[Unicode]  Technical Reports
 

Unicode Technical Standard #35

Unicode Locale Data Markup Language (LDML)
Part 9: Message Format

Version 46 (draft)
Editors Addison Phillips and other CLDR committee members

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

Summary

This specification defines the data model, syntax, processing, and conformance requirements for the next generation of dynamic messages.

This is a partial document, describing only those parts of the LDML that are relevant for message format. For the other parts of the LDML see the main LDML document and the links above.

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This is a draft document which may be updated, replaced, or superseded by other documents at any time.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].

Parts

The LDML specification is divided into the following parts:

Contents of Part 9, Message Format

Introduction

One of the challenges in adapting software to work for users with different languages and cultures is the need for dynamic messages. Whenever a user interface needs to present data as part of a larger string, that data needs to be formatted (and the message may need to be altered) to make it culturally accepted and grammatically correct.

For example, if your US English (en-US) interface has a message like:

Your item had 1,023 views on April 3, 2023

You want the translated message to be appropriately formatted into French:

Votre article a eu 1 023 vues le 3 avril 2023

Or Japanese:

あなたのアイテムは 2023 年 4 月 3 日に 1,023 回閲覧されました。

This specification defines the data model, syntax, processing, and conformance requirements for the next generation of dynamic messages. It is intended for adoption by programming languages and APIs. This will enable the integration of existing internationalization APIs (such as the date and number formats shown above), grammatical matching (such as plurals or genders), as well as user-defined formats and message selectors.

The document is the successor to ICU MessageFormat, henceforth called ICU MessageFormat 1.0.

Conformance

Everything in this specification is normative except for: sections marked as non-normative, all authoring guidelines, diagrams, examples, and notes.

The key words "MUST", "MUST NOT", "REQUIRED", "SHALL", "SHALL NOT", "SHOULD", "SHOULD NOT", "RECOMMENDED", "NOT RECOMMENDED", "MAY", and "OPTIONAL" in this document are to be interpreted as described in BCP 14 [RFC2119] [RFC8174] when, and only when, they appear in all capitals, as shown here.

Terminology and Conventions

A term looks like this when it is defined in this specification.

A reference to a term looks like this.

Examples are non-normative and styled like this.

Stability Policy

Important

The provisions of the stability policy are not in effect until the conclusion of the technical preview and adoption of this specification.

Updates to this specification will not make any valid message invalid.

Updates to this specification will not remove any syntax provided in this version.

Updates to this specification MUST NOT specify an error for any message that previously did not specify an error.

Updates to this specification MUST NOT specify the use of a fallback value for any message that previously did not specify a fallback value.

Updates to this specification will not change the syntactical meaning of any syntax defined in this specification.

Updates to this specification will not remove any functions defined in the default registry.

Updates to this specification will not remove any options or option values defined in the default registry.

Note

The foregoing policies are not a guarantee that the results of formatting will never change. Even when this specification or its implementation do not change, the functions for date formatting, number formatting and so on can change their results over time or behave differently due to local runtime differences in implementation or changes to locale data (such as due to the release of new CLDR versions).

Updates to this specification will only reserve, define, or require function names or function option names consisting of characters in the ranges a-z, A-Z, and 0-9. All other names in these categories are reserved for the use of implementations or users.

Note

Users defining custom names SHOULD include at least one character outside these ranges to ensure that they will be compatible with future versions of this specification. They SHOULD also use the namespace feature to avoid collisions with other implementations.

Future versions of this specification will not introduce changes to the data model that would result in a data model representation based on this version being invalid.

For example, existing interfaces or fields will not be removed.

Important

This stability policy allows any of the following, non-exhaustive list, of changes in future versions of this specification:

Syntax

This section defines the formal grammar describing the syntax of a single message.

Design Goals

This section is non-normative.

The design goals of the syntax specification are as follows:

  1. The syntax should leverage the familiarity with ICU MessageFormat 1.0 in order to lower the barrier to entry and increase the chance of adoption. At the same time, the syntax should fix the pain points of ICU MessageFormat 1.0.

    • Non-Goal: Be backwards-compatible with the ICU MessageFormat 1.0 syntax.
  2. The syntax inside translatable content should be easy to understand for humans. This includes making it clear which parts of the message body are translatable content, which parts inside it are placeholders for expressions, as well as making the selection logic predictable and easy to reason about.

    • Non-Goal: Make the syntax intuitive enough for non-technical translators to hand-edit. Instead, we assume that most translators will work with MessageFormat 2 by means of GUI tooling, CAT workbenches etc.
  3. The syntax surrounding translatable content should be easy to write and edit for developers, localization engineers, and easy to parse by machines.

  4. The syntax should make a single message easily embeddable inside many container formats: .properties, YAML, XML, inlined as string literals in programming languages, etc. This includes a future MessageResource specification.

    • Non-Goal: Support unnecessary escape sequences, which would theirselves require additional escaping when embedded. Instead, we tolerate direct use of nearly all characters (including line breaks, control characters, etc.) and rely upon escaping in those outer formats to aid human comprehension (e.g., depending upon container format, a U+000A LINE FEED might be represented as \n, \012, \x0A, \u000A, \U0000000A, &#xA;, &NewLine;, %0A, <LF>, or something else entirely).

Design Restrictions

This section is non-normative.

The syntax specification takes into account the following design restrictions:

  1. Whitespace outside the translatable content should be insignificant. It should be possible to define a message entirely on a single line with no ambiguity, as well as to format it over multiple lines for clarity.

  2. The syntax should define as few special characters and sigils as possible. Note that this necessitates extra care when presenting messages for human consumption, because they may contain invisible characters such as U+200B ZERO WIDTH SPACE, control characters such as U+0000 NULL and U+0009 TAB, permanently reserved noncharacters (U+FDD0 through U+FDEF and U+nFFFE and U+nFFFF where n is 0x0 through 0x10), private-use code points (U+E000 through U+F8FF, U+F0000 through U+FFFFD, and U+100000 through U+10FFFD), unassigned code points, and other potentially confusing content.

Messages and their Syntax

The purpose of MessageFormat is to allow content to vary at runtime. This variation might be due to placing a value into the content or it might be due to selecting a different bit of content based on some data value or it might be due to a combination of the two.

MessageFormat calls the template for a given formatting operation a message.

The values passed in at runtime (which are to be placed into the content or used to select between different content items) are called external variables. The author of a message can also assign local variables, including variables that modify external variables.

This part of the MessageFormat specification defines the syntax for a message, along with the concepts and terminology needed when processing a message during the formatting of a message at runtime.

The complete formal syntax of a message is described by the ABNF.

Well-formed vs. Valid Messages

A message is well-formed if it satisfies all the rules of the grammar. Attempting to parse a message that is not well-formed will result in a Syntax Error.

A message is valid if it is well-formed and also meets the additional content restrictions and semantic requirements about its structure defined below for declarations, matcher, and options. Attempting to parse a message that is not valid will result in a Data Model Error.

The Message

A message is the complete template for a specific message formatting request.

A variable is a name associated to a resolved value.

An external variable is a variable whose name and initial value are supplied by the caller to MessageFormat or available in the formatting context. Only an external variable can appear as an operand in an input declaration.

A local variable is a variable created as the result of a local declaration.

Note

This syntax is designed to be embeddable into many different programming languages and formats. As such, it avoids constructs, such as character escapes, that are specific to any given file format or processor. In particular, it avoids using quote characters common to many file formats and formal languages so that these do not need to be escaped in the body of a message.

Note

In general (and except where required by the syntax), whitespace carries no meaning in the structure of a message. While many of the examples in this spec are written on multiple lines, the formatting shown is primarily for readability.

Example This message:

.local $foo   =   { |horse| }
{{You have a {$foo}!}}

Can also be written as:

.local $foo={|horse|}{{You have a {$foo}!}}

An exception to this is: whitespace inside a pattern is always significant.

Note

The MessageFormat 2 syntax assumes that each message will be displayed with a left-to-right display order and be processed in the logical character order. The syntax permits the use of right-to-left characters in identifiers, literals, and other values. This can result in confusion when viewing the message or users might incorrectly insert bidi controls or marks that negatively affect the output of the message.

To assist with this, the syntax permits the use of various controls and strongly-directional markers in both optional and required whitespace in a message, as well was encouraging the use of isolating controls with expressions and quoted patterns. See: whitespace (below) for more information.

Additional restrictions or requirements might be added during the Tech Preview to better manage bidirectional text.

A message can be a simple message or it can be a complex message.

message = simple-message / complex-message

A simple message contains a single pattern, with restrictions on its first non-whitespace character. An empty string is a valid simple message.

Whitespace at the start or end of a simple message is significant, and a part of the text of the message.

simple-message = o [simple-start pattern]
simple-start   = simple-start-char / escaped-char / placeholder

A complex message is any message that contains declarations, a matcher, or both. A complex message always begins with either a keyword that has a . prefix or a quoted pattern and consists of:

  1. an optional list of declarations, followed by
  2. a complex body

Whitespace at the start or end of a complex message is not significant, and does not affect the processing of the message.

complex-message = o *(declaration o) complex-body o

Declarations

A declaration binds a variable identifier to a value within the scope of a message. This variable can then be used in other expressions within the same message. Declarations are optional: many messages will not contain any declarations.

An input-declaration binds a variable to an external input value. The variable-expression of an input-declaration MAY include a function that is applied to the external value.

A local-declaration binds a variable to the resolved value of an expression.

declaration       = input-declaration / local-declaration
input-declaration = input o variable-expression
local-declaration = local s variable o "=" o expression

Variables, once declared, MUST NOT be redeclared. A message that does any of the following is not valid and will produce a Duplicate Declaration error during processing:

A local-declaration MAY overwrite an external input value as long as the external input value does not appear in a previous declaration.

Note

These restrictions only apply to declarations. A placeholder can apply a different function to a variable than one applied to the same variable named in a declaration. For example, this message is valid:

.input {$var :number maximumFractionDigits=0}
.local $var2 = {$var :number maximumFractionDigits=2}
.match $var2
0 {{The selector can apply a different function to {$var} for the purposes of selection}}
* {{A placeholder in a pattern can apply a different function to {$var :number maximumFractionDigits=3}}}

(See the Errors section for examples of invalid messages)

Complex Body

The complex body of a complex message is the part that will be formatted. The complex body consists of either a quoted pattern or a matcher.

complex-body = quoted-pattern / matcher

Pattern

A pattern contains a sequence of text and placeholders to be formatted as a unit. Unless there is an error, resolving a message always results in the formatting of a single pattern.

pattern = *(text-char / escaped-char / placeholder)

A pattern MAY be empty.

A pattern MAY contain an arbitrary number of placeholders to be evaluated during the formatting process.

Quoted Pattern

A quoted pattern is a pattern that is "quoted" to prevent interference with other parts of the message. A quoted pattern starts with a sequence of two U+007B LEFT CURLY BRACKET {{ and ends with a sequence of two U+007D RIGHT CURLY BRACKET }}.

quoted-pattern = o "{{" pattern "}}"

A quoted pattern MAY be empty.

