How to enhance XKB configuration

By Kamil Toman and Ivan U. Pascal
25th Nov 2002
18 min read

Abstract

This guide is aimed at alleviating one’s labour when creating a new (internationalized) keyboard layout. Unlike other documents, this guide emphasizes the keymap developer’s point of view.

Overview

The developer of a new layout should read the XKB protocol specification (The X Keyboard Extension: Protocol Specification) at least to clarify for themselves some XKB-specific terms used in this document and elsewhere in XKB configuration. It is also wise to understand how the X server and a client digest their keyboard inputs (with and without XKB).

Another useful source is Ivan Pascal’s text about XKB configuration.

Note that this document covers only enhancements which are to be made to XFree86 versions 4.3.x and newer.

The Basics

At boottime (or later at the user’s command) the X server starts its XKB keyboard extension module and reads data from a compiled configuration file.

This compiled configuration file is prepared by the program xkbcomp which behaves altogether as an ordinary compiler (see man xkbcomp). Its input are human-readable XKB configuration files which are verified and then composed into a useful XKB configuration. Users don’t need to mess with xkbcomp them- selves, for them it is invisible. Usually, it is run upon X server startup.

As you probably already know, XKB configuration consists of five main modules:

Keycodes

Tables that define the translation from keyboard scan codes into reasonably symbolic names, maximum and minimum valid keycodes, symbolic aliases, and a description of physically present LED-indica- tors. The primary sense of this component is to allow definitions of maps of symbols (see below) to be independent of physical key- board scancodes. There are two main conventions for symbolic names (always four bytes long):

names which express some traditional meaning, like <SPCE> (which stands for space bar)
names which express a relative position on the keyboard, for example <AE01> (the exclamation mark on US keyboards), with on its right the keys <AE02>, <AE03>, etc.

Types

Types describe how the pressed key is affected by active modifiers (like Shift, Control, Alt, …). There are several predefined types which cover most of the usual combinations.

Compat

The compatibility component defines the internal behaviour of modifiers. Using the compat component you can assign various actions (elaborately described in the XKB specification) to key events. This is also the place where LED-indicators behaviour is defined.

Symbols

For i18n purposes, this is the most important table. It defines what values (=symbols) are assigned to what keycodes (represented by their symbolic name, see above). More than one value may be defined for each key and then it depends on the key type and on the modifiers state (respective compat component) which value will be the resulting one when the key is pressed.

Geometry

Geometry files aren’t used by XKB itself but they may be used by some external programs to depict a keyboard image.

All these components have their files located in the XKB configuration tree, in subdirectories with the same name (usually in /usr/share/X11/xkb).

Enhancing the XKB Configuration

Most of XKB enhancements are about a need to define new output symbols for some input key events. In other words, a need to define a new symbol map (for a new language, or standard, or just to feel more comfortable when typing text).

What do you need to do? Generally, you have to define the following things:

the map of symbols itself
the rules to allow users to select the new mapping
the description of the new layout

First of all, it is good to go through existing layouts and to examine them to see if there is something you could easily adjust to fit your needs. Even if there is nothing similar, you may get some ideas about the basic concepts and used tricks.

Levels and Groups

Since XFree86 4.3.0, you can use multiple layouts in the XKB configuration. Though still within the boundaries of the XKB protocol and its general ideas, the keymap designer must obey new rules when creating new maps. In exchange we get a more powerful and cleaner configuration system.

Remember that it is the application which must decide which symbol matches which keycode according to the effective modifier state. The X server itself sends only an input event message. Of course, usually the interpretation is done by Xlib, Xaw, Motif, Qt, Gtk, or similar libraries. The X server only supplies its mapping table (usually upon application startup).

You can think of the X server’s symbol table as of an irregular table where each keycode has its row and where each combination of modifiers determines exactly one column. The resulting cell then gives the proper symbolic value. Not all keycodes need to bind different values for different combinations of modifiers. The <ENTER> key, for instance, usually doesn’t depend on any modi- fiers so it has in its row only one column defined.

Note that in XKB there is no prior assumption that certain modifiers are bound to certain columns. By editing the proper files (see Key Types, below) this mapping can be changed as well.

Unlike the original X protocol, the XKB approach is far more flexible. XKB introduces one additional term: the group. You can think of a group as of a vector of columns per keycode (naturally the dimension of this vector may differ for different keycodes). What is it good for? The group is not very useful unless you intend to use more than one logically different set of symbols (like more than one alphabet) defined in a single mapping ta- ble. But then the group has a natural meaning: each symbol set has its own group and changing it means selecting a different one. The XKB approach allows up to four different groups. The columns inside each group are called (shift) levels. The X server knows what the current group is and reports it together with the modifier state and the keycode in key events.

