LDP/LDP/howto/linuxdoc/Unicode-HOWTO.sgml

<!doctype linuxdoc system>

<article>

<title>The Unicode HOWTO
<author>Bruno Haible,<htmlurl url="mailto:haible@clisp.cons.org"
name="&lt;haible@clisp.cons.org&gt;">
<date>v0.12, 19 October 1999
<abstract>
This document describes how to change your Linux system so it uses UTF-8
as text encoding. -
This is work in progress. Any tips, patches, pointers, URLs are very welcome.
</abstract>

<toc>


<sect>Introduction
<p>

<sect1>Why Unicode?
<p>

People in different countries use different characters to represent the
words of their native languages. Nowadays most applications, including
email systems and web browsers, are 8-bit clean, i.e. they can operate on
and display text correctly provided that it is represented in an 8-bit
character set, like ISO-8859-1.

There are far more than 256 characters in the world - think of cyrillic,
hebrew, arabic, chinese, japanese, korean and thai -, and new characters
are being invented now and then. The problems that come up for users are:

<itemize>
<item>
    It is impossible to store text with characters from different character
    sets in the same document. For example, I can cite russian papers in
    a German or French publication if I use TeX, xdvi and PostScript,
    but I cannot do it in plain text.
<item>
    As long as every document has its own character set, and recognition
    of the character set is not automatic, manual user intervention is
    inevitable. For example, in order to view the homepage of the
    XTeamLinux distribution
    <htmlurl url="http://www.xteamlinux.com.cn/"
            name="http://www.xteamlinux.com.cn/">
    I have to tell Netscape that the web page is coded in GB2312.
<item>
    New symbols like the Euro are being invented. ISO has issued a new
    standard ISO-8859-15, which is mostly like ISO-8859-1 except that it
    removes some rarely used characters (the old currency sign) and
    replaced it with the Euro sign. If users adopt this standard, they
    have documents in different character sets on their disk, and they
    start having to think about it daily. But computers should make things
    simpler, not more complicated.
</itemize>

The solution of this problem is the adoption of a world-wide usable character
set. This character set is Unicode
<htmlurl url="http://www.unicode.org/"
        name="http://www.unicode.org/">.
For more info about Unicode, do `<tt>man 7 unicode</tt>' (manpage contained
in the ldpman-1.20 package).

<sect1>Unicode encodings
<p>

This reduces the user's problem of dealing with character sets to a technical
problem: How to transport Unicode characters using the 8-bit bytes?
8-bit units are the smallest addressing units of most computers and also the
unit used by TCP/IP network connections. The use of 1 byte to represent
1 character is, however, an accident of history, caused by the fact that
computer development started in Europe and the U.S. where 96 characters were
found to be sufficient for a long time.

There are basically four ways to encode Unicode characters in bytes:

<descrip>
<tag>UTF-8</tag>
    128 characters are encoded using 1 byte (the ASCII characters).
    1920 characters are encoded using 2 bytes (Roman, Greek, Cyrillic,
    Coptic, Armenian, Hebrew, Arabic characters).
    63488 characters are encoded using 3 bytes (Chinese and Japanese among
    others).
    The other 2147418112 characters (not assigned yet) can be encoded
    using 4, 5 or 6 characters.
    For more info about UTF-8, do `<tt>man 7 utf-8</tt>' (manpage contained
    in the ldpman-1.20 package).
<tag>UCS-2</tag>
    Every character is represented as two bytes.
    This encoding can only represent the first 65536 Unicode characters.
<tag>UTF-16</tag>
    This is an extension of UCS-2 which can represent 1114112 Unicode
    characters. The first 65536 Unicode characters are represented as two
    bytes, the other ones as four bytes.
<tag>UCS-4</tag>
    Every character is represented as four bytes.
</descrip>

The space requirements for encoding a text, compared to encodings currently
in use (8 bit per character for European languages, more for
Chinese/Japanese/Korean), is as follows. This has an influence on disk
storage space and network download speed.

<descrip>
<tag>UTF-8</tag>
    No change for US ASCII, just a few percent more for ISO-8859-1,
    50% more for Chinese/Japanese/Korean, 100% more for Greek and Cyrillic.
<tag>UCS-2 and UTF-16</tag>
    No change for Chinese/Japanese/Korean. 100% more for
    US ASCII and ISO-8859-1, Greek and Cyrillic.
<tag>UCS-4</tag>
    100% more for Chinese/Japanese/Korean. 300% more for US ASCII and
    ISO-8859-1, Greek and Cyrillic.
</descrip>

Given the penalty for US and European documents caused by UCS-2, UTF-16, and
UCS-4, it seems unlikely that these encodings have a potential for wide-scale
use. The Microsoft Win32 API supports the UCS-2 encoding since 1995 (at
least), yet this encoding has not been widely adopted for documents - SJIS
remains prevalent in Japan.

UTF-8 on the other hand has the potential for wide-scale use, since it
doesn't penalize US and European users, and since many text processing
programs don't need to be changed for UTF-8 support.

In the following, we will describe how to change your Linux system so
it uses UTF-8 as text encoding.

<sect2>Footnotes for C/C++ developers
<p>

The Microsoft Win32 approach makes it easy for developers to produce
Unicode versions of their programs: You "#define UNICODE" at the top
of your program and then change many occurrences of `<tt>char</tt>' to
`<tt>TCHAR</tt>', until your program compiles without warnings. The problem
with it is that you end up with two versions of your program: one which
understands UCS-2 text but no 8-bit encodings, and one which understands
only old 8-bit encodings.

Moreover, there is an endianness issue with UCS-2 and UCS-4. The IANA
character set registry says about ISO-10646-UCS-2: "this needs to specify
network byte order: the standard does not specify". Network byte order is
big endian. Whereas Microsoft, in its C/C++ development tools, recommends
to use machine-dependent endianness (i.e. little endian on ix86 processors)
and either a byte-order mark at the beginning of the document, or some
statistical heuristics(!).

The UTF-8 approach on the other hand keeps `<tt>char*</tt>' as the standard C
string type. As a result, your program will handle US ASCII text,
independently of any environment variables, and will handle both
ISO-8859-1 and UTF-8 encoded text provided the LANG environment variable
is set accordingly.

<sect1>Related resources
<p>

Markus Kuhn's very up-to-date resource list:
<itemize>
<item>
<htmlurl url="http://www.cl.cam.ac.uk/~mgk25/unicode.html"
        name="http://www.cl.cam.ac.uk/~mgk25/unicode.html">
<item>
<htmlurl url="http://www.cl.cam.ac.uk/~mgk25/ucs-fonts.html"
        name="http://www.cl.cam.ac.uk/~mgk25/ucs-fonts.html">
</itemize>

Roman Czyborra's overview of Unicode, UTF-8 and UTF-8 aware programs:
<htmlurl url="http://czyborra.com/utf/#UTF-8"
        name="http://czyborra.com/utf/#UTF-8">

Some example UTF-8 files:
<itemize>
<item>
The files <tt>quickbrown.txt</tt>, <tt>utf-8-test.txt</tt>,
<tt>utf-8-demo.txt</tt> in the <tt>examples</tt> directory of Markus Kuhn's
ucs-fonts package
<htmlurl url="http://www.cl.cam.ac.uk/~mgk25/download/ucs-fonts.tar.gz"
        name="http://www.cl.cam.ac.uk/~mgk25/download/ucs-fonts.tar.gz">
<item>
<htmlurl url="ftp://ftp.cs.su.oz.au/gary/x-utf8.html"
        name="ftp://ftp.cs.su.oz.au/gary/x-utf8.html">
<item>
The file <tt>iso10646</tt> in the Kosta Kostis' trans-1.1.1 package
<htmlurl url="ftp://ftp.nid.ru/pub/os/unix/misc/trans111.tar.gz"
        name="ftp://ftp.nid.ru/pub/os/unix/misc/trans111.tar.gz">
<item>
<htmlurl url="ftp://ftp.dante.de/pub/tex/info/lwc/apc/utf8.html"
        name="ftp://ftp.dante.de/pub/tex/info/lwc/apc/utf8.html">
<item>
<htmlurl url="http://www.cogsci.ed.ac.uk/~richard/unicode-sample.html"
        name="http://www.cogsci.ed.ac.uk/~richard/unicode-sample.html">
</itemize>


<sect>Display setup
<p>

We assume you have already adapted your Linux console and X11 configuration
to your keyboard and locale. This is explained in the Danish/International
HOWTO, and in the other national HOWTOs: Finnish, French, German, Italian,
Polish, Slovenian, Spanish, Cyrillic, Hebrew, Chinese, Thai, Esperanto. But
please do not follow the advice given in the Thai HOWTO, to pretend you
were using ISO-8859-1 characters (U0000..U00FF) when what you are typing
are actually Thai characters (U0E01..U0E5B). Doing so will only cause
problems when you switch to Unicode.