An empty quoted pattern:

{{}}

Text

text is the translateable content of a pattern. Any Unicode code point is allowed, except for U+0000 NULL and the surrogate code points U+D800 through U+DFFF inclusive. The characters U+005C REVERSE SOLIDUS \, U+007B LEFT CURLY BRACKET {, and U+007D RIGHT CURLY BRACKET } MUST be escaped as \\, \{, and \} respectively.

In the ABNF, text is represented by non-empty sequences of simple-start-char, text-char, escaped-char, and s. The production simple-start-char represents the first non-whitespace in a simple message and matches text-char except for not allowing U+002E FULL STOP .. The ABNF uses content-char as a shared base for text and quoted literal characters.

Whitespace in text, including tabs, spaces, and newlines is significant and MUST be preserved during formatting.

simple-start-char = content-char / "@" / "|"
text-char         = content-char / ws / "." / "@" / "|"
quoted-char       = content-char / ws / "." / "@" / "{" / "}"
content-char      = %x01-08        ; omit NULL (%x00), HTAB (%x09) and LF (%x0A)
                  / %x0B-0C        ; omit CR (%x0D)
                  / %x0E-1F        ; omit SP (%x20)
                  / %x21-2D        ; omit . (%x2E)
                  / %x2F-3F        ; omit @ (%x40)
                  / %x41-5B        ; omit \ (%x5C)
                  / %x5D-7A        ; omit { | } (%x7B-7D)
                  / %x7E-2FFF      ; omit IDEOGRAPHIC SPACE (%x3000)
                  / %x3001-D7FF    ; omit surrogates
                  / %xE000-10FFFF

When a pattern is quoted by embedding the pattern in curly brackets, the resulting message can be embedded into various formats regardless of the container's whitespace trimming rules. Otherwise, care must be taken to ensure that pattern-significant whitespace is preserved.

Example In a Java .properties file, the values hello and hello2 both contain an identical message which consists of a single pattern. This pattern consists of text with exactly three spaces before and after the word "Hello":

hello = {{   Hello   }}
hello2=\   Hello  \ 

Placeholder

A placeholder is an expression or markup that appears inside of a pattern and which will be replaced during the formatting of a message.

placeholder = expression / markup

Matcher

A matcher is the complex body of a message that allows runtime selection of the pattern to use for formatting. This allows the form or content of a message to vary based on values determined at runtime.

A matcher consists of the keyword .match followed by at least one selector and at least one variant.

When the matcher is processed, the result will be a single pattern that serves as the template for the formatting process.

A message can only be considered valid if the following requirements are satisfied; otherwise, a corresponding Data Model Error will be produced during processing:

matcher         = match-statement s variant *(o variant)
match-statement = match 1*(s selector)

A message with a matcher:

.input {$count :number}
.match $count
one {{You have {$count} notification.}}
*   {{You have {$count} notifications.}}

A message containing a matcher formatted on a single line:

.local $os = {:platform} .match $os windows {{Settings}} * {{Preferences}}

Selector

A selector is a variable whose resolved value ranks or excludes the variants based on the value of the corresponding key in each variant. The combination of selectors in a matcher thus determines which pattern will be used during formatting.

selector = variable

There MUST be at least one selector in a matcher. There MAY be any number of additional selectors.

A message with a single selector that uses a custom function :hasCase which is a selector that allows the message to choose a pattern based on grammatical case:

.local $hasCase = {$userName :hasCase}
.match $hasCase
vocative {{Hello, {$userName :person case=vocative}!}}
accusative {{Please welcome {$userName :person case=accusative}!}}
* {{Hello!}}

A message with two selectors:

.input {$numLikes :integer}
.input {$numShares :integer}
.match $numLikes $numShares
0   0   {{Your item has no likes and has not been shared.}}
0   one {{Your item has no likes and has been shared {$numShares} time.}}
0   *   {{Your item has no likes and has been shared {$numShares} times.}}
one 0   {{Your item has {$numLikes} like and has not been shared.}}
one one {{Your item has {$numLikes} like and has been shared {$numShares} time.}}
one *   {{Your item has {$numLikes} like and has been shared {$numShares} times.}}
*   0   {{Your item has {$numLikes} likes and has not been shared.}}
*   one {{Your item has {$numLikes} likes and has been shared {$numShares} time.}}
*   *   {{Your item has {$numLikes} likes and has been shared {$numShares} times.}}

Variant

A variant is a quoted pattern associated with a list of keys in a matcher. Each variant MUST begin with a sequence of keys, and terminate with a valid quoted pattern. The number of keys in each variant MUST match the number of selectors in the matcher.

Each key is separated from each other by whitespace. Whitespace is permitted but not required between the last key and the quoted pattern.

variant = key *(s key) quoted-pattern
key     = literal / "*"

Key

A key is a value in a variant for use by a selector when ranking or excluding variants during the matcher process. A key can be either a literal value or the "catch-all" key *.

The catch-all key is a special key, represented by *, that matches all values for a given selector.

The value of each key MUST be treated as if it were in Unicode Normalization Form C ("NFC"). Two keys are considered equal if they are canonically equivalent strings, that is, if they consist of the same sequence of Unicode code points after Unicode Normalization Form C has been applied to both.

Expressions

An expression is a part of a message that will be determined during the message's formatting.

An expression MUST begin with U+007B LEFT CURLY BRACKET { and end with U+007D RIGHT CURLY BRACKET }. An expression MUST NOT be empty. An expression cannot contain another expression. An expression MAY contain one more attributes.

A literal-expression contains a literal, optionally followed by a function.

A variable-expression contains a variable, optionally followed by a function.

A function-expression contains a function without an operand.

expression          = literal-expression
                    / variable-expression
                    / function-expression
literal-expression  = "{" o literal [s function] *(s attribute) o "}"
variable-expression = "{" o variable [s function] *(s attribute) o "}"
function-expression = "{" o function *(s attribute) o "}"

There are several types of expression that can appear in a message. All expressions share a common syntax. The types of expression are:

  1. The value of a local-declaration
  2. A kind of placeholder in a pattern

Additionally, an input-declaration can contain a variable-expression.

Examples of different types of expression

Declarations:

.input {$x :function option=value}
.local $y = {|This is an expression|}

Placeholders:

This placeholder contains a literal expression: {|literal|}
This placeholder contains a variable expression: {$variable}
This placeholder references a function on a variable: {$variable :function with=options}
This placeholder contains a function expression with a variable-valued option: {:function option=$variable}

Operand

An operand is the literal of a literal-expression or the variable of a variable-expression.

Function

A function is named functionality in an expression. Functions are used to evaluate, format, select, or otherwise process data values during formatting.

A function can appear in an expression by itself or following a single operand. When following an operand, the operand serves as input to the function.

Each function is defined by the runtime's function registry. A function's entry in the function registry will define whether the function is a selector or formatter (or both), whether an operand is required, what form the values of an operand can take, what options and option values are acceptable, and what outputs might result. See function registry for more information.

A function starts with a prefix sigil : followed by an identifier. The identifier MAY be followed by one or more options. Options are not required.

function = ":" identifier *(s option)

A message with a function operating on the variable $now:

It is now {$now :datetime}.
Options

An option is a key-value pair containing a named argument that is passed to a function.

An option has an identifier and a value. The identifier is separated from the value by an U+003D EQUALS SIGN = along with optional whitespace. The value of an option can be either a literal or a variable.

Multiple options are permitted in a function. Options are separated from the preceding function identifier and from each other by whitespace. Each option's identifier MUST be unique within the function: a function with duplicate option identifiers is not valid and will produce a Duplicate Option Name error during processing.

The order of options is not significant.

option = identifier o "=" o (literal / variable)

Examples of functions with options

A message using the :datetime function. The option weekday has the literal long as its value:

Today is {$date :datetime weekday=long}!

A message using the :datetime function. The option weekday has a variable $dateStyle as its value:

Today is {$date :datetime weekday=$dateStyle}!

Markup

Markup placeholders are pattern parts that can be used to represent non-language parts of a message, such as inline elements or styling that should apply to a span of parts.

Markup MUST begin with U+007B LEFT CURLY BRACKET { and end with U+007D RIGHT CURLY BRACKET }. Markup MAY contain one more attributes.

Markup comes in three forms:

Markup-open starts with U+0023 NUMBER SIGN # and represents an opening element within the message, such as markup used to start a span. It MAY include options.

Markup-standalone starts with U+0023 NUMBER SIGN # and has a U+002F SOLIDUS / immediately before its closing } representing a self-closing or standalone element within the message. It MAY include options.

Markup-close starts with U+002F SOLIDUS / and is a pattern part ending a span.

markup = "{" o "#" identifier *(s option) *(s attribute) o ["/"] "}"  ; open and standalone
       / "{" o "/" identifier *(s option) *(s attribute) o "}"  ; close

A message with one button markup span and a standalone img markup element:

{#button}Submit{/button} or {#img alt=|Cancel| /}.

A message containing markup that uses options to pair two closing markup placeholders to the one open markup placeholder:

{#ansi attr=|bold,italic|}Bold and italic{/ansi attr=|bold|} italic only {/ansi attr=|italic|} no formatting.}

A markup-open can appear without a corresponding markup-close. A markup-close can appear without a corresponding markup-open. Markup placeholders can appear in any order without making the message invalid. However, specifications or implementations defining markup might impose requirements on the pairing, ordering, or contents of markup during formatting.

Attributes

An attribute is an identifier with an optional value that appears in an expression or in markup. During formatting, attributes have no effect, and they can be treated as code comments.

Attributes are prefixed by a U+0040 COMMERCIAL AT @ sign, followed by an identifier. An attribute MAY have a literal value which is separated from the identifier by an U+003D EQUALS SIGN = along with optional whitespace.

Multiple attributes are permitted in an expression or markup. Each attribute is separated by whitespace.

Each attribute's identifier SHOULD be unique within the expression or markup: all but the last attribute with the same identifier are ignored. The order of attributes is not otherwise significant.

attribute = "@" identifier [o "=" o literal]

Examples of expressions and markup with attributes:

A message including a literal that should not be translated:

In French, "{|bonjour| @translate=no}" is a greeting

A message with markup that should not be copied:

Have a {#span @can-copy}great and wonderful{/span @can-copy} birthday!

Other Syntax Elements

This section defines common elements used to construct messages.

Keywords

A keyword is a reserved token that has a unique meaning in the message syntax.

The following three keywords are defined: .input, .local, and .match. Keywords are always lowercase and start with U+002E FULL STOP ..

input = %s".input"
local = %s".local"
match = %s".match"

Literals

A literal is a character sequence that appears outside of text in various parts of a message. A literal can appear as a key value, as the operand of a literal-expression, or in the value of an option. A literal MAY include any Unicode code point except for U+0000 NULL or the surrogate code points U+D800 through U+DFFF.

All code points are preserved.

Important

Most text, including that produced by common keyboards and input methods, is already encoded in the canonical form known as Unicode Normalization Form C ("NFC"). A few languages, legacy character encoding conversions, or operating environments can result in literal values that are not in this form. Some uses of literals in MessageFormat, notably as the value of keys, apply NFC to the literal value during processing or comparison. While there is no requirement that the literal value actually be entered in a normalized form, users are cautioned to employ the same character sequences for equivalent values and, whenever possible, ensure literals are in NFC.

