@geometra/core

DOM-free UI framework core. Build declarative interfaces rendered to Canvas, terminal, or server-streamed layouts.

Install

npm install @geometra/core

Key Exports

• signal, computed, effect, batch -- reactive primitives
• box, text, image -- element constructors
• createApp -- application entry point
• toSemanticHTML -- SEO-friendly HTML generation
• toAccessibilityTree -- runtime accessibility tree from geometry
• insertInputText, backspaceInput, deleteInput -- text-input editing primitives
• getInputCaretGeometry -- caret x/y/height from measured text lines
• createTextInputHistory, undoTextInputHistory -- undo/redo state helpers
• transition, spring, createTweenTimeline, createPropertyTimeline, setMotionPreference -- animation utilities

Usage

import { box, text, createApp, signal } from '@geometra/core'

const count = signal(0)

const app = createApp(() =>
  box({ width: 300, height: 200, padding: 20, gap: 10 }, [
    text({ text: `Count: ${count.value}`, font: 'bold 24px sans-serif', lineHeight: 30 }),
    box({ width: 100, height: 40, backgroundColor: '#07f', onClick: () => count.set(count.peek() + 1) }, [
      text({ text: 'Click me', font: '16px sans-serif', lineHeight: 20, color: '#fff' }),
    ]),
  ])
)

Text input contract

• TextInputState is immutable editor state: { nodes: string[], selection }.
• selection is node-local and may be reversed; helpers normalize internally.
• insertInputText/replaceInputSelection always collapse caret after replacement.
• backspaceInput/deleteInput delete active selections first, then apply boundary merge behavior across nodes.
• moveInputCaret supports range extension via extendSelection=true.
• moveInputCaret and moveInputCaretByWord accept an optional reading direction (ltr default, rtl supported) for horizontal key semantics.
• moveInputCaretToLineBoundary also accepts optional reading direction so Home/End semantics can follow visual direction in RTL contexts.
• getInputCaretGeometry is defined for collapsed selections with measured line metrics; expanded ranges return null.

Direction model baseline

• Elements may provide dir: 'ltr' | 'rtl' | 'auto' on box(), text(), and image() props.
• dir='auto' currently inherits from parent direction (script-level auto detection is deferred).
• Omitted dir inherits from parent; root fallback is ltr.
• Direction metadata is kept in the declarative tree and intentionally stripped from Yoga layout props.

Accessibility guarantees and limits

• toAccessibilityTree(tree, layout) emits deterministic role/name/bounds/focusable nodes for geometry-rendered UIs.
• Semantic overrides (semantic.tag, semantic.role, semantic.ariaLabel, semantic.alt) are preserved when projecting accessibility and semantic HTML output.
• The core package does not directly expose native accessibility APIs; renderer integrations (for example canvas mirror strategies) are responsible for host-level assistive technology bridging.

Links

• Main repo