perluniintro - Perl Unicode introduction

NAME

perluniintro - Perl Unicode introduction

DESCRIPTION

This document gives a general idea of Unicode and how to use Unicode in Perl. See Further Resources for references to more in-depth treatments of Unicode.

Unicode

Unicode is a character set standard which plans to codify all of the writing systems of the world, plus many other symbols.

Unicode and ISO/IEC 10646 are coordinated standards that unify almost all other modern character set standards, covering more than 80 writing systems and hundreds of languages, including all commercially-important modern languages. All characters in the largest Chinese, Japanese, and Korean dictionaries are also encoded. The standards will eventually cover almost all characters in more than 250 writing systems and thousands of languages. Unicode 1.0 was released in October 1991, and 6.0 in October 2010.

A Unicode character is an abstract entity. It is not bound to any particular integer width, especially not to the C language char. Unicode is language-neutral and display-neutral: it does not encode the language of the text, and it does not generally define fonts or other graphical layout details. Unicode operates on characters and on text built from those characters.

Unicode defines characters like LATIN CAPITAL LETTER A or GREEK SMALL LETTER ALPHA and unique numbers for the characters, in this case 0x0041 and 0x03B1, respectively. These unique numbers are called code points. A code point is essentially the position of the character within the set of all possible Unicode characters, and thus in Perl, the term ordinal is often used interchangeably with it.

The Unicode standard prefers using hexadecimal notation for the code points. If numbers like 0x0041 are unfamiliar to you, take a peek at a later section, Hexadecimal Notation. The Unicode standard uses the notation U+0041 LATIN CAPITAL LETTER A, to give the hexadecimal code point and the normative name of the character.

Unicode also defines various properties for the characters, like ``uppercase'' or ``lowercase'', ``decimal digit'', or ``punctuation''; these properties are independent of the names of the characters. Furthermore, various operations on the characters like uppercasing, lowercasing, and collating (sorting) are defined.

A Unicode logical ``character'' can actually consist of more than one internal actual ``character'' or code point. For Western languages, this is adequately modelled by a base character (like LATIN CAPITAL LETTER A) followed by one or more modifiers (like COMBINING ACUTE ACCENT). This sequence of base character and modifiers is called a combining character sequence. Some non-western languages require more complicated models, so Unicode created the grapheme cluster concept, which was later further refined into the extended grapheme cluster. For example, a Korean Hangul syllable is considered a single logical character, but most often consists of three actual Unicode characters: a leading consonant followed by an interior vowel followed by a trailing consonant.

Whether to call these extended grapheme clusters ``characters'' depends on your point of view. If you are a programmer, you probably would tend towards seeing each element in the sequences as one unit, or ``character''. However from the user's point of view, the whole sequence could be seen as one ``character'' since that's probably what it looks like in the context of the user's language. In this document, we take the programmer's point of view: one ``character'' is one Unicode code point.

For some combinations of base character and modifiers, there are precomposed characters. There is a single character equivalent, for example, for the sequence LATIN CAPITAL LETTER A followed by COMBINING ACUTE ACCENT. It is called LATIN CAPITAL LETTER A WITH ACUTE. These precomposed characters are, however, only available for some combinations, and are mainly meant to support round-trip conversions between Unicode and legacy standards (like ISO 8859). Using sequences, as Unicode does, allows for needing fewer basic building blocks (code points) to express many more potential grapheme clusters. To support conversion between equivalent forms, various normalization forms are also defined. Thus, LATIN CAPITAL LETTER A WITH ACUTE is in Normalization Form Composed, (abbreviated NFC), and the sequence LATIN CAPITAL LETTER A followed by COMBINING ACUTE ACCENT represents the same character in Normalization Form Decomposed (NFD).

Because of backward compatibility with legacy encodings, the ``a unique number for every character'' idea breaks down a bit: instead, there is ``at least one number for every character''. The same character could be represented differently in several legacy encodings. The converse is not true: some code points do not have an assigned character. Firstly, there are unallocated code points within otherwise used blocks. Secondly, there are special Unicode control characters that do not represent true characters.

