package Encoding::FixLatin;
  $Encoding::FixLatin::VERSION = '1.04';

use warnings;
use strict;

require 5.008;

use Carp     qw(croak);
use Exporter qw(import);
use Encode   qw(is_utf8 encode_utf8);

our @EXPORT_OK = qw(fix_latin);

my $xs_loaded  = undef; # no attempt to load yet

my $byte_map;

my $ascii_str = qr{\A([\x00-\x7F]+)(.*)\z}s;

my $cont_byte = '[\x80-\xBF]';
my $utf8_2    = qr{\A([\xC0-\xDF])($cont_byte)(.*)\z}s;
my $utf8_3    = qr{\A([\xE0-\xEF])($cont_byte)($cont_byte)(.*)\z}s;
my $utf8_4    = qr{\A([\xF0-\xF7])($cont_byte)($cont_byte)($cont_byte)(.*)\z}s;
my $utf8_5    = qr{\A([\xF8-\xFB])($cont_byte)($cont_byte)($cont_byte)($cont_byte)(.*)\z}s;

my %known_opt = map { $_ => 1 } qw(bytes_only ascii_hex overlong_fatal use_xs);

my %non_1252  = (
    "\x81" => '%81',
    "\x8D" => '%8D',
    "\x8F" => '%8F',
    "\x90" => '%90',
    "\x9D" => '%9D',

sub fix_latin {
    my $input = shift;
    my %opt   = (
        ascii_hex      => 1,
        bytes_only     => 0,
        overlong_fatal => 0,
        use_xs         => 'auto',

    foreach (keys %opt) {
        croak "Unknown option '$_'" unless $known_opt{$_};

    return unless defined($input);
    _init_byte_map(\%opt) unless $byte_map;

    if(is_utf8($input)) {       # input string already has utf8 flag set
        if($opt{bytes_only}) {
            return encode_utf8($input);
        else {
            return $input;

    if($xs_loaded and $opt{use_xs} ne 'never') {
        my $olf = $opt{overlong_fatal} ? 1 : 0;
        my $asx = $opt{ascii_hex}      ? 1 : 0;
        $input = eval {   # assign back to $input to avoid copying if all ASCII
            Encoding::FixLatin::XS::_fix_latin_xs($input, $olf, $asx);
        if(my $msg = $@) {
            croak $msg;
        if($opt{bytes_only}) {
            return encode_utf8($input);
        else {
            return $input;
    return _fix_latin_pp($input, \%opt);

sub _fix_latin_pp {
    my($input, $opt) = @_;

    my $output = '';
    my $char   = '';
    my $rest   = '';
    my $olf    = $opt->{overlong_fatal};
    while(length($input) > 0) {
        if($input =~ $ascii_str) {
            $output .= $1;
            $rest = $2;
        elsif($input =~ $utf8_2) {
            $output .= _decode_utf8($olf, ord($1) & 0x1F, $1, $2);
            $rest = $3;
        elsif($input =~ $utf8_3) {
            $output .= _decode_utf8($olf, ord($1) & 0x0F, $1, $2, $3);
            $rest = $4;
        elsif($input =~ $utf8_4) {
            $output .= _decode_utf8($olf, ord($1) & 0x07, $1, $2, $3, $4);
            $rest = $5;
        elsif($input =~ $utf8_5) {
            $output .= _decode_utf8($olf, ord($1) & 0x03, $1, $2, $3, $4, $5);
            $rest = $6;
        else {
            ($char, $rest) = $input =~ /^(.)(.*)$/s;
            if($opt->{ascii_hex} && exists $non_1252{$char}) {
                $output .= $non_1252{$char};
            else {
                $output .= $byte_map->{$char};
        $input = $rest;
    utf8::decode($output) unless $opt->{bytes_only};
    return $output;

sub _decode_utf8 {
    my $overlong_fatal = shift;
    my $c              = shift;
    my $byte_count     = @_;
    foreach my $i (1..$#_) {
        $c = ($c << 6) + (ord($_[$i]) & 0x3F);
    my $bytes = encode_utf8(chr($c));
    if($overlong_fatal and $byte_count > length($bytes)) {
        my $hex_bytes= join ' ', map { sprintf('%02X', ord($_)) } @_;
        croak "Over-long UTF-8 byte sequence: $hex_bytes";
    return $bytes;

sub _init_byte_map {
    foreach my $i (0x80..0xFF) {
        my $utf_char = chr($i);
        $byte_map->{pack('C', $i)} = $utf_char;

sub _init_xs {
    my($use_xs) = @_;

    if($use_xs eq 'never' or $xs_loaded) {
    if(!defined($xs_loaded)) {
        $xs_loaded = eval { require 'Encoding/FixLatin/' } ? 1 : 0;
    if(!$xs_loaded and $use_xs eq 'always') {
        croak "Failed to load Encoding::FixLatin::XS";

