Generators

A generator is a back-end: it reads the Milda document and emits something concrete - real code for a framework, or a design file. Generators are where platform-specific decisions finally get made, and where a target says what it could not express.

#The React generator

The React generator is the one that ships first and defines the bar. It emits self-contained component source plus a shared theme - output you could read, review, and ship by hand, except you don't have to. It realizes:

• Structure - neutral node kinds become the right elements.
• Behavior - state machines become idiomatic React state and interaction.
• Style - facets and foundations become a theme expressed as CSS custom properties.

The goal is a drop-in replacement for your existing components - matching API and behavior closely enough to swap in without rewrites.

#The Figma generator

Figma is a design target rather than a code target. It exports the foundations as Figma Variables with a mode per context and a W3C DTCG token file, and the components as component sets with variants and bound variables. The Figma plugin is a persistent bridge rather than a one-way dump: Milda stays the source, and the plugin re-syncs. Its output unit is the library, not the component, and it returns a list of what it skipped.

#One model, many back-ends

Every generator implements the same small contract: turn a document into files, and say what you could not express. A component target adds per-component emission, a theme and an icon module; a library target, like Figma, emits one artifact for the whole system. New targets slot in as additional back-ends without changing the document or the editor.

#What a target could not express

Each target reports its limits rather than approximating: the Figma target's skipped list, the React target's report of a constraint no type can say. foreign content is the author's side of the same honesty.

#Realization differs per target

The same behavior can execute differently on different platforms, and the same facet token can resolve to a different value per platform. That divergence lives entirely in each generator's realization - never in the shared model. See The document.

Which generators a given project runs is controlled by its targets.