Fix afterAll skipped in --ci

[DmitrySharabin]

Closes #168.

In --ci (non-interactive) mode, process.exit() fires synchronously the moment stats.pending === 0, before the trailing .finally(() => afterAll()) microtask in TestResult#runAll can resolve. As a result, afterAll hooks never run.

Fix

In src/env/node.js: process.exit(N) → process.exitCode = N. Setting the exit code lets the event loop drain — Node runs the queued .finally(afterAll) microtask, then exits with the set code because nothing else keeps it alive.

No changes to src/classes/TestResult.js. result.finished and afterAll keep their existing meanings (per the discussion above).

Drive-by

tests/run.js had a pre-existing leak: setInterval(() => resolve("foo"), 100) was used where setTimeout was intended, leaving a recurring 100 ms timer that's never cleared. process.exit() masked it; process.exitCode exposes it (the runner would hang). One-line fix: setInterval → setTimeout.

Out of scope

Bottom-up afterAll ordering for nested groups (which an earlier draft of this PR also addressed via TestResult.js changes) — tracked separately as #178.

Test plan

• All 119 existing tests pass
• Repro from afterAll never runs in non-interactive mode (process.exit races the .finally chain) #168 (tmp-dir cleanup in --ci) works
• New regression test in tests/run.js (spawns htest --ci on a fixture with async afterAll, asserts cleanup output reached stderr)

[netlify]

✅ Deploy Preview for h-test ready!

Name	Link
🔨 Latest commit	20bc39b
🔍 Latest deploy log	https://app.netlify.com/projects/h-test/deploys/6a3eb0a69150290008066cfb
😎 Deploy Preview	https://deploy-preview-170--h-test.netlify.app
📱 Preview on mobile	Toggle QR Code... Use your smartphone camera to open QR code link.
🤖 Make changes	Run an agent on this branch

---

To edit notification comments on pull requests, go to your Netlify project configuration.

[DmitrySharabin]

Reviewer guide — the diff is noisier than the change

The semantic change is 3 hunks. The rest is re-indentation and code movement. Skim those 3 hunks; everything else is byte-identical to main, just shifted.

---

1. src/classes/TestResult.js — finish listener now awaits children + runs own afterAll

Was a 1-line resolve; now wraps the lifecycle.

main L190–191 → PR L190–204:

 		this.finished = new Promise(resolve =>
-			this.addEventListener("finish", resolve, { once: true }));
+			this.addEventListener(
+				"finish",
+				async () => {
+					await Promise.allSettled((this.tests ?? []).map(t => t.finished));
+					if (!this.parent?.error) {
+						try {
+							await this.test.afterAll?.();
+						}
+						catch {}
+					}
+					resolve();
+				},
+				{ once: true },
+			));

2. src/classes/TestResult.js — .then(() => this.finished).finally(...afterAll) deleted

Its body moved into hunk (1). Deleted from main L254–262:

 			return Promise.allSettled((this.tests ?? []).map(test => test.runAll()));
-		})
-		.then(() => this.finished)
-		.finally(async () => {
-			if (!this.parent?.error) {
-				try {
-					await this.test.afterAll?.();
-				}
-				catch {}
-			}
-		});
+		});

The rest of the delay(1).then(async () => { ... }) body — main L222–252 vs PR L235–264 — is byte-identical, just unindented by one tab because the chained .then().finally() is gone. Nothing to review there.

---

3. src/env/node.js — CI exit moves from done → start, wrapped in root.finished.then(...)

This is the actual bug fix: exit can no longer fire before afterAll resolves.

start handler — main L371–380 → PR L371–393:

 	start (target, options, event, root) {
 		// `start` bubbles — skip descendants so we only fire once, on the root's own start.
-		if (options.signal?.aborted || !isInteractive || target !== root) {
+		if (options.signal?.aborted || target !== root) {
 			return;
 		}

+		if (!isInteractive) {
+			root.finished.then(() => {
+				let messages = root.toString(options);
+				let tree = getTree(messages).toString();
+				tree = process.stdout.isTTY ? format(tree) : stripFormatting(tree);
+
+				console[root.stats.fail > 0 ? "error" : "log"](tree);
+
+				process.exit(root.stats.fail > 0 ? 1 : 0);
+			});
+			return;
+		}
+
 		currentRoot = root;
 		// Open the interactive tree now so progress shows immediately — otherwise nothing renders until the first `done` event.
 		interactiveTree(root, options, { rerun: () => rerun(options) });
 	},

done handler — main L381–397 → PR L394–401: the print/exit block deleted (lifted into start above), and the !isInteractive short-circuit folded into the top guard.

 	done (result, options, event, root) {
-		if (options.signal?.aborted) {
+		if (options.signal?.aborted || !isInteractive) {
 			return;
 		}

-		if (!isInteractive) {
-			if (root.stats.pending === 0) {
-				let messages = root.toString(options);
-				let tree = getTree(messages).toString();
-				tree = process.stdout.isTTY ? format(tree) : stripFormatting(tree);
-
-				console[root.stats.fail > 0 ? "error" : "log"](tree);
-
-				process.exit(root.stats.fail > 0 ? 1 : 0);
-			}
-			return;
-		}
-
 		currentRoot = root;
 		interactiveTree(root, options, { rerun: () => rerun(options) });
 	},

---

TL;DR: review the 3 hunks above. Everything else GitHub flags is either unindentation (TestResult.js) or unchanged identifiers around the moved block (node.js).

[LeaVerou]

I'm not sure about this behavior. I think I'd expect afterAll() to run after the test is done. This seems to be fixing a bug by changing afterAll() to mean something else, which doesn't seem necessary?

[DmitrySharabin]

I'm not sure about this behavior. I think I'd expect afterAll() to run after the test is done. This seems to be fixing a bug by changing afterAll() to mean something else, which doesn't seem necessary?

Thinking about it more, I tend to agree with you. The test IS actually done before we run afterAll. I wonder if the test is not started while we do beforeAll. Hm. Do we consider beforeEach/afterEach parts of the test? It's like, should we dispatch start/done before/after them, respectively? I don't remember what we do now. But we probably need to establish the correct behavior. I'll research what others do.

[DmitrySharabin]

Drastically simplified to address the particular bug.