To sum it up:

for each keycode the XKB keyboard map contains up to four one-dimensional tables - groups (logically different symbol sets)
for each group of a keycode the XKB keyboard map contains some columns - shift levels (values reached by combinations of Shift, Ctrl, Alt, … modifiers)
different keycodes can have different number of groups
different groups of one keycode can have different number of shift levels
the current group number is tracked by the X server

It is clear that if you sanely define levels and groups, and sanely bind modi- fiers and associated actions, you can have loaded simultaneously up to four different symbol sets where each of them would reside in its own group.

The multi-layout concept provides a facility to manipulate XKB groups and symbol definitions in a way that allows almost arbitrary composition of predefined symbol tables. To keep it fully functional you have to:

define all symbols only in the first group
(re)define any modifiers with extra care to avoid strange (anisometric) behaviour

Defining New Layouts

See “Some Words About XKB internals” for an explanation of used XKB terms and problems addressed by the XKB extension.

See “Common notes about XKB configuration files language” for a more precise explanation of the syntax of XKB configuration files.

Predefined XKB Symbol Sets

If you are about to define some European symbol map extension, you might want to use one of four predefined Latin alphabet layouts.

Okay, let’s assume you want to extend an existing keymap and you want to over- ride a few keys. Let’s take a simple U.K. keyboard as an example (defined in pc/gb):

default partial alphanumeric_keys
xkb_symbols "basic" {

    include "pc/latin"

    name[Group1]="Great Britain";

    key <AE02>  { [         2,   quotedbl,  twosuperior,    oneeighth ] };
    key <AE03>  { [         3,   sterling, threesuperior,    sterling ] };
    key <AC11>  { [apostrophe,         at, dead_circumflex, dead_caron] };
    key <TLDE>  { [     grave,    notsign,          bar,          bar ] };
    key <BKSL>  { [numbersign, asciitilde,   dead_grave,   dead_breve ] };

    key <RALT>  { type[Group1]="TWO_LEVEL",
                  [ ISO_Level3_Shift, Multi_key ] };

    modifier_map Mod5   { <RALT> };
};

It defines a new layout in the basic variant as an extension of a common latin alphabet layout. The layout (symbol set) name is set to “Great Britain”. Then there are redefinitions of a few keycodes and a modifier binding. As you can see, the number of shift levels is the same for the <AE02>, <AE03>, <AC11>, <TLDE> and <BKSL> keys but it differs from the number of shift levels of <RALT>.

Note that the <RALT> key itself is a binding key for Mod5 and that it serves like a shift modifier for LevelThree, and together with Shift as a Compose key. It is a good habit to respect this rule in a new similar layout.

Okay, you could now define more variants of your new layout besides basic simply by including (augmenting/overriding/…) the basic definition and altering what may be needed.

Key Types

The differences in the number of columns (shift levels) are caused by the different types of the keys (see the Types definition in section The Basics). Most keycodes have implicitly set the keytype in the included pc/latin file to FOUR_LEVEL_ALPHABETIC. The only exception is the <RALT> keycode which is explicitly set TWO_LEVEL keytype.

All those names refer to pre-defined shift level schemes. Usually you can choose a suitable shift level scheme from the default types scheme list in the proper XKB component’s subdirectory.

The most used schemes are:

ONE_LEVEL: The key does not depend on any modifiers. The symbol from the first level is always chosen.
TWO_LEVEL: The key uses the modifier Shift and may have two possible values. The second level is chosen by the Shift modifier. If the Lock modifier (usually Caps-lock) applies, the symbol is further processed using system-specific capitalization rules. If both the Shift and Lock modifiers apply, the symbol from the second level is taken and capitalization rules are applied (but usually have no effect).
ALPHABETIC: The key uses the modifiers Shift and Lock. It may have two possible values. The second level is chosen by Shift. When the Lock modifier applies, the symbol from the first level is taken and further processed using system-specific capitalization rules. If both the Shift and Lock modifiers apply, the symbol from the first level is taken and no capitalization rules are applied. This is often called shift-cancels-caps behaviour.
THREE_LEVEL: Is the same as TWO_LEVEL but it considers an extra modifier: LevelThree, which can be used to gain the symbol value from the third level. If both the Shift and LevelThree modifiers apply, the value from the third level is taken. As in TWO_LEVEL, the Lock modifier doesn’t influence the resulting level - only Shift and LevelThree are taken into consideration. If the Lock modifier is active, capitalization rules are applied to the resulting symbol.
FOUR_LEVEL: Is the same as THREE_LEVEL but, unlike THREE_LEVEL, if both the Shift and LevelThree modifiers apply, the symbol is taken from the fourth level.
FOUR_LEVEL_ALPHABETIC: Is similar to FOUR_LEVEL but also defines shift-cancels-caps behaviour as in ALPHABETIC. If both Lock and LevelThree apply, the symbol from the third level is taken and the capitalization rules are applied. If all three modifiers (Lock and Shift and LevelThree) apply, the symbol from the third level is taken and no capitalization rules are applied
KEYPAD: As the name suggest, this scheme is primarily used for numeric keypads. The scheme considers two modifiers: Shift and NumLock. If none of the modifiers applies, the symbol from the first level is taken. If either the Shift or the NumLock modifier apply, the symbol from the second level is taken. If both the Shift and the NumLock modifier apply, the symbol from the first level is taken. Again, a shift-cancels-caps variant.
FOUR_LEVEL_KEYPAD: Is similar to the KEYPAD scheme but considers also the LevelThree modifier. If the LevelThree modifier applies, the symbol from the third level is taken. If both Shift and LevelThree or NumLock and LevelThree apply, the symbol from the fourth level is taken. If all three (Shift+NumLock+LevelThree) apply, the symbol from the third level is taken. This also is a shift-cancels-caps variant.
FOUR_LEVEL_MIXED_KEYPAD: A four-level keypad scheme where the first two levels behave like the KEYPAD scheme (with Shift and NumLock). The LevelThree modifier acts as an override, providing access to two normally Shift-ed levels: when LevelThree is active we ignore the NumLock state. Intended for the digit area of the keypad.
FOUR_LEVEL_X: A four-level scheme where the base level accepts no modifier, `LevelThree` provides two more `Shift`-ed levels (levels 2 and 3), and `Ctrl` plus `Alt` command the fourth level. Intended for the operator part of a keypad, though since `NumLock` plays no part, it is not keypad-specific.

Besides that, there are some schemes for special purposes:

PC_CONTROL_LEVEL2: Similar to the TWO_LEVEL scheme but it considers the Control modifier rather than Shift. That means, the symbol from the second level is chosen by Control rather than by Shift.
PC_ALT_LEVEL2: Similar to the TWO_LEVEL scheme but it considers the Alt modifier rather than Shift. That means, the symbol from the second level is chosen by Alt rather than by Shift.
CTRL+ALT: The key uses the modifiers Alt and Control. It may have two possible values. If just one modifier (Alt or Control) applies, the symbol from the first level is chosen. Only if both the Alt and Control modifiers apply, the symbol from the second level is chosen.
SHIFT+ALT: The key uses the modifiers Shift and Alt. It may have two possible values. If just one modifier (Alt or Shift) applies, the symbol from the first level is chosen. Only if both the Alt and Shift modifiers apply, the symbol from the second level is chosen.

If needed, special caps schemes may be used. They redefine the standard behaviour of all *ALPHABETIC types. The layouts (maps of symbols) with keys defined in respective types then automatically change their behaviour accord- ingly. Possible redefinitions are:

internal
internal_nocancel
shift
shift_nocancel

None of these schemes should be used directly. They are defined merely for the caps: XKB option (used to globally change the layouts behaviour).

Don’t alter any of the existing key types. If you need a different behaviour, create a new type.

More on Definitions of Types

When the XKB software deals with a separate type description, it gets a com- plete list of modifiers that should be taken into account from the modifiers=<list of modifiers> list and expects a set of map[<combination of modifiers>]=<level indication> instructions that contain the mapping for each combination of modifiers mentioned in that list. Modifiers that are not explicitly listed are NOT taken into account when the resulting shift level is computed. If some combination is omitted, the program (subroutine) should choose the first level for this combination (a quite reasonable behavior).

Let’s consider an example with two modifiers, ModOne and ModTwo:

type "…" {
    modifiers = ModOne+ModTwo;
    map[None] = Level1;
    map[ModOne] = Level2;
};

In this case the map has a statement for ModOne only and ModOne+ModTwo is omitted. This means that if ModTwo is active, the subroutine can’t find an explicit mapping for this combination and will use the default level, i.e. Level1.

But in the case that the type is described as:

type "…" {
    modifiers = ModOne;
    map[None] = Level1;
    map[ModOne] = Level2;
};

the ModTwo will not be taken into account and the resulting level depends on the ModOne state only. That means, ModTwo alone produces the Level1 but the combination ModOne+ModTwo (as well as ModOne alone) produces the Level2.

What does it mean if the second modifier is not ModTwo but Lock? It means that in the first case (Lock itself is included in the list of modifiers but combina- tions with this modifier aren’t mentioned in the map statements) the internal capitalization rules will be applied to the symbol from the first level. But in the second case the capitalization will be applied to the symbol chosen accordingly to the first modifier - and this can be the symbol from the first as well as from the second level.

