How to further enhance XKB configuration

                              Abstract

     This guide is aimed to relieve one's labour to create a new
     (internationalized) keyboard layout. Unlike other documents
      this guide accents the keymap developer's point of view.
           _______________________________________________

   Table of Contents
   Overview
   The Basics
   Enhancing XKB Configuration
   Defining New Layouts

Overview

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

   A useful source is also Ivan Pascal's text about xkb
   configuration often referenced throughout this document.

   Note that this document covers only enhancements which are to
   be made to XFree86 version 4.3 and X11R6.7.0 and above.
     __________________________________________________________

The Basics

   At the startup (or at later at user's command) X server starts
   its xkb keyboard module extension 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
   themselves, for them it is invisible. Usually, it is started
   upon X server startup.

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

   Keycodes
          Tables that defines translation from keyboard scan codes
          into reasonable symbolic names, maximum, minimum legal
          keycodes, symbolic aliases and description of physically
          present LED-indicators. The primary sence of this
          component is to allow definitions of maps of symbols
          (see below) to be independent of physical keyboard
          scancodes. There are two main naming conventions for
          symbolic names (always four bytes long):

          + names which express some traditional meaning like
            <SPCE> (stands for space bar) or
          + names which express some relative positioning on a
            keyboard, for example <AE01> (an exclamation mark on
            US keyboards), on the right there are keys <AE02>,
            <AE03> etc.

   Types
          Types describe how the produced key is changed by active
          modifiers (like Shift, Control, Alt, ...). There are
          several predefined types which cover most of used
          combinations.

   Compat
          Compatibility component defines internal behaviour of
          modifiers. Using compat component you can assign various
          actions (elaborately described in 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). There may be defined more than one value for
          each key and then it depends on a key type and on
          modifiers state (respective compat component) which
          value will be the resulting one.

   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 the files located in xkb
   configuration tree in subdirectories with the same names
   (usually in /usr/lib/X11/xkb).
     __________________________________________________________

Enhancing XKB Configuration

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

   What do you need to do? Generally, you have to define 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 if there is something you could easily adjust to
   fit your needs. Even if there is nothing similar you may get
   some ideas about basic concepts and used tricks.
     __________________________________________________________

Levels And Groups

   Since XFree86 4.3.0 and X11R6.7.0 you can use multi-layout
   concept of xkb configuration. Though it is still in boundaries
   of xkb protocol and 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 effective modifier
   state. The X server itself sends only an input event message
   to. Of course, usually the general interpretation is processed
   by Xlib, Xaw, Motif, Qt, Gtk and similar libraries. The X
   server only supplies its mapping table (usually upon an
   application startup).

   You can think of the X server's symbol table as of a 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 combination of modifiers.
   <ENTER> key, for instance, usually doesn't depend on any
   modifiers so it its row has only one column defined.

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

   Unlike the original X protocol the XKB approach is far more
   flexible. It is comfortable to add one additional XKB term -
   group. You can think of a group as of a vector of columns per
   each 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 table. But then, the group has a natural
   meaning - each symbol set has its own group and changing it
   means selecting a different one. XKB approach allows up to four
   different groups. The columns inside each group are called
   (shift) levels. The X server knows the current group and
   reports it together with modifier set and with a keycode in key
   events.

   To sum it up:

     * for each keycode XKB keyboard map contains up to four
       one-dimensional tables - groups (logically different symbol
       sets)
     * for each group of a keycode 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 X server

   It is clear that if you sanely define levels, groups and sanely
   bind modifiers and associated actions you can have
   simultaneously loaded 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 explanation of used xkb
   terms and problems addressed by XKB extension.

   See Common notes about XKB configuration files language for
   more precise explanation of 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 on of four predefined latin alphabet
   layouts.

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

partial default 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 basic variant as an extension of
   common latin alphabet layout. The layout (symbol set) name is
   set to "Great Britain". Then there are redefinitions of a few
   keycodes and a modifiers binding. As you can see the number of
   shift levels is the same for <AE02>, <AE03>, <AC11>, <TLDE> and
   <BKSL> keys but it differs from 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, together
   with Shift as a multi-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 a different types of keys (see the types definition
   in section basics). Most keycodes have implicitly set the
   keytype in the included "pc/latin" file to
   "FOUR_LEVEL_ALPHABETIC". The only exception is <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
   default types scheme list in proper xkb component's
   subdirectory.

   The most used schemes are:

   ONE_LEVEL
          The key does not depend on any modifiers. The symbol
          from first level is always chosen.

   TWO_LEVEL
          The key uses a modifier Shift and may have two possible
          values. The second level may be chosen by Shift
          modifier. If Lock modifier (usually Caps-lock) applies
          the symbol is further processed using system-specific
          capitalization rules. If both Shift+Lock modifier apply
          the symbol from the second level is taken and
          capitalization rules are applied (and usually have no
          effect).

