[Unicode]  Technical Reports
 

Unicode Technical Standard #35 Tech Preview

Unicode Locale Data Markup Language (LDML)
Part 7: Keyboards

Version 45 (draft)
Editors Steven Loomis (srloomis@unicode.org) and other CLDR committee members

For the full header, summary, and status, see Part 1: Core.

Summary

This document describes parts of an XML format (vocabulary) for the exchange of structured locale data. This format is used in the Unicode Common Locale Data Repository.

This is a partial document, describing keyboards. For the other parts of the LDML see the main LDML document and the links above.

Note: Some links may lead to in-development or older versions of the data files. See https://cldr.unicode.org for up-to-date CLDR release data.

Status

This is a draft document which may be updated, replaced, or superseded by other documents at any time. Publication does not imply endorsement by the Unicode Consortium. This is not a stable document; it is inappropriate to cite this document as other than a work in progress.

A Unicode Technical Standard (UTS) is an independent specification. Conformance to the Unicode Standard does not imply conformance to any UTS.

Please submit corrigenda and other comments with the CLDR bug reporting form [Bugs]. Related information that is useful in understanding this document is found in the References. For the latest version of the Unicode Standard see [Unicode]. For a list of current Unicode Technical Reports see [Reports]. For more information about versions of the Unicode Standard, see [Versions].

See also Compatibility Notice.

Parts

The LDML specification is divided into the following parts:

Contents of Part 7, Keyboards

Keyboards

The Unicode Standard and related technologies such as CLDR have dramatically improved the path to language support. However, keyboard support remains platform and vendor specific, causing inconsistencies in implementation as well as timeline.

More and more language communities are determining that digitization is vital to their approach to language preservation and that engagement with Unicode is essential to becoming fully digitized. For many of these communities, however, getting new characters or a new script added to The Unicode Standard is not the end of their journey. The next, often more challenging stage is to get device makers, operating systems, apps and services to implement the script requirements that Unicode has just added to support their language.

However, commensurate improvements to streamline new language support on the input side have been lacking. CLDR’s Keyboard specification has been updated in an attempt to address this gap.

This document specifies an interchange format for the communication of keyboard mapping data independent of vendors and platforms. Keyboard authors can then create a single mapping file for their language, which implementations can use to provide that language’s keyboard mapping on their own platform.

Additionally, the standardized identifier for keyboards can be used to communicate, internally or externally, a request for a particular keyboard mapping that is to be used to transform either text or keystrokes. The corresponding data can then be used to perform the requested actions. For example, a remote screen-access application (such as used for customer service or server management) would be able to communicate and choose the same keyboard layout on the remote device as is used in front of the user, even if the two systems used different platforms.

The data can also be used in analysis of the capabilities of different keyboards. It also allows better interoperability by making it easier for keyboard designers to see which characters are generally supported on keyboards for given languages.

For complete examples, see the XML files in the CLDR source repository.

Attribute values should be evaluated considering the DTD and DTD Annotations.

Goals and Non-goals

Some goals of this format are:

  1. Physical and virtual keyboard layouts defined in a single file.
  2. Provide definitive platform-independent definitions for new keyboard layouts.
    • For example, a new French standard keyboard layout would have a single definition which would be usable across all implementations.
  3. Allow platforms to be able to use CLDR keyboard data for the character-emitting keys (non-frame) aspects of keyboard layouts.
  4. Deprecate & archive existing LDML platform-specific layouts so they are not part of future releases.

Some non-goals (outside the scope of the format) currently are:

  1. Adaptation for screen scaling resolution. Instead, keyboards should define layouts based on physical size. Platforms may interpret physical size definitions and adapt for different physical screen sizes with different resolutions.
  2. Unification of platform-specific virtual key and scan code mapping tables.
  3. Unification of pre-existing platform layouts themselves (e.g. existing fr-azerty on platform a, b, c).
  4. Support for prior (pre 3.0) CLDR keyboard files. See Compatibility Notice.
  5. Run-time efficiency. LDML is explicitly an interchange format, and so it is expected that data will be transformed to a more compact format for use by a keystroke processing engine.
  6. Platform-specific frame keys such as Fn, Numpad, IME swap keys, and cursor keys are out of scope. (This also means that in this specification, modifier (frame) keys cannot generate output, such as capslock producing backslash.)

Note that in parts of this document, the format @x is used to indicate the attribute x.

Compatibility Notice

A major rewrite of this specification, called "Keyboard 3.0", was introduced in CLDR v45. The changes required were too extensive to maintain compatibility. For this reason, the ldmlKeyboard3.dtd DTD is not compatible with DTDs from prior versions of CLDR such as v43 and prior.

To process earlier XML files, use the data and specification from v43.1, found at https://www.unicode.org/reports/tr35/tr35-69/tr35.html

ldmlKeyboard.dtd continues to be made available in CLDR, however, it will not be updated.

Accessibility

Keyboard use can be challenging for individuals with various types of disabilities. For this revision, features or architectural designs specifically for the purpose of improving accessibility are not yet included. However:

  1. Having an industry-wide standard format for keyboards will enable accessibility software to make use of keyboard data with a reduced dependence on platform-specific knowledge.
  2. Features which require certain levels of mobility or speed of entry should be considered for their impact on accessibility. This impact could be mitigated by means of additional, accessible methods of generating the same output.
  3. Public feedback is welcome on any aspects of this document which might hinder accessibility.

Definitions

Arrangement: The relative position of the rectangles that represent keys, either physically or virtually. A hardware keyboard has a static arrangement while a touch keyboard may have a dynamic arrangement that changes per language and/or layer. While the arrangement of keys on a keyboard may be fixed, the mapping of those keys may vary.

Base character: The character emitted by a particular key when no modifiers are active. In ISO 9995-1:2009 terms, this is Group 1, Level 1.

Core keys: also known as “alphanumeric” section. The primary set of key values on a keyboard that are used for typing the target language of the keyboard. For example, the three rows of letters on a standard US QWERTY keyboard (QWERTYUIOP, ASDFGHJKL, ZXCVBNM) together with the most significant punctuation keys. Usually this equates to the minimal set of keys for a language as seen on mobile phone keyboards. Distinguished from the frame keys.

Dead keys: These are keys which do not emit normal characters by themselves. They are so named because to the user, they may appear to be “dead,” i.e., non-functional. However, they do produce a change to the input context. For example, in many Latin keyboards hitting the ^ dead-key followed by the e key produces ê. The ^ by itself may be invisible or presented in a special way by the platform.

Frame keys: These are keys which are outside of the area of the core keys and typically do not emit characters. These keys include modifier keys, such as Shift or Ctrl, but also include platform specific keys: Fn, IME and layout-switching keys, cursor keys, insert emoji keys etc.

Hardware keyboard: an input device which has individual keys that are pressed. Each key has a unique identifier and the arrangement doesn't change, even if the mapping of those keys does. Also known as a physical keyboard.

Implementation: see Keyboard implementation

Input Method Editor (IME): a component or program that supports input of large character sets. Typically, IMEs employ contextual logic and candidate UI to identify the Unicode characters intended by the user.

Keyboard implementation: Software which implements the present specification, such that keyboard XML files can be used to interpret keystrokes from a Hardware keyboard or an on-screen Touch keyboard.

Keyboard implementations will typically consist of two parts:

  1. A compile/build tool part used by Keyboard authors to parse the XML file and produce a compact runtime format, and
  2. A runtime part which interprets the runtime format when the keyboard is selected by the end user, and delivers the output plain text to the platform or application.

Key: A physical key on a hardware keyboard, or a virtual key on a touch keyboard.

Key code: The integer code sent to the application on pressing a key.

Key map: The basic mapping between hardware or on-screen positions and the output characters for each set of modifier combinations associated with a particular layout. There may be multiple key maps for each layout.

Keyboard: A particular arrangement of keys for the inputting of text, such as a hardware keyboard or a touch keyboard.

Keyboard author: The person or group of people designing and producing a particular keyboard layout designed to support one or more languages. In the context of this specification, that author may be editing the LDML XML file directly or by means of software tools.

Keyboard layout: A layout is the overall keyboard configuration for a particular locale. Within a keyboard layout, there is a single base map, one or more key maps and zero or more transforms.

Layer is an arrangement of keys on a touch keyboard. A touch keyboard is made up of a set of layers. Each layer may have a different key layout, unlike with a hardware keyboard, and may not correspond directly to a hardware keyboard's modifier keys. A layer is accessed via a switch key. See also touch keyboard, modifier, switch.

Long-press key: also known as a “child key”. A secondary key that is invoked from a top level key on a touch keyboard. Secondary keys typically provide access to variants of the top level key, such as accented variants (a => á, à, ä, ã)

Modifier: A key that is held to change the behavior of a hardware keyboard. For example, the "Shift" key allows access to upper-case characters on a US keyboard. Other modifier keys include but are not limited to: Ctrl, Alt, Option, Command and Caps Lock. On a touch keyboard, keys that appear to be modifier keys should be considered to be layer-switching keys.

Physical keyboard: see Hardware keyboard

Touch keyboard: A keyboard that is rendered on a, typically, touch surface. It has a dynamic arrangement and contrasts with a hardware keyboard. This term has many synonyms: software keyboard, SIP (Software Input Panel), virtual keyboard. This contrasts with other uses of the term virtual keyboard as an on-screen keyboard for reference or accessibility data entry.

Transform: A transform is an element that specifies a set of conversions from sequences of code points into one (or more) other code points. Transforms may reorder or replace text. They may be used to implement “dead key” behaviors, simple orthographic corrections, visual (typewriter) type input etc.

Virtual keyboard: see Touch keyboard

Notation

Escaping

When explicitly specified, attribute values can contain escaped characters. This specification uses two methods of escaping, the UnicodeSet notation and the \u{…usv} notation.

UnicodeSet Escaping

The UnicodeSet notation is described in UTS #35 section 5.3.3 and allows for comprehensive character matching, including by character range, properties, names, or codepoints.

Note that the \u1234 and \x{C1} format escaping is not supported, only the \u{…} format (using bracketedHex).