<sect1>Linux console
<p>

I'm not talking much about the Linux console here, because on those machines
on which I don't have xdm running, I use it only to type my login name,
my password, and "xinit".

Anyway, the kbd-0.99 package
<htmlurl url="ftp://sunsite.unc.edu/pub/Linux/system/keyboards/kbd-0.99.tar.gz"
        name="ftp://sunsite.unc.edu/pub/Linux/system/keyboards/kbd-0.99.tar.gz">
and a heavily extended version, the console-tools-0.2.2 package
<htmlurl url="ftp://sunsite.unc.edu/pub/Linux/system/keyboards/console-tools-0.2.2.tar.gz"
        name="ftp://sunsite.unc.edu/pub/Linux/system/keyboards/console-tools-0.2.2.tar.gz">
contains in the kbd-0.99/src/ (or console-tools-0.2.2/screenfonttools/)
directory two programs: `unicode_start' and `unicode_stop'. When you call
`unicode_start', the console's screen output is interpreted as UTF-8. Also,
the keyboard is put into Unicode mode (see "man kbd_mode"). In this mode,
Unicode characters typed as Alt-x1 ... Alt-xn (where x1,...,xn are digits on
the numeric keypad) will be emitted in UTF-8. If your keyboard or, more
precisely, your normal keymap has non-ASCII letter keys (like the German
Umlaute) which you would like to be CapsLockable, you need to apply the kernel
patch
<htmlurl url="linux-2.2.9-keyboard.diff"
        name="linux-2.2.9-keyboard.diff">
or
<htmlurl url="linux-2.3.12-keyboard.diff"
        name="linux-2.3.12-keyboard.diff">.

You will want to use display characters from different scripts on the same
screen. For this, you need a Unicode console font. The
<htmlurl url="ftp://sunsite.unc.edu/pub/Linux/system/keyboards/console-data-1999.08.29.tar.gz"
        name="ftp://sunsite.unc.edu/pub/Linux/system/keyboards/console-data-1999.08.29.tar.gz">
package contains a font (LatArCyrHeb-{08,14,16,19}.psf) which
covers Latin, Cyrillic, Hebrew, Arabic scripts. It covers ISO 8859 parts
1,2,3,4,5,6,8,9,10 all at once. To install it, copy it to
/usr/lib/kbd/consolefonts/ and execute
"/usr/bin/setfont /usr/lib/kbd/consolefonts/LatArCyrHeb-14.psf".

If you want cut&amp;paste to work with UTF-8 consoles, you need the patch
<htmlurl url="linux-2.3.12-console.diff"
        name="linux-2.3.12-console.diff">
from Edmund Thomas Grimley Evans and Stanislav Voronyi.

<sect1>X11 Foreign fonts
<p>

Don't hesitate to install Cyrillic, Chinese, Japanese etc. fonts. Even
if they are not Unicode fonts, they will help in displaying Unicode
documents: at least Netscape Communicator 4 and Java will make use of
foreign fonts when available.

The following programs are useful when installing fonts:
<itemize>
<item>
    "mkfontdir directory"
    prepares a font directory for use by the X server, needs to be executed
    after installing fonts in a directory.
<item>
    "xset fp+ directory"
    adds a directory to the X server's current font path.
    To add a directory permanently, add a "FontPath" line to your
    /etc/XF86Config file, in section "Files".
<item>
    "xset fp rehash"
    needs to be executed after calling mkfontdir on a directory that is
    already contained in the X server's current font path.
<item>
    "xfontsel"
    allows you to browse the installed fonts by selecting various font
    properties.
<item>
    "xlsfonts -fn fontpattern"
    lists all fonts matching a font pattern. Also displays various font
    properties. In particular, "xlsfonts -ll -fn font" lists the font
    properties CHARSET_REGISTRY and CHARSET_ENCODING, which together
    determine the font's encoding.
<item>
    "xfd -fn font"
    displays a font page by page.
</itemize>

The following fonts are freely available (not a complete list):
<itemize>
<item>
    The ones contained in XFree86, sometimes packaged in separate packages.
    For example, SuSE has only normal 75dpi fonts in the base `xf86' package.
    The other fonts are in the packages `xfnt100', `xfntbig', `xfntcyr',
    `xfntscl'.
<item>
    The Emacs international fonts,
    <htmlurl url="ftp://ftp.gnu.org/pub/gnu/intlfonts/intlfonts-1.1.tar.gz"
            name="ftp://ftp.gnu.org/pub/gnu/intlfonts/intlfonts-1.1.tar.gz">
    As already mentioned, they are useful even if you prefer XEmacs to
    GNU Emacs or don't use any Emacs at all.
</itemize>

<sect1>X11 Unicode fonts
<p>

Application wishing to display text belonging to different scripts (like
Cyrillic and Greek) at the same time, can do so by using different X fonts
for the various pieces of text. This is what Netscape Communicator and Java
do. However, this approach is more complicated, because instead of working
with `Font' and `XFontStruct', the programmer has to deal with `XFontSet',
and also because not all fonts in the font set need to have the same
dimensions.

<itemize>
<item>
  Markus Kuhn has assembled fixed-width 75dpi fonts with Unicode encoding
  covering Latin, Greek, Cyrillic, Armenian, Georgian, Hebrew, Symbol scripts.
  They cover ISO 8859 parts 1,2,3,4,5,7,8,9,10,13,14,15 all at once.
  This font is required for running xterm in utf-8 mode.
  <htmlurl url="http://www.cl.cam.ac.uk/~mgk25/download/ucs-fonts.tar.gz"
          name="http://www.cl.cam.ac.uk/~mgk25/download/ucs-fonts.tar.gz">
<item>
  Roman Czyborra has assembled an 8x16 / 16x16 75dpi font with Unicode encoding
  covering a huge part of Unicode. Download unifont.hex.gz and hex2bdf from
  <htmlurl url="http://czyborra.com/unifont/"
          name="http://czyborra.com/unifont/">.
  It is not fixed-width: 8 pixels wide for European characters, 16 pixels wide
  for Chinese characters. Installation instructions:
<tscreen><verb>
$ gunzip unifont.hex.gz
$ hex2bdf < unifont.hex > unifont.bdf
$ bdftopcf -o unifont.pcf unifont.bdf
$ gzip -9 unifont.pcf
# cp unifont.pcf.gz /usr/X11R6/lib/X11/fonts/misc
# cd /usr/X11R6/lib/X11/fonts/misc
# mkfontdir
# xset fp rehash
</verb></tscreen>
<item>
  Primoz Peterlin has assembled an ETL family fonts covering Latin, Greek,
  Cyrillic, Armenian, Georgian, Hebrew scripts.
  <htmlurl url="ftp://ftp.x.org/contrib/fonts/etl-unicode.tar.gz"
          name="ftp://ftp.x.org/contrib/fonts/etl-unicode.tar.gz">
  Use the "bdftopcf" program in order to install it.
</itemize>

<sect1>Unicode xterm
<p>

xterm is part of X11R6 and XFree86, but is maintained separately by Tom
Dickey.
<htmlurl url="http://www.clark.net/pub/dickey/xterm/xterm.html"
        name="http://www.clark.net/pub/dickey/xterm/xterm.html">
Newer versions (patch level 109 and above) contain support for converting
keystrokes to UTF-8 before sending them to the application running in the
xterm, and for displaying Unicode characters that the application outputs
as UTF-8 byte sequence.