sub _add_cp1252_mappings {
    # From
    my %ms_map = (
        "\x80" => "\xE2\x82\xAC",  # EURO SIGN
        "\x82" => "\xE2\x80\x9A",  # SINGLE LOW-9 QUOTATION MARK
        "\x83" => "\xC6\x92",      # LATIN SMALL LETTER F WITH HOOK
        "\x84" => "\xE2\x80\x9E",  # DOUBLE LOW-9 QUOTATION MARK
        "\x85" => "\xE2\x80\xA6",  # HORIZONTAL ELLIPSIS
        "\x86" => "\xE2\x80\xA0",  # DAGGER
        "\x87" => "\xE2\x80\xA1",  # DOUBLE DAGGER
        "\x88" => "\xCB\x86",      # MODIFIER LETTER CIRCUMFLEX ACCENT
        "\x89" => "\xE2\x80\xB0",  # PER MILLE SIGN
        "\x8A" => "\xC5\xA0",      # LATIN CAPITAL LETTER S WITH CARON
        "\x8B" => "\xE2\x80\xB9",  # SINGLE LEFT-POINTING ANGLE QUOTATION MARK
        "\x8C" => "\xC5\x92",      # LATIN CAPITAL LIGATURE OE
        "\x8E" => "\xC5\xBD",      # LATIN CAPITAL LETTER Z WITH CARON
        "\x91" => "\xE2\x80\x98",  # LEFT SINGLE QUOTATION MARK
        "\x92" => "\xE2\x80\x99",  # RIGHT SINGLE QUOTATION MARK
        "\x93" => "\xE2\x80\x9C",  # LEFT DOUBLE QUOTATION MARK
        "\x94" => "\xE2\x80\x9D",  # RIGHT DOUBLE QUOTATION MARK
        "\x95" => "\xE2\x80\xA2",  # BULLET
        "\x96" => "\xE2\x80\x93",  # EN DASH
        "\x97" => "\xE2\x80\x94",  # EM DASH
        "\x98" => "\xCB\x9C",      # SMALL TILDE
        "\x99" => "\xE2\x84\xA2",  # TRADE MARK SIGN
        "\x9A" => "\xC5\xA1",      # LATIN SMALL LETTER S WITH CARON
        "\x9C" => "\xC5\x93",      # LATIN SMALL LIGATURE OE
        "\x9E" => "\xC5\xBE",      # LATIN SMALL LETTER Z WITH CARON
        "\x9F" => "\xC5\xB8",      # LATIN CAPITAL LETTER Y WITH DIAERESIS
    while(my($k, $v) = each %ms_map) {
        $byte_map->{$k} = $v;



=head1 NAME

Encoding::FixLatin - takes mixed encoding input and produces UTF-8 output


    use Encoding::FixLatin qw(fix_latin);

    my $utf8_string = fix_latin($mixed_encoding_string);


Most encoding conversion tools take input in one encoding and produce output in
another encoding.  This module takes input which may contain characters in more
than one encoding and makes a best effort to convert them all to UTF-8 output.

=head1 EXPORTS

Nothing is exported by default.  The only public function is C<fix_latin> which
will be exported on request (as per SYNOPSIS).


=head2 fix_latin( string, options ... )

Decodes the supplied 'string' and returns a UTF-8 version of the string.  The
following rules are used:

=over 4

=item *

ASCII characters (single bytes in the range 0x00 - 0x7F) are passed through

=item *

Well-formed UTF-8 multi-byte characters are also passed through unchanged.

=item *

UTF-8 multi-byte character which are over-long but otherwise well-formed are
converted to the shortest UTF-8 normal form.

=item *

Bytes in the range 0xA0 - 0xFF are assumed to be Latin-1 characters (ISO8859-1
encoded) and are converted to UTF-8.

=item *

Bytes in the range 0x80 - 0x9F are assumed to be Win-Latin-1 characters (CP1252
encoded) and are converted to UTF-8.  Except for the five bytes in this range
which are not defined in CP1252 (see the C<ascii_hex> option below).


The achilles heel of these rules is that it's possible for certain combinations
of two consecutive Latin-1 characters to be misinterpreted as a single UTF-8
character - ie: there is some risk of data corruption.  See the 'LIMITATIONS'
section below to quantify this risk for the type of data you're working with.

If you pass in a string that is already a UTF-8 character string (the utf8 flag
is set on the Perl scalar) then the string will simply be returned unchanged.
However if the 'bytes_only' option is specified (see below), the returned
string will be a byte string rather than a character string.  The rules
described above will not be applied in either case.

The C<fix_latin> function accepts options as name => value pairs.  Recognised
options are:

=over 4

=item bytes_only => 1/0

The value returned by fix_latin is normally a Perl character string and will
have the utf8 flag set if it contains non-ASCII characters.  If you set the
C<bytes_only> option to a true value, the returned string will be a binary
string of UTF-8 bytes.  The utf8 flag will not be set.  This is useful if
you're going to immediately use the string in an IO operation and wish to avoid
the overhead of converting to and from Perl's internal representation.