A quoted literal begins and ends with U+005E VERTICAL BAR |. The characters \ and | within a quoted literal MUST be escaped as \\ and \|.

An unquoted literal is a literal that does not require the | quotes around it to be distinct from the rest of the message syntax. An unquoted literal MAY be used when the content of the literal contains no whitespace and otherwise matches the unquoted production. Implementations MUST NOT distinguish between quoted literals and unquoted literals that have the same sequence of code points.

Unquoted literals can contain a name or consist of a number-literal. A number-literal uses the same syntax as JSON and is intended for the encoding of number values in operands or options, or as keys for variants.

literal          = quoted-literal / unquoted-literal
quoted-literal   = "|" *(quoted-char / escaped-char) "|"
unquoted-literal = name / number-literal
number-literal   = ["-"] (%x30 / (%x31-39 *DIGIT)) ["." 1*DIGIT] [%i"e" ["-" / "+"] 1*DIGIT]

Names and Identifiers

A name is a character sequence used in an identifier or as the name for a variable or the value of an unquoted literal.

A name can be preceded or followed by bidirectional marks or isolating controls to aid in presenting names that contain right-to-left or neutral characters. These characters are not part of the value of the name and MUST be treated as if they were not present when matching name or identifier strings or unquoted literal values.

Variable names are prefixed with $.

Two names are considered equal if they are canonically equivalent strings, that is, if they consist of the same sequence of Unicode code points after Unicode Normalization Form C ("NFC") has been applied to both.

Note

Implementations are not required to normalize all names. Comparisons of name values only need be done "as-if" normalization has occured. Since most text in the wild is already in NFC and since checking for NFC is fast and efficient, implementations can often substitute checking for actually applying normalization to name values.

Valid content for names is based on Namespaces in XML 1.0's NCName. This is different from XML's Name in that it MUST NOT contain a U+003A COLON :. Otherwise, the set of characters allowed in a name is large.

Note

External variables can be passed in that are not valid names. Such variables cannot be referenced in a message, but are not otherwise errors.

An identifier is a character sequence that identifies a function, markup, or option. Each identifier consists of a name optionally preceeded by a namespace. When present, the namespace is separated from the name by a U+003A COLON :. Built-in functions and their options do not have a namespace identifier.

The namespace u (U+0075 LATIN SMALL LETTER U) is reserved for future standardization.

Function identifiers are prefixed with :. Markup identifiers are prefixed with # or /. Option identifiers have no prefix.

Examples:

A variable:

This has a {$variable}

A function:

This has a {:function}

An add-on function from the icu namespace:

This has a {:icu:function}

An option and an add-on option:

This has {:options option=value icu:option=add_on}

Support for namespaces and their interpretation is implementation-defined in this release.

variable   = "$" name
option     = identifier o "=" o (literal / variable)

identifier = [namespace ":"] name
namespace  = name
name       = [bidi] name-start *name-char [bidi]
name-start = ALPHA / "_"
           / %xC0-D6 / %xD8-F6 / %xF8-2FF
           / %x370-37D / %x37F-61B / %x61D-1FFF / %x200C-200D
           / %x2070-218F / %x2C00-2FEF / %x3001-D7FF
           / %xF900-FDCF / %xFDF0-FFFC / %x10000-EFFFF
name-char  = name-start / DIGIT / "-" / "."
           / %xB7 / %x300-36F / %x203F-2040

Escape Sequences

An escape sequence is a two-character sequence starting with U+005C REVERSE SOLIDUS \.

An escape sequence allows the appearance of lexically meaningful characters in the body of text or quoted literal sequences. Each escape sequence represents the literal character immediately following the initial \.

escaped-char = backslash ( backslash / "{" / "|" / "}" )
backslash    = %x5C ; U+005C REVERSE SOLIDUS "\"

Note

The escaped-char rule allows escaping some characters in places where they do not need to be escaped, such as braces in a quoted literal. For example, |foo {bar}| and |foo \{bar\}| are synonymous.

When writing or generating a message, escape sequences SHOULD NOT be used unless required by the syntax. That is, inside literals only escape | and inside patterns only escape { and }.

Whitespace

The syntax limits whitespace characters outside of a pattern to the following: U+0009 CHARACTER TABULATION (tab), U+000A LINE FEED (new line), U+000D CARRIAGE RETURN, U+3000 IDEOGRAPHIC SPACE, or U+0020 SPACE.

Inside patterns and quoted literals, whitespace is part of the content and is recorded and stored verbatim. Whitespace is not significant outside translatable text, except where required by the syntax.

There are two whitespace productions in the syntax. Optional whitespace is whitespace that is not required by the syntax, but which users might want to include to increase the readability of a message. Required whitespace is whitespace that is required by the syntax.

Both types of whitespace optionally permit the use of the bidirectional isolate controls and certain strongly directional marks. These can assist users in presenting messages that contain right-to-left text, literals, or names (including those for functions, options, option values, and keys)

Messages that contain right-to-left (aka RTL) characters SHOULD use one of the following mechanisms to make messages display intelligibly in plain-text editors:

  1. Use paired isolating bidi controls U+2066 LEFT-TO-RIGHT ISOLATE ("LRI") and U+2069 POP DIRECTIONAL ISOLATE ("PDI") as permitted by the ABNF around parts of any message containing RTL characters:
    • inside of placeholder markers { and }
    • outside quoted-pattern markers {{ and }}
    • outside of variable, function, markup, or attribute, including the identifying sigil (e.g. <LRI>$var</PDI> or <LRI>:ns:name</PDI>)
  2. Use the 'local-effect' bidi marks U+061C ARABIC LETTER MARK, U+200E LEFT-TO-RIGHT MARK or U+200F RIGHT-TO-LEFT MARK as permitted by the ABNF before or after identifiers, names, unquoted literals, or option values, especially when the values contain a mix of neutral, weakly directional, and strongly directional characters.

Important

Always take care not to add bidirectional controls or marks where they would be semantically significant or where they would unintentionally become part of the message's output:

Controls placed inside literal quotes or quoted patterns are part of the literal or pattern. Controls in a pattern will appear in the output of the message. Controls inside literal quotes are part of the literal and will be considered in operations such as matching a key to a selector.

Note

Users cannot be expected to create or manage bidirectional controls or marks in messages, since the characters are invisible and can be difficult to manage. Tools (such as resource editors or translation editors) and other implementations of MessageFormat 2 serialization are strongly encouraged to provide paired isolates around any right-to-left syntax as described above so that messages display appropriately as plain text.

These definitions of whitespace implement UAX#31 Requirement R3a-2. It is a profile of R3a-1 in that specification because:

Note

The character U+3000 IDEOGRAPHIC SPACE is included in whitespace for compatibility with certain East Asian keyboards and input methods, in which users might accidentally create these characters in a message.

; Required whitespace
s = *bidi ws o

; Optional whitespace
o = *(s / bidi)

; Bidirectional marks and isolates
; ALM / LRM / RLM / LRI, RLI, FSI & PDI
bidi = %x061C / %x200E / %x200F / %x2066-2069

; Whitespace characters
ws = SP / HTAB / CR / LF / %x3000

Complete ABNF

The grammar is formally defined below using the ABNF notation [STD68], including the modifications found in RFC 7405.

RFC7405 defines a variation of ABNF that is case-sensitive. Some ABNF tools are only compatible with the specification found in RFC 5234. To make message.abnf compatible with that version of ABNF, replace the rules of the same name with this block:

input = %x2E.69.6E.70.75.74  ; ".input"
local = %x2E.6C.6F.63.61.6C  ; ".local"
match = %x2E.6D.61.74.63.68  ; ".match"

message.abnf

message           = simple-message / complex-message

simple-message    = o [simple-start pattern]
simple-start      = simple-start-char / escaped-char / placeholder
pattern           = *(text-char / escaped-char / placeholder)
placeholder       = expression / markup

complex-message   = o *(declaration o) complex-body o
declaration       = input-declaration / local-declaration
complex-body      = quoted-pattern / matcher

input-declaration = input o variable-expression
local-declaration = local s variable o "=" o expression

quoted-pattern    = o "{{" pattern "}}"

matcher           = match-statement s variant *(o variant)
match-statement   = match 1*(s selector)
selector          = variable
variant           = key *(s key) quoted-pattern
key               = literal / "*"

; Expressions
expression          = literal-expression
                    / variable-expression
                    / function-expression
literal-expression  = "{" o literal [s function] *(s attribute) o "}"
variable-expression = "{" o variable [s function] *(s attribute) o "}"
function-expression = "{" o function *(s attribute) o "}"

markup = "{" o "#" identifier *(s option) *(s attribute) o ["/"] "}"  ; open and standalone
       / "{" o "/" identifier *(s option) *(s attribute) o "}"  ; close

; Expression and literal parts
function       = ":" identifier *(s option)
option         = identifier o "=" o (literal / variable)

attribute      = "@" identifier [o "=" o literal]

variable       = "$" name

literal          = quoted-literal / unquoted-literal
quoted-literal   = "|" *(quoted-char / escaped-char) "|"
unquoted-literal = name / number-literal
; number-literal matches JSON number (https://www.rfc-editor.org/rfc/rfc8259#section-6)
number-literal   = ["-"] (%x30 / (%x31-39 *DIGIT)) ["." 1*DIGIT] [%i"e" ["-" / "+"] 1*DIGIT]

; Keywords; Note that these are case-sensitive
input = %s".input"
local = %s".local"
match = %s".match"

; Names and identifiers
; identifier matches https://www.w3.org/TR/REC-xml-names/#NT-QName
; name matches https://www.w3.org/TR/REC-xml-names/#NT-NCName but excludes U+FFFD and U+061C
identifier = [namespace ":"] name
namespace  = name
name       = [bidi] name-start *name-char [bidi]
name-start = ALPHA / "_"
           / %xC0-D6 / %xD8-F6 / %xF8-2FF
           / %x370-37D / %x37F-61B / %x61D-1FFF / %x200C-200D
           / %x2070-218F / %x2C00-2FEF / %x3001-D7FF
           / %xF900-FDCF / %xFDF0-FFFC / %x10000-EFFFF
name-char  = name-start / DIGIT / "-" / "."
           / %xB7 / %x300-36F / %x203F-2040

; Restrictions on characters in various contexts
simple-start-char = content-char / "@" / "|"
text-char         = content-char / ws / "." / "@" / "|"
quoted-char       = content-char / ws / "." / "@" / "{" / "}"
content-char      = %x01-08        ; omit NULL (%x00), HTAB (%x09) and LF (%x0A)
                  / %x0B-0C        ; omit CR (%x0D)
                  / %x0E-1F        ; omit SP (%x20)
                  / %x21-2D        ; omit . (%x2E)
                  / %x2F-3F        ; omit @ (%x40)
                  / %x41-5B        ; omit \ (%x5C)
                  / %x5D-7A        ; omit { | } (%x7B-7D)
                  / %x7E-2FFF      ; omit IDEOGRAPHIC SPACE (%x3000)
                  / %x3001-D7FF    ; omit surrogates
                  / %xE000-10FFFF

; Character escapes
escaped-char = backslash ( backslash / "{" / "|" / "}" )
backslash    = %x5C ; U+005C REVERSE SOLIDUS "\"

; Required whitespace
s = *bidi ws o

; Optional whitespace
o = *(ws / bidi)

; Bidirectional marks and isolates
; ALM / LRM / RLM / LRI, RLI, FSI & PDI
bidi = %x061C / %x200E / %x200F / %x2066-2069

; Whitespace characters
ws = SP / HTAB / CR / LF / %x3000

Errors

Errors can occur during the processing of a message. Some errors can be detected statically, such as those due to problems with message syntax, violations of requirements in the data model, or requirements defined by a function. Other errors might be detected during selection or formatting of a given message. Where available, the use of validation tools is recommended, as early detection of errors makes their correction easier.

