Your IP : 3.144.89.197
Current Path : /usr/share/perl5/URI/ |
|
Current File : //usr/share/perl5/URI/_punycode.pm |
package URI::_punycode;
use strict;
use warnings;
our $VERSION = '1.73';
$VERSION = eval $VERSION;
use Exporter 'import';
our @EXPORT = qw(encode_punycode decode_punycode);
use integer;
our $DEBUG = 0;
use constant BASE => 36;
use constant TMIN => 1;
use constant TMAX => 26;
use constant SKEW => 38;
use constant DAMP => 700;
use constant INITIAL_BIAS => 72;
use constant INITIAL_N => 128;
my $Delimiter = chr 0x2D;
my $BasicRE = qr/[\x00-\x7f]/;
sub _croak { require Carp; Carp::croak(@_); }
sub digit_value {
my $code = shift;
return ord($code) - ord("A") if $code =~ /[A-Z]/;
return ord($code) - ord("a") if $code =~ /[a-z]/;
return ord($code) - ord("0") + 26 if $code =~ /[0-9]/;
return;
}
sub code_point {
my $digit = shift;
return $digit + ord('a') if 0 <= $digit && $digit <= 25;
return $digit + ord('0') - 26 if 26 <= $digit && $digit <= 36;
die 'NOT COME HERE';
}
sub adapt {
my($delta, $numpoints, $firsttime) = @_;
$delta = $firsttime ? $delta / DAMP : $delta / 2;
$delta += $delta / $numpoints;
my $k = 0;
while ($delta > ((BASE - TMIN) * TMAX) / 2) {
$delta /= BASE - TMIN;
$k += BASE;
}
return $k + (((BASE - TMIN + 1) * $delta) / ($delta + SKEW));
}
sub decode_punycode {
my $code = shift;
my $n = INITIAL_N;
my $i = 0;
my $bias = INITIAL_BIAS;
my @output;
if ($code =~ s/(.*)$Delimiter//o) {
push @output, map ord, split //, $1;
return _croak('non-basic code point') unless $1 =~ /^$BasicRE*$/o;
}
while ($code) {
my $oldi = $i;
my $w = 1;
LOOP:
for (my $k = BASE; 1; $k += BASE) {
my $cp = substr($code, 0, 1, '');
my $digit = digit_value($cp);
defined $digit or return _croak("invalid punycode input");
$i += $digit * $w;
my $t = ($k <= $bias) ? TMIN
: ($k >= $bias + TMAX) ? TMAX : $k - $bias;
last LOOP if $digit < $t;
$w *= (BASE - $t);
}
$bias = adapt($i - $oldi, @output + 1, $oldi == 0);
warn "bias becomes $bias" if $DEBUG;
$n += $i / (@output + 1);
$i = $i % (@output + 1);
splice(@output, $i, 0, $n);
warn join " ", map sprintf('%04x', $_), @output if $DEBUG;
$i++;
}
return join '', map chr, @output;
}
sub encode_punycode {
my $input = shift;
my @input = split //, $input;
my $n = INITIAL_N;
my $delta = 0;
my $bias = INITIAL_BIAS;
my @output;
my @basic = grep /$BasicRE/, @input;
my $h = my $b = @basic;
push @output, @basic;
push @output, $Delimiter if $b && $h < @input;
warn "basic codepoints: (@output)" if $DEBUG;
while ($h < @input) {
my $m = min(grep { $_ >= $n } map ord, @input);
warn sprintf "next code point to insert is %04x", $m if $DEBUG;
$delta += ($m - $n) * ($h + 1);
$n = $m;
for my $i (@input) {
my $c = ord($i);
$delta++ if $c < $n;
if ($c == $n) {
my $q = $delta;
LOOP:
for (my $k = BASE; 1; $k += BASE) {
my $t = ($k <= $bias) ? TMIN :
($k >= $bias + TMAX) ? TMAX : $k - $bias;
last LOOP if $q < $t;
my $cp = code_point($t + (($q - $t) % (BASE - $t)));
push @output, chr($cp);
$q = ($q - $t) / (BASE - $t);
}
push @output, chr(code_point($q));
$bias = adapt($delta, $h + 1, $h == $b);
warn "bias becomes $bias" if $DEBUG;
$delta = 0;
$h++;
}
}
$delta++;
$n++;
}
return join '', @output;
}
sub min {
my $min = shift;
for (@_) { $min = $_ if $_ <= $min }
return $min;
}
1;
__END__
=encoding utf8
=head1 NAME
URI::_punycode - encodes Unicode string in Punycode
=head1 SYNOPSIS
use strict;
use warnings;
use utf8;
use URI::_punycode qw(encode_punycode decode_punycode);
# encode a unicode string
my $punycode = encode_punycode('http://☃.net'); # http://.net-xc8g
$punycode = encode_punycode('bücher'); # bcher-kva
$punycode = encode_punycode('他们为什么不说中文'); # ihqwcrb4cv8a8dqg056pqjye
# decode a punycode string back into a unicode string
my $unicode = decode_punycode('http://.net-xc8g'); # http://☃.net
$unicode = decode_punycode('bcher-kva'); # bücher
$unicode = decode_punycode('ihqwcrb4cv8a8dqg056pqjye'); # 他们为什么不说中文
=head1 DESCRIPTION
L<URI::_punycode> is a module to encode / decode Unicode strings into
L<Punycode|https://tools.ietf.org/html/rfc3492>, an efficient
encoding of Unicode for use with L<IDNA|https://tools.ietf.org/html/rfc5890>.
=head1 FUNCTIONS
All functions throw exceptions on failure. You can C<catch> them with
L<Syntax::Keyword::Try> or L<Try::Tiny>. The following functions are exported
by default.
=head2 encode_punycode
my $punycode = encode_punycode('http://☃.net'); # http://.net-xc8g
$punycode = encode_punycode('bücher'); # bcher-kva
$punycode = encode_punycode('他们为什么不说中文') # ihqwcrb4cv8a8dqg056pqjye
Takes a Unicode string (UTF8-flagged variable) and returns a Punycode
encoding for it.
=head2 decode_punycode
my $unicode = decode_punycode('http://.net-xc8g'); # http://☃.net
$unicode = decode_punycode('bcher-kva'); # bücher
$unicode = decode_punycode('ihqwcrb4cv8a8dqg056pqjye'); # 他们为什么不说中文
Takes a Punycode encoding and returns original Unicode string.
=head1 AUTHOR
Tatsuhiko Miyagawa <F<miyagawa@bulknews.net>> is the author of
L<IDNA::Punycode> which was the basis for this module.
=head1 SEE ALSO
L<IDNA::Punycode>, L<RFC 3492|https://tools.ietf.org/html/rfc3492>,
L<RFC 5891|https://tools.ietf.org/html/rfc5891>
=head1 COPYRIGHT AND LICENSE
This library is free software; you can redistribute it and/or modify
it under the same terms as Perl itself.
=cut