);
		}else{
		    $str .= pack($endian.2, ensurrogate($ord));
		}
	    }else{  # not supposed to happen
		$str .= pack($endian, FBCHAR);
	    }
	}else{
	    $str .= pack($endian, $ord);
	}
    }
    return $str;
}

sub BOMB {
    my ($size, $bom) = @_;
    my $N = $size == 2 ? 'n' : 'N';
    my $ord = unpack($N, $bom);
    return ($ord eq BOM_BE) ? $N : 
	($ord eq BOM16LE) ? 'v' : ($ord eq BOM32LE) ? 'V' : undef;
}

sub poisoned2death{
    my $obj = shift;
    my $msg = shift;
    my $pair = join(", ", map {sprintf "\\x%x", $_} @_);
    require Carp;
    Carp::croak $obj->name, ":", $msg, "<$pair>.", caller;
}

1;
__END__

=head1 NAME

Encode::Unicode -- Various Unicode Transformation Formats

=cut

=head1 SYNOPSIS

    use Encode qw/encode decode/; 
    $ucs2 = encode("UCS-2BE", $utf8);
    $utf8 = decode("UCS-2BE", $ucs2);

=head1 ABSTRACT

This module implements all Character Encoding Schemes of Unicode that
are officially documented by Unicode Consortium (except, of course,
for UTF-8, which is a native format in perl).

=over 4

=item L<http://www.unicode.org/glossary/> says:

I<Character Encoding Scheme> A character encoding form plus byte
serialization. There are seven character encoding schemes in Unicode:
UTF-8, UTF-16, UTF-16BE, UTF-16LE, UTF-32 (UCS-4), UTF-32BE (UCS-4BE) and
UTF-32LE (UCS-4LE).

=item Quick Reference

                Decodes from ord(N)           Encodes chr(N) to...
       octet/char BOM S.P d800-dfff  ord > 0xffff     \x{1abcd} ==
  ---------------+-----------------+------------------------------
  UCS-2BE	2   N   N  is bogus                  Not Available
  UCS-2LE       2   N   N     bogus                  Not Available
  UTF-16      2/4   Y   Y  is   S.P           S.P            BE/LE
  UTF-16BE    2/4   N   Y       S.P           S.P    0xd82a,0xdfcd
  UTF-16LE	2   N   Y       S.P           S.P    0x2ad8,0xcddf
  UTF-32	4   Y   -  is bogus         As is            BE/LE
  UTF-32BE	4   N   -     bogus         As is       0x0001abcd
  UTF-32LE	4   N   -     bogus         As is       0xcdab0100
  UTF-8       1-4   -   -     bogus   >= 4 octets   \xf0\x9a\af\8d
  ---------------+-----------------+------------------------------

=back

=head1 Size, Endianness, and BOM

You can categorize these CES by 3 criteria:  size of each character,
endianness, and Byte Order Mark.

=head2 by size

UCS-2 is a fixed-length encoding with each character taking 16 bits.
It B<does not> support I<surrogate pairs>.  When a surrogate pair
is encountered during decode(), its place is filled with \x{FFFD}
if I<CHECK> is 0, or the routine croaks if I<CHECK> is 1.  When a
character whose ord value is larger than 0xFFFF is encountered,
its place is filled with \x{FFFD} if I<CHECK> is 0, or the routine
croaks if I<CHECK> is 1.

UTF-16 is almost the same as UCS-2 but it supports I<surrogate pairs>.
When it encounters a high surrogate (0xD800-0xDBFF), it fetches the
following low surrogate (0xDC00-0xDFFF) and C<desurrogate>s them to
form a character.  Bogus surrogates result in death.  When \x{10000}
or above is encountered during encode(), it C<ensurrogate>s them and
pushes the surrogate pair to the output stream.

UTF-32 (UCS-4) is a fixed-length encoding with each character taking 32 bits.
Since it is 32-bit, there is no need for I<surrogate pairs>.

