[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: 1.35 $ =~ /\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__;

}

sub name { shift->{'Name'} }
sub new_sequence
{
    my $self = shift;
    # Return the original if endian known
    return $self if ($self->{endian});
    # Return a clone
    return bless {%$self},ref($self);
}

sub needs_lines { 0 };

sub perlio_ok { 
    eval{ require PerlIO::encoding };
    if ($@){
	return 0;
    }else{
	return 1;
    }
}


#
# 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, UTF-32BE and UTF-32LE.

=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 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/>,

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

L<http://www.unicode.org/unicode/faq/utf_bom.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
011b2d2f	6	our $VERSION = do { my @r = (q$Revision: 1.35 $ =~ /\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;
	16	for my $name (qw(UTF-16 UTF-16BE UTF-16LE
	17	UTF-32 UTF-32BE UTF-32LE
	18	UCS-2BE UCS-2LE))
df1df145	19	{
f2a2953c	20	my ($size, $endian, $ucs2, $mask);
	21	$name =~ /^(\w+)-(\d+)(\w*)$/o;
	22	if ($ucs2 = ($1 eq 'UCS')){
	23	$size = 2;
	24	}else{
	25	$size = $2/8;
df1df145	26	}
f2a2953c	27	$endian = ($3 eq 'BE') ? 'n' : ($3 eq 'LE') ? 'v' : '' ;
	28	$size == 4 and $endian = uc($endian);
	29
	30	$Encode::Encoding{$name} =
	31	bless {
	32	Name => $name,
	33	size => $size,
	34	endian => $endian,
	35	ucs2 => $ucs2,
c731e18e	36	} => __PACKAGE__;
f2a2953c	37
df1df145	38	}
df1df145	39
f2a2953c	40	sub name { shift->{'Name'} }
aae85ceb	41	sub new_sequence
	42	{
	43	my $self = shift;
	44	# Return the original if endian known
	45	return $self if ($self->{endian});
	46	# Return a clone
	47	return bless {%$self},ref($self);
	48	}
	49
0ab8f81e	50	sub needs_lines { 0 };
	51
	52	sub perlio_ok {
011b2d2f	53	eval{ require PerlIO::encoding };
	54	if ($@){
	55	return 0;
	56	}else{
	57	return 1;
	58	}
0ab8f81e	59	}
0ab8f81e	60
f2a2953c	61
f2a2953c	62	#
0ab8f81e	63	# three implementations of (en\|de)code exist. The XS version is the
	64	# fastest. _modern uses an array and _classic sticks with substr.
	65	# *_classic is much slower but more memory conservative.
	66	# *_xs is the default.
f2a2953c	67
	68	sub set_transcoder{
	69	no warnings qw(redefine);
	70	my $type = shift;
aae85ceb	71	if ($type eq "xs"){
	72	*decode = \&decode_xs;
	73	*encode = \&encode_xs;
	74	}elsif($type eq "modern"){
f2a2953c	75	*decode = \&decode_modern;
	76	*encode = \&encode_modern;
	77	}elsif($type eq "classic"){
	78	*decode = \&decode_classic;
	79	*encode = \&encode_classic;
	80	}else{
fcb875d4	81	require Carp;
aae85ceb	82	Carp::croak __PACKAGE__, "::set_transcoder(modern\|classic\|xs)";
f2a2953c	83	}
	84	}
	85
aae85ceb	86	set_transcoder("xs");
f2a2953c	87
f2a2953c	88	#
85982a32	89	# Aux. subs & constants
	90	#
	91
	92	sub FBCHAR(){ 0xFFFd }
	93	sub BOM_BE(){ 0xFeFF }
	94	sub BOM16LE(){ 0xFFFe }
