Language extensions

Status: accepted design (decisions 182 to 184, 2026-10-07), not released. Layers 1 and 2 are planned; layer 3 is an idea kept reachable, not scheduled.

MX is extended at three layers. Each one uses the layer below it and adds nothing to the parser beyond what the layer below already allows.

Layer Who What changes Parser involvement Status
1: tags and contracts any project custom tags, sidecars, contracts, mx.tags, mx.contracts none planned, mostly shipped
2: syntax a project, or a package it depends on delimiters, triggers, block forms, filters; lowering hooks for the forms it adds a data-only table read once per parse planned
3: languages a language author packaging: own extension, name, manifest key, editor language, pinned core none of its own idea

The core page states what stays fixed: Marko’s grammar on the default row, the intermediate representation, the event boundary of the parser, and the one runtime shape every host shares for attribute tags.

flowchart LR src[".mx source"] --> parser["parser<br/>SyntaxTable = default row ⊕ package.json#mx.syntax"] parser --> events["events<br/>text · tags · placeholders · triggers · blocks · filters"] events --> lower["core.lower<br/>tags · contracts · hooks"] lower --> ir["IR"] ir --> emit["target emitter"] l1["layer 1<br/>tags, sidecars, contracts"] -.-> lower l2["layer 2<br/>syntax table, lowerTrigger, lowerBlockTag, lowerFilter, afterLower"] -.-> parser l2 -.-> lower l3["layer 3<br/>extension, name, manifest key, editor"] -.-> parser l3 -.-> emit

The two rules every layer obeys

Grammar is parametrized as data, never as code. Everything a layer may change about parsing is a value in the syntax table, read once at the start of a parse. The table holds no functions. Per character the parser does what it does today, one lookup on a character code; a matcher runs only after its trigger character hit. A native lexer can take the same table and emit the same events.

Hooks run after parsing. A layer contributes functions only at lowering (lowerTrigger, lowerBlockTag, lowerFilter, afterLower) and, for layer 3, at emit. No parser state waits on a callback.

What leaves the core

Everything an extension of any layer can express leaves the core, so that the .mx default row is Marko’s grammar and nothing else. Atoms and every name sugar (:name, spaced #id and .class) become layer-2 entries of the Mesh syntax set. Attribute-tag cardinality from a callee’s Input type is dropped for Marko’s runtime record with a richer prototype. The core page lists each item with the Marko behavior it was checked against and the upstream proposal it allows.

Not a layer: the expression language

Replacing TypeScript inside expressions is a separate dimension a language may or may not set. It cuts through the expression scanner, the parser behind it, the binding rewrites, the type projection and every emitter. The table reserves expressionLanguage: "ts" so a second value is additive; nothing else is designed, because nothing uses it.

Previous Custom tags build spec Next Language extensions: the core