=head2 by endianness

The first (and now failed) goal of Unicode was to map all character
repertoires into a fixed-length integer so that programmers are happy.
Since each character is either a I<short> or I<long> in C, you have to
pay attention to the endianness of each platform when you pass data
to one another.

Anything marked as BE is Big Endian (or network byte order) and LE is
Little Endian (aka VAX byte order).  For anything not marked either
BE or LE, a character called Byte Order Mark (BOM) indicating the
endianness is prepended to the string.

=over 4

=item BOM as integer when fetched in network byte order

              16         32 bits/char
  -------------------------
  BE      0xFeFF 0x0000FeFF
  LE      0xFFeF 0xFFFe0000
  -------------------------

=back

This modules handles the BOM as follows.

=over 4

=item *

When BE or LE is explicitly stated as the name of encoding, BOM is
simply treated as a normal character (ZERO WIDTH NO-BREAK SPACE).

=item *

When BE or LE is omitted during decode(), it checks if BOM is at the
beginning of the string; if one is found, the endianness is set to
what the BOM says.  If no BOM is found, the routine dies.

=item *

When BE or LE is omitted during encode(), it returns a BE-encoded
string with BOM prepended.  So when you want to encode a whole text
file, make sure you encode() the whole text at once, not line by line
or each line, not file, will have a BOM prepended.

=item *

C<UCS-2> is an exception.  Unlike others, this is an alias of UCS-2BE.
UCS-2 is already registered by IANA and others that way.

=back

=head1 Surrogate Pairs

To say the least, surrogate pairs were the biggest mistake of the
Unicode Consortium.  But according to the late Douglas Adams in I<The
Hitchhiker's Guide to the Galaxy> Trilogy, C<In the beginning the
Universe was created. This has made a lot of people very angry and
been widely regarded as a bad move>.  Their mistake was not of this
magnitude so let's forgive them.

(I don't dare make any comparison with Unicode Consortium and the
Vogons here ;)  Or, comparing Encode to Babel Fish is completely
appropriate -- if you can only stick this into your ear :)

Surrogate pairs were born when the Unicode Consortium finally
admitted that 16 bits were not big enough to hold all the world's
character repertoires.  But they already made UCS-2 16-bit.  What
do we do?

Back then, the range 0xD800-0xDFFF was not allocated.  Let's split
that range in half and use the first half to represent the C<upper
half of a character> and the second half to represent the C<lower
half of a character>.  That way, you can represent 1024 * 1024 =
1048576 more characters.  Now we can store character ranges up to
\x{10ffff} even with 16-bit encodings.  This pair of half-character is
now called a I<surrogate pair> and UTF-16 is the name of the encoding
that embraces them.

Here is a formula to ensurrogate a Unicode character \x{10000} and
above;

  $hi = ($uni - 0x10000) / 0x400 + 0xD800;
  $lo = ($uni - 0x10000) % 0x400 + 0xDC00;

And to desurrogate;

 $uni = 0x10000 + ($hi - 0xD800) * 0x400 + ($lo - 0xDC00);

Note this move has made \x{D800}-\x{DFFF} into a forbidden zone but
perl does not prohibit the use of characters within this range.  To perl, 
every one of \x{0000_0000} up to \x{ffff_ffff} (*) is I<a character>.

  (*) or \x{ffff_ffff_ffff_ffff} if your perl is compiled with 64-bit
  integer support!

=head1 SEE ALSO

L<Encode>, L<http://www.unicode.org/glossary/>,
L<http://www.unicode.org/unicode/faq/utf_bom.html>,

RFC 2781 L<http://rfc.net/rfc2781.html>,

The whole Unicode standard L<http://www.unicode.org/unicode/uni2book/u2.html>

Ch. 15, pp. 403 of C<Programming Perl (3rd Edition)>
by Larry Wall, Tom Christiansen, Jon Orwant; 
O'Reilly & Associates; ISBN 0-596-00027-8