When Unicode was first conceived, it was thought that all the world's characters could be represented using a 16-bit word; that is a maximum of 0x10000 (or 65,536) characters would be needed, from 0x0000 to 0xFFFF. This soon proved to be wrong, and since Unicode 2.0 (July 1996), Unicode has been defined all the way up to 21 bits (0x10FFFF), and Unicode 3.1 (March 2001) defined the first characters above 0xFFFF. The first 0x10000 characters are called the Plane 0, or the Basic Multilingual Plane (BMP). With Unicode 3.1, 17 (yes, seventeen) planes in all were defined--but they are nowhere near full of defined characters, yet.

When a new language is being encoded, Unicode generally will choose a block of consecutive unallocated code points for its characters. So far, the number of code points in these blocks has always been evenly divisible by 16. Extras in a block, not currently needed, are left unallocated, for future growth. But there have been occasions when a later release needed more code points than the available extras, and a new block had to allocated somewhere else, not contiguous to the initial one, to handle the overflow. Thus, it became apparent early on that ``block'' wasn't an adequate organizing principle, and so the Script property was created. (Later an improved script property was added as well, the Script_Extensions property.) Those code points that are in overflow blocks can still have the same script as the original ones. The script concept fits more closely with natural language: there is Latin script, Greek script, and so on; and there are several artificial scripts, like Common for characters that are used in multiple scripts, such as mathematical symbols. Scripts usually span varied parts of several blocks. For more information about scripts, see Scripts in the perlunicode manpage. The division into blocks exists, but it is almost completely accidental--an artifact of how the characters have been and still are allocated. (Note that this paragraph has oversimplified things for the sake of this being an introduction. Unicode doesn't really encode languages, but the writing systems for them--their scripts; and one script can be used by many languages. Unicode also encodes things that aren't really about languages, such as symbols like BAGGAGE CLAIM.)

The Unicode code points are just abstract numbers. To input and output these abstract numbers, the numbers must be encoded or serialised somehow. Unicode defines several character encoding forms, of which UTF-8 is the most popular. UTF-8 is a variable length encoding that encodes Unicode characters as 1 to 4 bytes. Other encodings include UTF-16 and UTF-32 and their big- and little-endian variants (UTF-8 is byte-order independent). The ISO/IEC 10646 defines the UCS-2 and UCS-4 encoding forms.

For more information about encodings--for instance, to learn what surrogates and byte order marks (BOMs) are--see the perlunicode manpage.

Perl's Unicode Support

Starting from Perl v5.6.0, Perl has had the capacity to handle Unicode natively. Perl v5.8.0, however, is the first recommended release for serious Unicode work. The maintenance release 5.6.1 fixed many of the problems of the initial Unicode implementation, but for example regular expressions still do not work with Unicode in 5.6.1. Perl v5.14.0 is the first release where Unicode support is (almost) seamlessly integrable without some gotchas. (There are a few exceptions. Firstly, some differences in quotemeta were fixed starting in Perl 5.16.0. Secondly, some differences in the range operator were fixed starting in Perl 5.26.0. Thirdly, some differences in split were fixed started in Perl 5.28.0.)

To enable this seamless support, you should use feature 'unicode_strings' (which is automatically selected if you use 5.012 or higher). See the feature manpage. (5.14 also fixes a number of bugs and departures from the Unicode standard.)

Before Perl v5.8.0, the use of use utf8 was used to declare that operations in the current block or file would be Unicode-aware. This model was found to be wrong, or at least clumsy: the ``Unicodeness'' is now carried with the data, instead of being attached to the operations. Starting with Perl v5.8.0, only one case remains where an explicit use utf8 is needed: if your Perl script itself is encoded in UTF-8, you can use UTF-8 in your identifier names, and in string and regular expression literals, by saying use utf8. This is not the default because scripts with legacy 8-bit data in them would break. See the utf8 manpage.