Currently, the following attribute values allow UnicodeSet notation:

UTS18 Escaping

The \u{…usv} notation, a subset of hex notation, is described in UTS #18 section 1.1. It can refer to one or multiple individual codepoints. Currently, the following attribute values allow the \u{…} notation:

Characters of general category of Mark (M), Control characters (Cc), Format characters (Cf), and whitespace other than space should be encoded using one of the notation above as appropriate.

Attribute values escaped in this manner are annotated with the <!--@ALLOWS_UESC--> DTD annotation, see DTD Annotations

File and Directory Structure

<keyboard3 xmlns="https://schemas.unicode.org/cldr/45/keyboard3" conformsTo="45"/>

Note: Unlike other LDML files, layouts are designed to be used outside of the CLDR source tree. As such, they do not contain DOCTYPE entries.

DTD and Schema (.xsd) files are available for use in validating keyboard files.

Extensibility

For extensibility, the <special> element will be allowed at nearly every level.

See Element special in Part 1.

Normalization

Unicode Normalization, as described in The Unicode Standard, is a process by which Unicode text is processed to eliminate unwanted distinctions.

This section discusses how conformant keyboards are affected by normalization, and the impact of normalization on keyboard authors and keyboard implmentations.

Keyboard implementations will usually apply normalization as appropriate when matching transform rules and <display> value matching. Output from the keyboard, following application of all transform rules, will be normalized to the appropriate form by the keyboard implementation.

Note: There are many existing software libraries which perform Unicode Normalization, including ICU, ICU4X, and JavaScript's String.prototype.normalize().

Keyboard authors will not typically need to perform normalization as part of the keyboard layout. However, authors should be aware of areas where normalization affects keyboard operation so that they may achieve their desired results.

Where Normalization Occurs

There are four stages where normalization must be performed by keyboard implementations.

  1. From the keyboard source .xml

    Keyboard source .xml files may be in any normalization form. However, in processing they are converted to NFD.

    • From any form to NFD: full normalization (decompose+reorder)
    • Markers must be processed as described below.
    • Regex patterns must be processed so that matching is performed in NFD.

    Example: <key output=, and <transform from= to= attribute contents will be normalized to NFD.

  2. From the input context

    The input context must be normalized for purposes of matching.

    • From any form to NFD: full normalization (decompose+reorder)
    • Markers in the cached context must be preserved.

    Example: The input context contains U+00E8 (è). The user clicks the cursor after the character, then presses a key which produces U+0320 (<key output="\u{0320}"/>). The implementation must normalize the context buffer to e\u{0320}\u{0300} (è̠) before matching.

  3. Before each transformGroup

    Text must be normalized before processing by the next transformGroup.

    • To NFD: no decomposition should be needed, because all of the input text (including transform rules) was already in NFD form. However, marker reordering may be needed if transforms insert segments out of order.
    • Markers must be preserved.

    Example: The input context contains U+00E8 (è). The user clicks the cursor after this character, then presses a key producing x. A transform rule <transform from='x' to='\u{0320}'/> matches. The implementation must normalize the intermediate buffer to e\u{0320}\u{0300} (è̠) before proceeding to the next transformGroup.

  4. Before output to the platform/application

    Text must be normalized into the output form requested by the platform or application. This will typically be NFC, but may not be.

    • If normalizing to NFC, full normalization (reorder+composition) will be required.
    • No markers are present in this text, they are removed prior to output but retained in the implementation's input context for subsequent keystrokes. See markers.

    Example: The result of keystrokes and transform processing produces the string e\u{0300}. The keyboard implementation normalizes this to a single NFC codepoint U+00E8 (è), which is returned to the application.

Normalization and Transform Matching

Regardless of the normalization form in the keyboard source file or in the edit buffer context, transform matching will be performed using NFD. For example, all of the following transforms will match the input strings è̠, whether the input is U+00E8 U+0320, U+0065 U+0320 U+0300, or U+0065 U+0300 U+0320.

<transform from="e\u{0320}\u{0300}" /> <!-- NFD -->
<transform from="\u{00E8}\u{0320}"  /> <!-- NFC: è + U+0320 -->
<transform from="e\u{0300}\u{0320}" /> <!-- Unnormalized -->

Normalization and Markers

A special issue occurs when markers are involved. Markers are not text, and so not themselves modified or reordered by the Unicode Normalization Algorithm. Existing Normalization APIs typically operate on plain text, and so those APIs can not be used with content containing markers.

However, the markers must be retained and processed by keyboard implementations in a manner which will be both consistent across implementations and predictable to keyboard authors. Inconsistencies would result in different user experiences — specifically, different or incorrect text output — on some implementations and not another. Unpredictability would make it challenging for the keyboard author to create a keyboard with expected behavior.

This section gives an algorithm for implementing normalization on a text stream including markers.

Note: When the algorithm is performed on a plain text stream that doesn't include markers, implementations may skip the removing/re-adding steps 1 and 3 because no markers are involved.

Rationale for 'gluing' markers

The processing described here describes an extension to Unicode normalization to account for the desired behavior of markers.

The algorithm described considers markers 'glued' (remaining with) the following character. If a context ends with a marker, that marker would be guaranteed to remain at the end after processing, consistently located with respect to the next keystroke to be input.

  1. Keyboard authors can keep a marker together with a character of interest by emitting the marker just previous to that character.

For example, given a key output="\m{marker}X", the marker will proceed X regardless of any normalization. (If output="X\m{marker}" were used, and X were to reorder with other characters, the marker would no longer be adjacent to the X.)

  1. Markers which are at the end of the input remain at the end of input during normalization.

For example, given input context which ends with a marker, such as ...ABCDX\m{marker}, the marker will remain at the end of the input context regardless of any normalization.

The 'gluing' is only applicable during one particular processing step. It does not persist or affect further processing steps or future keystrokes.

Data Model: Marker

For purposes of this algorithm, a Marker is an opaque data type which has one property, its ID. See Markers for a discussion of the marker ID.

Data Model: string

For purposes of this algorithm, a string is an array of elements, where each element is either a codepoint or a Marker. For example, a key in the XML such as <key id="sha" output="𐓯\m{mymarker}x" /> would produce a string with three elements:

  1. The codepoint U+104EF
  2. The Marker named mymarker
  3. The codepoint U+0078

If this string were output to an application, it would be converted to plain text by removing all markers, which would yield the plain text string with only two codepoints: 𐓯x.

Data Model: MarkerEntry

This algorithm uses a temporary data structure which is an ordered array of MarkerEntry elements.

Each MarkerEntry element has the following properties:

Marker Algorithm Overview

This algorithm has three main phases to it.

  1. Parsing/Removing Markers

    In this phase, the input string is analyzed to locate all markers. Metadata about each marker is stored in a temporary MarkerArray data structure. Markers are removed from the input string, leaving only plain text.

  2. Plain Text Processing

    This phase is performed on the plain text string, such as NFD normalization.

  3. Re-Adding Markers

    Finally, markers are re-added to the plain text string using the MarkerEntry metadata from step 1. This phase results in a string which contains both codepoints and markers.

Phase 1: Parsing/Removing Markers

Given an input string s

  1. Initialize an empty MarkerEntry array e
  2. Initialize an empty Marker array pending
  3. Loop through each element i of the input s
    1. If i is a Marker:
      1. add the marker i to the end of pending
      2. remove the marker from the input string s
    2. else if i is a codepoint:
      1. Decompose i into NFD form into a plain text string array of codepoints d
      2. Add an element with glue=d[0] (the first codepoint of d) and divider? = true to the end of e
      3. For every marker m in pending:
        1. Add an element with glue=d[0] and marker=m and divider? = false to the end of e
      4. Clear the pending array.
      5. Finally, for every codepoint c in d following the initial codepoint: (d[1]..):
        1. Add an element with glue=c and divider? = true to the end of e
  4. At the end of text,
    1. Add an element with glue=END and divider?=true to the end of e
    2. For every marker m in pending:
      1. Add an element with glue=END and marker=m and divider? = false to the end of e

The string s is now plain text and can be processed by the next phase.

The array e will be used in Phase 3.

Phase 2: Plain Text Processing

See UAX #15 for an overview of the process. An existing Unicode-compliant API can be used here.

Phase 3: Adding Markers

  1. Initialize an empty output string o
  2. Loop through the elements p of the array e from end to beginning (backwards)
    1. If p.glue isn't END:
      1. break out of the loop
    2. If p.divider? == false:
      1. Prepend marker p.marker to the output string o
    3. Set p.processed?=true (so we don't process this again)
  3. Loop through each codepoint i ( in the plain text input string ) from end to beginning (backwards)
    1. Prepend i to output o
    2. Loop through the elements p of the array e from end to beginning (backwards)
      1. If p.processed? == true:
        1. Continue the inner loop (was already processed)
      2. If p.glue isn't i
        1. Continue the inner loop (wrong glue, not applicable)
      3. If p.divider? == true:
        1. Break out of the inner loop (reached end of this 'glue' char)
      4. Prepend marker p.marker to the output string o
      5. Set p.processed?=true (so we don't process this again)
  4. o is now the output string including markers.

Example Normalization with Markers

Example 1

Consider this example, without markers:

The combining marks are reordered.

Example 2

If we add markers:

Note that the marker is 'glued' to the following character. In the above example, \m{marker} was 'glued' to the \u{0320}.

Example 2

A second example:

Here \m{marker2} is 'glued' to the end of the string. However, if additional text is added such as by a subsequent keystroke (which may add an additional combining character, for example), this marker may be 'glued' to that following text.

Markers remain in the same normalization-safe segment during normalization. Consider:

Example 3

There are two normalization-safe segments here:

  1. e\u{0300}\m{marker1}\u{0320}
  2. a\u{0300}\m{marker2}\u{0320}

Normalization (and marker rearranging) effectively occurs within each segment. While \m{marker1} is 'glued' to the \u{0320}, it is glued within the first segment and has no effect on the second segment.

Normalization and Character Classes

If pre-composed (non-NFD) characters are used in character classes, such as [á-é], these may not match as keyboard authors expect, as the U+00E1 character (á) will not occur in NFD form. Thus this may be masking serious errors in the data.

Tools that process keyboard data must reject the data when character classes include non-NFD characters.