=cut
Commit	Line	Data
f2a2953c	1	package Encode::Unicode;
f2a2953c	2
df1df145	3	use strict;
f2a2953c	4	use warnings;
f2a2953c	5
151b5d36	6	our $VERSION = do { my @r = (q$Revision: 1.38 $ =~ /\d+/g); sprintf "%d."."%02d" x $#r, @r };
f2a2953c	7
85982a32	8	use XSLoader;
85982a32	9	XSLoader::load(__PACKAGE__,$VERSION);
df1df145	10
f2a2953c	11	#
	12	# Object Generator 8 transcoders all at once!
	13	#
df1df145	14
f2a2953c	15	require Encode;
10c5ecbb	16
f2a2953c	17	for my $name (qw(UTF-16 UTF-16BE UTF-16LE
	18	UTF-32 UTF-32BE UTF-32LE
	19	UCS-2BE UCS-2LE))
df1df145	20	{
f2a2953c	21	my ($size, $endian, $ucs2, $mask);
	22	$name =~ /^(\w+)-(\d+)(\w*)$/o;
	23	if ($ucs2 = ($1 eq 'UCS')){
	24	$size = 2;
	25	}else{
	26	$size = $2/8;
df1df145	27	}
f2a2953c	28	$endian = ($3 eq 'BE') ? 'n' : ($3 eq 'LE') ? 'v' : '' ;
	29	$size == 4 and $endian = uc($endian);
	30
	31	$Encode::Encoding{$name} =
	32	bless {
	33	Name => $name,
	34	size => $size,
	35	endian => $endian,
	36	ucs2 => $ucs2,
c731e18e	37	} => __PACKAGE__;
f2a2953c	38
df1df145	39	}
df1df145	40
10c5ecbb	41	use base qw(Encode::Encoding);
f2a2953c	42
f2a2953c	43	#
0ab8f81e	44	# three implementations of (en\|de)code exist. The XS version is the
	45	# fastest. _modern uses an array and _classic sticks with substr.
	46	# *_classic is much slower but more memory conservative.
	47	# *_xs is the default.
f2a2953c	48
	49	sub set_transcoder{
	50	no warnings qw(redefine);
	51	my $type = shift;
aae85ceb	52	if ($type eq "xs"){
	53	*decode = \&decode_xs;
	54	*encode = \&encode_xs;
	55	}elsif($type eq "modern"){
f2a2953c	56	*decode = \&decode_modern;
	57	*encode = \&encode_modern;
	58	}elsif($type eq "classic"){
	59	*decode = \&decode_classic;
	60	*encode = \&encode_classic;
	61	}else{
fcb875d4	62	require Carp;
4b291ae6	63	Carp::croak __PACKAGE__, "::set_transcoder(modern\|classic\|xs)";
f2a2953c	64	}
	65	}
	66
aae85ceb	67	set_transcoder("xs");
f2a2953c	68
f2a2953c	69	#
85982a32	70	# Aux. subs & constants
	71	#
	72
	73	sub FBCHAR(){ 0xFFFd }
	74	sub BOM_BE(){ 0xFeFF }
	75	sub BOM16LE(){ 0xFFFe }
	76	sub BOM32LE(){ 0xFFFe0000 }
	77
	78	sub valid_ucs2($){
	79	return
	80	(0 <= $_[0] && $_[0] < 0xD800)
	81	\|\| ( 0xDFFF < $_[0] && $_[0] <= 0xFFFF);
	82	}
	83
	84	sub issurrogate($){ 0xD800 <= $_[0] && $_[0] <= 0xDFFF }
	85	sub isHiSurrogate($){ 0xD800 <= $_[0] && $_[0] < 0xDC00 }
	86	sub isLoSurrogate($){ 0xDC00 <= $_[0] && $_[0] <= 0xDFFF }
	87
	88	sub ensurrogate($){
	89	use integer; # we have divisions
	90	my $uni = shift;
	91	my $hi = ($uni - 0x10000) / 0x400 + 0xD800;
	92	my $lo = ($uni - 0x10000) % 0x400 + 0xDC00;
	93	return ($hi, $lo);
	94	}
	95
	96	sub desurrogate($$){
	97	my ($hi, $lo) = @_;
	98	return 0x10000 + ($hi - 0xD800)*0x400 + ($lo - 0xDC00);
	99	}
	100
	101	sub Mask { {2 => 0xffff, 4 => 0xffffffff} }
	102
	103	#
f2a2953c	104	# *_modern are much faster but guzzle more memory
	105	#
	106
aae85ceb	107	sub decode_modern($$;$)
df1df145	108	{
f2a2953c	109	my ($obj, $str, $chk ) = @_;
	110	my ($size, $endian, $ucs2) = @$obj{qw(size endian ucs2)};