	95	sub BOM32LE(){ 0xFFFe0000 }
	96
	97	sub valid_ucs2($){
	98	return
	99	(0 <= $_[0] && $_[0] < 0xD800)
	100	\|\| ( 0xDFFF < $_[0] && $_[0] <= 0xFFFF);
	101	}
	102
	103	sub issurrogate($){ 0xD800 <= $_[0] && $_[0] <= 0xDFFF }
	104	sub isHiSurrogate($){ 0xD800 <= $_[0] && $_[0] < 0xDC00 }
	105	sub isLoSurrogate($){ 0xDC00 <= $_[0] && $_[0] <= 0xDFFF }
	106
	107	sub ensurrogate($){
	108	use integer; # we have divisions
	109	my $uni = shift;
	110	my $hi = ($uni - 0x10000) / 0x400 + 0xD800;
	111	my $lo = ($uni - 0x10000) % 0x400 + 0xDC00;
	112	return ($hi, $lo);
	113	}
	114
	115	sub desurrogate($$){
	116	my ($hi, $lo) = @_;
	117	return 0x10000 + ($hi - 0xD800)*0x400 + ($lo - 0xDC00);
	118	}
	119
	120	sub Mask { {2 => 0xffff, 4 => 0xffffffff} }
	121
	122	#
f2a2953c	123	# *_modern are much faster but guzzle more memory
	124	#
	125
aae85ceb	126	sub decode_modern($$;$)
df1df145	127	{
f2a2953c	128	my ($obj, $str, $chk ) = @_;
	129	my ($size, $endian, $ucs2) = @$obj{qw(size endian ucs2)};
	130
	131	# warn "$size, $endian, $ucs2";
	132	$endian \|\|= BOMB($size, substr($str, 0, $size, ''))
	133	or poisoned2death($obj, "Where's the BOM?");
	134	my $mask = Mask->{$size};
	135	my $utf8 = '';
	136	my @ord = unpack("$endian*", $str);
	137	undef $str; # to conserve memory
	138	while (@ord){
	139	my $ord = shift @ord;
	140	unless ($size == 4 or valid_ucs2($ord &= $mask)){
	141	if ($ucs2){
fcb875d4	142	$chk and
f2a2953c	143	poisoned2death($obj, "no surrogates allowed", $ord);
	144	shift @ord; # skip the next one as well
	145	$ord = FBCHAR;
	146	}else{
	147	unless (isHiSurrogate($ord)){
	148	poisoned2death($obj, "Malformed HI surrogate", $ord);
	149	}
	150	my $lo = shift @ord;
	151	unless (isLoSurrogate($lo &= $mask)){
	152	poisoned2death($obj, "Malformed LO surrogate", $ord, $lo);
	153	}
	154	$ord = desurrogate($ord, $lo);
	155	}
	156	}
	157	$utf8 .= chr($ord);
df1df145	158	}
f2a2953c	159	utf8::upgrade($utf8);
	160	return $utf8;
	161	}
	162
aae85ceb	163	sub encode_modern($$;$)
f2a2953c	164	{
	165	my ($obj, $utf8, $chk) = @_;
	166	my ($size, $endian, $ucs2) = @$obj{qw(size endian ucs2)};
	167	my @str = ();
	168	unless ($endian){
	169	$endian = ($size == 4) ? 'N' : 'n';
	170	push @str, BOM_BE;
	171	}
	172	my @ord = unpack("U*", $utf8);
	173	undef $utf8; # to conserve memory
	174	for my $ord (@ord){
	175	unless ($size == 4 or valid_ucs2($ord)) {
	176	unless(issurrogate($ord)){
	177	if ($ucs2){
fcb875d4	178	$chk and
f2a2953c	179	poisoned2death($obj, "code point too high", $ord);
	180
	181	push @str, FBCHAR;
	182	}else{
fcb875d4	183
f2a2953c	184	push @str, ensurrogate($ord);
	185	}
	186	}else{ # not supposed to happen
	187	push @str, FBCHAR;
	188	}
	189	}else{
	190	push @str, $ord;
	191	}
	192	}
	193	return pack("$endian*", @str);
	194	}
	195
	196	#
	197	# *_classic are slower but more memory conservative
	198	#
	199
aae85ceb	200	sub decode_classic($$;$)
f2a2953c	201	{
	202	my ($obj, $str, $chk ) = @_;