The above should be written instead as a regex (á|â|ã|ä|å|æ|ç|è|é). Alternatively, it could be written as a set variable <set id="Example" value="á â ã ä å æ ç è é"/> and matched as $[Example].

There is another case where there is no explicit mention of a non-NFD character, but the character class could include non-NFD characters, such as the range [\u{0020}-\u{01FF}]. For these, the tools should raise a warning by default.

Normalization and Reorder elements

reorder elements operate on NFD codepoints.

Normalization-safe Segments

For purposes of this algorithm, "normalization-safe segments" are defined as a string of codepoints which are

  1. already in NFD, and
  2. begin with a character with Canonical Combining Class of 0.

See UAX #15 Section 9.1: Stable Code Points for related discussion. Text under consideration can be segmented by locating such characters.

Normalization and Output

On output, text will be normalized into a specified normalization form. That form will typically be NFC, but an implementation may allow a calling application to override the choice of normalization form. For example, many platforms may request NFC as the output format. In such a case, all text emitted via the keyboard will be transformed into NFC.

Existing text in a document will only have normalization applied within a single normalization-safe segment from the caret. The output will not contain any markers, thus any normalization is unaffected by any markers embedded within the segment.

For example, the sequence e\m{marker}\u{300} would be output in NFC as è. The marker is removed and has no effect on the output.

Disabling Normalization

The attribute value normalization="disabled" can be used to indicate that no automatic normalization is to be applied in input, matching, or output. Using this setting should be done with caution:

The majority of the above section only applies when normalization="disabled" is not used.

Element Hierarchy

This section describes the XML elements in a keyboard layout file, beginning with the top level element <keyboard3>.

Element: keyboard3

This is the top level element. All other elements defined below are under this element.

Syntax

<keyboard3 locale="…localeId">
    <!-- …definition of the layout as described by the elements defined below -->
</keyboard3>

Parents: none

Children: displays, import, info, keys, flicks, layers, locales, settings, special, transforms, variables, version

Occurrence: required, single

Attribute: conformsTo (required)

This attribute value distinguishes the keyboard from prior versions, and it also specifies the minimum CLDR major version required.

This attribute value must be a whole number of 45 or greater. See cldrVersion

<keyboard3 … conformsTo="45"/>

Attribute: locale (required)

This attribute value contains the primary locale of the keyboard using BCP 47 Unicode locale identifiers - for example "el" for Greek. Sometimes, the locale may not specify the base language. For example, a Devanagari keyboard for many languages could be specified by BCP-47 code: "und-Deva". However, it is better to list out the languages explicitly using the locales element.

For further details about the choice of locale ID, see Keyboard IDs.

Example (for illustrative purposes only, not indicative of the real data)

<keyboard3 locale="ka">
  …
</keyboard3>

<keyboard3 locale="fr-CH-t-k0-azerty">
  …
</keyboard3>

Element: locales

The optional <locales> element allows specifying additional or alternate locales. Denotes intentional support for an extra language, not just that a keyboard incidentally supports a language’s orthography.

Syntax

<locales>
    <locale id="…"/>
    <locale id="…"/>
</locales>

Parents: keyboard3

Children: locale

Occurrence: optional, single

Element: locale

The optional <locales> element allows specifying additional or alternate locales. Denotes intentional support for an extra language, not just that a keyboard incidentally supports a language’s orthography.

Syntax

<locale id="…id"/>

Parents: locales

Children: none

Occurrence: optional, multiple

Attribute: id (required)

The BCP 47 locale ID of an additional language supported by this keyboard. Must not include the -k0- subtag for this additional language.

Example

See Principles for Keyboard IDs for discussion and further examples.

<!-- Pan Nigerian Keyboard-->
<keyboard3 locale="mul-Latn-NG-t-k0-panng">
    <locales>
        <locale id="ha"/>
        <locale id="ig"/>
        <!-- others … -->
    </locales>
</keyboard3>

Element: version

Element used to keep track of the source data version.

Syntax

<version number="…number">

Parents: keyboard3

Children: none

Occurrence: optional, single

Attribute: number (required)

Must be a [SEMVER] compatible version number, such as 1.0.0 or 38.0.0-beta.11

Attribute: cldrVersion (fixed by DTD)

The CLDR specification version that is associated with this data file. This value is fixed and is inherited from the DTD file and therefore does not show up directly in the XML file.

Example

<keyboard3 locale="tok">
    …
    <version number="1"/>
    …
</keyboard3>

Element: info

Element containing informative properties about the layout, for displaying in user interfaces etc.

Syntax

<info
      name="…name"
      author="…author"
      layout="…hint of the layout"
      indicator="…short identifier" />

Parents: keyboard3

Children: none

Occurrence: required, single

Attribute: name (required)

Note that this is the only required attribute for the <info> element.

This attribute is an informative name for the keyboard.

<keyboard3 locale="bg-t-k0-phonetic-trad">
    …
    <info name="Bulgarian (Phonetic Traditional)" />
    …
</keyboard3>

Attribute: author

The author attribute value contains the name of the author of the layout file.

Attribute: layout

The layout attribute describes the layout pattern, such as QWERTY, DVORAK, INSCRIPT, etc. typically used to distinguish various layouts for the same language.

This attribute is not localized, but is an informative identifier for implementation use.

Attribute: indicator

The indicator attribute describes a short string to be used in currently selected layout indicator, such as US, SI9 etc. Typically, this is shown on a UI element that allows switching keyboard layouts and/or input languages.

This attribute is not localized.

Element: settings

An element used to keep track of layout-specific settings by implementations. This element may or may not show up on a layout. These settings reflect the normal practice by the implementation. However, an implementation using the data may customize the behavior.

Syntax

<settings normalization="disabled" />

Parents: keyboard3

Children: none

Occurrence: optional, single

Attribute: normalization="disabled"

The presence of this attribute indicates that normalization will not be applied to the input text, matching, or the output. See Normalization for additional details.

Note: while this attribute is allowed by the specification, it should be used with caution.

Example

<keyboard3 locale="bg">
    …
    <settings normalization="disabled" />
    …
</keyboard3>

Element: keys

This element defines the properties of all possible keys via <key> elements used in all layouts. It is a “bag of keys” without specifying any ordering or relation between the keys. There is only a single <keys> element in each layout.

Syntax

<keys>
    <key … />
    <key … />
    <key … />
</keys>

Parents: keyboard3 Children: key Occurrence: optional, single

Element: key

This element defines a mapping between an abstract key and its output. This element must have the keys element as its parent. The key element is referenced by the keys= attribute of the row element.

Syntax

<key
 id="…keyId"
 flickId="…flickId"
 gap="true"
 longPressKeyIds="…list of keyIds"
 longPressDefaultKeyId="…keyId"
 multiTapKeyIds="…listId"
 stretch="true"
 layerId="…layerId"
 output="…string"
 width="…number"
 />

Parents: keys

Children: none

Occurrence: optional, multiple

Note: The id attribute is required.

Note: at least one of layerId, gap, or output are required.

Attribute: id

The id attribute uniquely identifies the key. NMTOKEN. It can (but needn't be) the key name (a, b, c, A, B, C, …), or any other valid token (e-acute, alef, alif, alpha, …).

In the future, this attribute’s definition is expected to be updated to align with UAX#31.

Attribute: flickId="…flickId" (optional)

The flickId attribute indicates that this key makes use of a flick set with the specified id.

Attribute: gap="true" (optional)

The gap attribute indicates that this key does not have any appearance, but causes a "gap" of the specified number of key widths. Can be used with width to set a width.

<key id="mediumgap" gap="true" width="1.5"/>

Attribute: longPressKeyIds="…list of keyIds" (optional)

A space-separated ordered list of key element ids, which keys which can be emitted by "long-pressing" this key. This feature is prominent in mobile devices.

In a list of keys specified by longPressKeyIds, the key matching longPressDefaultKeyId attribute (if present) specifies the default long-press target, which could be different than the first element. It is an error if the longPressDefaultKeyId key is not in the longPressKeyIds list.

Implementations shall ignore any gestures (such as flick, multiTap, longPress) defined on keys in the longPressKeyIds list.

For example, if the default key is a key whose display value is {, an implementation might render the key as follows:

keycap hint

Example:

<displays>
   <displays output="\m{marker}" display="{" />
</displays>

<keys>
   <key id="o" output="o" longPressKeyIds="o-acute marker" longPressDefaultKeyId="marker">
   <key id="o-acute" output="ó"/>
   <key id="marker" display="{"/>
</key>

Attribute: longPressDefaultKeyId="…keyId" (optional)

Specifies the default key, by id, in a list of long-press keys. See the discussion of LongPressKeyIds, above.

Attribute: multiTapKeyIds (optional)

A space-separated ordered list of key element ids, which keys, where each successive key in the list is produced by the corresponding number of quick taps. It is an error for a key to reference itself in the multiTapKeyIds list.

Implementations shall ignore any gestures (such as flick, multiTap, longPress) defined on keys in the multiTapKeyIds list.

Example:

<keys>
   <key id="a" output="a" multiTapKeyIds="bb c d">
   <key id="bb" output="bb" />
   <key id="c" output="c" />
   <key id="d" output="d" />
</key>

Note: Behavior past the end of the multiTap list is implementation specific.

Attribute: stretch="true" (optional)

The stretch attribute indicates that a touch layout may stretch this key to fill available horizontal space on the row. This is used, for example, on the spacebar. Note that stretch= is ignored for hardware layouts.

Attribute: layerId="shift" (optional)

The layerId attribute indicates that this key switches to another layer with the specified id (such as <layer id="shift"/> in this example). Note that a key may have both a layerId= and a output= attribute, indicating that the key outputs prior to switching layers. Also note that layerId= is ignored for hardware layouts: their shifting is controlled via the modifier keys.

This attribute is an NMTOKEN.

In the future, this attribute’s definition is expected to be updated to align with UAX#31.

Attribute: output

The output attribute value contains the sequence of characters that is emitted when pressing this particular key. Control characters, whitespace (other than the regular space character) and combining marks in this attribute are escaped using the \u{…} notation. More than one key may output the same output.

The output attribute may also contain the \m{…markerId} syntax to insert a marker. See the definition of markers.