	111
	112	# warn "$size, $endian, $ucs2";
	113	$endian \|\|= BOMB($size, substr($str, 0, $size, ''))
	114	or poisoned2death($obj, "Where's the BOM?");
	115	my $mask = Mask->{$size};
	116	my $utf8 = '';
	117	my @ord = unpack("$endian*", $str);
	118	undef $str; # to conserve memory
	119	while (@ord){
	120	my $ord = shift @ord;
	121	unless ($size == 4 or valid_ucs2($ord &= $mask)){
	122	if ($ucs2){
fcb875d4	123	$chk and
f2a2953c	124	poisoned2death($obj, "no surrogates allowed", $ord);
	125	shift @ord; # skip the next one as well
	126	$ord = FBCHAR;
	127	}else{
	128	unless (isHiSurrogate($ord)){
	129	poisoned2death($obj, "Malformed HI surrogate", $ord);
	130	}
	131	my $lo = shift @ord;
	132	unless (isLoSurrogate($lo &= $mask)){
	133	poisoned2death($obj, "Malformed LO surrogate", $ord, $lo);
	134	}
	135	$ord = desurrogate($ord, $lo);
	136	}
	137	}
	138	$utf8 .= chr($ord);
df1df145	139	}
f2a2953c	140	utf8::upgrade($utf8);
	141	return $utf8;
	142	}
	143
aae85ceb	144	sub encode_modern($$;$)
f2a2953c	145	{
	146	my ($obj, $utf8, $chk) = @_;
	147	my ($size, $endian, $ucs2) = @$obj{qw(size endian ucs2)};
	148	my @str = ();
	149	unless ($endian){
	150	$endian = ($size == 4) ? 'N' : 'n';
	151	push @str, BOM_BE;
	152	}
	153	my @ord = unpack("U*", $utf8);
	154	undef $utf8; # to conserve memory
	155	for my $ord (@ord){
	156	unless ($size == 4 or valid_ucs2($ord)) {
	157	unless(issurrogate($ord)){
	158	if ($ucs2){
fcb875d4	159	$chk and
f2a2953c	160	poisoned2death($obj, "code point too high", $ord);
	161
	162	push @str, FBCHAR;
	163	}else{
fcb875d4	164
f2a2953c	165	push @str, ensurrogate($ord);
	166	}
	167	}else{ # not supposed to happen
	168	push @str, FBCHAR;
	169	}
	170	}else{
	171	push @str, $ord;
	172	}
	173	}
	174	return pack("$endian*", @str);
	175	}
	176
	177	#
	178	# *_classic are slower but more memory conservative
	179	#
	180
aae85ceb	181	sub decode_classic($$;$)
f2a2953c	182	{
	183	my ($obj, $str, $chk ) = @_;
	184	my ($size, $endian, $ucs2) = @$obj{qw(size endian ucs2)};
	185
	186	# warn "$size, $endian, $ucs2";
	187	$endian \|\|= BOMB($size, substr($str, 0, $size, ''))
	188	or poisoned2death($obj, "Where's the BOM?");
	189	my $mask = Mask->{$size};
	190	my $utf8 = '';
	191	my @ord = unpack("$endian*", $str);
	192	while (length($str)){
	193	my $ord = unpack($endian, substr($str, 0, $size, ''));
	194	unless ($size == 4 or valid_ucs2($ord &= $mask)){
	195	if ($ucs2){
fcb875d4	196	$chk and
f2a2953c	197	poisoned2death($obj, "no surrogates allowed", $ord);
	198	substr($str,0,$size,''); # skip the next one as well
	199	$ord = FBCHAR;
	200	}else{
	201	unless (isHiSurrogate($ord)){
	202	poisoned2death($obj, "Malformed HI surrogate", $ord);
	203	}
	204	my $lo = unpack($endian ,substr($str,0,$size,''));
	205	unless (isLoSurrogate($lo &= $mask)){
	206	poisoned2death($obj, "Malformed LO surrogate", $ord, $lo);
	207	}
	208	$ord = desurrogate($ord, $lo);
	209	}
	210	}
	211	$utf8 .= chr($ord);
	212	}
	213	utf8::upgrade($utf8);
	214	return $utf8;