Error Handling

Syntax Errors and Data Model Errors apply to all message processors, and MUST be emitted as soon as possible. The other error categories are only emitted during formatting, but it might be possible to detect them with validation tools.

During selection and formatting, expression handlers MUST only emit Message Function Errors.

Implementations do not have to check for or emit Resolution Errors or Message Function Errors in expressions that are not otherwise used by the message, such as placeholders in unselected patterns or declarations that are never referenced during formatting.

When formatting a message with one or more errors, an implementation MUST provide a mechanism to discover and identify at least one of the errors. The exact form of error signaling is implementation defined. Some examples include throwing an exception, returning an error code, or providing a function or method for enumerating any errors.

For all valid messages, an implementation MUST enable a user to get a formatted result. The formatted result might include fallback values such as when a placeholder's expression produced an error during formatting.

The two above requirements MAY be fulfilled by a single formatting method, or separately by more than one such method.

When a message contains more than one error, or contains some error which leads to further errors, an implementation which does not emit all of the errors SHOULD prioritise Syntax Errors and Data Model Errors over others.

When an error occurs while resolving a selector or calling MatchSelectorKeys with its resolved value, the selector MUST NOT match any variant key other than the catch-all * and a Bad Selector error MUST be emitted.

Syntax Errors

Syntax Errors occur when the syntax representation of a message is not well-formed.

Example invalid messages resulting in a Syntax Error:

{{Missing end braces
{{Missing one end brace}
Unknown {{expression}}
.local $var = {|no message body|}

Data Model Errors

Data Model Errors occur when a message is not valid due to violating one of the semantic requirements on its structure.

Variant Key Mismatch

A Variant Key Mismatch occurs when the number of keys on a variant does not equal the number of selectors.

Example invalid messages resulting in a Variant Key Mismatch error:

.input {$one :func}
.match $one
1 2 {{Too many}}
* {{Otherwise}}
.input {$one :func}
.input {$two :func}
.match $one $two
1 2 {{Two keys}}
* {{Missing a key}}
* * {{Otherwise}}

Missing Fallback Variant

A Missing Fallback Variant error occurs when the message does not include a variant with only catch-all keys.

Example invalid messages resulting in a Missing Fallback Variant error:

.input {$one :func}
.match $one
1 {{Value is one}}
2 {{Value is two}}
.input {$one :func}
.input {$two :func}
.match $one $two
1 * {{First is one}}
* 1 {{Second is one}}

Missing Selector Annotation

A Missing Selector Annotation error occurs when the message contains a selector that does not directly or indirectly reference a declaration with a function.

Examples of invalid messages resulting in a Missing Selector Annotation error:

.match $one
1 {{Value is one}}
* {{Value is not one}}
.local $one = {|The one|}
.match $one
1 {{Value is one}}
* {{Value is not one}}
.input {$one}
.match $one
1 {{Value is one}}
* {{Value is not one}}

Duplicate Declaration

A Duplicate Declaration error occurs when a variable is declared more than once. Note that an input variable is implicitly declared when it is first used, so explicitly declaring it after such use is also an error.

Examples of invalid messages resulting in a Duplicate Declaration error:

.input {$var :number maximumFractionDigits=0}
.input {$var :number minimumFractionDigits=0}
{{Redeclaration of the same variable}}

.local $var = {$ext :number maximumFractionDigits=0}
.input {$var :number minimumFractionDigits=0}
{{Redeclaration of a local variable}}

.input {$var :number minimumFractionDigits=0}
.local $var = {$ext :number maximumFractionDigits=0}
{{Redeclaration of an input variable}}

.input {$var :number minimumFractionDigits=$var2}
.input {$var2 :number}
{{Redeclaration of the implicit input variable $var2}}

.local $var = {$ext :someFunction}
.local $var = {$error}
.local $var2 = {$var2 :error}
{{{$var} cannot be redefined. {$var2} cannot refer to itself}}

Duplicate Option Name

A Duplicate Option Name error occurs when the same identifier appears on the left-hand side of more than one option in the same expression.

Examples of invalid messages resulting in a Duplicate Option Name error:

Value is {42 :number style=percent style=decimal}
.local $foo = {horse :func one=1 two=2 one=1}
{{This is {$foo}}}

Duplicate Variant

A Duplicate Variant error occurs when the same list of keys is used for more than one variant.

Examples of invalid messages resulting in a Duplicate Variant error:

.input {$var :string}
.match $var
* {{The first default}}
* {{The second default}}
.input {$x :string}
.input {$y :string}
.match $x $y
*   foo   {{The first "foo" variant}}
bar *     {{The "bar" variant}}
*   |foo| {{The second "foo" variant}}
*   *     {{The default variant}}

Resolution Errors

Resolution Errors occur when the runtime value of a part of a message cannot be determined.

Unresolved Variable

An Unresolved Variable error occurs when a variable reference cannot be resolved.

For example, attempting to format either of the following messages would result in an Unresolved Variable error if done within a context that does not provide for the variable reference $var to be successfully resolved:

The value is {$var}.
.input {$var :func}
.match $var
1 {{The value is one.}}
* {{The value is not one.}}

Unknown Function

An Unknown Function error occurs when an expression includes a reference to a function which cannot be resolved.

For example, attempting to format either of the following messages would result in an Unknown Function error if done within a context that does not provide for the function :func to be successfully resolved:

The value is {horse :func}.
.local $horse = {|horse| :func}
.match $horse
1 {{The value is one.}}
* {{The value is not one.}}

Bad Selector

A Bad Selector error occurs when a message includes a selector with a resolved value which does not support selection.

For example, attempting to format this message would result in a Bad Selector error:

.local $day = {|2024-05-01| :date}
.match $day
* {{The due date is {$day}}}

Message Function Errors

A Message Function Error is any error that occurs when calling a message function implementation or which depends on validation associated with a specific function.

Implementations SHOULD provide a way for functions to emit (or cause to be emitted) any of the types of error defined in this section. Implementations MAY also provide implementation-defined Message Function Error types.

For example, attempting to format any of the following messages might result in a Message Function Error if done within a context that

  1. Provides for the variable reference $user to resolve to an object { name: 'Kat', id: 1234 },
  2. Provides for the variable reference $field to resolve to a string 'address', and
  3. Uses a :get message function which requires its argument to be an object and an option field to be provided with a string value.

The exact type of Message Function Error is determined by the message function implementation.

Hello, {horse :get field=name}!
Hello, {$user :get}!
.local $id = {$user :get field=id}
{{Hello, {$id :get field=name}!}}
Your {$field} is {$id :get field=$field}

Bad Operand

A Bad Operand error is any error that occurs due to the content or format of the operand, such as when the operand provided to a function during function resolution does not match one of the expected implementation-defined types for that function; or in which a literal operand value does not have the required format and thus cannot be processed into one of the expected implementation-defined types for that specific function.

For example, the following messages each produce a Bad Operand error because the literal |horse| does not match the number-literal production, which is a requirement of the function :number for its operand:

.local $horse = {|horse| :number}
{{You have a {$horse}.}}
.local $horse = {|horse| :number}
.match $horse
1 {{The value is one.}}
* {{The value is not one.}}

Bad Option

A Bad Option error is an error that occurs when there is an implementation-defined error with an option or its value. These might include:

For example, the following message might produce a Bad Option error because the literal foo does not match the production digit-size-option, which is a requirement of the function :number for its minimumFractionDigits option:

The answer is {42 :number minimumFractionDigits=foo}.

Bad Variant Key

A Bad Variant Key error is an error that occurs when a variant key does not match the expected implementation-defined format.

For example, the following message produces a Bad Variant Key error because horse is not a recognized plural category and does not match the number-literal production, which is a requirement of the :number function:

.local $answer = {42 :number}
.match $answer
1     {{The value is one.}}
horse {{The value is a horse.}}
*     {{The value is not one.}}

Function Registry

This section describes the functions which each implementation MUST provide to be conformant with this specification.

Implementations MAY implement additional functions or additional options. In particular, implementations are encouraged to provide feedback on proposed options and their values.

Note

The Stability Policy allows for updates to Default Registry functions to add support for new options. As implementations are permitted to ignore options that they do not support, it is possible to write messages using options not defined below which currently format with no error, but which could produce errors when formatted with a later edition of the Default Registry. Therefore, using options not explicitly defined here is NOT RECOMMENDED.

String Value Selection and Formatting

The :string function

The function :string provides string selection and formatting.

Operands

The operand of :string is either any implementation-defined type that is a string or for which conversion to a string is supported, or any literal value. All other values produce a Bad Operand error.

For example, in Java, implementations of the java.lang.CharSequence interface (such as java.lang.String or java.lang.StringBuilder), the type char, or the class java.lang.Character might be considered as the "implementation-defined types". Such an implementation might also support other classes via the method toString(). This might be used to enable selection of a enum value by name, for example.

Other programming languages would define string and character sequence types or classes according to their local needs, including, where appropriate, coercion to string.

Options

The function :string has no options.

Note

Proposals for string transformation options or implementation experience with user requirements is desired during the Tech Preview.

Selection

When implementing MatchSelectorKeys(resolvedSelector, keys) where resolvedSelector is the resolved value of a selector and keys is a list of strings, the :string selector function performs as described below.

  1. Let compare be the string value of resolvedSelector.
  2. Let result be a new empty list of strings.
  3. For each string key in keys:
    1. If key and compare consist of the same sequence of Unicode code points, then
      1. Append key as the last element of the list result.
  4. Return result.

Note

Matching of key and compare values is sensitive to the sequence of code points in each string. As a result, variations in how text can be encoded can affect the performance of matching. The function :string does not perform case folding or Unicode Normalization of string values. Users SHOULD encode messages and their parts (such as keys and operands), in Unicode Normalization Form C (NFC) unless there is a very good reason not to. See also: String Matching

Note

Unquoted string literals in a variant do not include spaces. If users wish to match strings that include whitespace (including U+3000 IDEOGRAPHIC SPACE) to a key, the key needs to be quoted.

For example:

.input {$string :string}
.match $string
| space key | {{Matches the string " space key "}}
*             {{Matches the string "space key"}}

Formatting

The :string function returns the string value of the resolved value of the operand.