To get an UTF-8 xterm running, you need to:
<itemize>
  Fetch
  <htmlurl url="http://www.clark.net/pub/dickey/xterm/xterm.tar.gz"
          name="http://www.clark.net/pub/dickey/xterm/xterm.tar.gz">,
<item>
  Configure it by calling "./configure --enable-wide-chars ...", then
  compile and install it.
<item>
  Have a Unicode fixed-width font installed. Markus Kuhn's ucs-fonts.tar.gz
  (see above) is made for this.
<item>
  Start "xterm -u8 -fn fixed".
  The option "-u8" turns on Unicode and UTF-8 handling. The "fixed" font
  is Markus Kuhn's Unicode font.
<item>
  Take a look at the sample files contained in Markus Kuhn's ucs-fonts
  package:
<tscreen><verb>
$ cd .../ucs-fonts
$ cat quickbrown.txt
$ cat utf-8-demo.txt
</verb></tscreen>
  You should be seeing (among others) greek and russian characters.
<item>
  To make xterm come up with UTF-8 handling each time it is started,
  add the line
    XTerm*utf8:   1
  to your $HOME/.Xdefaults (for yourself only). I don't recommend changing
  the system-wide /usr/X11R6/lib/X11/app-defaults/XTerm, because then your
  changes will be erased next time you upgrade to a new XFree86 version.
<item>
  If you also changed the font name, you need a line of the form
    *VT100*font:  your-font
  in your $HOME/.Xdefaults. For "fixed" it is not needed, because "fixed"
  is the default anyway.
</itemize>

<sect1>Miscellaneous
<p>

A small program which tests whether a Linux console or xterm is in UTF-8 mode
can be found in the
<htmlurl url="ftp://sunsite.unc.edu/pub/Linux/system/keyboards/x-lt-1.18.tar.gz"
        name="ftp://sunsite.unc.edu/pub/Linux/system/keyboards/x-lt-1.18.tar.gz">
package by Ricardas Cepas, files testUTF-8.c and testUTF8.c.


<sect>Locale setup
<p>

<sect1>Files &amp; the kernel
<p>

You can now already use any Unicode characters in file names. No kernel
or file utilities need modifications. This is because file names in the
kernel can be anything not containing a null byte, and '/' is used to
delimit subdirectories. When encoded using UTF-8, non-ASCII characters
will never be encoded using null bytes or slashes. All that happens is
that file and directory names occupy more bytes than they contain characters.
For example, a filename consisting of five greek characters will appear
to the kernel as a 10-byte filename. The kernel does not know (and does
not need to know) that these bytes are displayed as greek.

This is the general theory, as long as your files stay inside Linux. On
filesystems which are used from other operating systems, you have mount
options to control conversion of filenames to/from UTF-8:
<itemize>
<item>
    The "vfat" filesystems has a mount option "utf8".
    See <htmlurl url="file:/usr/src/linux/Documentation/filesystems/vfat.txt"
                name="file:/usr/src/linux/Documentation/filesystems/vfat.txt">.
    When you give an "iocharset" mount option different from the default
    (which is "iso8859-1"), the results with and without "utf8" are not
    consistent. Therefore I don't recommend the "iocharset" mount option.
<item>
    The "msdos", "umsdos" filesystems have the same mount option, but it
    appears to have no effect.
<item>
    The "iso9660" filesystem has a mount option "utf8".
    See <htmlurl url="file:/usr/src/linux/Documentation/filesystems/isofs.txt"
                name="file:/usr/src/linux/Documentation/filesystems/isofs.txt">.
<item>
    Since Linux 2.2.x kernels, the "ntfs" filesystem has a mount option
    "utf8". See
    <htmlurl url="file:/usr/src/linux/Documentation/filesystems/ntfs.txt"
            name="file:/usr/src/linux/Documentation/filesystems/ntfs.txt">.
</itemize>
The other filesystems (nfs, smbfs, ncpfs, hpfs, etc.) don't convert
filenames; therefore they support Unicode file names in UTF-8 encoding only
if the other operating system supports them.
Recall that to enable a mount option for all future remount, you add to
the fourth column of the corresponding /etc/fstab line.

<sect1>Ttys &amp; the kernel
<p>

Ttys are some kind of bidirectional pipes between two program, allowing
fancy features like echoing or command-line editing. When in an xterm,
you execute the "cat" command without arguments, you can enter and edit
any number of lines, and they will be echoed back line by line. The
kernel's editing actions are not correct, especially the Backspace (erase)
key and the tab bey are not treated correctly.

To fix this, you need to:
<itemize>
<item>
  apply the kernel patch
  <htmlurl url="linux-2.0.35-tty.diff"
          name="linux-2.0.35-tty.diff"> or
  <htmlurl url="linux-2.2.9-tty.diff"
          name="linux-2.2.9-tty.diff"> or
  <htmlurl url="linux-2.3.12-tty.diff"
          name="linux-2.3.12-tty.diff">
  and recompile your kernel,
<item>
  if you are using glibc2, apply the patch
  <htmlurl url="glibc211-tty.diff"
          name="glibc211-tty.diff">
  and recompile your libc (or if you are not so adventurous, it is sufficient
  to patch an already installed include file:
  <htmlurl url="glibc-tty.diff"
          name="glibc-tty.diff">),
<item>
  apply the patch <htmlurl url="stty.diff"
                          name="stty.diff">
  to GNU sh-utils-1.16b, and rebuild the "stty" program, then test it using
  "stty -a" and "stty iutf8".
<item>
  add the command "stty iutf8" to the "unicode_start" script, and
  add the command "stty -iutf8" to the "unicode_stop script.
<item>
  apply the patch <htmlurl url="xterm.diff"
                          name="xterm.diff">
  to xterm-109, and rebuild "xterm", then test it by starting
  "xterm -u8"/"xterm +u8" and running "stty -a" and interactive "cat" inside it.
</itemize>

To make this fix persistent across rlogin and telnet, you also need to:
<itemize>
<item>
  Define new values for the TERM environment variable, "linux-utf8" as an
  alias to "linux", and "xterm-utf8" as an alias to "xterm".
  If your system has the ncurses library and the /usr/lib/terminfo (or
  /usr/share/terminfo) database, do this by running
<tscreen><verb>
$ tic linux-utf8.terminfo
$ tic xterm-utf8.terminfo
</verb></tscreen>
  as non-root (this will create the terminfo entries in your $HOME/.terminfo
  directory). Here are
  <htmlurl url="linux-utf8.terminfo" name="linux-utf8.terminfo"> and
  <htmlurl url="xterm-utf8.terminfo" name="xterm-utf8.terminfo">.
  I don't recommend running this as root, because it will create
  the terminfo entries in /usr/lib/terminfo where they might be erased next
  time you upgrade your system.
  If your system has an /etc/termcap file, you should also edit that file:
  copy the linux and xterm entries and give them the new names "linux-utf8"
  and "xterm-utf8". For an example, see
  <htmlurl url="termcap.diff"
          name="termcap.diff">.
<item>
  Each time you call "unicode_start" or "unicode_stop" from the console, also
  execute "export TERM=linux-utf8" or "export TERM=linux", respectively.
<item>
  Apply the patch <htmlurl url="xterm2.diff"
                          name="xterm2.diff">
  to xterm-109, rebuild "xterm", and remove any "XTerm*termName" line from
  /usr/X11R6/lib/X11/app-defaults/XTerm and $HOME/.Xdefaults. Now xterm sets
  the TERM variable to "xterm-utf8" instead of "xterm" when running in UTF-8
  mode.
<item>
  Apply the patches
  <htmlurl url="netkit.diff"
          name="netkit.diff">,
  <htmlurl url="netkitb.diff"
          name="netkitb.diff"> and
  <htmlurl url="telnet.diff"
          name="telnet.diff"> and
  rebuild "rlogind" and "telnetd". Now rlogin and telnet put the tty into
  UTF-8 editing mode whenever the TERM environment variable is "linux-utf8"
  or "xterm-utf8".
</itemize>

<sect1>General data conversion
<p>

You will need a program to convert your locally (probably ISO-8859-1) encoded
texts to UTF-8. (The alternative would be to keep using texts in different
encodings on the same machine; this is not fun in the long run.)
One such program is `iconv', which comes with glibc-2.1. Simply use
<tscreen><verb>
$ iconv --from-code=ISO-8859-1 --to-code=UTF-8 &lt; old_file &gt; new_file
</verb></tscreen>

Here are two handy shell scripts, called
"i2u" <htmlurl url="i2u.sh" name="i2u.sh"> (for ISO to UTF conversion) and
"u2i" <htmlurl url="u2i.sh" name="u2i.sh"> (for UTF to ISO conversion).
Adapt according to your current 8-bit character set.

If you don't have glibc-2.1 and iconv installed, you can use GNU recode 3.5
instead.
"i2u" <htmlurl url="i2u_recode.sh" name="i2u_recode.sh"> is
"recode ISO-8859-1..UTF-8", and
"u2i" <htmlurl url="u2i_recode.sh" name="u2i_recode.sh"> is
"recode UTF-8..ISO-8859-1".
<htmlurl url="ftp://ftp.iro.umontreal.ca/pub/recode/recode-3.5.tar.gz"
        name="ftp://ftp.iro.umontreal.ca/pub/recode/recode-3.5.tar.gz">
<htmlurl url="ftp://ftp.gnu.org/pub/gnu/recode/recode-3.5.tar.gz"
        name="ftp://ftp.gnu.org/pub/gnu/recode/recode-3.5.tar.gz">
Notes: You need GNU recode 3.5 or newer. To compile GNU recode 3.5 on
platforms without glibc2 (i.e. on all platforms except recent Linux systems),
you need to configure it with --disable-nls, otherwise it won't link.