Attribute: width="1.2" (optional, default "1.0")

The width attribute indicates that this key has a different width than other keys, by the specified number of key widths.

<key id="wide-a" output="a" width="1.2"/>
<key id="wide-gap" gap="true" width="2.5"/>
Implied Keys

Not all keys need to be listed explicitly. The following two can be assumed to already exist:

<key id="gap" gap="true" width="1"/>
<key id="space" output=" " stretch="true" width="1"/>

In addition, these 62 keys, comprising 10 digit keys, 26 Latin lower-case keys, and 26 Latin upper-case keys, where the id is the same as the to, are assumed to exist:

<key id="0" output="0"/>
<key id="1" output="1"/>
<key id="2" output="2"/>
…
<key id="A" output="A"/>
<key id="B" output="B"/>
<key id="C" output="C"/>
…
<key id="a" output="a"/>
<key id="b" output="b"/>
<key id="c" output="c"/>
…

These implied keys are available in a data file named keyboards/import/keys-Latn-implied.xml in the CLDR distribution for the convenience of implementations.

Thus, the implied keys behave as if the following import were present.

<keyboard3>
    <keys>
        <import base="cldr" path="45/keys-Latn-implied.xml" />
    </keys>
</keyboard3>

Note: All implied keys may be overridden, as with all other imported data items. See the import element for more details.

Element: flicks

The flicks element is a collection of flick elements.

Parents: keyboard3

Children: flick, import, special

Occurrence: optional, single

Element: flick

The flick element is used to generate results from a "flick" of the finger on a mobile device.

Syntax

<keyboard3>
    <keys>
        <key id="a" flicks="a-flicks" output="a" />
    </keys>
    <flicks>
        <flick id="a-flicks">
            <flickSegment … />
            <flickSegment … />
            <flickSegment … />
        </flick>
    </flicks>
</keyboard3>

Parents: flicks

Children: flickSegment, special

Occurrence: optional, multiple

Attribute: id (required)

The id attribute identifies the flicks. It can be any NMTOKEN.

The flick elements do not share a namespace with the keys, so it would also be allowed to have <key id="a" flick="a"/>

In the future, this attribute’s definition is expected to be updated to align with UAX#31.

Element: flickSegment

Parents: flick

Children: none

Occurrence: required, multiple

Attribute: directions (required)

The directions attribute value is a space-delimited list of keywords, that describe a path, currently restricted to the cardinal and intercardinal directions {n e s w ne nw se sw}.

Attribute: keyId (required)

The keyId attribute value is the result of (one or more) flicks.

Implementations shall ignore any gestures (such as flick, multiTap, longPress) defined on the key specified by keyId.

Example where a flick to the Northeast then South produces Å.

<keys>
    <key id="something" flickId="a" output="Something" />
    <key id="A-ring" output="A-ring" />
</keys>

<flicks>
    <flick id="a">
        <flickSegment directions="ne s" keyId="A-ring" />
    </flick>
</flicks>

Element: import

The import element is used to reference another xml file so that elements are imported from another file. The use case is to be able to import a standard set of transforms and similar from the CLDR repository, especially to be able to share common information relevant to a particular script. The intent is for each single XML file to contain all that is needed for a keyboard layout, other than required standard import data from the CLDR repository.

<import> can be used as a child of a number of elements (see the Parents section immediately below). Multiple <import> elements may be used, however, <import> elements must come before any other sibling elements. If two identical elements are defined, the later element will take precedence, that is, override.

Note: imported files do not have any indication of their normalization mode. For this reason, the keyboard author must verify that the imported file is of a compatible normalization mode. See the settings element for further details.

Syntax

<import base="cldr" path="45/keys-Zyyy-punctuation.xml"/>

Parents: displays, keyboard3, keys, flicks, layers, transformGroup, transforms, variables Children: none

Occurrence: optional, multiple

Attribute: base

The base may be omitted (indicating a local import) or have the value "cldr".

Note: base="cldr" is required for all <import> statements within keyboard files in the CLDR repository.

Attribute: path (required)

If base is cldr, then the path must start with a CLDR major version (such as 45) representing the CLDR version to pull imports from. The imports are located in the keyboard/import subdirectory of the CLDR source repository. Implementations are not required to have all CLDR versions available to them.

If base is omitted, then path is an absolute or relative file path.

Further Examples

<!-- in a keyboard xml file-->
…
<transforms type="simple">
    <import base="cldr" path="45/transforms-example.xml"/>
    <transform from="` " to="`" />
    <transform from="^ " to="^" />
</transforms>
…


<!-- contents of transforms-example.xml -->
<?xml version="1.0" encoding="UTF-8"?>
<!DOCTYPE transforms SYSTEM "../dtd/ldmlKeyboard3.dtd">
<transforms>
    <!-- begin imported part-->
    <transform from="`a" to="à" />
    <transform from="`e" to="è" />
    <transform from="`i" to="ì" />
    <transform from="`o" to="ò" />
    <transform from="`u" to="ù" />
    <!-- end imported part -->
</transforms>

Note: The DOCTYPE and root element, here transforms, is the same as the parent of the <import/> element. It is an error to import an XML file whose root element is different than the parent element of the <import/> element.

After loading, the above example will be the equivalent of the following.

<transforms type="simple">
    <!-- begin imported part-->
    <transform from="`a" to="à" />
    <transform from="`e" to="è" />
    <transform from="`i" to="ì" />
    <transform from="`o" to="ò" />
    <transform from="`u" to="ù" />
    <!-- end imported part -->

    <!-- this line is after the import -->
    <transform from="^ " to="^" />
    <transform from="` " to="`" />
</transforms>

Element: displays

The displays element consists of a list of display subelements.

Syntax

<displays>
    <display … />
    <display … />
    …
</displays>

Parents: keyboard3

Children: display, displayOptions, special

Occurrence: optional, single

Element: display

The display elements can be used to describe what is to be displayed on the keytops for various keys. For the most part, such explicit information is unnecessary since the @to element from the keys/key element will be used for keytop display.

Note: displays elements are designed to be shared across many different keyboard layout descriptions, and imported with <import> where needed.

Non-spacing marks on keytops

For non-spacing marks, U+25CC ◌ is used as a base. It is an error to use a nonspacing character without a base in the display attribute. For example, display="\u{0303}" would produce an error.

A key which outputs a combining tilde (U+0303) could be represented as either of the following:

    <display output="\u{0303}" display="◌̃" />  <!-- \u{25CC} \u{0303}-->
    <display output="\u{0303}" display="\u{25cc}\u{0303}" />  <!-- also acceptable -->

This way, a key which outputs a combining tilde (U+0303) will be represented as ◌̃ (a tilde on a dotted circle).

Users of some scripts/languages may prefer a different base than U+25CC. See <displayOptions baseCharacter=…/>.

Syntax

<display output="…string" display="…string" />

Parents: displays

Children: none

Occurrence: required, multiple

One of the output or id attributes is required.

Note: There is currently no way to indicate a custom display for a key without output (i.e. without a to= attribute), nor is there a way to indicate that such a key has a standardized identity (e.g. that a key should be identified as a “Shift”). These may be addressed in future versions of this standard.

Attribute: output (optional)

Specifies the character or character sequence from the keys/key element that is to have a special display. This attribute may be escaped with \u notation, see Escaping. The output attribute may also contain the \m{…} syntax to reference a marker. See Markers. Implementations may highlight a displayed marker, such as with a lighter text color, or a yellow highlight. String variables may be substituted. See String variables

Attribute: id (optional)

Specifies the key id. This is useful for keys which do not produce any output (no output= value), such as a shift key.

This attribute must match [A-Za-z0-9][A-Za-z0-9-]*

Attribute: display (required)

Required and specifies the character sequence that should be displayed on the keytop for any key that generates the @output sequence or has the @id. (It is an error if the value of the display attribute is the same as the value of the output attribute, this would be an extraneous entry.)

String variables may be substituted. See String variables

This attribute may be escaped with \u notation, see Escaping.

Example

<keyboard3>
    <keys>
        <key id="grave" output="\u{0300}" /> <!-- combining grave -->
        <key id="marker" output="\m{acute}" /> <!-- generates a marker-->
        <key id="numeric" layerId="numeric" /> <!-- changes layers-->
    </keys>
    <displays>
        <display output="\u{0300}" display="ˋ" /> <!-- \u{02CB} -->
        <display keyId="numeric"  display="#" /> <!-- display the layer shift key as # -->
        <display output="\m{acute}" display="´" /> <!-- Display \m{acute} as ´ -->
    </displays>
</keyboard3>

To allow displays elements to be shared across keyboards, there is no requirement that @output in a display element matches any @output/@id in any keys/key element in the keyboard description.

Element: displayOptions

The displayOptions is an optional singleton element providing additional settings on this displays. It is structured so as to provide for future flexibility in such options.

Syntax

<displays>
    <display …/>
    <displayOptions baseCharacter="x"/>
</displays>

Parents: displays

Children: none

Occurrence: optional, single

Attribute: baseCharacter (optional)

Note: At present, this is the only option settable in the displayOptions.

Some scripts/languages may prefer a different base than U+25CC. For Lao for example, x is often used as a base instead of ◌. Setting baseCharacter="x" (for example) is a hint to the implementation which requests U+25CC to be substituted with x on display. As a hint, the implementation may ignore this option.

Note that not all base characters will be suitable as bases for combining marks.

This attribute may be escaped with \u notation, see Escaping.

Element: forms

This element contains a set of form elements which define the layout of a particular hardware form.

Parents: keyboard3

Children: import, form, special

Occurrence: optional, single

Syntax

<forms>
    <form id="iso">
        <!-- … -->
    </form>
    <form id="us">
        <!-- … -->
    </form>
</forms>

Element: form

This element contains a specific form element which defines the layout of a particular hardware form.

Note: Most keyboards will not need to use this element directly, and the CLDR repository will not accept keyboards which define a custom form element. This element is provided for two reasons:

  1. To formally specify the standard hardware arrangements used with CLDR for implementations. Implementations can verify the arrangement, and validate keyboards against the number of rows and the number of keys per row.