Numeric Value Selection and Formatting

The :number function

The function :number is a selector and formatter for numeric values.

Operands

The function :number requires a Number Operand as its operand.

Options

Some options do not have default values defined in this specification. The defaults for these options are implementation-dependent. In general, the default values for such options depend on the locale, the value of other options, or both.

Note

The names of options and their values were derived from the options in JavaScript's Intl.NumberFormat.

The following options and their values are required to be available on the function :number:

Note

The following options and option values are being developed during the Technical Preview period.

The following values for the option style are not part of the default registry. Implementations SHOULD avoid creating options that conflict with these, but are encouraged to track development of these options during Tech Preview:

The following options are not part of the default registry. Implementations SHOULD avoid creating options that conflict with these, but are encouraged to track development of these options during Tech Preview:

Default Value of select Option

The value plural is the default for the option select because it is the most common use case for numeric selection. It can be used for exact value matches but also allows for the grammatical needs of languages using CLDR's plural rules. This might not be noticeable in the source language (particularly English), but can cause problems in target locales that the original developer is not considering.

For example, a naive developer might use a special message for the value 1 without considering a locale's need for a one plural:

.input {$var :number}
.match $var
1   {{You have one last chance}}
one {{You have {$var} chance remaining}}
*   {{You have {$var} chances remaining}}

The one variant is needed by languages such as Polish or Russian. Such locales typically also require other keywords such as two, few, and many.

Percent Style

When implementing style=percent, the numeric value of the operand MUST be multiplied by 100 for the purposes of formatting.

For example,

The total was {0.5 :number style=percent}.

should format in a manner similar to:

The total was 50%.

Selection

The function :number performs selection as described in Number Selection below.

The :integer function

The function :integer is a selector and formatter for matching or formatting numeric values as integers.

Operands

The function :integer requires a Number Operand as its operand.

Options

Some options do not have default values defined in this specification. The defaults for these options are implementation-dependent. In general, the default values for such options depend on the locale, the value of other options, or both.

Note

The names of options and their values were derived from the options in JavaScript's Intl.NumberFormat.

The following options and their values are required in the default registry to be available on the function :integer:

Note

The following options and option values are being developed during the Technical Preview period.

The following values for the option style are not part of the default registry. Implementations SHOULD avoid creating options that conflict with these, but are encouraged to track development of these options during Tech Preview:

The following options are not part of the default registry. Implementations SHOULD avoid creating options that conflict with these, but are encouraged to track development of these options during Tech Preview:

Default Value of select Option

The value plural is the default for the option select because it is the most common use case for numeric selection. It can be used for exact value matches but also allows for the grammatical needs of languages using CLDR's plural rules. This might not be noticeable in the source language (particularly English), but can cause problems in target locales that the original developer is not considering.

For example, a naive developer might use a special message for the value 1 without considering a locale's need for a one plural:

.input {$var :integer}
.match $var
1   {{You have one last chance}}
one {{You have {$var} chance remaining}}
*   {{You have {$var} chances remaining}}

The one variant is needed by languages such as Polish or Russian. Such locales typically also require other keywords such as two, few, and many.

Percent Style

When implementing style=percent, the numeric value of the operand MUST be multiplied by 100 for the purposes of formatting.

For example,

The total was {0.5 :number style=percent}.

should format in a manner similar to:

The total was 50%.

Selection

The function :integer performs selection as described in Number Selection below.

Number Operands

The operand of a number function is either an implementation-defined type or a literal whose contents match the number-literal production in the ABNF. All other values produce a Bad Operand error.

For example, in Java, any subclass of java.lang.Number plus the primitive types (byte, short, int, long, float, double, etc.) might be considered as the "implementation-defined numeric types". Implementations in other programming languages would define different types or classes according to their local needs.

Note

String values passed as variables in the formatting context's input mapping can be formatted as numeric values as long as their contents match the number-literal production in the ABNF.

For example, if the value of the variable num were the string -1234.567, it would behave identically to the local variable in this example:

.local $example = {|-1234.567| :number}
{{{$num :number} == {$example}}}

Note

Implementations are encouraged to provide support for compound types or data structures that provide additional semantic meaning to the formatting of number-like values. For example, in ICU4J, the type com.ibm.icu.util.Measure can be used to communicate a value that includes a unit or the type com.ibm.icu.util.CurrencyAmount can be used to set the currency and related options (such as the number of fraction digits).

Digit Size Options

Some options of number functions are defined to take a "digit size option". Implementations of number functions use these options to control aspects of numeric display such as the number of fraction, integer, or significant digits.

A "digit size option" is an option value that the function interprets as a small integer value greater than or equal to zero. Implementations MAY define an upper limit on the resolved value of a digit size option option consistent with that implementation's practical limits.

In most cases, the value of a digit size option will be a string that encodes the value as a non-negative integer. Implementations MAY also accept implementation-defined types as the value. When provided as a string, the representation of a digit size option matches the following ABNF:

digit-size-option = "0" / (("1"-"9") [DIGIT])

If the value of a digit size option does not evaluate as a non-negative integer, or if the value exceeds any implementation-defined upper limit or any option-specific lower limit, a Bad Option Error is emitted.

Number Selection

Number selection has three modes:

When implementing MatchSelectorKeys(resolvedSelector, keys) where resolvedSelector is the resolved value of a selector and keys is a list of strings, numeric selectors perform as described below.

  1. Let exact be the JSON string representation of the numeric value of resolvedSelector. (See Determining Exact Literal Match for details)
  2. Let keyword be a string which is the result of rule selection on resolvedSelector.
  3. Let resultExact be a new empty list of strings.
  4. Let resultKeyword be a new empty list of strings.
  5. For each string key in keys:
    1. If the value of key matches the production number-literal, then
      1. If key and exact consist of the same sequence of Unicode code points, then
        1. Append key as the last element of the list resultExact.
    2. Else if key is one of the keywords zero, one, two, few, many, or other, then
      1. If key and keyword consist of the same sequence of Unicode code points, then
        1. Append key as the last element of the list resultKeyword.
    3. Else, emit a Bad Variant Key error.
  6. Return a new list whose elements are the concatenation of the elements (in order) of resultExact followed by the elements (in order) of resultKeyword.

Note

Implementations are not required to implement this exactly as written. However, the observed behavior must be consistent with what is described here.

Rule Selection

Rule selection is intended to support the grammatical matching needs of different languages/locales in order to support plural or ordinal numeric values.

If the option select is set to exact, rule-based selection is not used. Otherwise rule selection matches the operand, as modified by function options, to exactly one of these keywords: zero, one, two, few, many, or other. The keyword other is the default.

Note

Since valid keys cannot be the empty string in a numeric expression, returning the empty string disables keyword selection.

The meaning of the keywords is locale-dependent and implementation-defined. A key that matches the rule-selected keyword is a stronger match than the fallback key * but a weaker match than any exact match key value.

The rules for a given locale might not produce all of the keywords. A given operand value might produce different keywords depending on the locale.

Apply the rules to the resolved value of the operand and the relevant function options, and return the resulting keyword. If no rules match, return other.

If the option select is set to plural, the rules applied to selection SHOULD be the CLDR plural rule data of type cardinal. See charts for examples.

If the option select is set to ordinal, the rules applied to selection SHOULD be the CLDR plural rule data of type ordinal. See charts for examples.

Example. In CLDR 44, the Czech (cs) plural rule set can be found here.

A message in Czech might be:

.input {$numDays :number}
.match $numDays
one  {{{$numDays} den}}
few  {{{$numDays} dny}}
many {{{$numDays} dne}}
*    {{{$numDays} dní}}

Using the rules found above, the results of various operand values might look like:

Operand value Keyword Formatted Message
1 one 1 den
2 few 2 dny
5 other 5 dní
22 few 22 dny
27 other 27 dní
2.4 many 2,4 dne

Determining Exact Literal Match

Important

The exact behavior of exact literal match is currently only well defined for non-zero-filled integer values. Functions that use fraction digits or significant digits might work in specific implementation-defined ways. Users should avoid depending on these types of keys in message selection in this release.

Number literals in the MessageFormat 2 syntax use the format defined for a JSON number. A resolvedSelector exactly matches a numeric literal key if, when the numeric value of resolvedSelector is serialized using the format for a JSON number, the two strings are equal.

Note

The above description of numeric matching contains open issues in the Technical Preview, since a given numeric value might be formatted in several different ways under RFC8259 and since the effect of formatting options, such as the number of fraction digits or significant digits, is not described. The Working Group intends to address these issues before final release with a number of design options being considered.

Users should avoid creating messages that depend on exact matching of non-integer numeric values. Feedback, including use cases encountered in message authoring, is strongly desired.

Date and Time Value Formatting

This subsection describes the functions and options for date/time formatting. Selection based on date and time values is not required in this release.

Note

Selection based on date/time types is not required by MF2. Implementations should use care when defining selectors based on date/time types. The types of queries found in implementations such as java.time.TemporalAccessor are complex and user expectations may be inconsistent with good I18N practices.

The :datetime function

The function :datetime is used to format date/time values, including the ability to compose user-specified combinations of fields.

If no options are specified, this function defaults to the following:

Note

The default formatting behavior of :datetime is inconsistent with Intl.DateTimeFormat in JavaScript and with {d,date} in ICU MessageFormat 1.0. This is because, unlike those implementations, :datetime is distinct from :date and :time.

Operands

The operand of the :datetime function is either an implementation-defined date/time type or a date/time literal value, as defined in Date and Time Operand. All other operand values produce a Bad Operand error.

Options

The :datetime function can use either the appropriate style options or can use a collection of field options (but not both) to control the formatted output.

If both are specified, a Bad Option error MUST be emitted and a fallback value used as the resolved value of the expression.

Note

The names of options and their values were derived from the options in JavaScript's Intl.DateTimeFormat.

Style Options

The function :datetime has these style options.

Field Options

Field options describe which fields to include in the formatted output and what format to use for that field. The implementation may use this function to configure which fields appear in the formatted output.

Note

Field options do not have default values because they are only to be used to compose the formatter.

The field options are defined as follows:

Important

The value 2-digit for some field options must be quoted in the MessageFormat syntax because it starts with a digit but does not match the number-literal production in the ABNF.

.local $correct = {$someDate :datetime year=|2-digit|}
.local $syntaxError = {$someDate :datetime year=2-digit}

The function :datetime has the following options:

Note

The following options do not have default values because they are only to be used as overrides for locale-and-value dependent implementation-defined defaults.

The following date/time options are not part of the default registry. Implementations SHOULD avoid creating options that conflict with these, but are encouraged to track development of these options during Tech Preview:

The :date function

The function :date is used to format the date portion of date/time values.

If no options are specified, this function defaults to the following:

Operands

The operand of the :date function is either an implementation-defined date/time type or a date/time literal value, as defined in Date and Time Operand. All other operand values produce a Bad Operand error.

Options

The function :date has these options:

The :time function

The function :time is used to format the time portion of date/time values.

If no options are specified, this function defaults to the following:

Operands

The operand of the :time function is either an implementation-defined date/time type or a date/time literal value, as defined in Date and Time Operand. All other operand values produce a Bad Operand error.