Or you can also use CLISP instead. Here are
"i2u" <htmlurl url="i2u.lsp" name="i2u.lsp"> and
"u2i" <htmlurl url="u2i.lsp" name="u2i.lsp">
written in Lisp. Note: You need a CLISP version from July 1999 or newer.
<htmlurl url="ftp://clisp.cons.org/pub/lisp/clisp/source/clispsrc.tar.gz"
        name="ftp://clisp.cons.org/pub/lisp/clisp/source/clispsrc.tar.gz">.

Other data conversion programs, less powerful than GNU recode, are
`trans'
<htmlurl url="ftp://ftp.informatik.uni-erlangen.de/pub/doc/ISO/charsets/trans113.tar.gz"
        name="ftp://ftp.informatik.uni-erlangen.de/pub/doc/ISO/charsets/trans113.tar.gz">,
`tcs' from the Plan9 operating system
<htmlurl url="ftp://ftp.informatik.uni-erlangen.de/pub/doc/ISO/charsets/tcs.tar.gz"
        name="ftp://ftp.informatik.uni-erlangen.de/pub/doc/ISO/charsets/tcs.tar.gz">,
and
`utrans'/`uhtrans'/`hutrans'
<htmlurl url="ftp://ftp.cdrom.com/pub/FreeBSD/distfiles/i18ntools-1.0.tar.gz"
        name="ftp://ftp.cdrom.com/pub/FreeBSD/distfiles/i18ntools-1.0.tar.gz">
by G. Adam Stanislav
<htmlurl url="mailto:adam@whizkidtech.net"
           name="&lt;adam@whizkidtech.net&gt;">.

<sect1>Locale environment variables
<p>

You may have the following environment variables set, containing locale
names:
<descrip>
<tag>LANGUAGE</tag>
   override for LC_MESSAGES, used by GNU gettext only
<tag>LC_ALL</tag>
   override for all other LC_* variables
<tag>LC_CTYPE, LC_MESSAGES, LC_COLLATE, LC_NUMERIC, LC_MONETARY, LC_TIME</tag>
   individual variables for:
   character types and encoding,
   natural language messages,
   sorting rules,
   number formatting,
   money amount formatting,
   date and time display
<tag>LANG</tag>
   default value for all LC_* variables
</descrip>
(See `<tt>man 7 locale</tt>' for a detailed description.)

Each of the LC_* and LANG variables can contain a locale name of the
following form:

<quote>
    language[_territory[.codeset]][@modifier]
</quote>

where language is an <htmlurl url="ISO_639" name="ISO 639"> language code
(lower case), territory is an <htmlurl url="ISO_3166" name="ISO 3166">
country code (upper case), codeset denotes a character set, and
modifier stands for other particular attributes (for example indicating
a particular language dialect, or a nonstandard orthography).

LANGUAGE can contain several locale names, separated by colons.

In order to tell your system and all applications that you are using UTF-8,
you need to add a codeset suffix of UTF-8 to your locale names. For example,
if you were using
<tscreen><verb>
LANGUAGE=de:fr:en
LC_CTYPE=de_DE
</verb></tscreen>
you would change this to
<tscreen><verb>
LANGUAGE=de.UTF-8:fr.UTF-8:en.UTF-8
LC_CTYPE=de_DE.UTF-8
</verb></tscreen>

<sect1>Creating the locale support files
<p>

If you have glibc-2.1 or glibc-2.1.1 or glibc-2.1.2 installed, first check
using "localedef --help" that the system directory for character maps is
/usr/share/i18n/charmaps. Then apply to the file /usr/share/i18n/charmaps/UTF8
the patch <htmlurl url="glibc21.diff" name="glibc21.diff">
or <htmlurl url="glibc211.diff" name="glibc211.diff">
or <htmlurl url="glibc212.diff" name="glibc212.diff">, respectively.
Then create the support files for each UTF-8 locale you intend to use, for
example:
<tscreen><verb>
$ localedef -v -c -i de_DE -f UTF8 /usr/share/locale/de_DE.UTF-8
</verb></tscreen>

You typically don't need to create locales named "de" or "fr" without
country suffix, because these locales are normally only used by the
LANGUAGE variable and not by the LC_* variables, and LANGUAGE is only
used as an override for LC_MESSAGES.

<sect1>Adding support to the C library
<p>

The glibc-2.2 will support multibyte locales, in particular the UTF-8 locales
created above. But glibc-2.1 or glibc-2.1.1 do not really support it.
Therefore the only real effect of the above creation of the
/usr/share/locale/de_DE.UTF-8/* files is that `setlocale(LC_ALL,"")'
will return "de_DE.UTF-8", according to your environment variables, instead
of stripping off the ".UTF-8" suffix.

To add support for the UTF-8 locale, you should build and install the
`libutf8_plug.so' library, from
<htmlurl url="libutf8-0.5.2.tar.gz"
        name="libutf8-0.5.2.tar.gz">.
Then you can set the LD_PRELOAD environment variable to point to the
installed library:
<tscreen><verb>
$ export LD_PRELOAD=/usr/local/lib/libutf8_plug.so
</verb></tscreen>
Then, in every program launched with this environment variable set, the
functions in libutf8_plug.so will override the original ones in
/lib/libc.so.6. For more info about LD_PRELOAD, see "man 8 ld.so".

This entire thing will not be necessary any more once glibc-2.2 comes out.

<sect1>Converting the message catalogs
<p>

Now let's fill these new locales with content. The following /bin/sh
commands can convert your message catalogs to UTF-8 format. They must
be run as root, and require the programs `msgfmt' and `msgunfmt' from
GNU gettext-0.10.35 to be installed.
<htmlurl url="convert-msgcat.sh"
        name="convert-msgcat.sh">

This too will not be necessary any more once glibc-2.2 comes out, because
by then, the gettext function will convert the strings appropriately from
the translator's character set to the user's character set, using either
iconv or librecode.

<sect>Specific applications
<p>

<sect1>Networking
<p>

<sect2>rlogin
<p>

is fine with the above mentioned patches.

<sect2>telnet
<p>

telnet is not 8-bit clean by default. In order to be able to send Unicode
keystrokes to the remote host, you need to set telnet into "outbinary" mode.
There are two ways to do this:
<tscreen><verb>
$ telnet -L <host>
</verb></tscreen>
and
<tscreen><verb>
$ telnet
telnet> set outbinary
telnet> open <host>
</verb></tscreen>

Additionally, use the above mentioned patches.

<sect1>Browsers
<p>

<sect2>Netscape
<p>

Netscape 4.05 or newer can display HTML documents in UTF-8 encoding. All a
document needs is the following line between the
&lt;head&gt; and &lt;/head&gt; tags:
<tscreen><verb>
<meta http-equiv="Content-Type" content="text/html; charset=UTF-8">
</verb></tscreen>

Netscape 4.05 or newer can also display HTML and text files in UCS-2
encoding with byte-order mark.

<htmlurl url="http://www.netscape.com/computing/download/"
        name="http://www.netscape.com/computing/download/">

<sect2>lynx
<p>

lynx-2.8 has an options screen (key 'O') which permits to set the display
character set. When running in an xterm or Linux console in UTF-8 mode,
set this to "UNICODE UTF-8".

Now, again, all a document needs is the following line between the
&lt;head&gt; and &lt;/head&gt; tags:
<tscreen><verb>
<meta http-equiv="Content-Type" content="text/html; charset=UTF-8">
</verb></tscreen>

When you are viewing text files in UTF-8 encoding, you also need to
pass the command-line option "-assume_local_charset=UTF-8" (affects only
file:/... URLs) or "-assume_charset=UTF-8" (affects all URLs).
In lynx-2.8.2 you can alternatively, in the options screen (key 'O'),
change the assumed document character set to "utf-8".

There is also an option in the options screen, to set the "preferred document
character set". But it has no effect, at least with file:/... URLs
and with http://... URLs served by apache-1.3.0.

There is a spacing and line-breaking problem, however. (Look at the
russian section of x-utf8.html, or at utf-8-demo.txt.)