   ALPHABETIC
          The key uses modifiers Shift and Lock. It may have two
          possible values. The second level may be chosen by Shift
          modifier. When Lock modifier applies, the symbol from
          the first level is taken and further processed using
          system-specific capitalization rules. If both Shift+Lock
          modifier apply the symbol from the first level is taken
          and no capitalization rules 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
          Shift+LevelThree modifiers apply the value from the
          third level is also taken. As in TWO_LEVEL, the Lock
          modifier doesn't influence the resulting level. Only
          Shift and LevelThree are taken into that consideration.
          If the Lock modifier is active capitalization rules are
          applied on the resulting symbol.

   FOUR_LEVEL
          Is the same as THREE_LEVEL but unlike LEVEL_THREE if
          both Shift+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
          Lock+LevelThree apply the symbol from the third level is
          taken and the capitalization rules are applied. If
          Lock+Shift+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 modifiers applies the
          symbol from the first level is taken. If either Shift or
          NumLock modifiers apply the symbol from the second level
          is taken. If both Shift+NumLock modifiers apply the
          symbol from the first level is taken. Again,
          shift-cancels-caps variant.

   FOUR_LEVEL_KEYPAD
          Is similar to KEYPAD scheme but considers also
          LevelThree modifier. If LevelThree modifier applies the
          symbol from the third level is taken. If
          Shift+LevelThree or NumLock+LevelThree apply the symbol
          from the fourth level is taken. If all
          Shift+NumLock+LevelThree modifiers apply the symbol from
          the third level is taken. This also, shift-cancels-caps
          variant.

   Besides that, there are several schemes for special purposes:

   PC_BREAK
          It is similar to 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_SYSRQ
          It is similar to 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 modifiers Alt and Control. It may have two
          possible values. If only one modifier (Alt or Control)
          applies the symbol from the first level is chosen. Only
          if both Alt+Control modifiers apply the symbol from the
          second level is chosen.

   SHIFT+ALT
          The key uses modifiers Shift and Alt. It may have two
          possible values. If only one modifier (Alt or Shift)
          applies the symbol from the first level is chosen. Only
          if both Alt+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 accordingly. Possible
   redefinitions are:

     * internal
     * internal_nocancel
     * shift
     * shift_nocancel

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

   Don't alter any of existing key types. If you need a different
   behaviour create a new one.
     __________________________________________________________

More On Definitions Of Types

   When the XKB software deals with a separate type description it
   gets a complete list of modifiers that should be taken into
   account from the 'modifiers=<list of modifiers>' list and
   expects that a set of 'map[<combination of modifiers>]=<list of
   modifiers>' 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).

   Lets consider an example with two modifiers ModOne and ModTwo:
type "..." {
    modifiers = ModOne+ModTwo;
    map[None] = Level1;
    map[ModOne] = Level2;
};

   In this case the map statements for ModTwo only and
   ModOne+ModTwo are omitted. It means that if the ModTwo is
   active the subroutine can't found explicit mapping for such
   combination an will use the default level i.e. Level1.

   But in the case the type 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
   produces the Level2 as well as ModOne alone.

   What does it mean if the second modifier is the Lock? It means
   that in the first case (the Lock itself is included in the list
   of modifiers but combinations 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 he 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 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
   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 can be declared (prior using them)
   directly in key type definition:
   virtual_modifiers <comma-separated list of modifiers>  ;

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

Rules

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

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

   A pattern in a rules file (often located in
   /usr/lib/X11/xkb/rules) can be parameterized with four other
   arguments: Model, Layout, Variant and Options. For most cases
   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. The pattern line starts with an exclamation mark
   ('!') and describes how will the xkb interpret the following
   lines (rules). A sample rules file looks like this:
! model                         =       keycodes
  macintosh_old                 =       macintosh
  ...
  *                             =       xorg

! 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 certain combination of values on the
   left side of equal 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 models (represented by a wild card '*') instructs it to
   take them from file keycodes/xorg. The wild card 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 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 models are left intact. Similarly, the
   last block overrides some key type definitions, so the common
   global behaviour ''shift cancels caps'' or ''shift doesn't
   cancel caps'' can be selected. The rest of rules produces
   special symbols for each variant us layout of 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 <rules>.xml
   description file so the other users (and external programs
   which often parse this file) know what is your work 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 cavities, 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 converted to old format '.lst'.

   For each parameter of rules file should be described its
   meaning. For the rules file described above the .lst file could
   look like:
! 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 Cap
s
  caps:internal_nocancel uses internal capitalization. Shift doesn't can
cel Caps

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