Options

The function :time has these options:

Date and Time Operands

The operand of a date/time function is either an implementation-defined date/time type or a date/time literal value, as defined below. All other operand values produce a Bad Operand error.

A date/time literal value is a non-empty string consisting of an ISO 8601 date, or an ISO 8601 datetime optionally followed by a timezone offset. As implementations differ slightly in their parsing of such strings, ISO 8601 date and datetime values not matching the following regular expression MAY also be supported. Furthermore, matching this regular expression does not guarantee validity, given the variable number of days in each month.

(?!0000)[0-9]{4}-(0[1-9]|1[0-2])-(0[1-9]|[12][0-9]|3[01])(T([01][0-9]|2[0-3]):[0-5][0-9]:[0-5][0-9](\.[0-9]{1,3})?(Z|[+-]((0[0-9]|1[0-3]):[0-5][0-9]|14:00))?)?

When the time is not present, implementations SHOULD use 00:00:00 as the time. When the offset is not present, implementations SHOULD use a floating time type (such as Java's java.time.LocalDateTime) to represent the time value. For more information, see Working with Timezones.

Important

The ABNF and syntax of MF2 do not formally define date/time literals. This means that a message can be syntactically valid but produce a Bad Operand error at runtime.

Note

String values passed as variables in the formatting context's input mapping can be formatted as date/time values as long as their contents are date/time literals.

For example, if the value of the variable now were the string 2024-02-06T16:40:00Z, it would behave identically to the local variable in this example:

.local $example = {|2024-02-06T16:40:00Z| :datetime}
{{{$now :datetime} == {$example}}}

Note

True time zone support in serializations is expected to coincide with the adoption of Temporal in JavaScript. The form of these serializations is known and is a de facto standard. Support for these extensions is expected to be required in the post-tech preview. See: https://datatracker.ietf.org/doc/draft-ietf-sedate-datetime-extended/

Formatting

This section defines the behaviour of a MessageFormat 2.0 implementation when formatting a message for display in a user interface, or for some later processing.

To start, we presume that a message has either been parsed from its syntax or created from a data model description. If the resulting message is not well-formed, a Syntax Error is emitted. If the resulting message is well-formed but is not valid, a Data Model Error is emitted.

The formatting of a message is defined by the following operations:

Implementations are not required to expose the expression resolution and pattern selection operations to their users, or even use them in their internal processing, as long as the final formatting result is made available to users and the observable behavior of the formatting matches that described here.

Attributes MUST NOT have any effect on the formatted output of a message, nor be made available to function implementations.

Important

This specification does not require either eager or lazy expression resolution of message parts; do not construe any requirement in this document as requiring either.

Implementations are not required to evaluate all parts of a message when parsing, processing, or formatting. In particular, an implementation MAY choose not to evaluate or resolve the value of a given expression until it is actually used by a selection or formatting process. However, when an expression is resolved, it MUST behave as if all preceding declarations affecting variables referenced by that expression have already been evaluated in the order in which the relevant declarations appear in the message.

Formatting Context

A message's formatting context represents the data and procedures that are required for the message's expression resolution, pattern selection and formatting.

At a minimum, it includes:

Implementations MAY include additional fields in their formatting context.

Expression and Markup Resolution

Expressions are used in declarations and patterns. Markup is only used in patterns.

In a declaration, the resolved value of the expression is bound to a variable, which is available for use by later expressions. Since a variable can be referenced in different ways later, implementations SHOULD NOT immediately fully format the value for output.

In an input-declaration, the variable operand of the variable-expression identifies not only the name of the external input value, but also the variable to which the resolved value of the variable-expression is bound.

In a pattern, the resolved value of an expression or markup is used in its formatting.

The form that resolved values take is implementation-dependent, and different implementations MAY choose to perform different levels of resolution.

For example, the resolved value of the expression {|0.40| :number style=percent} could be an object such as

{ value: Number('0.40'),
  formatter: NumberFormat(locale, { style: 'percent' }) }

Alternatively, it could be an instance of an ICU4J FormattedNumber, or some other locally appropriate value.

Depending on the presence or absence of a variable or literal operand and a function, the resolved value of the expression is determined as follows:

If the expression contains a function, its resolved value is defined by function resolution.

Else, if the expression consists of a variable, its resolved value is defined by variable resolution. An implementation MAY perform additional processing when resolving the value of an expression that consists only of a variable.

For example, it could apply function resolution using a function and a set of options chosen based on the value or type of the variable. So, given a message like this:

Today is {$date}

If the value passed in the variable were a date object, such as a JavaScript Date or a Java java.util.Date or java.time.Temporal, the implementation could interpret the placeholder {$date} as if the pattern included the function :datetime with some set of default options.

Else, the expression consists of a literal. Its resolved value is defined by literal resolution.

Note This means that a literal value with no function is always treated as a string. To represent values that are not strings as a literal, a function needs to be provided:

.local $aNumber = {1234 :number}
.local $aDate = {|2023-08-30| :datetime}
.local $aFoo = {|some foo| :foo}
{{You have {42 :number}}}

Literal Resolution

The resolved value of a text or a literal is the character sequence of the text or literal after any character escape has been converted to the escaped character.

When a literal is used as an operand or on the right-hand side of an option, the formatting function MUST treat its resolved value the same whether its value was originally a quoted literal or an unquoted literal.

For example, the option foo=42 and the option foo=|42| are treated as identical.

The resolution of a text or literal MUST resolve to a string.

Variable Resolution

To resolve the value of a variable, its name is used to identify either a local variable or an input variable. If a declaration exists for the variable, its resolved value is used. Otherwise, the variable is an implicit reference to an input value, and its value is looked up from the formatting context input mapping.

The resolution of a variable fails if no value is identified for its name. If this happens, an Unresolved Variable error is emitted. If a variable would resolve to a fallback value, this MUST also be considered a failure.

Function Resolution

To resolve an expression with a function, the following steps are taken:

  1. If the expression includes an operand, resolve its value. If this fails, use a fallback value for the expression.

  2. Resolve the identifier of the function and, based on the starting sigil, find the appropriate function implementation to call. If the implementation cannot find the function, or if the identifier includes a namespace that the implementation does not support, emit an Unknown Function error and use a fallback value for the expression.

    Implementations are not required to implement namespaces or installable function registries.

  3. Perform option resolution.

  4. Call the function implementation with the following arguments:

    • The current locale.
    • The resolved mapping of options.
    • If the expression includes an operand, its resolved value.

    The form that resolved operand and option values take is implementation-defined.

    A declaration binds the resolved value of an expression to a variable. Thus, the result of one function is potentially the operand of another function, or the value of one of the options for another function. For example, in

    .input {$n :number minimumIntegerDigits=3}
    .local $n1 = {$n :number maximumFractionDigits=3}
    

    the value bound to $n is the resolved value used as the operand of the :number function when resolving the value of the variable $n1.

    Implementations that provide a means for defining custom functions SHOULD provide a means for function implementations to return values that contain enough information (e.g. a representation of the resolved operand and option values that the function was called with) to be used as arguments to subsequent calls to the function implementations. For example, an implementation might define an interface that allows custom function implementation. Such an interface SHOULD define an implementation-specific argument type T and return type U for implementations of functions such that U can be coerced to T. Implementations of a function SHOULD emit a Bad Operand error for operands whose resolved value or type is not supported.

Note

The behavior of the previous example is currently implementation-dependent. Supposing that the external input variable n is bound to the string "1", and that the implementation formats to a string, the formatted result of the following message:

.input {$n :number minimumIntegerDigits=3}
.local $n1 = {$n :number maximumFractionDigits=3}
{{What is the value of: {$n1}}}

is currently implementation-dependent. Depending on whether the options are preserved between the resolution of the first :number function and the resolution of the second :number function, a conformant implementation could produce either "001.000" or "1.000"

Each function specification MAY have its own rules to preserve some options in the returned structure and discard others. In instances where a function specification does not determine whether an option is preserved or discarded, each function implementation of that specification MAY have its own rules to preserve some options in the returned structure and discard others.

Note

During the Technical Preview, feedback on how the registry describes the flow of resolved values and options from one function to another, and on what requirements this specification should impose, is highly desired.

An implementation MAY pass additional arguments to the function, as long as reasonable precautions are taken to keep the function interface simple and minimal, and avoid introducing potential security vulnerabilities.

An implementation MAY define its own functions. An implementation MAY allow custom functions to be defined by users.

Function access to the formatting context MUST be minimal and read-only, and execution time SHOULD be limited.

Implementation-defined functions SHOULD use an implementation-defined namespace.

  1. If the call succeeds, resolve the value of the expression as the result of that function call.

    If the call fails or does not return a valid value, emit the appropriate Message Function Error for the failure.

    Implementations MAY provide a mechanism for the function to provide additional detail about internal failures. Specifically, if the cause of the failure was that the datatype, value, or format of the operand did not match that expected by the function, the function might cause a Bad Operand error to be emitted.

    In all failure cases, use the fallback value for the expression as the resolved value.

Option Resolution

The result of resolving option values is an unordered mapping of string identifiers to values.

For each option:

Errors MAY be emitted during option resolution, but it always resolves to some mapping of string identifiers to values. This mapping can be empty.

Markup Resolution

Unlike functions, the resolution of markup is not customizable.

The resolved value of markup includes the following fields:

The resolution of markup MUST always succeed.

Fallback Resolution

A fallback value is the resolved value for an expression that fails to resolve.

An expression fails to resolve when:

The fallback value depends on the contents of the expression:

Option identifiers and values are not included in the fallback value.

Pattern selection is not supported for fallback values.

Pattern Selection

If the message being formatted is not well-formed and valid, the result of pattern selection is a pattern consisting of a single fallback value using the message's fallback string defined in the formatting context or if this is not available or empty, the U+FFFD REPLACEMENT CHARACTER .

If the message being formatted does not contain a matcher, the result of pattern selection is its pattern value.

When a message contains a matcher with one or more selectors, the implementation needs to determine which variant will be used to provide the pattern for the formatting operation. This is done by ordering and filtering the available variant statements according to their key values and selecting the first one.

Note

At least one variant is required to have all of its keys consist of the fallback value *. Some selectors might be implemented in a way that the key value * cannot be selected in a valid message. In other cases, this key value might be unreachable only in certain locales. This could result in the need in some locales to create one or more variants that do not make sense grammatically for that language.

For example, in the pl (Polish) locale, this message cannot reach the * variant:

.input {$num :integer}
.match $num
0    {{ }}
one  {{ }}
few  {{ }}
many {{ }}
*    {{Only used by fractions in Polish.}}

In the Tech Preview, feedback from users and implementers is desired about whether to relax the requirement that such a "fallback variant" appear in every message, versus the potential for a message to fail at runtime because no matching variant is available.

The number of keys in each variant MUST equal the number of selectors.

Each key corresponds to a selector by its position in the variant.

For example, in this message:

.input {$one :number}
.input {$two :number}
.input {$three :number}
.match $one $two $three
1 2 3 {{ ... }}

The first key 1 corresponds to the first selector ($one), the second key 2 to the second selector ($two), and the third key 3 to the third selector ($three).