	215	}
	216
aae85ceb	217	sub encode_classic($$;$)
f2a2953c	218	{
	219	my ($obj, $utf8, $chk) = @_;
	220	my ($size, $endian, $ucs2) = @$obj{qw(size endian ucs2)};
	221	# warn join ", ", $size, $ucs2, $endian, $mask;
	222	my $str = '';
	223	unless ($endian){
	224	$endian = ($size == 4) ? 'N' : 'n';
	225	$str .= pack($endian, BOM_BE);
	226	}
	227	while (length($utf8)){
	228	my $ord = ord(substr($utf8,0,1,''));
	229	unless ($size == 4 or valid_ucs2($ord)) {
	230	unless(issurrogate($ord)){
	231	if ($ucs2){
fcb875d4	232	$chk and
f2a2953c	233	poisoned2death($obj, "code point too high", $ord);
	234	$str .= pack($endian, FBCHAR);
	235	}else{
	236	$str .= pack($endian.2, ensurrogate($ord));
	237	}
	238	}else{ # not supposed to happen
	239	$str .= pack($endian, FBCHAR);
	240	}
	241	}else{
	242	$str .= pack($endian, $ord);
	243	}
	244	}
	245	return $str;
	246	}
	247
	248	sub BOMB {
	249	my ($size, $bom) = @_;
	250	my $N = $size == 2 ? 'n' : 'N';
	251	my $ord = unpack($N, $bom);
fcb875d4	252	return ($ord eq BOM_BE) ? $N :
f2a2953c	253	($ord eq BOM16LE) ? 'v' : ($ord eq BOM32LE) ? 'V' : undef;
	254	}
	255
	256	sub poisoned2death{
	257	my $obj = shift;
	258	my $msg = shift;
	259	my $pair = join(", ", map {sprintf "\\x%x", $_} @_);
	260	require Carp;
4b291ae6	261	Carp::croak $obj->name, ":", $msg, "<$pair>.", caller;
df1df145	262	}
	263
	264	1;
	265	__END__
67d7b5ef	266
	267	=head1 NAME
	268
0ab8f81e	269	Encode::Unicode -- Various Unicode Transformation Formats
67d7b5ef	270
67d7b5ef	271	=cut
f2a2953c	272
	273	=head1 SYNOPSIS
	274
fcb875d4	275	use Encode qw/encode decode/;
f2a2953c	276	$ucs2 = encode("UCS-2BE", $utf8);
	277	$utf8 = decode("UCS-2BE", $ucs2);
	278
	279	=head1 ABSTRACT
	280
	281	This module implements all Character Encoding Schemes of Unicode that
	282	are officially documented by Unicode Consortium (except, of course,
	283	for UTF-8, which is a native format in perl).
	284
	285	=over 4
	286
	287	=item L<http://www.unicode.org/glossary/> says:
	288
	289	I<Character Encoding Scheme> A character encoding form plus byte
	290	serialization. There are seven character encoding schemes in Unicode:
11067275	291	UTF-8, UTF-16, UTF-16BE, UTF-16LE, UTF-32 (UCS-4), UTF-32BE (UCS-4BE) and
11067275	292	UTF-32LE (UCS-4LE).
f2a2953c	293
	294	=item Quick Reference
	295
	296	Decodes from ord(N) Encodes chr(N) to...
	297	octet/char BOM S.P d800-dfff ord > 0xffff \x{1abcd} ==
	298	---------------+-----------------+------------------------------
	299	UCS-2BE 2 N N is bogus Not Available
	300	UCS-2LE 2 N N bogus Not Available
	301	UTF-16 2/4 Y Y is S.P S.P BE/LE
	302	UTF-16BE 2/4 N Y S.P S.P 0xd82a,0xdfcd
	303	UTF-16LE 2 N Y S.P S.P 0x2ad8,0xcddf
	304	UTF-32 4 Y - is bogus As is BE/LE
0ab8f81e	305	UTF-32BE 4 N - bogus As is 0x0001abcd
0ab8f81e	306	UTF-32LE 4 N - bogus As is 0xcdab0100
f2a2953c	307	UTF-8 1-4 - - bogus >= 4 octets \xf0\x9a\af\8d
	308	---------------+-----------------+------------------------------
	309
	310	=back
	311
	312	=head1 Size, Endianness, and BOM