Also, in lynx-2.8.2, configured with --enable-prettysrc, the nice colour
scheme does not work correctly any more when the display character set
has been set to "UNICODE UTF-8". This is fixed by a simple patch
<htmlurl url="lynx282.diff" name="lynx282.diff">.

The Lynx developers say: "For any serious use of UTF-8 screen output with
lynx, compiling with slang lib and -DSLANG_MBCS_HACK is still recommended."

<htmlurl url="ftp://ftp.gnu.org/pub/gnu/lynx/lynx-2.8.2.tar.gz"
        name="ftp://ftp.gnu.org/pub/gnu/lynx/lynx-2.8.2.tar.gz">
<htmlurl url="http://lynx.browser.org/"
        name="http://lynx.browser.org/">
<htmlurl url="http://www.slcc.edu/lynx/"
        name="http://www.slcc.edu/lynx/">
<htmlurl url="ftp://lynx.isc.org/"
        name="ftp://lynx.isc.org/">

<sect2>Test pages
<p>

Some test pages for browsers can be found at the pages of Alan Wood
<htmlurl url="http://www.hclrss.demon.co.uk/unicode/#links"
        name="http://www.hclrss.demon.co.uk/unicode/#links">
and James Kass
<htmlurl url="http://home.att.net/~jameskass/"
        name="http://home.att.net/~jameskass/">.

<sect1>Editors
<p>

<sect2>yudit
<p>

yudit by G&aacute;sp&aacute;r Sinai
<htmlurl url="http://czyborra.com/yudit/"
        name="http://czyborra.com/yudit/">
is a first-class unicode text editor for the X Window System.
It supports simultaneous processing of many languages, input methods,
conversions for local character standards.
It has facilities for entering text in all languages with only
an English keyboard, using keyboard configuration maps.

It can be compiled in three versions: Xlib GUI, KDE GUI, or Motif GUI.

Customization is very easy. Typically you will first customize your font.
From the font menu I chose "Unicode". Then, since the command
"xlsfonts '*-*-iso10646-1'" still showed some ambiguity, I chose a font
size of 13 (to match Markus Kuhn's 13-pixel fixed font).

Next, you will customize your input method. The input methods "Straight",
"Unicode" and "SGML" are most remarkable. For details about the other
built-in input methods, look in /usr/local/share/yudit/data/.

To make a change the default for the next session, edit your $HOME/.yuditrc
file.

The general editor functionality is limited to editing, cut&amp;paste
and search&amp;replace. No undo.

<sect2>mined98
<p>

mined98 is a small text editor by Michiel Huisjes, Achim M&uuml;ller and
Thomas Wolff.
<htmlurl url="http://www.inf.fu-berlin.de/~wolff/mined.html"
        name="http://www.inf.fu-berlin.de/~wolff/mined.html">
It lets you edit UTF-8 or 8-bit encoded files, in an UTF-8 or 8-bit xterm.
It also has powerful capabilities for entering Unicode characters.

When it is running in xterm or Linux console in UTF-8 mode, you should
set the environment variable <tt>utf8_term</tt>, or call mined with the
command-line option <tt>-U</tt>.

mined lets you edit both 8-bit encoded and UTF-8 encoded files. By default
it uses an autodetection heuristic. If you don't want to rely on heuristics,
pass the command-line option <tt>-u</tt> when editing an UTF-8 file, or
<tt>+u</tt> when editing an 8-bit encoded file. You can change the
interpretation at any time from within the editor: It displays the encoding
("L:h" for 8-bit, "U:h" for UTF-8) in the menu line. Click on the first
of these characters to change it.

A few caveats:
<itemize>
<item>
  The Linux binary in the distribution is out of date and does not support
  UTF-8. You have to rebuild a binary from the source. Then
  install src/mined as /usr/local/bin/mined, and
  install doc/mined.help as /usr/local/man/cat1/mined.1, so that
  <tt>ESC h</tt> will find it.
<item>
  mined ignores your "stty erase" setting. When your backspace key sends
  ASCII code 127, and you have set "stty erase ^?", -- these are the ultimately
  correct settings -- you have to call mined with option <tt>-B</tt> in order
  to get the backspace key erase a character to the left of the cursor.
<item>
  The "Home", "End", "Delete" keys do not work.
</itemize>

<sect2>vim
<p>

vim (as of version 5.4m) has some support for multi-byte locales, but only
as far as the X library has the same support, and only for encodings with
at most two bytes per character (i.e. ISO-2022 encodings). No UTF-8 support.

<sect2>emacs
<p>

First of all, you should read the section "International Character Set Support"
(node "International") in the Emacs manual. In particular, note that you need
to start Emacs using the command
<tscreen><verb>
$ emacs -fn fontset-standard
</verb></tscreen>
so that it will use a font set comprising a lot of international characters.

In the short term, the emacs-utf package
<htmlurl url="http://www.cs.ust.hk/faculty/otfried/Mule/"
        name="http://www.cs.ust.hk/faculty/otfried/Mule/">
by Otfried Cheong provides a "unicode-utf8" encoding to Emacs.
After compiling the program "utf2mule" and installing it somewhere in your
$PATH, also install unicode.el, muleuni-1.el, unicode-char.el somewhere, and
add the lines
<tscreen><verb>
(setq load-path (cons "/home/user/somewhere/emacs" load-path))
(if (not (string-match "XEmacs" emacs-version))
  (progn
    (require 'unicode)
    (if (eq window-system 'x)
      (progn
        (create-fontset-from-fontset-spec
          "-misc-fixed-medium-r-normal-*-12-*-*-*-*-*-fontset-standard")
        (create-fontset-from-fontset-spec
          "-misc-fixed-medium-r-normal-*-13-*-*-*-*-*-fontset-standard")
        (create-fontset-from-fontset-spec
          "-misc-fixed-medium-r-normal-*-14-*-*-*-*-*-fontset-standard")
        (create-fontset-from-fontset-spec
          "-misc-fixed-medium-r-normal-*-15-*-*-*-*-*-fontset-standard")
        (create-fontset-from-fontset-spec
          "-misc-fixed-medium-r-normal-*-16-*-*-*-*-*-fontset-standard")
        (create-fontset-from-fontset-spec
          "-misc-fixed-medium-r-normal-*-18-*-*-*-*-*-fontset-standard")))))
</verb></tscreen>
to your $HOME/.emacs file. To activate any of the font sets, use the Mule
menu item "Set Font/FontSet" or Shift-down-mouse-1. Currently the font sets
with height 15 and 13 have the best Unicode coverage, due to Markus Kuhn's
9x15 and 6x13 fonts.
In order to open an UTF-8 encoded file, you will type
<tscreen><verb>
M-x universal-coding-system-argument unicode-utf8 RET
M-x find-file filename RET
</verb></tscreen>
or
<tscreen><verb>
C-x RET c unicode-utf8 RET
C-x C-f filename RET
</verb></tscreen>

Note that this works with Emacs in windowing mode only, not in terminal mode.

Richard Stallman plans to add integrated UTF-8 support to Emacs in the long
term.

<sect2>xemacs
<p>

(This section is written by Gilbert Baumann.)

Here is how to teach XEmacs (20.4 configured with MULE) the UTF-8 encoding.
Unfortunately you need its sources to be able to patch it.

First you need these files provided by Tomohiko Morioka:

<htmlurl url="http://turnbull.sk.tsukuba.ac.jp/Tools/XEmacs/xemacs-21.0-b55-emc-b55-ucs.diff"
        name="http://turnbull.sk.tsukuba.ac.jp/Tools/XEmacs/xemacs-21.0-b55-emc-b55-ucs.diff">
and
<htmlurl url="http://turnbull.sk.tsukuba.ac.jp/Tools/XEmacs/xemacs-ucs-conv-0.1.tar.gz"
        name="http://turnbull.sk.tsukuba.ac.jp/Tools/XEmacs/xemacs-ucs-conv-0.1.tar.gz">

The .diff is a diff against the C sources. The tar ball is elisp code, 
which provides lots of code tables to map to and from Unicode. As the
name of the diff file suggests it is against XEmacs-21; I needed to
help `patch' a bit. The most notable difference to my XEmacs-20.4
sources is that file-coding.[ch] was called mule-coding.[ch].

For those unfamilar with the XEmacs-MULE stuff (as I am) a quick
guide:

What we call an encoding is called by MULE a `coding-system'. The most
important commands are:

<tscreen><verb>
M-x set-file-coding-system
M-x set-buffer-process-coding-system   [comint buffers]
</verb></tscreen>

and the variable `file-coding-system-alist', which guides `find-file'
to guess the encoding used. After stuff was running, the very first
thing I did was <htmlurl url="gb-hacks.el" name="this">.