  2. To allow a way to customize the scancode layout for keyboards not intended to be included in the common CLDR repository.

See Implied Form Values, below.

Parents: forms

Children: scanCodes, special

Occurrence: optional, multiple

Attribute: id (required)

This attribute specifies the form id. The value may not be touch.

Syntax

<form id="us">
    <scanCodes codes="00 01 02"/>
    <scanCodes codes="03 04 05"/>
</form>
Implied Form Values

There is an implied set of <form> elements corresponding to the default forms, thus implementations must behave as if there was the following import statement:

<keyboard3>
    <forms>
        <import base="cldr" path="45/scanCodes-implied.xml" /> <!-- the version will match the current conformsTo of the file -->
    </forms>
</keyboard3>

Here is a summary of the implied form elements. Keyboards included in the CLDR Repository must only use these form= values and may not override the scanCodes.

Element: scanCodes

This element contains a keyboard row, and defines the scan codes for the non-frame keys in that row.

Parents: form

Children: none

Occurrence: required, multiple

Attribute: codes (required)

The codes attribute is a space-separated list of 2-digit hex bytes, each representing a scan code.

Syntax

<scanCodes codes="29 02 03 04 05 06 07 08 09 0A 0B 0C 0D" />

Element: layers

This element contains a set of layer elements with a specific physical form factor, whether hardware or touch layout.

Parents: keyboard3

Children: import, layer, special

Occurrence: required, multiple

Attribute: form (required)

This attribute specifies the physical layout of a hardware keyboard, or that the form is a touch layout.

When using an on-screen touch keyboard, if the keyboard does not specify a <layers form="touch"> element, a <layers form="…formId"> element can be used as an fallback alternative. If there is no hardware form, the implementation may need to choose a different keyboard file, or use some other fallback behavior when using a hardware keyboard.

Because a hardware keyboard facilitates non-trivial amounts of text input, and many touch devices can also be connected to a hardware keyboard, it is recommended to always have at least one hardware (non-touch) form.

Multiple <layers form="touch"> elements are allowed with distinct minDeviceWidth values. At most one hardware (non-touch) <layers> element is allowed. If a different key arrangement is desired between, for example, us and iso formats, these should be separated into two different keyboards.

The typical keyboard author will be designing a keyboard based on their circumstances and the hardware that they are using. So, for example, if they are in South East Asia, they will almost certainly be using an 101 key hardware keyboard with US key caps. So we want them to be able to reference that (<layers form="us">) in their design, rather than having to work with an unfamiliar form.

A mismatch between the hardware layout in the keyboard file, and the actual hardware used by the user could result in some keys being inaccessible to the user if their hardware cannot generate the scancodes corresponding to the layout specified by the form= attribute. Such keys could be accessed only via an on-screen keyboard utility. Conversely, a user with hardware keys that are not present in the specified form= will result in some hardware keys which have no function when pressed.

The value of the form= attribute may be touch, or correspond to a form element. See form.

Attribute: minDeviceWidth

This attribute specifies the minimum required width, in millimeters (mm), of the touch surface. The layers entry with the greatest matching width will be selected. This attribute is intended for form="touch", but is supported for hardware forms.

This must be a whole number between 1 and 999, inclusive.

Element: layer

A layer element describes the configuration of keys on a particular layer of a keyboard. It contains one or more row elements to describe which keys exist in each row.

Syntax

<layer id="…layerId" modifiers="…modifier modifier, …modifier modifier, …">
    <row …/>
    <row …/>
    …
</layer>

Parents: keyboard3

Children: row, special

Occurrence: optional, multiple

Attribute id (required for touch)

The id attribute identifies the layer for touch layouts. This identifier specifies the layout as the target for layer switching, as specified by the switch= attribute on the <key> element. Touch layouts must have one layer with id="base" to serve as the base layer.

Must match [A-Za-z0-9][A-Za-z0-9-]*

Attribute: modifiers (required for hardware)

This has two roles. It acts as an identifier for the layer element for hardware keyboards (in the absence of the id= element) and also provides the linkage from the hardware modifiers into the correct layer.

For hardware layouts, the use of @modifiers as an identifier for a layer is sufficient since it is always unique among the set of layer elements in each form.

This attribute value is a list of lists. It is a comma-separated (,) list of modifier sets, and each modifier set is a space-separated list of modifier components.

Each modifier component must match [A-Za-z0-9]+. Extra whitespace is ignored.

To indicate that no modifiers apply, the reserved name of none is used.

Syntax

<layer id="base"        modifiers="none">
    <row keys="a" />
</layer>

<layer id="upper"       modifiers="shift">
    <row keys="A" />
</layer>

<layer id="altgr"       modifiers="altR">
    <row keys="a-umlaut" />
</layer>

<layer id="upper-altgr" modifiers="altR shift">
    <row keys="A-umlaut" />
</layer>

Layer Modifier Sets

The @modifiers attribute value contains one or more Layer Modifier Sets, separated by commas. For example, in the element <layer … modifiers="ctrlL altL, altR" … the attribute value consists of two sets:

The order of the sets and the order of the components within each set is not significant. However, for clarity in reading, the canonical order within a set is in the order listed in Layout Modifier Components; the canonical order for the sets should be first by the cardinality of the sets (least first), then alphabetical.

Layer Modifier Components

Within a Layer Modifier Set, the following modifier components can be used, separated by spaces.

  1. alt in this specification is referred to on some platforms as "opt" or "option".

  2. none and other may not be combined with any other components.

Modifier Left- and Right- keys

  1. L or R indicates a left- or right- side modifier only (such as altL) whereas alt indicates either left or right alt key (that is, altL or altR). ctrl indicates either left or right ctrl key (that is, ctrlL or ctrlR).

  2. Keyboard implementations must warn if a keyboard mixes alt with altL/altR, or ctrl with ctrlL/ctrlR.

  3. Left- and right- side modifiers may not be mixed together in a single modifier attribute value, so neither altL ctrlR" nor altL altR are allowed.

  4. shift indicates either shift key. The left and right shift keys are not distinguishable in this specification.

Layer Modifier Matching

Layers are matched exactly based on the modifier keys which are down. For example:

Multiple modifier sets are separated by commas. For example, none, shift caps will match either no modifiers or shift and caps. ctrlL altL, altR will match either left-control and left-alt, or right-alt.

Keystrokes must be ignored where there isn’t a layer that explicitly matches nor a layer with other. Example: If there is a ctrl and shift layer, but no ctrl shift nor other layer, no output will result from ctrl shift X.

Layers are not allowed to overlap in their matching. For example, the keyboard author will receive an error if one layer specifies alt shift and another layer specifies altR shift.

There is one special case: the other layer matches if and only if no other layer matches. Thus logically the other layer is matched after all other layers have been checked.

Because there is no overlap allowed between layers, the order of <layer> elements is not significant.

Note: The modifier syntax may be enhanced in the future, but will remain backwards compatible with the syntax described here.

Element: row

A row element describes the keys that are present in the row of a keyboard.

Syntax

<row keys="…keyId …keyId …" />

Parents: layer

Children: none

Occurrence: required, multiple

Attribute: keys (required)

This is a string that lists the id of key elements for each of the keys in a row, whether those are explicitly listed in the file or are implied. See the key documentation for more detail.

For non-touch forms, the number of keys in each row may not exceed the number of scan codes defined for that row, and the number of rows may not exceed the defined number of rows for that form. See scanCodes;

Example

Here is an example of a row element:

<row keys="a z e r t y u i o p caret dollar" />

Element: variables

Parents: keyboard3

Children: import, special, string, set, uset

Occurrence: optional, single

This is a container for variables to be used with transform, display and key elements.

Note that the id= attribute value must be unique across all children of the variables element.

Example

<variables>
    <string id="y" value="yes" /> <!-- a simple string-->
    <set id="upper" value="A B C D E FF" /> <!-- a set with 6 items -->
    <uset id="consonants" value="[कसतनमह]" /> <!-- a UnicodeSet -->
</variables>

Element: string

Parents: variables

Children: none

Occurrence: optional, multiple

This element contains a single string which is used by the transform elements for string matching and substitution, as well as by the key and display elements.

Attribute: id (required)

Specifies the identifier (name) of this string. All ids must be unique across all types of variables.

id must match [0-9A-Za-z_]{1,32}

Attribute: value (required)

Strings may contain whitespaces. However, for clarity, it is recommended to escape spacing marks, even in strings. This attribute value may be escaped with \u notation, see Escaping. Variables may refer to other string variables if they have been previously defined, using ${string} syntax. Markers may be included with the \m{…} notation.

Example

<variables>
    <string id="cluster_hi" value="हि" /> <!-- a string -->
    <string id="zwnj" value="\u{200C}"/> <!-- single codepoint -->
    <string id="acute" value="\m{acute}"/> <!-- refer to a marker -->
    <string id="backquote" value="`"/>
    <string id="zwnj_acute" value="${zwnj}${acute}"  /> <!-- Combine two variables -->
    <string id="zwnj_sp_acute" value="${zwnj}\u{0020}${acute}"  /> <!-- Combine two variables -->
</variables>

These may be then used in multiple contexts:

<!-- as part of a regex -->
<transform from="${cluster_hi}X" to="X" />
<transform from="Y" to="${cluster_hi}" />
…
<!-- as part of a key bag  -->
<key id="hi_key" output="${cluster_hi}" />
<key id="acute_key" output="${acute}" />
…
<!-- Display ´ instead of the non-displayable marker -->
<display output="${acute}" display="${backquote}" />

Element: set

Parents: variables

Children: none

Occurrence: optional, multiple

This element contains a set of strings used by the transform elements for string matching and substitution.

Attribute: id (required)

Specifies the identifier (name) of this set. All ids must be unique across all types of variables.

id must match [0-9A-Za-z_]{1,32}

Attribute: value (required)

The value attribute value is always a set of strings separated by whitespace, even if there is only a single item in the set, such as "A". Leading and trailing whitespace is ignored. This attribute value may be escaped with \u notation, see Escaping. Sets may refer to other string variables if they have been previously defined, using ${string} syntax, or to other previously-defined sets using $[set] syntax. Set references must be separated by whitespace: $[set1]$[set2] is an error; instead use $[set1] $[set2]. Markers may be included with the \m{…} notation.