	203	my ($size, $endian, $ucs2) = @$obj{qw(size endian ucs2)};
	204
	205	# warn "$size, $endian, $ucs2";
	206	$endian \|\|= BOMB($size, substr($str, 0, $size, ''))
	207	or poisoned2death($obj, "Where's the BOM?");
	208	my $mask = Mask->{$size};
	209	my $utf8 = '';
	210	my @ord = unpack("$endian*", $str);
	211	while (length($str)){
	212	my $ord = unpack($endian, substr($str, 0, $size, ''));
	213	unless ($size == 4 or valid_ucs2($ord &= $mask)){
	214	if ($ucs2){
fcb875d4	215	$chk and
f2a2953c	216	poisoned2death($obj, "no surrogates allowed", $ord);
	217	substr($str,0,$size,''); # skip the next one as well
	218	$ord = FBCHAR;
	219	}else{
	220	unless (isHiSurrogate($ord)){
	221	poisoned2death($obj, "Malformed HI surrogate", $ord);
	222	}
	223	my $lo = unpack($endian ,substr($str,0,$size,''));
	224	unless (isLoSurrogate($lo &= $mask)){
	225	poisoned2death($obj, "Malformed LO surrogate", $ord, $lo);
	226	}
	227	$ord = desurrogate($ord, $lo);
	228	}
	229	}
	230	$utf8 .= chr($ord);
	231	}
	232	utf8::upgrade($utf8);
	233	return $utf8;
	234	}
	235
aae85ceb	236	sub encode_classic($$;$)
f2a2953c	237	{
	238	my ($obj, $utf8, $chk) = @_;
	239	my ($size, $endian, $ucs2) = @$obj{qw(size endian ucs2)};
	240	# warn join ", ", $size, $ucs2, $endian, $mask;
	241	my $str = '';
	242	unless ($endian){
	243	$endian = ($size == 4) ? 'N' : 'n';
	244	$str .= pack($endian, BOM_BE);
	245	}
	246	while (length($utf8)){
	247	my $ord = ord(substr($utf8,0,1,''));
	248	unless ($size == 4 or valid_ucs2($ord)) {
	249	unless(issurrogate($ord)){
	250	if ($ucs2){
fcb875d4	251	$chk and
f2a2953c	252	poisoned2death($obj, "code point too high", $ord);
	253	$str .= pack($endian, FBCHAR);
	254	}else{
	255	$str .= pack($endian.2, ensurrogate($ord));
	256	}
	257	}else{ # not supposed to happen
	258	$str .= pack($endian, FBCHAR);
	259	}
	260	}else{
	261	$str .= pack($endian, $ord);
	262	}
	263	}
	264	return $str;
	265	}
	266
	267	sub BOMB {
	268	my ($size, $bom) = @_;
	269	my $N = $size == 2 ? 'n' : 'N';
	270	my $ord = unpack($N, $bom);
fcb875d4	271	return ($ord eq BOM_BE) ? $N :
f2a2953c	272	($ord eq BOM16LE) ? 'v' : ($ord eq BOM32LE) ? 'V' : undef;
	273	}
	274
	275	sub poisoned2death{
	276	my $obj = shift;
	277	my $msg = shift;
	278	my $pair = join(", ", map {sprintf "\\x%x", $_} @_);
	279	require Carp;
	280	Carp::croak $obj->name, ":", $msg, "<$pair>.", caller;
df1df145	281	}
	282
	283	1;
	284	__END__
67d7b5ef	285
	286	=head1 NAME
	287
0ab8f81e	288	Encode::Unicode -- Various Unicode Transformation Formats
67d7b5ef	289
67d7b5ef	290	=cut
f2a2953c	291
	292	=head1 SYNOPSIS
	293
fcb875d4	294	use Encode qw/encode decode/;
f2a2953c	295	$ucs2 = encode("UCS-2BE", $utf8);
	296	$utf8 = decode("UCS-2BE", $ucs2);
	297
	298	=head1 ABSTRACT
	299
	300	This module implements all Character Encoding Schemes of Unicode that
	301	are officially documented by Unicode Consortium (except, of course,
	302	for UTF-8, which is a native format in perl).