This code looks into the special mode line introduced by -*- somewhere
in the first 600 bytes of the file about to opened; if now there is a
field "Encoding: xyz;" and the xyz encoding ("coding system" in Emacs speak)
exists, choose that. So now you could do e.g.

<tscreen><verb>
;;; -*- Mode: Lisp; Syntax: Common-Lisp; Package: CLEX; Encoding: utf-8; -*-
</verb></tscreen>

and XEmacs goes into utf-8 mode here.

Atfer everything was running I defined \u03BB (greek lambda) as a
macro like:

<tscreen><verb>
(defmacro \u03BB (x) `(lambda .,x))
</verb></tscreen>

<sect2>nedit
<p>

<sect2>xedit
<p>

In theory, xedit should be able to edit UTF-8 files if you set the locale
accordingly (see above), and add the line "Xedit*international: true" to
your $HOME/.Xdefaults file. In practice, it will recognize UTF-8 encoding of
non-ASCII characters, but will display them as sequences of "@" characters.

<sect2>axe
<p>

In theory, axe should be able to edit UTF-8 files if you set the locale
accordingly (see above), and add the line "Axe*international: true" to
your $HOME/.Xdefaults file. In practice, it will simply dump core.

<sect2>pico
<p>

<sect2>TeX
<p>

The teTeX 0.9 (and newer) distribution contains an Unicode adaptation of TeX,
called Omega
(<htmlurl url="http://www.gutenberg.eu.org/omega/"
         name="http://www.gutenberg.eu.org/omega/">,
<htmlurl url="ftp://ftp.ens.fr/pub/tex/yannis/omega"
        name="ftp://ftp.ens.fr/pub/tex/yannis/omega">),
but can someone point me to a tutorial for using this system?

Other maybe related links:
<htmlurl url="http://www.dante.de/projekte/nts/NTS-FAQ.html"
        name="http://www.dante.de/projekte/nts/NTS-FAQ.html">,
<htmlurl url="ftp://ftp.dante.de/pub/tex/language/chinese/CJK/"
        name="ftp://ftp.dante.de/pub/tex/language/chinese/CJK/">.

<sect1>Mailers
<p>

MIME: RFC 2279 defines UTF-8 as a MIME charset, which can be transported
under the 8bit, quoted-printable and base64 encodings. The older MIME
UTF-7 proposal (RFC 2152) is considered to be deprecated and should not
be used any further.

Mail clients released after January 1, 1999, should be capable of sending and
displaying UTF-8 encoded mails, otherwise they are considered deficient.
But these mails have to carry the MIME labels
<tscreen><verb>
Content-Type: text/plain; charset=UTF-8
Content-Transfer-Encoding: 8bit
</verb></tscreen>
Simply piping an UTF-8 file into "mail" without caring about the MIME labels
will not work.

Mail client implementors should take a look at
<htmlurl url="http://www.imc.org/imc-intl/"
        name="http://www.imc.org/imc-intl/">
and
<htmlurl url="http://www.imc.org/mail-i18n.html"
        name="http://www.imc.org/mail-i18n.html">.

Now about the individual mail clients (or "mail user agents"):

<sect2>pine
<p>

The situation for an unpatched pine version 4.10 is as follows.

Pine does not do character set conversions. But it allows you to view
UTF-8 mails in an UTF-8 text window (Linux console or xterm).

Normally, Pine will warn about different character sets each time you view
an UTF-8 encoded mail. To get rid of this warning, choose S (setup), then
C (config), then change the value of "character-set" to UTF-8. This option
will not do anything, except to reduce the warnings, as Pine has no built-in
knowledge of UTF-8.

Also note that Pine's notion of Unicode characters is pretty limited: It
will display Latin and Greek characters, but not other kinds of Unicode
characters.

A patch by Robert Brady
<htmlurl url="http://www.ents.susu.soton.ac.uk/~robert/pine-utf8-0.1.diff"
        name="http://www.ents.susu.soton.ac.uk/~robert/pine-utf8-0.1.diff">
adds UTF-8 support to Pine. With this patch, it decodes and prints headers
and bodies properly. The patch depends on the GNOME libunicode
<htmlurl url="http://cvs.gnome.org/lxr/source/libunicode/"
        name="http://cvs.gnome.org/lxr/source/libunicode/">.

However, alignment remains broken in many places; replying to a mail does
not cause the character set to be converted as appropriate; and the editor,
pico, cannot deal with multibyte characters.

<sect2>kmail
<p>

kmail (as of KDE 1.0) does not support UTF-8 mails at all.

<sect2>Netscape Communicator
<p>

Netscape Communicator's Messenger can send and display mails in UTF-8
encoding, but it needs a little bit of manual user intervention.

To send an UTF-8 encoded mail: After opening the "Compose" window, but before
starting to compose the message, select from the menu
"View -> Character Set -> Unicode (UTF-8)". Then compose the message and
send it.

When you receive an UTF-8 encoded mail, Netscape unfortunately does not
display it in UTF-8 right away, and does not even give a visual clue that
the mail was encoded in UTF-8. You have to manually select from the menu
"View -> Character Set -> Unicode (UTF-8)".

For displaying UTF-8 mails, Netscape uses different fonts. You can adjust
your font settings in the "Edit -> Preferences -> Fonts" dialog; choose
the "Unicode" font category.

<sect2>emacs (rmail, vm)
<p>

<sect1>Other text-mode applications
<p>

<sect2>less
<p>

Get
<htmlurl url="ftp://ftp.gnu.org/pub/gnu/less/less-340.tar.gz"
        name="ftp://ftp.gnu.org/pub/gnu/less/less-340.tar.gz">
and apply the patch
<htmlurl url="less-340-utf-2.diff"
        name="less-340-utf-2.diff">
by Robert Brady
<htmlurl url="mailto:rwb197@ecs.soton.ac.uk"
           name="&lt;rwb197@ecs.soton.ac.uk&gt;">.
Then set the environment variable LESSCHARSET:
<tscreen><verb>
$ export LESSCHARSET=utf-8
</verb></tscreen>
If you also have a LESSKEY environment variable set, also make sure that
the file it points to does not define LESSCHARSET. If necessary, regenerate
this file using the `lesskey' command, or unset the LESSKEY environment
variable.

<sect2>expand, wc
<p>

Get the GNU textutils-2.0 and apply the patch
<htmlurl url="textutils-2.0.diff"
        name="textutils-2.0.diff">,
then configure, add "#define HAVE_MBRTOWC 1", "#define HAVE_FGETWC 1",
"#define HAVE_FPUTWC 1" to config.h. In src/Makefile, modify CFLAGS and LDFLAGS
so that they include the directories where libutf8 is installed. Then rebuild.

<sect2>col, colcrt, colrm, column, rev, ul
<p>

Get the util-linux-2.9y package, configure it, then define ENABLE_WIDECHAR in
defines.h, change the "#if 0" to "#if 1" in lib/widechar.h. In
text-utils/Makefile, modify CFLAGS and LDFLAGS so that they include the
directories where libutf8 is installed. Then rebuild.

<sect2>figlet
<p>

figlet 2.2 has an option for UTF-8 input: "figlet -C utf8"

<sect2>kermit
<p>

The serial communications program C-Kermit
<htmlurl url="http://www.columbia.edu/kermit/"
        name="http://www.columbia.edu/kermit/">,
in versions 7.0beta10 or newer, understands the file and transfer encodings
UTF-8 and UCS-2, and understands the terminal encoding UTF-8. Documentation
of these features can be found in
<htmlurl url="ftp://kermit.columbia.edu/kermit/test/text/ckermit2.txt"
        name="ftp://kermit.columbia.edu/kermit/test/text/ckermit2.txt">.

<sect1>Other X11 applications
<p>

X11 Xlib kann leider noch kein UTF-8 locale, das m<>sste auch noch dringend
mal gemacht werden.


<sect>Making your programs Unicode aware
<p>

<sect1>C/C++
<p>

The C `<tt>char</tt>' type is 8-bit and will stay 8-bit because it denotes
the smallest addressable data unit. Various facilities are available:

<sect2>For normal text handling
<p>

The ISO/ANSI C standard contains, in an amendment which was added in 1995,
a "wide character" type `<tt>wchar_t</tt>', a set of functions like those
found in <tt>&lt;string.h&gt;</tt> and <tt>&lt;ctype.h&gt;</tt> (declared in
<tt>&lt;wchar.h&gt;</tt> and <tt>&lt;wctype.h&gt;</tt>, respectively), and
a set of conversion functions between `<tt>char *</tt>' and
`<tt>wchar_t *</tt>' (declared in <tt>&lt;stdlib.h&gt;</tt>).