	313
0ab8f81e	314	You can categorize these CES by 3 criteria: size of each character,
0ab8f81e	315	endianness, and Byte Order Mark.
f2a2953c	316
0ab8f81e	317	=head2 by size
f2a2953c	318
f2a2953c	319	UCS-2 is a fixed-length encoding with each character taking 16 bits.
0ab8f81e	320	It B<does not> support I<surrogate pairs>. When a surrogate pair
	321	is encountered during decode(), its place is filled with \x{FFFD}
	322	if I<CHECK> is 0, or the routine croaks if I<CHECK> is 1. When a
	323	character whose ord value is larger than 0xFFFF is encountered,
	324	its place is filled with \x{FFFD} if I<CHECK> is 0, or the routine
	325	croaks if I<CHECK> is 1.
	326
	327	UTF-16 is almost the same as UCS-2 but it supports I<surrogate pairs>.
f2a2953c	328	When it encounters a high surrogate (0xD800-0xDBFF), it fetches the
0ab8f81e	329	following low surrogate (0xDC00-0xDFFF) and C<desurrogate>s them to
	330	form a character. Bogus surrogates result in death. When \x{10000}
	331	or above is encountered during encode(), it C<ensurrogate>s them and
	332	pushes the surrogate pair to the output stream.
f2a2953c	333
11067275	334	UTF-32 (UCS-4) is a fixed-length encoding with each character taking 32 bits.
0ab8f81e	335	Since it is 32-bit, there is no need for I<surrogate pairs>.
f2a2953c	336
0ab8f81e	337	=head2 by endianness
f2a2953c	338
0ab8f81e	339	The first (and now failed) goal of Unicode was to map all character
	340	repertoires into a fixed-length integer so that programmers are happy.
	341	Since each character is either a I<short> or I<long> in C, you have to
	342	pay attention to the endianness of each platform when you pass data
	343	to one another.
f2a2953c	344
f2a2953c	345	Anything marked as BE is Big Endian (or network byte order) and LE is
0ab8f81e	346	Little Endian (aka VAX byte order). For anything not marked either
	347	BE or LE, a character called Byte Order Mark (BOM) indicating the
	348	endianness is prepended to the string.
f2a2953c	349
	350	=over 4
	351
fcb875d4	352	=item BOM as integer when fetched in network byte order
f2a2953c	353
fcb875d4	354	16 32 bits/char
	355	-------------------------
	356	BE 0xFeFF 0x0000FeFF
	357	LE 0xFFeF 0xFFFe0000
	358	-------------------------
f2a2953c	359
f2a2953c	360	=back
151b5d36	361
0ab8f81e	362	This modules handles the BOM as follows.
f2a2953c	363
	364	=over 4
	365
	366	=item *
	367
	368	When BE or LE is explicitly stated as the name of encoding, BOM is
0ab8f81e	369	simply treated as a normal character (ZERO WIDTH NO-BREAK SPACE).
f2a2953c	370
	371	=item *
	372
0ab8f81e	373	When BE or LE is omitted during decode(), it checks if BOM is at the
	374	beginning of the string; if one is found, the endianness is set to
	375	what the BOM says. If no BOM is found, the routine dies.
f2a2953c	376
	377	=item *
	378
	379	When BE or LE is omitted during encode(), it returns a BE-encoded
	380	string with BOM prepended. So when you want to encode a whole text
0ab8f81e	381	file, make sure you encode() the whole text at once, not line by line