Examples

<variables>
    <set id="upper" value="A B CC D E FF " /> <!-- 6 items -->
    <set id="lower" value="a b c  d e  f " /> <!-- 6 items -->
    <set id="upper_or_lower" value="$[upper] $[lower]"  /> <!-- Concatenate two sets -->
    <set id="lower_or_upper" value="$[lower] $[upper]"  /> <!-- Concatenate two sets -->
    <set id="a" value="A"/> <!-- Just one element, an 'A'-->
    <set id="cluster_or_zwnj" value="${hi_cluster} ${zwnj}"/> <!-- 2 items: "हि \u${200C}"-->
</variables>

Match "X" followed by any uppercase letter:

<transform from="X$[upper]" to="…" />

Map from upper to lower:

<transform from="($[upper])" to="$[1:lower]" />

See transform for further details and syntax.

Element: uset

Parents: variables

Children: none

Occurrence: optional, multiple

This element contains a set, using a subset of the UnicodeSet format, used by the transform elements for string matching and substitution. Note important restrictions on the syntax below.

Attribute: id (required)

Specifies the identifier (name) of this uset. All ids must be unique across all types of variables.

id must match [0-9A-Za-z_]{1,32}

Attribute: value (required)

String value in a subset of UnicodeSet format. Leading and trailing whitespace is ignored. Variables may refer to other string variables if they have been previously defined, using ${string} syntax, or to other previously-defined uset elements (not set elements) using $[...usetId] syntax.

Rationale: allowing property notation would make keyboard implementations dependent on a particular version of Unicode. However, implementations and tools may wish to pre-calculate the value of a particular uset, and "freeze" it as explicit code points. The example below of $[KhmrMn] matches nonspacing marks in the Khmr script.

Examples

<variables>
  <uset id="consonants" value="[कसतनमह]" /> <!-- unicode set range -->
  <uset id="range" value="[a-z D E F G \u{200A}]" /> <!-- a through z, plus a few others -->
  <uset id="newrange" value="[$[range]-[G]]" /> <!-- The above range, but not including G -->
  <uset id="KhmrMn" value="[\u{17B4}\u{17B5}\u{17B7}-\u{17BD}\u{17C6}\u{17C9}-\u{17D3}\u{17DD}]"> <!--  [[:Khmr:][:Mn:]] as of Unicode 15.0-->
</variables>

Element: transforms

This element defines a group of one or more transform elements associated with this keyboard layout. This is used to support features such as dead-keys, character reordering, backspace behavior, etc. using a straightforward structure that works for all the keyboards tested, and that results in readable source data.

There can be multiple <transforms> elements, but only one for each type.

Syntax

<transforms type="…type">
    <transformGroup …/>
    <transformGroup …/>
    …
</transforms>

Parents: keyboard3

Children: import, special, transformGroup

Occurrence: optional, multiple

Attribute: type (required)

Values: simple, backspace

There are other keying behaviors that are needed particularly in handing complex orthographies from various parts of the world. The behaviors intended to be covered by the transforms are:

Markers

Markers are placeholders which record some state, but without producing normal visible text output. They were designed particularly to support dead-keys.

The marker ID is any valid NMTOKEN.

Consider the following abbreviated example:

    <display output="\m{circ_marker}" display="^" />
…
    <key id="circ_key" output="\m{circ_marker}" />
    <key id="e" output="e" />
…
    <transform from="\m{circ_marker}e" to="ê" />

  1. The user presses the circ_key key. The key can be shown with the keycap ^ due to the <display> element.

  2. The special marker, circ_marker, is added to the end of the input context.

    The input context does not match any transforms.

    The input context has:

    • …
    • marker circ_marker

  3. Also due to the <display> element, implementations can opt to display a visible ^ (perhaps visually distinct from a plain ^ carat). Implementations may opt to display nothing and only store the marker in the input context.

  4. The user now presses the e key, which is also added to the input context. The input context now has:

    • …
    • character e
    • marker circ_marker

  5. Now, the input context matches the transform. The e and the marker are replaced with ê.

    The input context now has:

    • …
    • character ê

Using markers to inhibit other transforms

Sometimes it is desirable to prevent transforms from having an effect. Perhaps two different keys output the same characters, with different key or modifier combinations, but only one of them is intended to participate in a transform.

Consider the following case, where pressing the keys X, e results in ^e, which is transformed into ê.

<keys>
    <key id="X" output="^"/>
    <key id="e" output="e" />
</keys>
<transforms>
    <transform from="^e" output="ê"/>
</transforms>

However, what if the user wanted to produce ^e without the transform taking effect? One strategy would be to use a marker, which won’t be visible in the output, but will inhibit the transform.

<keys>
    <key id="caret" output="^\m{no_transform}"/>
    <key id="X" output="^" />
    <key id="e" output="e" />
</keys>
…
<transforms>
    <!-- this wouldn't match the key caret output because of the marker -->
    <transform from="^e" output="ê"/>
</transforms>

Pressing caret e will result in ^e (with an invisible no_transform marker — note that any name could be used). The ^e won’t have the transform applied, at least while the marker’s context remains valid.

Another strategy might be to use a marker to indicate where transforms are desired, instead of where they aren't desired.

<keys>
    <key id="caret" output="^"/>
    <key id="X" output="^\m{transform}"/>
    <key id="e" output="e" />
</keys>
…
<transforms …>
    <!-- Won't match ^e without marker. -->
    <transform from="^\m{transform}e" output="ê"/>
</transforms>

In this way, only the X, e keys will produce ^e with a transform marker (again, any name could be used) which will cause the transform to be applied. One benefit is that navigating to an existing ^ in a document and adding an e will result in ^e, and this output will not be affected by the transform, because there will be no marker present there (remember that markers are not stored with the document but only recorded in memory temporarily during text input).

Please note important considerations for Normalization and Markers.

Effect of markers on final text

All markers must be removed before text is returned to the application from the input context. If the input context changes, such as if the cursor or mouse moves the insertion point somewhere else, all markers in the input context are removed.

Implementation Notes

Ideally, markers are implemented entirely out-of-band from the normal text stream. However, implementations may choose to map each marker to a Unicode private-use character for use only within the implementation’s processing and temporary storage in the input context.

For example, the first marker encountered could be represented as U+E000, the second by U+E001 and so on. If a regex processing engine were used, then those PUA characters could be processed through the existing regex processing engine. [^\u{E000}-\u{E009}] could be used as an expression to match a character that is not a marker, and [Ee]\u{E000} could match E or e followed by the first marker.

Such implementations must take care to remove all such markers (see prior section) from the resultant text. As well, implementations must take care to avoid conflicts if applications themselves are using PUA characters, such as is often done with not-yet-encoded scripts or characters.

Element: transformGroup

Parents: transforms

Children: import, reorder, special, transform

Occurrence: optional, multiple

A transformGroup contains a set of transform elements or reorder elements.

Each transformGroup is processed entirely before proceeding to the next one.

Each transformGroup element, after imports are processed, must have either reorder elements or transform elements, but not both. The <transformGroup> element may not be empty.

Examples

Example: transformGroup with transform elements

This is a transformGroup that consists of one or more transform elements, prefaced by one or more import elements. See the discussion of those elements for details. import elements in this group may not import reorder elements.

<transformGroup>
    <import path="…"/> <!-- optional import elements-->
    <transform />
    <!-- other <transform/> elements -->
</transformGroup>

Example: transformGroup with reorder elements

This is a transformGroup that consists of one or more transform elements, optionally prefaced by one or more import elements that import transform elements. See the discussion of those elements for details.

import elements in this group may not import transform elements.

<transformGroup>
    <import path="…"/> <!-- optional import elements-->
    <reorder … />
    <!-- other <reorder> elements -->
</transformGroup>

Element: transform

This element contains a single transform that may be performed using the keyboard layout. A transform is an element that specifies a set of conversions from sequences of code points into (one or more) other code points. For example, in most French keyboards hitting the ^ dead-key followed by the e key produces ê.

Matches are processed against the "input context", a temporary buffer containing all relevant text up to the insertion point. If the user moves the insertion point, the input context is discarded and recreated from the application’s text buffer. Implementations may discard the input context at any time.

The input context may contain, besides regular text, any Markers as a result of keys or transforms, since the insertion point was moved.

Using regular expression terminology, matches are done as if there was an implicit $ (match end of buffer) at the end of each pattern. In other words, <transform from="ke" …> will not match an input context ending with …keyboard, but it will match the last two codepoints of an input context ending with …awake.

All of the transform elements in a transformGroup are tested for a match, in order, until a match is found. Then, the matching element is processed, and then processing proceeds to the next transformGroup. If none of the transform elements match, processing proceeds without modification to the buffer to the next transformGroup.

Syntax

<transform from="…matching pattern" to="…output pattern"/>

Parents: transformGroup Children: none Occurrence: required, multiple

Attribute: from (required)

The from attribute value consists of an input rule for matching the input context.

The transform rule and output pattern uses a modified, mostly subsetted, regular expression syntax, with EcmaScript syntax (with the u Unicode flag) as its baseline reference (see MDN-REGEX). Differences from regex implementations will be noted.

Regex-like Syntax

Additional Features

The following are additions to standard Regex syntax.

Disallowed Regex Features

Attribute: to

This attribute value represents the characters that are output from the transform.

If this attribute is absent, it indicates that the no characters are output, such as with a backspace transform.

A final rule such as <transform from=".*"/> will remove all context which doesn’t match one of the prior rules.

Replacement syntax

Used in the to=

Element: reorder

The reorder transform consists of a <transformGroup> element containing <reorder> elements. Multiple such <transformGroup> elements may be contained in an enclosing <transforms> element.

One or more <import> elements are allowed to precede the <reorder> elements.

This transform has the job of reordering sequences of characters that have been typed, from their typed order to the desired output order. The primary concern in this transform is to sort combining marks into their correct relative order after a base, as described in this section. The reorder transforms can be quite complex, keyboard layouts will almost always import them.