Perl's Unicode Model

Perl supports both pre-5.6 strings of eight-bit native bytes, and strings of Unicode characters. The general principle is that Perl tries to keep its data as eight-bit bytes for as long as possible, but as soon as Unicodeness cannot be avoided, the data is transparently upgraded to Unicode. Prior to Perl v5.14.0, the upgrade was not completely transparent (see The ``Unicode Bug'' in the perlunicode manpage), and for backwards compatibility, full transparency is not gained unless use feature 'unicode_strings' (see the feature manpage) or use 5.012 (or higher) is selected.

Internally, Perl currently uses either whatever the native eight-bit character set of the platform (for example Latin-1) is, defaulting to UTF-8, to encode Unicode strings. Specifically, if all code points in the string are 0xFF or less, Perl uses the native eight-bit character set. Otherwise, it uses UTF-8.

A user of Perl does not normally need to know nor care how Perl happens to encode its internal strings, but it becomes relevant when outputting Unicode strings to a stream without a PerlIO layer (one with the ``default'' encoding). In such a case, the raw bytes used internally (the native character set or UTF-8, as appropriate for each string) will be used, and a ``Wide character'' warning will be issued if those strings contain a character beyond 0x00FF.

For example,

      perl -e 'print "\x{DF}\n", "\x{0100}\x{DF}\n"'

produces a fairly useless mixture of native bytes and UTF-8, as well as a warning:

     Wide character in print at ...

To output UTF-8, use the :encoding or :utf8 output layer. Prepending

      binmode(STDOUT, ":utf8");

to this sample program ensures that the output is completely UTF-8, and removes the program's warning.

You can enable automatic UTF-8-ification of your standard file handles, default open() layer, and @ARGV by using either the -C command line switch or the PERL_UNICODE environment variable, see the perlrun manpage for the documentation of the -C switch.

Note that this means that Perl expects other software to work the same way: if Perl has been led to believe that STDIN should be UTF-8, but then STDIN coming in from another command is not UTF-8, Perl will likely complain about the malformed UTF-8.

All features that combine Unicode and I/O also require using the new PerlIO feature. Almost all Perl 5.8 platforms do use PerlIO, though: you can see whether yours is by running ``perl -V'' and looking for useperlio=define.

Unicode and EBCDIC

Perl 5.8.0 added support for Unicode on EBCDIC platforms. This support was allowed to lapse in later releases, but was revived in 5.22. Unicode support is somewhat more complex to implement since additional conversions are needed. See the perlebcdic manpage for more information.

On EBCDIC platforms, the internal Unicode encoding form is UTF-EBCDIC instead of UTF-8. The difference is that as UTF-8 is ``ASCII-safe'' in that ASCII characters encode to UTF-8 as-is, while UTF-EBCDIC is ``EBCDIC-safe'', in that all the basic characters (which includes all those that have ASCII equivalents (like "A", "0", "%", etc.) are the same in both EBCDIC and UTF-EBCDIC. Often, documentation will use the term ``UTF-8'' to mean UTF-EBCDIC as well. This is the case in this document.

Creating Unicode

This section applies fully to Perls starting with v5.22. Various caveats for earlier releases are in the Earlier releases caveats subsection below.

To create Unicode characters in literals, use the \N{...} notation in double-quoted strings:

 my $smiley_from_name = "\N{WHITE SMILING FACE}";
 my $smiley_from_code_point = "\N{U+263a}";

Similarly, they can be used in regular expression literals

 $smiley =~ /\N{WHITE SMILING FACE}/;
 $smiley =~ /\N{U+263a}/;

At run-time you can use:

 use charnames ();
 my $hebrew_alef_from_name
                      = charnames::string_vianame("HEBREW LETTER ALEF");
 my $hebrew_alef_from_code_point = charnames::string_vianame("U+05D0");

Naturally, ord() will do the reverse: it turns a character into a code point.