Usually, all modifiers introduced in modifiers=<list of modifiers> list are used for shift level calculation and then discarded. Sometimes this is not desirable. If you want to use a modifier for shift level calculation but you don’t want to discard it, you may list it in preserve[<combination of modifiers>]=<list of modifiers>. That means, for a given combination all listed modifiers will be preserved. If the Lock modifier is preserved then the resulting symbol is passed to the internal capitalization routine regardless whether it has been used for a shift level calculation or not.

Any key type description can use both real and virtual modifiers. Since real modifiers always have standard names it is not necessary to explicitly declare them. Virtual modifiers can have arbitrary names and must be declared (prior to using them) directly in the key type definition:

virtual_modifiers <comma-separated list of modifiers> ;

as seen in for example the basic, pc, or mousekeys key type definitions.

Rules

Once you are finished with your symbol map you need to add it to the rules file. The rules file describes how all the five basic components (keycodes, types, compat, symbols, and geometry) should be composed to give a sensible resulting XKB configuration.

The main advantage of rules over formerly used keymaps is the possibility to simply parameterize (once) fixed patterns of configurations and thus to ele- gantly allow substitutions of various local configurations into predefined templates.

A pattern in a rules file (often located in /usr/share/X11/xkb/rules) can be parameterized with four other arguments: Model, Layout, Variant, and Options. For most cases the parameters Model and Layout should be sufficient for choosing a functional keyboard mapping.

The rules file itself is composed of pattern lines and lines with rules. Each pattern line starts with an exclamation mark (!) and describes how XKB will interpret the subsequent lines (rules). A sample rules file looks like this:

 ! model                   =   keycodes
   macintosh_old           =   macintosh
   …
   *                       =   xfree86

 ! model                   =   symbols
   hp                      =   +inet(%m)
   microsoftpro            =   +inet(%m)
   geniuscomfy             =   +inet(%m)

 ! model       layout[1]   =   symbols
   macintosh   us          =   macintosh/us%(v[1])
   *           *           =   pc/pc(%m)+pc/%l[1]%(v[1])

 ! model       layout[2]   =   symbols
   macintosh   us          =   +macintosh/us[2]%(v[2]):2
   *           *           =   +pc/%l[2]%(v[2]):2

 ! option                  =   types
   caps:internal           =   +caps(internal)
   caps:internal_nocancel  =   +caps(internal_nocancel)

Each rule defines what a certain combination of values on the left side of the equals sign (=) results in. For example, a (keyboard) model macintosh_old instructs XKB to take definitions of keycodes from file keycodes/macintosh while the rest of the models (represented by a wildcard *) instructs it to take them from file keycodes/xfree86. The wildcard represents all possible values on the left side which were not found in any of the previous rules. The more specialized (more complete) rules have higher precedence than general ones, i.e. the more general rules supply reasonable default values.

As you can see some lines contain substitution parameters – the parameters preceded by the percent sign (%). The first alphabetical character after the percent sign expands to the value which has been found on the left side. For example +%l%(v) expands into +cz(bksl) if the respective values on the left side were cz layout in its bksl variant. More, if the layout resp. variant parameter is followed by a pair of brackets ([, ]) it means that XKB should place the layout resp. variant into the specified XKB group. If the brackets are omitted, the first group is the default value.

So the second block of rules enhances symbol definitions for some particular keyboard models with extra keys (for internet, multimedia, …) . Other mod- els are left intact. Similarly, the last block overrides some key type defi- nitions, so the common global behaviour “shift cancels caps” or “shift doesn’t cancel caps” can be selected. The rest of the rules produce special symbols for each US variant of the macintosh keyboard, and standard pc symbols in appropriate variants as a default.

Descriptive Files of Rules

Now you just need to add a detailed description to the <rules>.xml description file so that other users (and external programs which often parse this file) know what your work is about.

Old Descriptive Files

The formerly used descriptive files were named <rules>.lst. Its structure is very simple and quite self descriptive but such simplicity had also some cav- ities, for example there was no way how to describe local variants of layouts and there were problems with the localization of descriptions. To preserve compatibility with some older programs, new XML descriptive files can be con- verted to the old .lst format.

The meaning of each possible parameter of the rules file should be described. For the sample rules file given above, the .lst file could look like this:

 ! model
   pc104        Generic 104-key PC
   microsoft    Microsoft Natural
   pc98         PC-98xx Series
   macintosh    Original Macintosh
   …

 ! layout
   us      U.S. English
   cz      Czech
   de      German
   …

 ! option
   caps:internal           uses internal capitalization, Shift cancels Caps
   caps:internal_nocancel  uses internal capitalization, Shift doesn’t cancel Caps

And that should be it. Enjoy creating your own XKB mapping.