The reordering algorithm consists of four parts:

  1. Create a sort key for each character in the input string. A sort key has 4 parts (primary, index, tertiary, quaternary):
    • The primary weight is the primary order value.
    • The secondary weight is the index, a position in the input string, usually of the character itself, but it may be of a character earlier in the string.
    • The tertiary weight is a tertiary order value (defaulting to 0).
    • The quaternary weight is the index of the character in the string. This is solely to ensure a stable sort for sequences of characters with the same tertiary weight.
  2. Mark each character as to whether it is a prebase character, one that is typed before the base and logically stored after. Thus it will have a primary order > 0.
  3. Use the sort key and the prebase mark to identify runs. A run starts with a prefix that contains any prebase characters and a single base character whose primary and tertiary key is 0. The run extends until, but not including, the start of the prefix of the next run or end of the string.
    • run := preBase* (primary=0 && tertiary=0) ((primary≠0 || tertiary≠0) && !preBase)*
  4. Sort the character order of each character in the run based on its sort key.

The primary order of a character with the Unicode property Canonical_Combining_Class (ccc) of 0 may well not be 0. In addition, a character may receive a different primary order dependent on context. For example, in the Devanagari sequence ka halant ka, the first ka would have a primary order 0 while the halant ka sequence would give both halant and the second ka a primary order > 0, for example 2. Note that “base” character in this discussion is not a Unicode base character. It is instead a character with primary=0.

In order to get the characters into the correct relative order, it is necessary not only to order combining marks relative to the base character, but also to order some combining marks in a subsequence following another combining mark. For example in Devanagari, a nukta may follow a consonant character, but it may also follow a conjunct consisting of consonant, halant, consonant. Notice that the second consonant is not, in this model, the start of a new run because some characters may need to be reordered to before the first base, for example repha. The repha would get primary < 0, and be sorted before the character with order = 0, which is, in the case of Devanagari, the initial consonant of the orthographic syllable.

The reorder transform consists of <reorder> elements encapsulated in a <transformGroup> element. Each element is a rule that matches against a string of characters with the action of setting the various ordering attributes (primary, tertiary, tertiaryBase, preBase) for the matched characters in the string.

The relative ordering of <reorder> elements is not significant.

Syntax

<transformGroup>
    <!-- one or more <import/> elements are allowed at this point -->
    <reorder from="…combination of characters"
    before="…look-behind required match"
    order="…list of weights"
    tertiary="…list of weights"
    tertiaryBase="…list of true/false"
    preBase="…list of true/false" />
    <!-- other <reorder/> elements… -->
</transformGroup>

Parents: transformGroup Children: none Occurrence: optional, multiple

Attribute: from (required)

This attribute value contains a string of elements. Each element matches one character and may consist of a codepoint or a UnicodeSet (both as defined in UTS #35 Part One).

Attribute: before

This attribute value contains the element string that must match the string immediately preceding the start of the string that the @from matches.

Attribute: order

This attribute value gives the primary order for the elements in the matched string in the @from attribute. The value is a simple integer between -128 and +127 inclusive, or a space separated list of such integers. For a single integer, it is applied to all the elements in the matched string. Details of such list type attributes are given after all the attributes are described. If missing, the order value of all the matched characters is 0. We consider the order value for a matched character in the string.

A character with a zero tertiary value is a primary character and receives a sort key consisting of:

Attribute: tertiary

This attribute value gives the tertiary order value to the characters matched. The value is a simple integer between -128 and +127 inclusive, or a space separated list of such integers. If missing, the value for all the characters matched is 0. We consider the tertiary value for a matched character in the string.

A tertiary character receives its primary order and index from a previous character, which it is intended to sort closely after. The sort key for a tertiary character consists of:

Attribute: tertiaryBase

This attribute value is a space separated list of "true" or "false" values corresponding to each character matched. It is illegal for a tertiary character to have a true tertiaryBase value. For a primary character it marks that this character may have tertiary characters moved after it. When calculating the secondary weight for a tertiary character, the most recently encountered primary character with a true tertiaryBase attribute value is used. Primary characters with an @order value of 0 automatically are treated as having tertiaryBase true regardless of what is specified for them.

Attribute: preBase

This attribute value gives the prebase attribute for each character matched. The value may be "true" or "false" or a space separated list of such values. If missing the value for all the characters matched is false. It is illegal for a tertiary character to have a true prebase value.

If a primary character has a true prebase value then the character is marked as being typed before the base character of a run, even though it is intended to be stored after it. The primary order gives the intended position in the order after the base character, that the prebase character will end up. Thus @order shall not be 0. These characters are part of the run prefix. If such characters are typed then, in order to give the run a base character after which characters can be sorted, an appropriate base character, such as a dotted circle, is inserted into the output run, until a real base character has been typed. A value of "false" indicates that the character is not a prebase.

For @from attribute values with a match string length greater than 1, the sort key information (@order, @tertiary, @tertiaryBase, @preBase) may consist of a space-separated list of values, one for each element matched. The last value is repeated to fill out any missing values. Such a list may not contain more values than there are elements in the @from attribute:

if len(@from) < len(@list) then error
else
    while len(@from) > len(@list)
        append lastitem(@list) to @list
    endwhile
endif

Example

For example, consider the Northern Thai (nod-Lana, Tai Tham script) word: ᨡ᩠ᩅᩫ᩶ 'roasted'. This is ideally encoded as the following:

name kha sakot wa o t2
code 1A21 1A60 1A45 1A6B 1A76
ccc 0 9 0 0 230

(That sequence is already in NFC format.)

Some users may type the upper component of the vowel first, and the tone before or after the lower component. Thus someone might type it as:

name kha o t2 sakot wa
code 1A21 1A6B 1A76 1A60 1A45
ccc 0 0 230 9 0

The Unicode NFC format of that typed value reorders to:

name kha o sakot t2 wa
code 1A21 1A6B 1A60 1A76 1A45
ccc 0 0 9 230 0

Finally, the user might also type in the sequence with the tone after the lower component.

name kha o sakot wa t2
code 1A21 1A6B 1A60 1A45 1A76
ccc 0 0 9 0 230

(That sequence is already in NFC format.)

We want all of these sequences to end up ordered as the first. To do this, we use the following rules:

<reorder from="\u{1A60}" order="127" />      <!-- max possible order -->
<reorder from="\u{1A6B}" order="42" />
<reorder from="[\u{1A75}-\u{1A79}]" order="55" />
<reorder before="\u{1A6B}" from="\u{1A60}\u{1A45}" order="10" />
<reorder before="\u{1A6B}[\u{1A75}-\u{1A79}]" from="\u{1A60}\u{1A45}" order="10" />
<reorder before="\u{1A6B}" from="\u{1A60}[\u{1A75}-\u{1A79}]\u{1A45}" order="10 55 10" />

The first reorder is the default ordering for the sakot which allows for it to be placed anywhere in a sequence, but moves any non-consonants that may immediately follow it, back before it in the sequence. The next two rules give the orders for the top vowel component and tone marks respectively. The next three rules give the sakot and wa characters a primary order that places them before the o. Notice particularly the final reorder rule where the sakot+wa is split by the tone mark. This rule is necessary in case someone types into the middle of previously normalized text.

<reorder> elements are priority ordered based first on the length of string their @from attribute value matches and then the sum of the lengths of the strings their @before attribute value matches.

Using <import> with <reorder> elements

This section describes the impact of using import elements with <reorder> elements.

The @from string in a <reorder> element describes a set of strings that it matches. This also holds for the @before attribute. The intersection of any two <reorder> elements consists of the intersections of their @from and @before string sets. Tooling should warn users if the intersection between any two <reorder> elements in the same <transformGroup> element to be non empty prior to processing imports.

If two <reorder> elements have a non empty intersection, then they are split and merged. They are split such that where there were two <reorder> elements, there are, in effect (but not actuality), three elements consisting of:

When merging the other attributes, the second rule is taken to have priority (being an override of the earlier element). Where the second rule does not define the value for a character but the first does, the value is taken from the first rule, otherwise it is taken from the second rule.

Notice that it is possible for two rules to match the same string, but for them not to merge because the distribution of the string across @before and @from is different. For example, the following would not merge:

<reorder before="ab" from="cd" />
<reorder before="a" from="bcd" />

After <reorder> elements merge, the resulting reorder elements are sorted into priority order for matching.

Consider this fragment from a shared reordering for the Myanmar script:

<!-- File: "myanmar-reordering.xml" -->
<transformGroup>
    <!-- medial-r -->
    <reorder from="\u{103C}" order="20" />

    <!-- [medial-wa or shan-medial-wa] -->
    <reorder from="[\u{103D}\u{1082}]" order="25" />

    <!-- [medial-ha or shan-medial-wa]+asat = Mon asat -->
    <reorder from="[\u{103E}\u{1082}]\u{103A}" order="27" />

    <!-- [medial-ha or mon-medial-wa] -->
    <reorder from="[\u{103E}\u{1060}]" order="27" />

    <!-- [e-vowel (U+1031) or shan-e-vowel (U+1084)] -->
    <reorder from="[\u{1031}\u{1084}]" order="30" />

    <reorder from="[\u{102D}\u{102E}\u{1033}-\u{1035}\u{1071}-\u{1074}\u{1085}\u{109D}\u{A9E5}]" order="35" />
</transformGroup>

A particular Myanmar keyboard layout can have these reorder elements:

<transformGroup>
    <import path="myanmar-reordering.xml"/> <!-- import the above transformGroup -->
    <!-- Kinzi -->
    <reorder from="\u{1004}\u{103A}\u{1039}" order="-1" />

    <!-- e-vowel -->
    <reorder from="\u{1031}" preBase="1" />

    <!-- medial-r -->
    <reorder from="\u{103C}" preBase="1" />
</transformGroup>

The effect of this is that the e-vowel will be identified as a prebase and will have an order of 30. Likewise a medial-r will be identified as a prebase and will have an order of 20. Notice that a shan-e-vowel (\u{1084}) will not be identified as a prebase (even if it should be!). The kinzi is described in the layout since it moves something across a run boundary. By separating such movements (prebase or moving to in front of a base) from the shared ordering rules, the shared ordering rules become a self-contained combining order description that can be used in other keyboards or even in other contexts than keyboarding.

