[perf] Terminal stream events lack client-side batching before xterm.write() #2241

Problem

The backend PTY output is well-batched at ~30Hz (pty-subprocess.ts, 32ms flush interval with 128KB cap). However, on the renderer side, each stream event from the tRPC subscription calls xterm.write() individually without further batching:

// Terminal.tsx line 262-268
electronTrpc.terminal.stream.useSubscription(paneId, {
  onData: (event) => {
    handleStreamData(event);  // writes to xterm per-message
  },
});

When agents produce rapid output (Codex streaming, Claude Code tool results, npm install logs), the renderer processes each IPC message as a separate xterm.write() call. Each call can trigger:

• xterm parser processing
• WebGL texture atlas updates
• Potential layout recalculations

Native terminals batch rendering at the display refresh rate. xterm.js does have internal write buffering, but coalescing writes before they enter xterm's pipeline would reduce overhead further.

Suggested Fix

Add a client-side write coalescing buffer using requestAnimationFrame:

const pendingWrites = useRef<string[]>([]);
const rafScheduled = useRef(false);

function scheduleWrite(data: string) {
  pendingWrites.current.push(data);
  if (!rafScheduled.current) {
    rafScheduled.current = true;
    requestAnimationFrame(() => {
      const batch = pendingWrites.current.join("");
      pendingWrites.current = [];
      rafScheduled.current = false;
      xterm.write(batch);
    });
  }
}

This coalesces all writes arriving within a single frame into one xterm.write() call, aligned with the display refresh rate.

Environment

• Superset 1.1.2
• macOS