To determine which variant best matches a given set of inputs, each selector is used in turn to order and filter the list of variants.

Each variant with a key that does not match its corresponding selector is omitted from the list of variants. The remaining variants are sorted according to the selector's key-ordering preference. Earlier selectors in the matcher's list of selectors have a higher priority than later ones.

When all of the selectors have been processed, the earliest-sorted variant in the remaining list of variants is selected.

This selection method is defined in more detail below. An implementation MAY use any pattern selection method, as long as its observable behavior matches the results of the method defined here.

Resolve Selectors

First, resolve the values of each selector:

  1. Let res be a new empty list of resolved values that support selection.
  2. For each selector sel, in source order,
    1. Let rv be the resolved value of sel.
    2. If selection is supported for rv:
      1. Append rv as the last element of the list res.
    3. Else:
      1. Let nomatch be a resolved value for which selection always fails.
      2. Append nomatch as the last element of the list res.
      3. Emit a Bad Selector error.

The form of the resolved values is determined by each implementation, along with the manner of determining their support for selection.

Resolve Preferences

Next, using res, resolve the preferential order for all message keys:

  1. Let pref be a new empty list of lists of strings.
  2. For each index i in res:
    1. Let keys be a new empty list of strings.
    2. For each variant var of the message:
      1. Let key be the var key at position i.
      2. If key is not the catch-all key '*':
        1. Assert that key is a literal.
        2. Let ks be the resolved value of key in Unicode Normalization Form C.
        3. Append ks as the last element of the list keys.
    3. Let rv be the resolved value at index i of res.
    4. Let matches be the result of calling the method MatchSelectorKeys(rv, keys)
    5. Append matches as the last element of the list pref.

The method MatchSelectorKeys is determined by the implementation. It takes as arguments a resolved selector value rv and a list of string keys keys, and returns a list of string keys in preferential order. The returned list MUST contain only unique elements of the input list keys. The returned list MAY be empty. The most-preferred key is first, with each successive key appearing in order by decreasing preference.

The resolved value of each key MUST be in Unicode Normalization Form C ("NFC"), even if the literal for the key is not.

If calling MatchSelectorKeys encounters any error, a Bad Selector error is emitted and an empty list is returned.

Filter Variants

Then, using the preferential key orders pref, filter the list of variants to the ones that match with some preference:

  1. Let vars be a new empty list of variants.
  2. For each variant var of the message:
    1. For each index i in pref:
      1. Let key be the var key at position i.
      2. If key is the catch-all key '*':
        1. Continue the inner loop on pref.
      3. Assert that key is a literal.
      4. Let ks be the resolved value of key.
      5. Let matches be the list of strings at index i of pref.
      6. If matches includes ks:
        1. Continue the inner loop on pref.
      7. Else:
        1. Continue the outer loop on message variants.
    2. Append var as the last element of the list vars.

Sort Variants

Finally, sort the list of variants vars and select the pattern:

  1. Let sortable be a new empty list of (integer, variant) tuples.
  2. For each variant var of vars:
    1. Let tuple be a new tuple (-1, var).
    2. Append tuple as the last element of the list sortable.
  3. Let len be the integer count of items in pref.
  4. Let i be len - 1.
  5. While i >= 0:
    1. Let matches be the list of strings at index i of pref.
    2. Let minpref be the integer count of items in matches.
    3. For each tuple tuple of sortable:
      1. Let matchpref be an integer with the value minpref.
      2. Let key be the tuple variant key at position i.
      3. If key is not the catch-all key '*':
        1. Assert that key is a literal.
        2. Let ks be the resolved value of key.
        3. Let matchpref be the integer position of ks in matches.
      4. Set the tuple integer value as matchpref.
    4. Set sortable to be the result of calling the method SortVariants(sortable).
    5. Set i to be i - 1.
  6. Let var be the variant element of the first element of sortable.
  7. Select the pattern of var.

SortVariants is a method whose single argument is a list of (integer, variant) tuples. It returns a list of (integer, variant) tuples. Any implementation of SortVariants is acceptable as long as it satisfies the following requirements:

  1. Let sortable be an arbitrary list of (integer, variant) tuples.
  2. Let sorted be SortVariants(sortable).
  3. sorted is the result of sorting sortable using the following comparator:
    1. (i1, v1) <= (i2, v2) if and only if i1 <= i2.
  4. The sort is stable (pairs of tuples from sortable that are equal in their first element have the same relative order in sorted).

Examples

This section is non-normative.

Example 1

Presuming a minimal implementation which only supports :string function which matches keys by using string comparison, and a formatting context in which the variable reference $foo resolves to the string 'foo' and the variable reference $bar resolves to the string 'bar', pattern selection proceeds as follows for this message:

.input {$foo :string}
.input {$bar :string}
.match $foo $bar
bar bar {{All bar}}
foo foo {{All foo}}
* * {{Otherwise}}
  1. For the first selector:
    The value of the selector is resolved to be 'foo'.
    The available keys « 'bar', 'foo' » are compared to 'foo',
    resulting in a list « 'foo' » of matching keys.

  2. For the second selector:
    The value of the selector is resolved to be 'bar'.
    The available keys « 'bar', 'foo' » are compared to 'bar',
    resulting in a list « 'bar' » of matching keys.

  3. Creating the list vars of variants matching all keys:
    The first variant bar bar is discarded as its first key does not match the first selector.
    The second variant foo foo is discarded as its second key does not match the second selector.
    The catch-all keys of the third variant * * always match, and this is added to vars,
    resulting in a list « * * » of variants.

  4. As the list vars only has one entry, it does not need to be sorted.
    The pattern Otherwise of the third variant is selected.

Example 2

Alternatively, with the same implementation and formatting context as in Example 1, pattern selection would proceed as follows for this message:

.input {$foo :string}
.input {$bar :string}
.match $foo $bar
* bar {{Any and bar}}
foo * {{Foo and any}}
foo bar {{Foo and bar}}
* * {{Otherwise}}
  1. For the first selector:
    The value of the selector is resolved to be 'foo'.
    The available keys « 'foo' » are compared to 'foo',
    resulting in a list « 'foo' » of matching keys.

  2. For the second selector:
    The value of the selector is resolved to be 'bar'.
    The available keys « 'bar' » are compared to 'bar',
    resulting in a list « 'bar' » of matching keys.

  3. Creating the list vars of variants matching all keys:
    The keys of all variants either match each selector exactly, or via the catch-all key,
    resulting in a list « * bar, foo *, foo bar, * * » of variants.

  4. Sorting the variants:
    The list sortable is first set with the variants in their source order and scores determined by the second selector:
    « ( 0, * bar ), ( 1, foo * ), ( 0, foo bar ), ( 1, * * ) »
    This is then sorted as:
    « ( 0, * bar ), ( 0, foo bar ), ( 1, foo * ), ( 1, * * ) ».
    To sort according to the first selector, the scores are updated to:
    « ( 1, * bar ), ( 0, foo bar ), ( 0, foo * ), ( 1, * * ) ».
    This is then sorted as:
    « ( 0, foo bar ), ( 0, foo * ), ( 1, * bar ), ( 1, * * ) ».

  5. The pattern Foo and bar of the most preferred foo bar variant is selected.

Example 3

A more-complex example is the matching found in selection APIs such as ICU's PluralFormat. Suppose that this API is represented here by the function :number. This :number function can match a given numeric value to a specific number literal and also to a plural category (zero, one, two, few, many, other) according to locale rules defined in CLDR.

Given a variable reference $count whose value resolves to the number 1 and an en (English) locale, the pattern selection proceeds as follows for this message:

.input {$count :number}
.match $count
one {{Category match for {$count}}}
1   {{Exact match for {$count}}}
*   {{Other match for {$count}}}
  1. For the selector:
    The value of the selector is resolved to an implementation-defined value that is capable of performing English plural category selection on the value 1.
    The available keys « 'one', '1' » are passed to the implementation's MatchSelectorKeys method,
    resulting in a list « '1', 'one' » of matching keys.

  2. Creating the list vars of variants matching all keys:
    The keys of all variants are included in the list of matching keys, or use the catch-all key,
    resulting in a list « one, 1, * » of variants.

  3. Sorting the variants:
    The list sortable is first set with the variants in their source order and scores determined by the selector key order:
    « ( 1, one ), ( 0, 1 ), ( 2, * ) »
    This is then sorted as:
    « ( 0, 1 ), ( 1, one ), ( 2, * ) »

  4. The pattern Exact match for {$count} of the most preferred 1 variant is selected.

Formatting

After pattern selection, each text and placeholder part of the selected pattern is resolved and formatted.

Resolved values cannot always be formatted by a given implementation. When such an error occurs during formatting, an appropriate Message Function Error is emitted and a fallback value is used for the placeholder with the error.

Implementations MAY represent the result of formatting using the most appropriate data type or structure. Some examples of these include:

Implementations SHOULD provide formatting result types that match user needs, including situations that require further processing of formatted messages. Implementations SHOULD encourage users to consider a formatted localised string as an opaque data structure, suitable only for presentation.

When formatting to a string, the default representation of all markup MUST be an empty string. Implementations MAY offer functionality for customizing this, such as by emitting XML-ish tags for each markup.

Examples

This section is non-normative.

  1. An implementation might choose to return an interstitial object so that the caller can "decorate" portions of the formatted value. In ICU4J, the NumberFormatter class returns a FormattedNumber object, so a pattern such as This is my number {42 :number} might return the character sequence This is my number followed by a FormattedNumber object representing the value 42 in the current locale.

  2. A formatter in a web browser could format a message as a DOM fragment rather than as a representation of its HTML source.

Formatting Fallback Values

If the resolved pattern includes any fallback values and the formatting result is a concatenated string or a sequence of strings, the string representation of each fallback value MUST be the concatenation of a U+007B LEFT CURLY BRACKET {, the fallback value as a string, and a U+007D RIGHT CURLY BRACKET }.

For example, a message that is not well-formed would format to a string as {�}, unless a fallback string is defined in the formatting context, in which case that string would be used instead.

Handling Bidirectional Text

Messages contain text. Any text can be bidirectional text. That is, the text can can consist of a mixture of left-to-right and right-to-left spans of text. The display of bidirectional text is defined by the Unicode Bidirectional Algorithm [UAX9].

The directionality of the formatted message as a whole is provided by the formatting context.

Note

Keep in mind the difference between the formatted output of a message, which is the topic of this section, and the syntax of message prior to formatting. The processing of a message depends on the logical sequence of Unicode code points, not on the presentation of the message. Affordances to allow users appropriate control over the appearance of the message's syntax have been provided.

When a message is formatted, placeholders are replaced with their formatted representation. Applying the Unicode Bidirectional Algorithm to the text of a formatted message (including its formatted parts) can result in unexpected or undesirable spillover effects. Applying bidi isolation to each affected formatted value helps avoid this spillover in a formatted message.

Note that both the message and, separately, each placeholder need to have direction metadata for this to work. If an implementation supports formatting to something other than a string (such as a sequence of parts), the directionality of each formatted placeholder needs to be available to the caller.

