Japanese Input and Display in X under Linux

Several years ago, I had a pretty hard time trying to get Japanese display and input working in gvim in XFree86 3. My goal was to have a system that was localized for english, but could also display Japanese text and allow me to input it. Thankfully it's become a lot simpler in recent years with advances in glibc and locales handling, as well as the introduction of XFree86 4.0. These instructions were developed and tested in Debian's unstable distribution, codenamed "Sid," but they should be appropriate for any linux installation with a minor amount of tweaking.

Installing the locales package
Setting up the environment
Setting up the conversion server and the input engine
Setting up .Xresources
Inputting Japanese
Tips
Helpful Software
References

Installing the locales package

In Debian, locales is a seperate package from glibc, so you'll have to install it in order to generate and use the Japanese locale in Liunx. During installation, debconf will ask you to choose which locales to generate. For starters, pick ja_JP.EUC-JP and en_US. You may also want to choose en_US.UTF-8 and ja_JP.UTF-8 for unicode support.

After installation, you can choose and generate locales manually by editing /etc/locale.gen. The contents of mine are as follows:

# This file lists locales that you wish to have built. You can find a list
# of valid supported locales at /usr/share/doc/locales/SUPPORTED.gz. Other
# combinations are possible, but may not be well tested. If you change
# this file, you need to rerun locale-gen.

en_US ISO-8859-1
en_US.UTF-8 UTF-8
ja_JP.EUC-JP EUC-JP
ja_JP.UTF-8 UTF-8

After editing the file, run locale-gen as root.

Setting up the environment

There are several special environment variables that determine how various locale-specific tasks should be handled.

LANG - Specifies the default locale for all unset locale variables
LANGUAGE - Most programs use this for the language of its interface
LINGUAS - (Obsolete?) The WindowMaker window manager used to use this instead of LANGUAGE.
LC_ALL - Overrides the locale for all LC_ variables and LANG
LC_CTYPE - Character handling functions
LC_COLLATE - Affects glob bracket ranges and sort order
LC_MESSAGES - Language that messages should be written in. Most programs use the value of LANGUAGE instead.
LC_NUMERIC - Determines the thousands seperator and how to write floating point numbers
LC_TIME - How to format dates and times

To run X with an English interface but the ability to display, input, and copy-paste Japanese text, set the environment variables as follows prior to running X or in your .Xsession file. These examples assume your are running a Bourne-like shell, such as sh, bash, or zsh:

export LANGUAGE=en
export LINGUAS=en
export LC_CTYPE=ja_JP    # May break certain things like window maker; untested
export LC_TIME=C         # Format time as English
export LC_NUMERIC=C      # Format numbers as English
export LC_MESSAGES=C     # Output messages in English
export LC_COLLATE=ja_JP  # Do sorting and collating of characters as Japanese
export LANG=ja_JP        # Use Japanese for all others

If you are using a c-shell like csh or tcsh, use the following instead:

setenv LANGUAGE en
setenv LINGUAS en
setenv LC_CTYPE ja_JP  
setenv LC_TIME C       
setenv LC_NUMERIC C    
setenv LC_MESSAGES C   
setenv LC_COLLATE ja_JP
setenv LANG ja_JP

Setting up the conversion server and the input engine

Install a conversion server. The two main choices are wnn and canna. I use canna. In Debian, installing canna will automatically run cannaserver for you and insert it into your rc scripts for automatic startup on boot. To install it, run
```
      # apt-get install canna canna-utils libcanna1g
      
```
On other systems, you may have to run cannaserver by hand and add it to your rc scripts yourself.
Install and run an input engine. The most popular, and also the one that I use, is kinput2. In Debian, install it by running
```
      # apt-get install kinput2-common kinput2-canna
```
To make sure you can always input Japanese in your X applications, you should run kinput2 in your .Xsession file. Insert the following lines after your environment variable setup section:
```
      export XMODIFIERS="@im=kinput2"
      kinput2 -canna &
```

Setting up .Xresources

Add the following lines to your .Xresources file (for non-Debian distributions, this might be called .Xdefaults):

*.inputMethod: kinput2
*.preeditType: OverTheSpot
Kinput2*useOverrideShellForMode: true
Kinput2*conversionStartKeys: Shift<Key>space

! for Kterm
KTerm*VT100*KanjiMode: euc

! for rxvt
Rxvt*multichar_encoding: eucj

! for Gvim
Vim*international: True
Vim*inputMethod: kinput2
Vim*preeditType: OnTheSpot

If you're in X already, be sure to run "% xrdb -merge .Xresources" or whatever to have your changes take effect.

Inputting Japanese

Now when you start X and a Japanese-aware application, you should be able to use shift-space to begin Japanese input. A little window containing a Japanese hiragana "A" should appear, and what you type will become hiragana as you go. Hit space to transform short segments of hiragana into Kanji, or ctrl-n to cycle through katakana, half-width katakana, double-width romaji, normal romaji, and back to hiragana. Hit enter to set your choice, space to choose the next kanji, or backspace to go back to entering more hiragana. Hit shift-space again to leave Japanese input mode, but beware that any text you haven't set by hitting enter will be lost.