0ab8f81e	382	or each line, not file, will have a BOM prepended.
f2a2953c	383
	384	=item *
	385
0ab8f81e	386	C<UCS-2> is an exception. Unlike others, this is an alias of UCS-2BE.
f2a2953c	387	UCS-2 is already registered by IANA and others that way.
f2a2953c	388
fdd579e2	389	=back
f2a2953c	390
fcb875d4	391	=head1 Surrogate Pairs
f2a2953c	392
fcb875d4	393	To say the least, surrogate pairs were the biggest mistake of the
	394	Unicode Consortium. But according to the late Douglas Adams in I<The
	395	Hitchhiker's Guide to the Galaxy> Trilogy, C<In the beginning the
	396	Universe was created. This has made a lot of people very angry and
	397	been widely regarded as a bad move>. Their mistake was not of this
	398	magnitude so let's forgive them.
f2a2953c	399
f2a2953c	400	(I don't dare make any comparison with Unicode Consortium and the
c731e18e	401	Vogons here ;) Or, comparing Encode to Babel Fish is completely
c731e18e	402	appropriate -- if you can only stick this into your ear :)
f2a2953c	403
0ab8f81e	404	Surrogate pairs were born when the Unicode Consortium finally
fcb875d4	405	admitted that 16 bits were not big enough to hold all the world's
0ab8f81e	406	character repertoires. But they already made UCS-2 16-bit. What
f2a2953c	407	do we do?
f2a2953c	408
0ab8f81e	409	Back then, the range 0xD800-0xDFFF was not allocated. Let's split
	410	that range in half and use the first half to represent the C<upper
	411	half of a character> and the second half to represent the C<lower
	412	half of a character>. That way, you can represent 1024 * 1024 =
	413	1048576 more characters. Now we can store character ranges up to
	414	\x{10ffff} even with 16-bit encodings. This pair of half-character is
	415	now called a I<surrogate pair> and UTF-16 is the name of the encoding
	416	that embraces them.
f2a2953c	417
448e90bb	418	Here is a formula to ensurrogate a Unicode character \x{10000} and
f2a2953c	419	above;
	420
	421	$hi = ($uni - 0x10000) / 0x400 + 0xD800;
	422	$lo = ($uni - 0x10000) % 0x400 + 0xDC00;
	423
	424	And to desurrogate;
	425
	426	$uni = 0x10000 + ($hi - 0xD800) * 0x400 + ($lo - 0xDC00);
	427
fcb875d4	428	Note this move has made \x{D800}-\x{DFFF} into a forbidden zone but
	429	perl does not prohibit the use of characters within this range. To perl,
	430	every one of \x{0000_0000} up to \x{ffff_ffff} (*) is I<a character>.
	431
	432	(*) or \x{ffff_ffff_ffff_ffff} if your perl is compiled with 64-bit
0ab8f81e	433	integer support!
f2a2953c	434
	435	=head1 SEE ALSO
	436
fdd579e2	437	L<Encode>, L<http://www.unicode.org/glossary/>,
11067275	438	L<http://www.unicode.org/unicode/faq/utf_bom.html>,
f2a2953c	439
fdd579e2	440	RFC 2781 L<http://rfc.net/rfc2781.html>,
fdd579e2	441
11067275	442	The whole Unicode standard L<http://www.unicode.org/unicode/uni2book/u2.html>
fdd579e2	443
fcb875d4	444	Ch. 15, pp. 403 of C<Programming Perl (3rd Edition)>
	445	by Larry Wall, Tom Christiansen, Jon Orwant;
	446	O'Reilly & Associates; ISBN 0-596-00027-8
	447
fdd579e2	448	=cut