IME support?

[diiyw]

Is the project support IME ?

[kolkov]

@diiyw — thanks for asking. IME support is planned but not yet implemented at the platform level.

What's ready

The event interfaces for IME composition are defined in gpucontext and wired through gogpu:

// gpucontext/events.go
OnIMECompositionStart(fn func())
OnIMECompositionUpdate(fn func(IMEState))  // preview text, cursor, selection
OnIMECompositionEnd(fn func(committed string))

type IMEState struct {
    Composing       bool
    CompositionText string   // e.g. pinyin "nihao"
    CursorPos       int
    SelectionStart  int
    SelectionEnd    int
}

Basic text input works on Windows — WM_CHAR from TranslateMessage() delivers final Unicode characters (including CJK committed via system IME, with surrogate pair handling for supplementary characters). On macOS, [NSEvent characters] delivers committed text. Linux and Browser deliver EventChar through xkb/DOM respectively. So committed CJK text reaches the application, but without inline composition preview (the underlined preedit text with candidate selection).

What's missing (platform-level)

Platform	What's needed	API
Windows	Handle WM_IME_STARTCOMPOSITION, WM_IME_COMPOSITION, WM_IME_ENDCOMPOSITION. Call ImmSetCompositionWindow to position the candidate window near the caret.	Win32 IMM32
Linux/X11	XIM or IBus/Fcitx integration via D-Bus	XIM / IBus
Linux/Wayland	zwp_text_input_v3 protocol	Wayland text-input
macOS	Implement setMarkedText:selectedRange:replacementRange: + insertText:replacementRange: in NSTextInputClient	Cocoa NSTextInputClient
Browser/WASM	compositionstart/compositionupdate/compositionend events	DOM CompositionEvent

What's missing (UI level)

The TextField widget needs composition rendering — underlined preview text with cursor, candidate window positioning. Currently it only handles committed text.

Timeline

This is on our roadmap for the internationalization track. Contributions welcome — the interface layer is ready, the platform wiring is the main work. Windows IME (WM_IME_* + IMM32) would be a good starting point since it's the most documented.

If you have a specific use case or platform priority, let us know — it helps us sequence the work.