Tips:

Old glibc and the Locale Hack - Older versions of glibc reqiured a locale hack to be able to handle Japanese. It's probably best to update your glibc and use the new locales facilities, but if that's unfeasible, you may have to do some digging for that hack...
Getting Ctrl-O back from Canna/Kinput2 - If ctrl-o also starts conversion, you can make it stop by putting the line
```
  (global-unbind-key-function 'japanese-mode)
```
in your ~/.canna file. If you want to do it system-wide, edit the file /etc/canna/default.canna as well. Note that a ~/.canna takes precedence, so if it exists, you must add the line to that file anyway. Also note that for non-debian systems or custom installations, the system-wide file is probably /var/lib/canna/default.canna instead. After editing the file, you should restart X, or at least restart kinput2 as described below. You don't have to restart canna. (Thanks to Norio-san for his page on the matter: http://www.susono.com/~norio/linux/x/c-o.html )
Restarting Kinput2 - If you make some setting changes for kinput2, sometimes it can be desirable to let them take effect without having to restart your entire X session. We shouldn't kill kinput2 in a window that has XIM active or it will lock up or exit out. The best way I've found is to run xterm with LC_ALL=C:
```
 $ LC_ALL=C xterm&
```
Now in the xterm, kill off the current kinput2 and start it up again:
```
 $ kill `pidof kinput2` && kinput2 -canna&
```
It's best to minimize the time for which kinput2 isn't running. Programs currently using the input method may become unstable if they suddenly find it's no longer there. Finally, disown the process in the xterm so the shell won't complain when you exit:
```
 $ disown ; exit
```
Notes:
Gvim doesn't like it when you kill kinput2 out from under it. You'll probably have to restart any instances in which you want to be able to input Japanese again.
Display Fonts for the Preediting and Choice windows - I run at a wacked resolution, so I like to have my kanji larger so their individual strokes are a bit more legible. My fonts are currently set as follows in my .Xresources file:
```
	Kinput2*defaultFontList: -ricoh-ms mincho-medium-r-normal-*-*-160-*-*-*-*-jisx0208.1983-0
	Kinput2*JpWcharDisplay.font: -ricoh-ms mincho-medium-r-normal-*-*-160-*-*-c-*-iso8859-1
	Kinput2*JpWcharDisplay.kanjiFont: -ricoh-ms mincho-medium-r-normal-*-*-160-*-*-c-*-jisx0208.1983-0
	Kinput2*JpWcharDisplay.kanaFont: -ricoh-ms mincho-medium-r-normal-*-*-160-*-*-c-*-jisx0201.1976-0
```
See 'man kinput2' for more information. Restart your X server to have the changes take effect, or use 'xrdb -merge ~/.Xresources' and follow the instructions above for restarting kinput2.
Japanese Messages - If you've got Japanese working and suddenly all your menus and shell error messages are in Japanese, then you'll have to remember to set LANGUAGE=en and LC_MESSAGES=C as described above in Setting up the environment.
Locale Name Differences - The locale for Japanese EUC encoding may be different on your machine. Check /usr/share/locale or /usr/lib/locale for likely ones. On my setup, ja_JP does the trick. You might need to use something like ja_JP.EUC, ja_JP.euc or even ja_JP.eucJP. Setting LANG to ja does not seem to work, though the multibyte helpfile for vim suggests that it might.
Map a kanji key - You can map a key to the keysym "Kanji" in your xkb map or your xmodmap. I prefer the xkb extensions of X11R6 because it loads keymaps (even uncompiled ones) infinately many times faster than xmodmap.

Helpful Software:

gjiten - a great gnome/gtk lookup program for edict-style dictionaries and kanjidic. Also has a kanji-lookup-by-grapheme mode.
Jim Breen's xjdic - console based dictionary. This hasn't been updated for a while, but according to the author, when it is, it'll use the new XML format for edict, called JMdict. This file has yet to make an appearance in Debian.
xyaku - the X translator. Select a word and hit a configurable key-combo, and xyaku pops up a little window with the definition. Uses edict by default, but is highly configurable via plugin scripts.
Vim - the Vi IMproved editor by Bram Moolenaar. Version 6 is starting to have some pretty excellent support for international character sets, but I've had Japanese editing working in Vim since version 5.3...
Rxvt - My favorite terminal for Japanese work. The debs are rxvt and rxvt-ml.
```
 # apt-get install rxvt rxvt-ml
```
Then run krxvt and you should be able to view and input Japanese.
Kterm - This tries to be as close to xterm as possible, but with Kanji. It has the standard ctrl-mouseclick menus for font selection and so forth. Also, kterm has ability tont information, including info on true-type fonts (ugh).

[ Home ]

Steven Mueller, diffusor@ugcs.caltech.edu
This page created: Saturday March 23, 2002
Last modified: Fri May 10 16:42:18 PDT 2002
(created/edited with Vim)