@telorun/message-format

Telo's message core: a closed profile of Unicode MessageFormat 2, read into a plain-JSON model and formatted with pinned CLDR 48.2 data. It has no dependencies and calls no Intl, so a message formats identically in Node, in Rust (telorun-message-format) and in every browser.

import { formatMessage, lookupLanguage, readMessage } from "@telorun/message-format";

const read = readMessage({
  plural: "count",
  cases: { "=0": "No files", one: "{$count} file", other: "{$count} files" },
});
if (!read.ok) throw new Error(read.problems[0].message);

formatMessage(read.message, "en", { count: 0n }); // "No files"
formatMessage(read.message, "en", { count: 1234n }); // "1,234 files"
lookupLanguage("pl-PL", ["en", "pl"]); // "pl"

The profile

• A pattern is text with {$name} placeholders and the escapes \{, \} and \\. A name follows the Telo name rule: a letter or _ first, then letters, digits or _, and not a CEL reserved word. Inside the braces, the whitespace and bidi marks MF2 allows are accepted. Anything else MF2 can write in braces is refused: a literal, a function, an option, an attribute or markup.
• A plural selection is written { plural: <name>, cases: { <category> | =<integer>: Message } }. other is required. A case is itself a Message, so selections nest. =N keys are canonical integers inside int64 ∪ uint64.
• A text is a Message, or { text: Message, context } with a non-empty context.

What it answers

| Function | Answer | | --- | --- | | parsePattern, serializePattern | A pattern's parts, and its one canonical spelling. | | readMessage, readText | The model, or every fault as MESSAGE_SYNTAX_INVALID with its path and, inside a pattern, its UTF-16 range. | | messageIdentity, textIdentity | A source text's key: the pattern reached by taking other at every selection, canonically written, plus its context. | | messageVariables | Every placeholder and selector a message reads. | | unknownVariables | MESSAGE_PLACEHOLDER_UNKNOWN for each name a translation reads that its source does not. | | pluralCategories, invalidPluralCategories | A language's CLDR categories, and MESSAGE_PLURAL_CATEGORY_INVALID for each case naming one it lacks. | | formatNumber, pluralCategory, pluralCategoryOfDecimal | A number's text and category in a language. | | formatMessage | The text. Refuses with MESSAGE_PLACEHOLDER_UNBOUND when a name the message reads has no value, on any branch. Refuses with MESSAGE_PLACEHOLDER_VALUE_UNSUPPORTED for a value other than a string or a number, and for a selector that is not a number. | | truncationChain, lookupLanguage | RFC 4647 lookup: pl-Latn-PL → pl-Latn → pl, case-insensitive. null when nothing matches, and the caller then uses its source. |

Numbers

A number's text depends on its value alone, never on its host type:

• A bigint inside int64 ∪ uint64 is written exactly, however large.
• A number is a double: its shortest round-trip decimal, rounded half-to-even to the pattern's maximum fraction digits, with trailing zeros dropped and a zero result unsigned. So 1.0 is "1", and in en it selects one.
• A non-finite double is refused.

The text then takes the language's CLDR standard decimal pattern, grouping (with minimumGroupingDigits), default-numbering digits, decimal separator and minus sign. Plural selection reads the operands of that rounded decimal. =N cases match numerically before the categories.

CLDR data

The data is CLDR 48.2 for every locale, generated into message-format/cldr/ from the exact-pinned cldr-core and cldr-numbers-full. A tag resolves through CLDR's own inheritance and ends at the root (other only, Latin digits, #,##0.###):

• Number data goes through parentLocales, then the nonlikelyScript rule, then truncation.
• Plural rules go through CLDR's plurals parent map.

The data is used under the Unicode License V3. The package ships the licence as LICENSE-CLDR, and every build of the core carries it as a /*! */ notice that survives bundling.