Good references for this API are
<itemize>
<item>
the GNU libc-2.1 manual, chapters 4 "Character Handling" and
6 "Character Set Handling",
<item>
the manual pages
<htmlurl url="man-mbswcs.tar.gz"
        name="man-mbswcs.tar.gz">,
<item>
the Dinkumware C library reference
<htmlurl url="http://www.dinkumware.com/htm_cl/"
        name="http://www.dinkumware.com/htm_cl/">,
<item>
the OpenGroup's Single Unix specification
<htmlurl url="http://www.UNIX-systems.org/online.html"
        name="http://www.UNIX-systems.org/online.html">.
</itemize>

Advantages of using this API:
<itemize>
<item>
It's a vendor independent standard.
<item>
The functions do the right thing, depending on the user's locale.
All a program needs to call is <tt>setlocale(LC_ALL,"");</tt>.
</itemize>

Drawbacks of this API:
<itemize>
<item>
Some of the functions are not multithread-safe, because they keep a hidden
internal state between function calls.
<item>
There is no first-class locale datatype. Therefore this API cannot reasonably
be used for anything that needs more than one locale or character set at the
same time.
<item>
The OS support for this API is not good on most OSes.
</itemize>

<sect3>Portability notes
<p>

A `<tt>wchar_t</tt>' may or may not be encoded in Unicode; this is
platform and sometimes also locale dependent. A multibyte sequence
`<tt>char *</tt>' may or may not be encoded in UTF-8; this is platform
and sometimes also locale dependent.

In detail, here is what the
<htmlurl url="http://www.UNIX-systems.org/online.html"
        name="Single Unix specification">
says about the `<tt>wchar_t</tt>' type:
<em>All wide-character codes in a given process consist of an equal number
  of bits. This is in contrast to characters, which can consist of a
  variable number of bytes. The byte or byte sequence that represents a 
  character can also be represented as a wide-character code.  
  Wide-character codes thus provide a uniform size for manipulating text
  data. A wide-character code having all bits zero is the null
  wide-character code, and terminates wide-character strings. The
  wide-character value for each member of the Portable Character Set
</em> (i.e. ASCII) <em>
  will equal its value when used as the lone character in an integer
  character constant. Wide-character codes for other characters are
  locale- and implementation-dependent. State shift bytes do not have a
  wide-character code representation.</em>

One particular consequence is that in portable programs you shouldn't use
non-ASCII characters in string literals. That means, even though you
know the Unicode double quotation marks have the codes U+201C and U+201D,
you shouldn't write a string literal <tt>L"\u201cHello\u201d, he said"</tt>
or <tt>"\xe2\x80\x9cHello\xe2\x80\x9d, he said"</tt> in C programs. Instead,
use GNU gettext, write it as <tt>gettext("'Hello', he said")</tt>, and create
a message database en.UTF-8.po which translates "'Hello', he said" to
"\u201cHello\u201d, he said".

Here is a survey of the portability of the ISO/ANSI C facilities on various
Unix flavours. GNU glibc-2.2 will support all of it, but for now we have
the following picture.

<descrip>
<tag>GNU glibc-2.0.x, glibc-2.1.x</tag>
  <itemize>
    <item>&lt;wchar.h&gt; and &lt;wctype.h&gt; exist.
    <item>Has wcs/mbs functions, but no fgetwc/fputwc/wprintf.
    <item>No UTF-8 locale.
    <item>mbrtowc returns EILSEQ for bytes &gt;= 0x80.
  </itemize>
<tag>Solaris 2.7</tag>
  <itemize>
    <item>&lt;wchar.h&gt; and &lt;wctype.h&gt; exist.
    <item>Has wcs/mbs functions, fgetwc/fputwc/wprintf, everything.
    <item>Has the following UTF-8 locales:
    en_US.UTF-8, de.UTF-8, es.UTF-8, fr.UTF-8, it.UTF-8, sv.UTF-8.
    <item>mbrtowc returns EILSEQ for bytes &gt;= 0x80.
  </itemize>
<tag>OSF/1 4.0d</tag>
  <itemize>
    <item>&lt;wchar.h&gt; and &lt;wctype.h&gt; exist.
    <item>Has wcs/mbs functions, fgetwc/fputwc/wprintf, everything.
    <item>Has an add-on universal.utf8@ucs4 locale, see "man 5 unicode".
    <item>mbrtowc does not know about UTF-8.
  </itemize>
<tag>Irix 6.5</tag>
  <itemize>
    <item>&lt;wchar.h&gt; and &lt;wctype.h&gt; exist.
    <item>Has wcs/mbs functions and fgetwc/fputwc, but not wprintf.
    <item>Has no multibyte locales.
    <item>Has only a dummy definition for mbstate_t.
    <item>Doesn't have mbrtowc.
  </itemize>
<tag>HP-UX 11.00</tag>
  <itemize>
    <item>&lt;wchar.h&gt; exists, &lt;wctype.h&gt; does not.
    <item>Has wcs/mbs functions and fgetwc/fputwc, but not wprintf.
    <item>Has a C.utf8 locale.
    <item>Doesn't have mbstate_t.
    <item>Doesn't have mbrtowc.
  </itemize>
<tag>AIX 4.2</tag>
  <itemize>
    <item>&lt;wchar.h&gt; exists, &lt;wctype.h&gt; does not - instead use &lt;ctype.h&gt; and &lt;wchar.h&gt;.
    <item>Has wcs/mbs functions and fgetwc/fputwc, but not wprintf.
    <item>Has the following UTF-8 locales:
    ET_EE.UTF-8, LT_LT.UTF-8, LV_LV.UTF-8, ZH_CN.UTF-8.
    <item>Doesn't have mbstate_t.
    <item>Doesn't have mbrtowc.
  </itemize>
</descrip>

As a consequence, I recommend to use the restartable and multithread-safe
wcsr/mbsr functions, forget about those systems which don't have them (Irix,
HP-UX, AIX), and use the UTF-8 locale plug-in libutf8_plug.so (see below)
on those systems which permit you to compile programs which use these
wcsr/mbsr functions (Linux, Solaris, OSF/1).

A similar advice, given by Sun in
<htmlurl url="http://www.sun.com/software/white-papers/wp-unicode/"
        name="http://www.sun.com/software/white-papers/wp-unicode/">,
section "Internationalized Applications with Unicode", is:

<em>To properly internationalize an application, use the following
guidelines:</em>
<enum>
  <item><em>Avoid direct access with Unicode. This is a task of the platform's
        internationalization framework.</em>
  <item><em>Use the POSIX model for multibyte and wide-character interfaces.</em>
  <item><em>Only call the APIs that the internationalization framework
        provides for language and cultural-specific operations.</em>
  <item><em>Remain code-set independent.</em>
</enum>

If, for some reason, in some piece of code, you really have to assume that
`wchar_t' is Unicode (for example, if you want to do special treatment of
some Unicode characters), you should make that piece of code conditional
upon the result of <tt>is_locale_utf8()</tt>. Otherwise you will mess up
your program's behaviour in different locales or other platforms. The
function <tt>is_locale_utf8</tt> is declared in
<htmlurl url="utf8locale.h" name="utf8locale.h">
and defined in
<htmlurl url="utf8locale.c" name="utf8locale.c">.

<sect3>The libutf8 library
<p>

A portable implementation of the ISO/ANSI C API, which supports 8-bit locales
and UTF-8 locales, can be found in
<htmlurl url="libutf8-0.5.2.tar.gz"
        name="libutf8-0.5.2.tar.gz">.

Advantages:
<itemize>
<item>
Unicode UTF-8 support now, portably, even on OSes whose multibyte character
support does not work or which don't have multibyte/wide character support
at all.
<item>
The same binary works in all OS supported 8-bit locales and in UTF-8 locales.
<item>
When an OS vendor adds proper multibyte character support, you can take
advantage of it by simply recompiling without -DHAVE_LIBUTF8 compiler option.
</itemize>

<sect3>The Plan9 way
<p>

The Plan9 operating system, a variant of Unix, uses UTF-8 as character
encoding in all applications. Its wide character type is called
`<tt>Rune</tt>', not `<tt>wchar_t</tt>'. Parts of its libraries, written by
Rob Pike and Howard Trickey, are available at
<htmlurl url="ftp://ftp.cdrom.com/pub/netlib/research/9libs/9libs-1.0.tar.gz"
        name="ftp://ftp.cdrom.com/pub/netlib/research/9libs/9libs-1.0.tar.gz">.
Another similar library, written by Alistair G. Crooks, is
<htmlurl url="ftp://ftp.cdrom.com/pub/NetBSD/packages/distfiles/libutf-2.10.tar.gz"
        name="ftp://ftp.cdrom.com/pub/NetBSD/packages/distfiles/libutf-2.10.tar.gz">.
In particular, each of these libraries contains an UTF-8 aware regular
expression matcher.

