snapshot of https://github.com/ueberdosis/hocuspocus.git @ cf3cfa355b4d3bb61c6b6c80071f71783e32470e (2026-09-15)

tests/server/debounce.ts

@@ -0,0 +1,31 @@
+import { useDebounce } from "@hocuspocus/server";
+import test from "ava";
+import { sleep } from "../utils/index.ts";
+
+test("maxDebounce does not queue an execution per call while a slow execution is in flight", async (t) => {
+	const { debounce } = useDebounce();
+
+	let executions = 0;
+	const func = async () => {
+		executions += 1;
+		// the first execution is slower than maxDebounce, simulating a
+		// database/S3 latency spike during a store
+		await sleep(executions === 1 ? 500 : 10);
+	};
+
+	// Continuous calls, like a busy collaborative document: each call resets
+	// the debounce timeout, so only the maxDebounce branch triggers executions
+	// (at most one per 200ms).
+	const end = Date.now() + 1_000;
+	while (Date.now() < end) {
+		debounce("doc", func, 50, 200);
+		await sleep(20);
+	}
+	await sleep(600); // let in-flight and queued executions finish
+
+	// ~1s of continuous calls with maxDebounce=200 should execute ~5 times.
+	// If run() suspends on the in-flight execution before deleting the timer
+	// entry, the stale entry keeps satisfying the maxDebounce age check and
+	// every call during the slow execution queues another run (~25 executions).
+	t.true(executions <= 8, `executed ${executions} times, expected ~5`);
+});