	303
	304	=over 4
	305
	306	=item L<http://www.unicode.org/glossary/> says:
	307
	308	I<Character Encoding Scheme> A character encoding form plus byte
	309	serialization. There are seven character encoding schemes in Unicode:
	310	UTF-8, UTF-16, UTF-16BE, UTF-16LE, UTF-32, UTF-32BE and UTF-32LE.
	311
	312	=item Quick Reference
	313
	314	Decodes from ord(N) Encodes chr(N) to...
	315	octet/char BOM S.P d800-dfff ord > 0xffff \x{1abcd} ==
	316	---------------+-----------------+------------------------------
	317	UCS-2BE 2 N N is bogus Not Available
	318	UCS-2LE 2 N N bogus Not Available
	319	UTF-16 2/4 Y Y is S.P S.P BE/LE
	320	UTF-16BE 2/4 N Y S.P S.P 0xd82a,0xdfcd
	321	UTF-16LE 2 N Y S.P S.P 0x2ad8,0xcddf
	322	UTF-32 4 Y - is bogus As is BE/LE
0ab8f81e	323	UTF-32BE 4 N - bogus As is 0x0001abcd
0ab8f81e	324	UTF-32LE 4 N - bogus As is 0xcdab0100
f2a2953c	325	UTF-8 1-4 - - bogus >= 4 octets \xf0\x9a\af\8d
	326	---------------+-----------------+------------------------------
	327
	328	=back
	329
	330	=head1 Size, Endianness, and BOM
	331
0ab8f81e	332	You can categorize these CES by 3 criteria: size of each character,
0ab8f81e	333	endianness, and Byte Order Mark.
f2a2953c	334
0ab8f81e	335	=head2 by size
f2a2953c	336
f2a2953c	337	UCS-2 is a fixed-length encoding with each character taking 16 bits.
0ab8f81e	338	It B<does not> support I<surrogate pairs>. When a surrogate pair
	339	is encountered during decode(), its place is filled with \x{FFFD}
	340	if I<CHECK> is 0, or the routine croaks if I<CHECK> is 1. When a
	341	character whose ord value is larger than 0xFFFF is encountered,
	342	its place is filled with \x{FFFD} if I<CHECK> is 0, or the routine
	343	croaks if I<CHECK> is 1.
	344
	345	UTF-16 is almost the same as UCS-2 but it supports I<surrogate pairs>.
f2a2953c	346	When it encounters a high surrogate (0xD800-0xDBFF), it fetches the
0ab8f81e	347	following low surrogate (0xDC00-0xDFFF) and C<desurrogate>s them to
	348	form a character. Bogus surrogates result in death. When \x{10000}
	349	or above is encountered during encode(), it C<ensurrogate>s them and
	350	pushes the surrogate pair to the output stream.
f2a2953c	351
f2a2953c	352	UTF-32 is a fixed-length encoding with each character taking 32 bits.
0ab8f81e	353	Since it is 32-bit, there is no need for I<surrogate pairs>.
f2a2953c	354
0ab8f81e	355	=head2 by endianness
f2a2953c	356
0ab8f81e	357	The first (and now failed) goal of Unicode was to map all character
	358	repertoires into a fixed-length integer so that programmers are happy.
	359	Since each character is either a I<short> or I<long> in C, you have to
	360	pay attention to the endianness of each platform when you pass data
	361	to one another.
f2a2953c	362
f2a2953c	363	Anything marked as BE is Big Endian (or network byte order) and LE is
0ab8f81e	364	Little Endian (aka VAX byte order). For anything not marked either
	365	BE or LE, a character called Byte Order Mark (BOM) indicating the
	366	endianness is prepended to the string.
f2a2953c	367
	368	=over 4
	369
fcb875d4	370	=item BOM as integer when fetched in network byte order
f2a2953c	371
fcb875d4	372	16 32 bits/char
	373	-------------------------
	374	BE 0xFeFF 0x0000FeFF
	375	LE 0xFFeF 0xFFFe0000
	376	-------------------------
f2a2953c	377
f2a2953c	378	=back
fcb875d4	379
0ab8f81e	380	This modules handles the BOM as follows.
f2a2953c	381
	382	=over 4
	383
	384	=item *
	385
	386	When BE or LE is explicitly stated as the name of encoding, BOM is
0ab8f81e	387	simply treated as a normal character (ZERO WIDTH NO-BREAK SPACE).