There are other runtime options as well. You can use pack():

 my $hebrew_alef_from_code_point = pack("U", 0x05d0);

Or you can use chr(), though it is less convenient in the general case:

 $hebrew_alef_from_code_point = chr(utf8::unicode_to_native(0x05d0));
 utf8::upgrade($hebrew_alef_from_code_point);

The utf8::unicode_to_native() and utf8::upgrade() aren't needed if the argument is above 0xFF, so the above could have been written as

 $hebrew_alef_from_code_point = chr(0x05d0);

since 0x5d0 is above 255.

\x{} and \o{} can also be used to specify code points at compile time in double-quotish strings, but, for backward compatibility with older Perls, the same rules apply as with chr() for code points less than 256.

utf8::unicode_to_native() is used so that the Perl code is portable to EBCDIC platforms. You can omit it if you're really sure no one will ever want to use your code on a non-ASCII platform. Starting in Perl v5.22, calls to it on ASCII platforms are optimized out, so there's no performance penalty at all in adding it. Or you can simply use the other constructs that don't require it.

See Further Resources for how to find all these names and numeric codes.

Earlier releases caveats

On EBCDIC platforms, prior to v5.22, using \N{U+...} doesn't work properly.

Prior to v5.16, using \N{...} with a character name (as opposed to a U+... code point) required a use charnames :full.

Prior to v5.14, there were some bugs in \N{...} with a character name (as opposed to a U+... code point).

charnames::string_vianame() was introduced in v5.14. Prior to that, charnames::vianame() should work, but only if the argument is of the form "U+...". Your best bet there for runtime Unicode by character name is probably:

 use charnames ();
 my $hebrew_alef_from_name
                  = pack("U", charnames::vianame("HEBREW LETTER ALEF"));

Handling Unicode

Handling Unicode is for the most part transparent: just use the strings as usual. Functions like index(), length(), and substr() will work on the Unicode characters; regular expressions will work on the Unicode characters (see the perlunicode manpage and the perlretut manpage).

Note that Perl considers grapheme clusters to be separate characters, so for example

 print length("\N{LATIN CAPITAL LETTER A}\N{COMBINING ACUTE ACCENT}"),
       "\n";

will print 2, not 1. The only exception is that regular expressions have \X for matching an extended grapheme cluster. (Thus \X in a regular expression would match the entire sequence of both the example characters.)

Life is not quite so transparent, however, when working with legacy encodings, I/O, and certain special cases:

Legacy Encodings

When you combine legacy data and Unicode, the legacy data needs to be upgraded to Unicode. Normally the legacy data is assumed to be ISO 8859-1 (or EBCDIC, if applicable).

The Encode module knows about many encodings and has interfaces for doing conversions between those encodings:

    use Encode 'decode';
    $data = decode("iso-8859-3", $data); # convert from legacy

Unicode I/O

Normally, writing out Unicode data

    print FH $some_string_with_unicode, "\n";

produces raw bytes that Perl happens to use to internally encode the Unicode string. Perl's internal encoding depends on the system as well as what characters happen to be in the string at the time. If any of the characters are at code points 0x100 or above, you will get a warning. To ensure that the output is explicitly rendered in the encoding you desire--and to avoid the warning--open the stream with the desired encoding. Some examples:

    open FH, ">:utf8", "file";
    open FH, ">:encoding(ucs2)",      "file";
    open FH, ">:encoding(UTF-8)",     "file";
    open FH, ">:encoding(shift_jis)", "file";

and on already open streams, use binmode():

    binmode(STDOUT, ":utf8");
    binmode(STDOUT, ":encoding(ucs2)");
    binmode(STDOUT, ":encoding(UTF-8)");
    binmode(STDOUT, ":encoding(shift_jis)");

The matching of encoding names is loose: case does not matter, and many encodings have several aliases. Note that the :utf8 layer must always be specified exactly like that; it is not subject to the loose matching of encoding names. Also note that currently :utf8 is unsafe for input, because it accepts the data without validating that it is indeed valid UTF-8; you should instead use :encoding(UTF-8) (with or without a hyphen).