Example Post-reorder transforms

It may be desired to perform additional processing following reorder operations. This may be aaccomplished by adding an additional <transformGroup> element after the group containing <reorder> elements.

First, a partial example from Khmer where split vowels are combined after reordering.

…
<transformGroup>
    <reorder … />
    <reorder … />
    <reorder … />
    …
</transformGroup>
<transformGroup>
    <transform from="\u{17C1}\u{17B8}" to="\u{17BE}" />
    <transform from="\u{17C1}\u{17B6}" to="\u{17C4}" />
</transformGroup>

Another partial example allows a keyboard implementation to prevent people typing two lower vowels in a Burmese cluster:

…
<transformGroup>
    <reorder … />
    <reorder … />
    <reorder … />
    …
</transformGroup>
<transformGroup>
    <transform from="[\u{102F}\u{1030}\u{1048}\u{1059}][\u{102F}\u{1030}\u{1048}\u{1059}]"  />
</transformGroup>

Reorder and Markers

Markers are not matched by reorder elements. However, if a character preceded by one or more markers is reordered due to a reorder element, those markers will be reordered with the characters, maintaining the same relative order. This is a similar process to the algorithm used to normalize strings processed by transform elements.

Keyboard implementations must process reorder elements using the following algorithm.

Note that steps 1 and 3 are identical to the steps used for normalization using markers in the Marker Algorithm Overview.

Given an input string from context or from a previous transformGroup:

  1. Parsing/Removing Markers

  2. Perform reordering (as in this section)

  3. Re-Adding Markers

Backspace Transforms

The <transforms type="backspace"> describe an optional transform that is not applied on input of normal characters, but is only used to perform extra backspace modifications to previously committed text.

When the backspace key is pressed, the <transforms type="backspace"> element (if present) is processed, and then the <transforms type="simple"> element (if processed) as with any other key.

Keyboarding applications typically work, but are not required to, in one of two modes:

text entry

text entry happens while a user is typing new text. A user typically wants the backspace key to undo whatever they last typed, whether or not they typed things in the 'right' order.

text editing

text editing happens when a user moves the cursor into some previously entered text which may have been entered by someone else. As such, there is no way to know in which order things were typed, but a user will still want appropriate behaviour when they press backspace. This may involve deleting more than one character or replacing a sequence of characters with a different sequence.

In text editing mode, different keyboard layouts may behave differently in the same textual context. The backspace transform allows the keyboard layout to specify the effect of pressing backspace in a particular textual context. This is done by specifying a set of backspace rules that match a string before the cursor and replace it with another string. The rules are expressed within a transforms type="backspace" element.

<transforms type="backspace">
    <transformGroup>
        <transform from="…match pattern" to="…output pattern" />
    </transformGroup>
</transforms>

Example

For example, consider deleting a Devanagari ksha क्श:

While this character is made up of three codepoints, the following rule causes all three to be deleted by a single press of the backspace.

<transforms type="backspace">
    <transformGroup>
        <transform from="\u{0915}\u{094D}\u{0936}"/>
    </transformGroup>
</transforms>

Note that the optional attribute @to is omitted, since the whole string is being deleted. This is not uncommon in backspace transforms.

A more complex example comes from a Burmese visually ordered keyboard:

<transforms type="backspace">
    <transformGroup>
        <!-- Kinzi -->
        <transform from="[\u{1004}\u{101B}\u{105A}]\u{103A}\u{1039}" />

        <!-- subjoined consonant -->
        <transform from="\u{1039}[\u{1000}-\u{101C}\u{101E}\u{1020}\u{1021}\u{1050}\u{1051}\u{105A}-\u{105D}]" />

        <!-- tone mark -->
        <transform from="\u{102B}\u{103A}" />

        <!-- Handle prebases -->
        <!-- diacritics stored before e-vowel -->
        <transform from="[\u{103A}-\u{103F}\u{105E}-\u{1060}\u{1082}]\u{1031}" to="\u{1031}" />

        <!-- diacritics stored before medial r -->
        <transform from="[\u{103A}-\u{103B}\u{105E}-\u{105F}]\u{103C}" to="\u{103C}" />

        <!-- subjoined consonant before e-vowel -->
        <transform from="\u{1039}[\u{1000}-\u{101C}\u{101E}\u{1020}\u{1021}]\u{1031}" to="\u{1031}" />

        <!-- base consonant before e-vowel -->
        <transform from="[\u{1000}-\u{102A}\u{103F}-\u{1049}\u{104E}]\u{1031}" to="\m{prebase}\u{1031}" />

        <!-- subjoined consonant before medial r -->
        <transform from="\u{1039}[\u{1000}-\u{101C}\u{101E}\u{1020}\u{1021}]\u{103C}" to="\u{103C}" />

        <!-- base consonant before medial r -->
        <transform from="[\u{1000}-\u{102A}\u{103F}-\u{1049}\u{104E}]\u{103C}" to="\m{prebase}\u{103C}" />

        <!-- delete lone medial r or e-vowel -->
        <transform from="\m{prebase}[\u{1031}\u{103C}]" />
    </transformGroup>
</transforms>

The above example is simplified, and doesn't fully handle the interaction between medial-r and e-vowel.

The character \m{prebase} does not represent a literal character, but is instead a special marker, used as a "filler string". When a keyboard implementation handles a user pressing a key that inserts a prebase character, it also has to insert a special filler string before the prebase to ensure that the prebase character does not combine with the previous cluster. See the reorder transform for details. See markers for the \m syntax.

The first three transforms above delete various ligatures with a single keypress. The other transforms handle prebase characters. There are two in this Burmese keyboard. The transforms delete the characters preceding the prebase character up to base which gets replaced with the prebase filler string, which represents a null base. Finally the prebase filler string + prebase is deleted as a unit.

If no specified transform among all transformGroups under the <transforms type="backspace"> element matches, a default will be used instead — an implied final transform that simply deletes the codepoint at the end of the input context. This implied transform is effectively similar to the following code sample, even though the * operator is not actually allowed in from=. See the documentation for Match a single Unicode codepoint under transform syntax and markers, above.

It is important that implementations do not by default delete more than one non-marker codepoint at a time, except in the case of emoji clusters. Note that implementations will vary in the emoji handling due to the iterative nature of successive Unicode releases. See UTS#51 §2.4.2: Emoji Modifiers in Text

<transforms type="backspace">
    <!-- Other explicit transforms -->

    <!-- Final implicit backspace transform: Delete the final codepoint. -->
    <transformGroup>
        <!-- (:?\m{.})*  - matches any number of contiguous markers -->
        <transform from="(:?\m{.})*.(:?\m{.})*" /> <!-- deletes any number of markers directly on either side of the final pre-caret codepoint -->
    </transformGroup>
</transforms>

Invariants

Beyond what the DTD imposes, certain other restrictions on the data are imposed on the data. Please note the constraints given under each element section above. DTD validation alone is not sufficient to verify a keyboard file.

Keyboard IDs

There is a set of subtags that help identify the keyboards. Each of these are used after the "t-k0" subtags to help identify the keyboards. The first tag appended is a mandatory platform tag followed by zero or more tags that help differentiate the keyboard from others with the same locale code.

Principles for Keyboard IDs

The following are the design principles for the IDs.

  1. BCP47 compliant.
    1. Eg, en, sr-Cyrl, or en-t-k0-extended.
  2. Use the minimal language id based on likelySubtags (see Part 1: Likely Subtags)
    1. Eg, instead of fa-Arab, use fa.
    2. The data is in https://github.com/unicode-org/cldr/blob/main/common/supplemental/likelySubtags.xml
  3. Keyboard files should be platform-independent, however, if included, a platform id is the first subtag after -t-k0-. If a keyboard on the platform changes over time, both are dated, eg bg-t-k0-chromeos-2011. When selecting, if there is no date, it means the latest one.
  4. Keyboards are only tagged that differ from the "standard for each language". That is, for each language on a platform, there will be a keyboard with no subtags. Subtags with common semantics across languages and platforms are used, such as -extended, -phonetic, -qwerty, -qwertz, -azerty, …
  5. In order to get to 8 letters, abbreviations are reused that are already in bcp47 -u/-t extensions and in language-subtag-registry variants, eg for Traditional use -trad or -traditio (both exist in bcp47).
  6. Multiple languages cannot be indicated in the locale id, so the predominant target is used.
    1. For Finnish + Sami, use fi-t-k0-smi or extended-smi
    2. The <locales> element may be used to identify additional languages.
  7. In some cases, there are multiple subtags, like en-US-t-k0-chromeos-intl-altgr.xml
  8. Otherwise, platform names are used as a guide.

Examples

<!-- Serbian Latin -->
<keyboard3 locale="sr-Latn"/>

<!-- Serbian Cyrillic -->
<keyboard3 locale="sr-Cyrl"/>

<!-- Pan Nigerian Keyboard-->
<keyboard3 locale="mul-Latn-NG-t-k0-panng">
    <locales>
    <locale id="ha"/>
    <locale id="ig"/>
    <!-- others … -->
    </locales>
</keyboard3>

<!-- Finnish Keyboard including Skolt Sami -->
<keyboard3 locale="fi-t-k0-smi">
    <locales>
    <locale id="sms"/>
    </locales>
</keyboard3>

Platform Behaviors in Edge Cases

Platform No modifier combination match is available No map match is available for key position Transform fails (i.e. if ^d is pressed when that transform does not exist)
Chrome OS Fall back to base Fall back to character in a keyMap with same "level" of modifier combination. If this character does not exist, fall back to (n-1) level. (This is handled data-generation-side.)
In the specification: No output		 No output at all
Mac OS X Fall back to base (unless combination is some sort of keyboard shortcut, e.g. cmd-c) No output Both keys are output separately
Windows No output No output Both keys are output separately

Copyright © 2001–2024 Unicode, Inc. All Rights Reserved. The Unicode Consortium makes no expressed or implied warranty of any kind, and assumes no liability for errors or omissions. No liability is assumed for incidental and consequential damages in connection with or arising out of the use of the information or programs contained or accompanying this technical report. The Unicode Terms of Use apply.

Unicode and the Unicode logo are trademarks of Unicode, Inc., and are registered in some jurisdictions.