/ 
Encoding-FixLatin-1.02
River stage two • 3 direct dependents • 37 total dependents
/ Encoding::FixLatin

NAME

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

SYNOPSIS

use Encoding::FixLatin qw(fix_latin);

my $utf8_string = fix_latin($mixed_encoding_string);

DESCRIPTION

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.

EXPORTS

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

FUNCTIONS

fix_latin( string, options ... )

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

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

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

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

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

  • 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 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 fix_latin function accepts options as name => value pairs. Recognised options are:

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 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.

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:

  • 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.

  • 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.

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.

  • 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>".

  • 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.

LIMITATIONS OF THIS 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.

  • 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.

  • 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 misinterpreted.

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'.

BUGS

Please report any bugs to bug-encoding-fixlatin at rt.cpan.org, or through the web interface at http://rt.cpan.org/NoAuth/ReportBug.html?Queue=Encoding-FixLatin. I will be notified, and then you'll automatically be notified of progress on your bug as I make changes.

SUPPORT

You can also look for information at:

AUTHOR

Grant McLean, <grantm at cpan.org>

COPYRIGHT & LICENSE

Copyright 2009-2010 Grant McLean

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