f2a2953c	388
	389	=item *
	390
0ab8f81e	391	When BE or LE is omitted during decode(), it checks if BOM is at the
	392	beginning of the string; if one is found, the endianness is set to
	393	what the BOM says. If no BOM is found, the routine dies.
f2a2953c	394
	395	=item *
	396
	397	When BE or LE is omitted during encode(), it returns a BE-encoded
	398	string with BOM prepended. So when you want to encode a whole text
0ab8f81e	399	file, make sure you encode() the whole text at once, not line by line
0ab8f81e	400	or each line, not file, will have a BOM prepended.
f2a2953c	401
	402	=item *
	403
0ab8f81e	404	C<UCS-2> is an exception. Unlike others, this is an alias of UCS-2BE.
f2a2953c	405	UCS-2 is already registered by IANA and others that way.
f2a2953c	406
fdd579e2	407	=back
f2a2953c	408
fcb875d4	409	=head1 Surrogate Pairs
f2a2953c	410
fcb875d4	411	To say the least, surrogate pairs were the biggest mistake of the
	412	Unicode Consortium. But according to the late Douglas Adams in I<The
	413	Hitchhiker's Guide to the Galaxy> Trilogy, C<In the beginning the
	414	Universe was created. This has made a lot of people very angry and
	415	been widely regarded as a bad move>. Their mistake was not of this
	416	magnitude so let's forgive them.
f2a2953c	417
f2a2953c	418	(I don't dare make any comparison with Unicode Consortium and the
c731e18e	419	Vogons here ;) Or, comparing Encode to Babel Fish is completely
c731e18e	420	appropriate -- if you can only stick this into your ear :)
f2a2953c	421
0ab8f81e	422	Surrogate pairs were born when the Unicode Consortium finally
fcb875d4	423	admitted that 16 bits were not big enough to hold all the world's
0ab8f81e	424	character repertoires. But they already made UCS-2 16-bit. What
f2a2953c	425	do we do?
f2a2953c	426
0ab8f81e	427	Back then, the range 0xD800-0xDFFF was not allocated. Let's split
	428	that range in half and use the first half to represent the C<upper
	429	half of a character> and the second half to represent the C<lower
	430	half of a character>. That way, you can represent 1024 * 1024 =
	431	1048576 more characters. Now we can store character ranges up to
	432	\x{10ffff} even with 16-bit encodings. This pair of half-character is
	433	now called a I<surrogate pair> and UTF-16 is the name of the encoding
	434	that embraces them.
f2a2953c	435
448e90bb	436	Here is a formula to ensurrogate a Unicode character \x{10000} and
f2a2953c	437	above;
	438
	439	$hi = ($uni - 0x10000) / 0x400 + 0xD800;
	440	$lo = ($uni - 0x10000) % 0x400 + 0xDC00;
	441
	442	And to desurrogate;
	443
	444	$uni = 0x10000 + ($hi - 0xD800) * 0x400 + ($lo - 0xDC00);
	445
fcb875d4	446	Note this move has made \x{D800}-\x{DFFF} into a forbidden zone but
	447	perl does not prohibit the use of characters within this range. To perl,
	448	every one of \x{0000_0000} up to \x{ffff_ffff} (*) is I<a character>.
	449
	450	(*) or \x{ffff_ffff_ffff_ffff} if your perl is compiled with 64-bit
0ab8f81e	451	integer support!
f2a2953c	452
	453	=head1 SEE ALSO
	454
fdd579e2	455	L<Encode>, L<http://www.unicode.org/glossary/>,
f2a2953c	456
fdd579e2	457	RFC 2781 L<http://rfc.net/rfc2781.html>,
	458
	459	L<http://www.unicode.org/unicode/faq/utf_bom.html>
	460
fcb875d4	461	Ch. 15, pp. 403 of C<Programming Perl (3rd Edition)>
	462	by Larry Wall, Tom Christiansen, Jon Orwant;
	463	O'Reilly & Associates; ISBN 0-596-00027-8
	464
fdd579e2	465	=cut