See the PerlIO manpage for the :utf8 layer, the PerlIO::encoding manpage and the Encode::PerlIO manpage for the :encoding() layer, and the Encode::Supported manpage for many encodings supported by the Encode module.

Reading in a file that you know happens to be encoded in one of the Unicode or legacy encodings does not magically turn the data into Unicode in Perl's eyes. To do that, specify the appropriate layer when opening files

    open(my $fh,'<:encoding(UTF-8)', 'anything');
    my $line_of_unicode = <$fh>;
    open(my $fh,'<:encoding(Big5)', 'anything');
    my $line_of_unicode = <$fh>;

The I/O layers can also be specified more flexibly with the open pragma. See the open manpage, or look at the following example.

    use open ':encoding(UTF-8)'; # input/output default encoding will be
                                 # UTF-8
    open X, ">file";
    print X chr(0x100), "\n";
    close X;
    open Y, "<file";
    printf "%#x\n", ord(<Y>); # this should print 0x100
    close Y;

With the open pragma you can use the :locale layer

    BEGIN { $ENV{LC_ALL} = $ENV{LANG} = 'ru_RU.KOI8-R' }
    # the :locale will probe the locale environment variables like
    # LC_ALL
    use open OUT => ':locale'; # russki parusski
    open(O, ">koi8");
    print O chr(0x430); # Unicode CYRILLIC SMALL LETTER A = KOI8-R 0xc1
    close O;
    open(I, "<koi8");
    printf "%#x\n", ord(<I>), "\n"; # this should print 0xc1
    close I;

These methods install a transparent filter on the I/O stream that converts data from the specified encoding when it is read in from the stream. The result is always Unicode.

The the open manpage pragma affects all the open() calls after the pragma by setting default layers. If you want to affect only certain streams, use explicit layers directly in the open() call.

You can switch encodings on an already opened stream by using binmode(); see binmode in the perlfunc manpage.

The :locale does not currently work with open() and binmode(), only with the open pragma. The :utf8 and :encoding(...) methods do work with all of open(), binmode(), and the open pragma.

Similarly, you may use these I/O layers on output streams to automatically convert Unicode to the specified encoding when it is written to the stream. For example, the following snippet copies the contents of the file ``text.jis'' (encoded as ISO-2022-JP, aka JIS) to the file ``text.utf8'', encoded as UTF-8:

    open(my $nihongo, '<:encoding(iso-2022-jp)', 'text.jis');
    open(my $unicode, '>:utf8',                  'text.utf8');
    while (<$nihongo>) { print $unicode $_ }

The naming of encodings, both by the open() and by the open pragma allows for flexible names: koi8-r and KOI8R will both be understood.

Common encodings recognized by ISO, MIME, IANA, and various other standardisation organisations are recognised; for a more detailed list see the Encode::Supported manpage.

read() reads characters and returns the number of characters. seek() and tell() operate on byte counts, as does sysseek().

sysread() and syswrite() should not be used on file handles with character encoding layers, they behave badly, and that behaviour has been deprecated since perl 5.24.

Notice that because of the default behaviour of not doing any conversion upon input if there is no default layer, it is easy to mistakenly write code that keeps on expanding a file by repeatedly encoding the data:

    # BAD CODE WARNING
    open F, "file";
    local $/; ## read in the whole file of 8-bit characters
    $t = <F>;
    close F;
    open F, ">:encoding(UTF-8)", "file";
    print F $t; ## convert to UTF-8 on output
    close F;

If you run this code twice, the contents of the file will be twice UTF-8 encoded. A use open ':encoding(UTF-8)' would have avoided the bug, or explicitly opening also the file for input as UTF-8.