=item ascii_hex => 1/0

Bytes in the range 0x80-0x9F are assumed to be CP1252, however CP1252 does not
define a mapping for 5 of these bytes (0x81, 0x8D, 0x8F, 0x90 and 0x9D).  Use
this option to specify how they should be handled:

=over 4

=item *

If the ascii_hex option is set to true (the default), these bytes will be
converted to 3 character ASCII hex strings of the form %XX.  For example the
byte 0x81 will become %81.

=item *

If the ascii_hex option is set to false, these bytes will be treated as Latin-1
control characters and converted to the equivalent UTF-8 multi-byte sequences.


When processing text strings you will almost certainly never encounter these
bytes at all.  The most likely reason you would see them is if a malicious
attacker was feeding random bytes to your application.  It is difficult to
conceive of a scenario in which it makes sense to change this option from its
default setting.

=item overlong_fatal => 1/0

An over-long UTF-8 byte sequence is one which uses more than the minimum number
of bytes required to represent the character.  Use this option to specify how
overlong sequences should be handled.

=over 4

=item *

If the overlong_fatal option is set to false (the default) over-long sequences
will be converted to the shortest normal UTF-8 sequence.  For example the input
byte string "\xC0\xBCscript>" would be converted to "<script>".

=item *

If the overlong_fatal option is set to true, this module will die with an
error when an overlong sequence is encountered.  You would probably want to
use eval to trap and handle this scenario.


There is a strong argument that overlong sequences are only ever encountered
in malicious input and therefore they should always be rejected.

=item use_xs => 'auto' | 'always' | 'never'

This option controls whether or not the XS (compiled C) implementation of
C<fix_latin> is used.  Note, the L<Encoding::FixLatin::XS> module must be
installed separately.  The three possible values for this option are:

=over 4

=item *

'auto' is the default behaviour - if L<Encoding::FixLatin::XS> is installed, it
will be loaded and used, otherwise the pure Perl implementation will be used.

=item *

'always' means the XS module will be used and a fatal exception will be thrown
if it is not available.

=item *

'never' means no attempt will be made to use the XS module.




This module is perfectly safe when handling data containing only ASCII and
UTF-8 characters.  Introducing ISO8859-1 or CP1252 characters does add a risk
of data corruption (ie: some characters in the input being converted to
incorrect characters in the output).  To quantify the risk it is necessary to
understand it's cause.  First, let's break the input bytes into two categories.

=over 4

=item *

ASCII bytes fall into the range 0x00-0x7F - the most significant bit is always
set to zero.  I'll use the symbol 'a' to represent these bytes.

=item *

Non-ASCII bytes fall into the range 0x80-0xFF - the most significant bit is
always set to one.  I'll use the symbol 'B' to represent these bytes.


A sequence of ASCII bytes ('aaa') is always unambiguous and will not be

Lone non-ASCII bytes within sequences of ASCII bytes ('aaBaBa') are also
unambiguous and will not be misinterpreted.

The potential for error occurs with two (or more) consecutive non-ASCII bytes.
For example the sequence 'BB' might be intended to represent two characters in
one of the legacy encodings or a single character in UTF-8.  Because this
module gives precedence to the UTF-8 characters it is possible that a random
pair of legacy characters may be misinterpreted as a single UTF-8 character.

The risk is reduced by the fact that not all pairs of non-ASCII bytes form
valid UTF-8 sequences.  Every non-ASCII UTF-8 character is made up of two or
more 'B' bytes and no 'a' bytes.  For a two-byte character, the first byte must
be in the range 0xC0-0xDF and the second must be in the range 0x80-0xBF.

Any pair of 'BB' bytes that do not fall into the required ranges are
unambiguous and will not be misinterpreted.

Pairs of 'BB' bytes that are actually individual Latin-1 characters but
happen to fall into the required ranges to be misinterpreted as a UTF-8
character are rather unlikely to appear in normal text.  If you look those
ranges up on a Latin-1 code chart you'll see that the first character would
need to be an uppercase accented letter and the second  would need to be a
non-printable control character or a special punctuation symbol.

One way to summarise the role of this module is that it guarantees to
produce UTF-8 output, possibly at the cost of introducing the odd 'typo'.

=head1 BUGS

Please report any bugs to C<bug-encoding-fixlatin at>, or through
the web interface at
L<>.  I will be
notified, and then you'll automatically be notified of progress on your bug as
I make changes.

=head1 SUPPORT

You can also look for information at:

=over 4

=item * Issue tracker


=item * AnnoCPAN: Annotated CPAN documentation


=item * CPAN Ratings


=item * Search CPAN


=item * Source code repository




Copyright 2009-2014 Grant McLean C<< <> >>

This program is free software; you can redistribute it and/or modify it
under the same terms as Perl itself.