If a formatted expression itself contains spans with differing directionality, its formatter SHOULD perform any necessary processing, such as inserting controls or isolating such parts to ensure that the formatted value displays correctly in a plain text context.

For example, an implementation could provide a :currency formatting function which inserts strongly directional characters, such as U+200F RIGHT-TO-LEFT MARK (RLM), U+200E LEFT-TO-RIGHT MARK (LRM), or U+061C ARABIC LETTER MARKER (ALM), to coerce proper display of the sign and currency symbol next to a formatted number. An example of this is formatting the value -1234.56 as the currency AED in the ar-AE locale. The formatted value appears like this:

‎-1,234.56 د.إ.‏

The code point sequence for this string, as produced by the ICU4J NumberFormat function, includes U+200F U+200E at the start and U+200F at the end of the string. If it did not do this, the same string would appear like this instead:

image

A bidirectional isolation strategy is functionality in the formatter's processing of a message that produces bidirectional output text that is ready for display.

The Default Bidi Strategy is a bidirectional isolation strategy that uses isolating Unicode control characters around placeholder's formatted values. It is primarily intended for use in plain-text strings, where markup or other mechanisms are not available. Implementations MUST provide the Default Bidi Strategy as one of the bidirectional isolation strategies.

Implementations MAY provide other bidirectional isolation strategies.

Implementations MAY supply a bidirectional isolation strategy that performs no processing.

The Default Bidi Strategy is defined as follows:

  1. Let msgdir be the directionality of the whole message, one of « 'LTR', 'RTL', 'unknown' ». These correspond to the message having left-to-right directionality, right-to-left directionality, and to the message's directionality not being known.
  2. For each expression exp in pattern:
    1. Let fmt be the formatted string representation of the resolved value of exp.
    2. Let dir be the directionality of fmt, one of « 'LTR', 'RTL', 'unknown' », with the same meanings as for msgdir.
    3. If dir is 'LTR':
      1. If msgdir is 'LTR' in the formatted output, let fmt be itself
      2. Else, in the formatted output, prefix fmt with U+2066 LEFT-TO-RIGHT ISOLATE and postfix it with U+2069 POP DIRECTIONAL ISOLATE.
    4. Else, if dir is 'RTL':
      1. In the formatted output, prefix fmt with U+2067 RIGHT-TO-LEFT ISOLATE and postfix it with U+2069 POP DIRECTIONAL ISOLATE.
    5. Else:
      1. In the formatted output, prefix fmt with U+2068 FIRST STRONG ISOLATE and postfix it with U+2069 POP DIRECTIONAL ISOLATE.

Interchange Data Model

This section defines a data model representation of MessageFormat 2 messages.

Implementations are not required to use this data model for their internal representation of messages. Neither are they required to provide an interface that accepts or produces representations of this data model.

The major reason this specification provides a data model is to allow interchange of the logical representation of a message between different implementations. This includes mapping legacy formatting syntaxes (such as MessageFormat 1) to a MessageFormat 2 implementation. Another use would be in converting to or from translation formats without the need to continually parse and serialize all or part of a message.

Implementations that expose APIs supporting the production, consumption, or transformation of a message as a data structure are encouraged to use this data model.

This data model provides these capabilities:

This data model might also be used to:

To ensure compatibility across all platforms, this interchange data model is defined here using TypeScript notation. Two equivalent definitions of the data model are also provided:

Note that while the data model description below is the canonical one, the JSON and DTD definitions are intended for interchange between systems and processors. To that end, they relax some aspects of the data model, such as allowing declarations, options, and attributes to be optional rather than required properties.

Note

Users relying on XML representations of messages should note that XML 1.0 does not allow for the representation of all C0 control characters (U+0000-U+001F). Except for U+0000 NULL , these characters are allowed in MessageFormat 2 messages, so systems and users relying on this XML representation for interchange might need to supply an alternate escape mechanism to support messages that contain these characters.

Important

The data model uses the field name name to denote various interface identifiers. In the MessageFormat 2 syntax, the source for these name fields sometimes uses the production identifier. This happens when the named item, such as a function, supports namespacing.

In the Tech Preview, feedback on whether to separate the namespace from the name and represent both separately, or just, as here, use an opaque single field name is desired.

Messages

A SelectMessage corresponds to a syntax message that includes selectors. A message without selectors and with a single pattern is represented by a PatternMessage.

In the syntax, a PatternMessage may be represented either as a simple message or as a complex message, depending on whether it has declarations and if its pattern is allowed in a simple message.

type Message = PatternMessage | SelectMessage;

interface PatternMessage {
  type: "message";
  declarations: Declaration[];
  pattern: Pattern;
}

interface SelectMessage {
  type: "select";
  declarations: Declaration[];
  selectors: VariableRef[];
  variants: Variant[];
}

Each message declaration is represented by a Declaration, which connects the name of a variable with its expression value. The name does not include the initial $ of the variable.

The name of an InputDeclaration MUST be the same as the name in the VariableRef of its VariableExpression value.

type Declaration = InputDeclaration | LocalDeclaration;

interface InputDeclaration {
  type: "input";
  name: string;
  value: VariableExpression;
}

interface LocalDeclaration {
  type: "local";
  name: string;
  value: Expression;
}

In a SelectMessage, the keys and value of each variant are represented as an array of Variant. For the CatchallKey, a string value may be provided to retain an identifier. This is always '*' in MessageFormat 2 syntax, but may vary in other formats.

interface Variant {
  keys: Array<Literal | CatchallKey>;
  value: Pattern;
}

interface CatchallKey {
  type: "*";
  value?: string;
}

Patterns

Each Pattern contains a linear sequence of text and placeholders corresponding to potential output of a message.

Each element of the Pattern MUST either be a non-empty string, an Expression, or a Markup object. String values represent literal text. String values include all processing of the underlying text values, including escape sequence processing. Expression wraps each of the potential expression shapes. Markup wraps each of the potential markup shapes.

Implementations MUST NOT rely on the set of Expression and Markup interfaces defined in this document being exhaustive. Future versions of this specification might define additional expressions or markup.

type Pattern = Array<string | Expression | Markup>;

type Expression =
  | LiteralExpression
  | VariableExpression
  | FunctionExpression;

interface LiteralExpression {
  type: "expression";
  arg: Literal;
  function?: FunctionRef;
  attributes: Attributes;
}

interface VariableExpression {
  type: "expression";
  arg: VariableRef;
  function?: FunctionRef;
  attributes: Attributes;
}

interface FunctionExpression {
  type: "expression";
  arg?: never;
  function: FunctionRef;
  attributes: Attributes;
}

Expressions

The Literal and VariableRef correspond to the the literal and variable syntax rules. When they are used as the body of an Expression, they represent expression values with no function.

Literal represents all literal values, both quoted literal and unquoted literal. The presence or absence of quotes is not preserved by the data model. The value of Literal is the "cooked" value (i.e. escape sequences are processed).

In a VariableRef, the name does not include the initial $ of the variable.

interface Literal {
  type: "literal";
  value: string;
}

interface VariableRef {
  type: "variable";
  name: string;
}

A FunctionRef represents a function. The name does not include the : starting sigil.

Options is a key-value mapping containing options, and is used to represent the function and markup options.

interface FunctionRef {
  type: "function";
  name: string;
  options: Options;
}

type Options = Map<string, Literal | VariableRef>;

Markup

A Markup object has a kind of either "open", "standalone", or "close", each corresponding to open, standalone, and close markup. The name in these does not include the starting sigils # and / or the ending sigil /. The options for markup use the same key-value mapping as FunctionRef.

interface Markup {
  type: "markup";
  kind: "open" | "standalone" | "close";
  name: string;
  options: Options;
  attributes: Attributes;
}

Attributes

Attributes is a key-value mapping used to represent the expression and markup attributes.

Attributes with no value are represented by true here.

type Attributes = Map<string, Literal | true>;

Extensions

Implementations MAY extend this data model with additional interfaces, as well as adding new fields to existing interfaces. When encountering an unfamiliar field, an implementation MUST ignore it. For example, an implementation could include a span field on all interfaces encoding the corresponding start and end positions in its source syntax.

In general, implementations MUST NOT extend the sets of values for any defined field or type when representing a valid message. However, when using this data model to represent an invalid message, an implementation MAY do so. This is intended to allow for the representation of "junk" or invalid content within messages.

Appendices

Security Considerations

MessageFormat 2.0 patterns are meant to allow a message to include any string value which users might normally wish to use in their environment. Programming languages and other environments vary in what characters are permitted to appear in a valid string. In many cases, certain types of characters, such as invisible control characters, require escaping by these host formats. In other cases, strings are not permitted to contain certain characters at all. Since messages are subject to the restrictions and limitations of their host environments, their serializations and resource formats, that might be sufficient to prevent most problems. However, MessageFormat itself does not supply such a restriction.

MessageFormat messages permit nearly all Unicode code points, with the exception of surrogates, to appear in literals, including the text portions of a pattern. This means that it can be possible for a message to contain invisible characters (such as bidirectional controls, ASCII control characters in the range U+0000 to U+001F, or characters that might be interpreted as escapes or syntax in the host format) that abnormally affect the display of the message when viewed as source code, or in resource formats or translation tools, but do not generate errors from MessageFormat parsers or processing APIs.

Bidirectional text containing right-to-left characters (such as used for Arabic or Hebrew) also poses a potential source of confusion for users. Since MessageFormat 2.0's syntax makes use of keywords and symbols that are left-to-right or consist of neutral characters (including characters subject to mirroring under the Unicode Bidirectional Algorithm), it is possible to create messages that, when displayed in source code, or in resource formats or translation tools, have a misleading appearance or are difficult to parse visually.

For more information, see [UTS#55] Unicode Source Code Handling.

MessageFormat 2.0 implementations might allow end-users to install selectors, functions, or markup from third-party sources. Such functionality can be a vector for various exploits, including buffer overflow, code injection, user tracking, fingerprinting, and other types of bad behavior. Any installed code needs to be appropriately sandboxed. In addition, end-users need to be aware of the risks involved.

Acknowledgements

Special thanks to the following people for their contributions to making MessageFormat v2. The following people contributed to our github repo and are listed in order by contribution size:

Addison Phillips, Eemeli Aro, Romulo Cintra, Stanisław Małolepszy, Elango Cheran, Richard Gibson, Tim Chevalier, Mihai Niță, Shane F. Carr, Mark Davis, Steven R. Loomis, Caleb Maclennan, David Filip, Daniel Minor, Christopher Dieringer, George Rhoten, Ujjwal Sharma, Daniel Ehrenberg, Markus Scherer, Zibi Braniecki, Matt Radbourne, Bruno Haible, and Rafael Xavier de Souza.

Addison Phillips was chair of the working group from January 2023. Prior to 2023, the group was governed by a chair group, consisting of Romulo Cintra, Elango Cheran, Mihai Niță, David Filip, Nicolas Bouvrette, Stanisław Małolepszy, Rafael Xavier de Souza, Addison Phillips, and Daniel Minor. Romulo Cintra chaired the chair group.


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.