Accessibility & IME

Glyx draws its entire UI to a GPU canvas, so unlike a DOM-based framework there's no free accessibility tree for the OS to enumerate. Glyx builds one explicitly via AccessKit (opens in a new tab) and wires IME composition into TextInput, so screen readers and CJK input both work.

Both are live-tested — accessibility against Windows Narrator, confirmed working across the full role set below.

Accessibility

Every committed layout pass walks the node tree into an accesskit::TreeUpdate, pushed to the OS accessibility API (Windows UIA, macOS NSAccessibility, AT-SPI on Linux) via accesskit_winit. This is compiled in behind the a11y Cargo feature — check __glyx_hasA11y() from JS if your app needs to detect whether a build has it.

Props

Prop	Type	Description
`role`	`string`	Explicit AccessKit role — see the table below. Overrides the inferred default.
`ariaLabel`	`string`	Accessible name announced by the screen reader.

<Pressable role="button" ariaLabel="Close dialog" onPress={onClose}>
  <Icon name="x" />
</Pressable>

Role mapping

`role` value	AccessKit role
`'button'`	`Button`
`'textbox'`	`TextInput`
`'checkbox'`	`CheckBox`
`'radio'`	`RadioButton`
`'switch'`	`Switch`
`'link'`	`Link`
`'image'`	`Image`
`'heading'`	`Heading`
`'list'`	`List`
`'listitem'`	`ListItem`
`'combobox'`	`ComboBox`
`'slider'`	`Slider`
`'none'` / `'presentation'`	`GenericContainer` (filtered from the platform tree, same as ARIA `none`)

Without an explicit role, Glyx infers one: <Text> → Label, <Image> → Image, any node with a visible caret → TextInput, any pressable node → Button, everything else → GenericContainer.

Actions

Wired: Focus, Click, Increment/Decrement (e.g. Slider), and SetValue for numeric values.

⚠️

Known gaps: Expand/Collapse, ScrollIntoView, and text-selection actions aren't wired yet. There's also no keyboard Tab-key focus cycling implementation — focus can be set programmatically and screen readers can navigate their own virtual cursor, but there's no built-in Tab/Shift+Tab traversal across focusable nodes yet.

IME / CJK text input

TextInput handles IME composition end-to-end — not just receiving the OS events, but actually rendering the preedit string inline and committing it into the field's value.

Composition (preedit) text renders inline in the field as you type, before you've committed a character.
On commit, the composed text is inserted into the real value at the caret — a proper two-phase insertion, not a naive append.
The OS candidate window is positioned automatically against the caret.

This covers Chinese, Japanese, Korean, and any other script that needs a composition window — no extra setup required beyond using <TextInput>.

What's genuinely still missing

Tab-key focus cycling (keyboard-only navigation between focusable nodes).
Expand/Collapse, ScrollIntoView, and text-selection AccessKit actions.
An ariaLabel-adjacent "hint" prop (e.g. a longer description alongside the accessible name) — not implemented.
Right-to-left TextInput cursor/selection directionality, and bidirectional text mixing within a single field.

See the remaining accessibility gaps roadmap page for more detail on the first two.

Animations Packaging