NOTE: the :utf8 and :encoding features work only if your Perl has been built with the PerlIO manpage, which is the default on most systems.

Displaying Unicode As Text

Sometimes you might want to display Perl scalars containing Unicode as simple ASCII (or EBCDIC) text. The following subroutine converts its argument so that Unicode characters with code points greater than 255 are displayed as \x{...}, control characters (like \n) are displayed as \x.., and the rest of the characters as themselves:

 sub nice_string {
        join("",
        map { $_ > 255                    # if wide character...
              ? sprintf("\\x{%04X}", $_)  # \x{...}
              : chr($_) =~ /[[:cntrl:]]/  # else if control character...
                ? sprintf("\\x%02X", $_)  # \x..
                : quotemeta(chr($_))      # else quoted or as themselves
        } unpack("W*", $_[0]));           # unpack Unicode characters
   }

For example,

   nice_string("foo\x{100}bar\n")

returns the string

   'foo\x{0100}bar\x0A'

which is ready to be printed.

(\\x{} is used here instead of \\N{}, since it's most likely that you want to see what the native values are.)

Special Cases

Advanced Topics

Miscellaneous

Questions With Answers

Hexadecimal Notation

The Unicode standard prefers using hexadecimal notation because that more clearly shows the division of Unicode into blocks of 256 characters. Hexadecimal is also simply shorter than decimal. You can use decimal notation, too, but learning to use hexadecimal just makes life easier with the Unicode standard. The U+HHHH notation uses hexadecimal, for example.

The 0x prefix means a hexadecimal number, the digits are 0-9 and a-f (or A-F, case doesn't matter). Each hexadecimal digit represents four bits, or half a byte. print 0x..., "\n" will show a hexadecimal number in decimal, and printf "%x\n", $decimal will show a decimal number in hexadecimal. If you have just the ``hex digits'' of a hexadecimal number, you can use the hex() function.

    print 0x0009, "\n";    # 9
    print 0x000a, "\n";    # 10
    print 0x000f, "\n";    # 15
    print 0x0010, "\n";    # 16
    print 0x0011, "\n";    # 17
    print 0x0100, "\n";    # 256
    print 0x0041, "\n";    # 65
    printf "%x\n",  65;    # 41
    printf "%#x\n", 65;    # 0x41
    print hex("41"), "\n"; # 65

Further Resources

UNICODE IN OLDER PERLS

If you cannot upgrade your Perl to 5.8.0 or later, you can still do some Unicode processing by using the modules Unicode::String, Unicode::Map8, and Unicode::Map, available from CPAN. If you have the GNU recode installed, you can also use the Perl front-end Convert::Recode for character conversions.

The following are fast conversions from ISO 8859-1 (Latin-1) bytes to UTF-8 bytes and back, the code works even with older Perl 5 versions.

    # ISO 8859-1 to UTF-8
    s/([\x80-\xFF])/chr(0xC0|ord($1)>>6).chr(0x80|ord($1)&0x3F)/eg;
    # UTF-8 to ISO 8859-1
    s/([\xC2\xC3])([\x80-\xBF])/chr(ord($1)<<6&0xC0|ord($2)&0x3F)/eg;

SEE ALSO

the perlunitut manpage, the perlunicode manpage, the Encode manpage, the open manpage, the utf8 manpage, the bytes manpage, the perlretut manpage, the perlrun manpage, the Unicode::Collate manpage, the Unicode::Normalize manpage, the Unicode::UCD manpage

ACKNOWLEDGMENTS

Thanks to the kind readers of the perl5-porters@perl.org, perl-unicode@perl.org, linux-utf8@nl.linux.org, and unicore@unicode.org mailing lists for their valuable feedback.

AUTHOR, COPYRIGHT, AND LICENSE

Copyright 2001-2011 Jarkko Hietaniemi <jhi@iki.fi>. Now maintained by Perl 5 Porters.

This document may be distributed under the same terms as Perl itself.

 perluniintro - Perl Unicode introduction