Drawback of this API:
<itemize>
<item>
UTF-8 is compiled in, not optional. Programs compiled in this universe lose
support for the 8-bit encodings which are still frequently used in Europe.
</itemize>

<sect2>For graphical user interface
<p>

The Qt-2.0 library
<htmlurl url="http://www.troll.no/"
        name="http://www.troll.no/">
contains a fully-Unicode QString class. You can use the member functions
QString::utf8 and QString::fromUtf8 to convert to/from UTF-8 encoded text.
The QString::ascii and QString::latin1 member functions should not be used
any more.

<sect2>For advanced text handling
<p>

The previously mentioned libraries implement Unicode aware versions of
the ASCII concepts. Here are libraries which deal with Unicode concepts,
such as titlecase (a third letter case, different from uppercase and
lowercase), distinction between punctuation and symbols, canonical
decomposition, combining classes, canonical ordering and the like.

<descrip>
<tag>ucdata-1.9</tag>
The ucdata library by Mark Leisher
<htmlurl url="http://crl.nmsu.edu/~mleisher/ucdata.html"
        name="http://crl.nmsu.edu/~mleisher/ucdata.html">
deals with character properties, case conversion, decomposition, combining
classes.

<tag>ICU</tag>
IBMs Classes for Unicode
<htmlurl url="http://www.alphaworks.ibm.com/tech/icu/"
        name="http://www.alphaworks.ibm.com/tech/icu/">.
A very comprehensive internationalization library featuring Unicode strings,
resource bundles, number formatters, date/time formatters, message formatters,
collation and more. Lots of supported locales. Portable to Unix and Win32,
but compiles out of the box only on Linux libc6, not libc5.

<tag>libunicode</tag>
The GNOME libunicode library
<htmlurl url="http://cvs.gnome.org/lxr/source/libunicode/"
        name="http://cvs.gnome.org/lxr/source/libunicode/">
by Tom Tromey and others. It covers character set conversion, character
properties, decomposition.
</descrip>

<sect2>For conversion
<p>

Two conversion libraries, which support UTF-8 and a large number of 8-bit
character sets, are available:

The iconv implementation by Ulrich Drepper, contained in the GNU glibc-2.1.1.
<htmlurl url="ftp://ftp.gnu.org/pub/gnu/glibc/glibc-2.1.1.tar.gz"
        name="ftp://ftp.gnu.org/pub/gnu/glibc/glibc-2.1.1.tar.gz">

Advantages:
<itemize>
<item>
iconv is POSIX standardized, programs using iconv to convert from/to UTF-8
will also run under Solaris. However, the names for the character sets differ
between platforms. For example, "EUC-JP" under glibc is "eucJP" under HP-UX.
(The official IANA name for this character set is "EUC-JP", so it's clearly
a HP-UX deficiency.)
<item>
No additional library needed.
</itemize>

librecode by Fran&ccedil;ois Pinard
<htmlurl url="ftp://ftp.gnu.org/pub/gnu/recode/recode-3.5.tar.gz"
        name="ftp://ftp.gnu.org/pub/gnu/recode/recode-3.5.tar.gz">.

Advantages:
<itemize>
<item>
Support for transliteration, i.e. conversion of non-ASCII characters
to sequences of ASCII characters in order to preserve readability by
humans, even when a lossless transformation is impossible.
</itemize>

Drawbacks:
<itemize>
<item>
Non-standard API.
</itemize>

<sect2>Other approaches
<p>

<descrip>
<tag>libutf-8</tag>
libutf-8 by G. Adam Stanislav
<htmlurl url="mailto:adam@whizkidtech.net"
           name="&lt;adam@whizkidtech.net&gt;">
contains a few functions for on-the-fly conversion from/to UTF-8 encoded
`FILE*' streams.
<htmlurl url="http://www.whizkidtech.net/i18n/libutf-8-1.0.tar.gz"
        name="http://www.whizkidtech.net/i18n/libutf-8-1.0.tar.gz">

Advantages:
<itemize>
<item>
Very small.
</itemize>

Drawbacks:
<itemize>
<item>
Non-standard API.
<item>
UTF-8 is compiled in, not optional. Programs compiled with this library
lose support for the 8-bit encodings which are still frequently used in Europe.
<item>
Installation is nontrivial: Makefile needs tweaking, not autoconfiguring.
</itemize>

</descrip>

<sect1>Java
<p>

Java has Unicode support built into the language. The type `char' denotes
a Unicode character, and the `java.lang.String' class denotes a string
built up from Unicode characters.

Java can display any Unicode characters through its windowing system AWT,
provided that
  1. you set the Java system property "user.language" appropriately,
  2. the /usr/lib/java/lib/font.properties.<it>language</it> font set
     definitions are appropriate, and
  3. the fonts specified in that file are installed.
For example, in order to display text containing japanese characters,
you would install japanese fonts and run "java -Duser.language=ja ...".
You can combine font sets: In order to display western european, greek
and japanese characters simultaneously, you would create a combination
of the files "font.properties" (covers ISO-8859-1), "font.properties.el"
(covers ISO-8859-7) and "font.properties.ja" into a single file.
??This is untested??

The interfaces java.io.DataInput and java.io.DataOutput have methods called
`readUTF' and `writeUTF' respectively. But note that they don't use UTF-8;
they use a modified UTF-8 encoding: the NUL character is encoded as the
two-byte sequence 0xC0 0x80 instead of 0x00, and a 0x00 byte is added at
the end. Encoded this way, strings can contain NUL characters and nevertheless
need not be prefixed with a length field - the C &lt;string.h&gt; functions
like strlen() and strcpy() can be used to manipulate them.

<sect1>Lisp
<p>

The Common Lisp standard specifies two character types: `base-char' and
`character'. It's up to the implementation to support Unicode or not.
The language also specifies a keyword argument `:external-format' to `open',
as the natural place to specify a character set or encoding.

Among the free Common Lisp implementations, only CLISP
<htmlurl url="http://clisp.cons.org/"
        name="http://clisp.cons.org/">
supports Unicode. You need a CLISP version from July 1999 or newer.
<htmlurl url="ftp://clisp.cons.org/pub/lisp/clisp/source/clispsrc.tar.gz"
        name="ftp://clisp.cons.org/pub/lisp/clisp/source/clispsrc.tar.gz">.
The types `base-char' and `character' are both equivalent to 16-bit Unicode.
The encoding used for file or socket/pipe I/O can be specified through the
`:external-format' argument. The encodings used for tty I/O and the default
encoding for file/socket/pipe I/O are locale dependent.

Among the commercial Common Lisp implementations, only Eclipse
<htmlurl url="http://www.elwood.com/eclipse/eclipse.htm"
        name="http://www.elwood.com/eclipse/eclipse.htm">
supports Unicode. See
<htmlurl url="http://www.elwood.com/eclipse/char.htm"
        name="http://www.elwood.com/eclipse/char.htm">.
The type `base-char' is equivalent
to ISO-8859-1, and the type `character' contains all Unicode characters.
The encoding used for file I/O can be specified through a combination of
the `:element-type' and `:external-format' arguments to `open'.
Limitations: Character attribute functions are locale dependent. Source and
compiled source files cannot contain Unicode string literals.

The commercial Common Lisp implementation Allegro CL does not support
Unicode yet, but Erik Naggum is working on it.

<sect1>Ada95
<p>

Ada95 was designed for Unicode support and the Ada95 standard library
features special ISO 10646-1 data types Wide_Character and Wide_String,
as well as numerous associated procedures and functions. The GNU Ada95
compiler (gnat-3.11 or newer) supports UTF-8 as the external encoding of
wide characters. This allows you to use UTF-8 in both source code and
application I/O. To activate it in the application, use "WCEM=8" in the
FORM string when opening a file, and use compiler option "-gnatW8" if
the source code is in UTF-8. See the GNAT and Ada95 reference manuals
for details.


</article>