test: fail fast when the DOM stub cannot find the confirm button

The stub auto-clicks the confirm button so customConfirm() resolves
headlessly, but had no branch for "did not match". A reworded button would
leave the client waiting forever and node would run to the caller's 120s
subprocess timeout, surfacing as TimeoutExpired with a traceback - a
multi-minute mystery instead of a diagnosis.

Report and exit non-zero instead, naming the buttons that were seen and
where to update the pattern. The branch is guarded on there being buttons at
all: the ordinary error-path dialog has none, and firing on that would break
every capture.

Also correct what the predicate's comments claim. The two false-lead patterns
name the regression that already happened; they are not what makes the guard
general. Any substitution displaces the server's own text, so the server-body
clause is what catches a reworded revival - the patterns alone would not.

Author: ltdrdata

tests/js/dom.mjs

@@ -39,10 +39,23 @@ const realAppend = body.appendChild.bind(body);
 body.appendChild = (child) => {
 	realAppend(child);
 	// The confirm modal resolves on a click; fire it so the flow continues.
-	const confirmBtn = walk(child).find(
-		(e) => e.tagName === 'BUTTON' && /confirm|yes|ok/i.test(e.textContent || '')
-	);
-	if (confirmBtn) setTimeout(() => confirmBtn.click(), 0);
+	const buttons = walk(child).filter((e) => e.tagName === 'BUTTON');
+	const confirmBtn = buttons.find((e) => /confirm|yes|ok/i.test(e.textContent || ''));
+	if (confirmBtn) {
+		setTimeout(() => confirmBtn.click(), 0);
+	} else if (buttons.length) {
+		// A dialog with buttons, none of which this stub recognizes: the client
+		// would wait forever on customConfirm() and node would hang until the
+		// caller's subprocess timeout, surfacing as a multi-minute mystery
+		// rather than a diagnosis. Name what was seen and stop now.
+		console.error(
+			'capture_dialog: no confirm button matched /confirm|yes|ok/i, so the '
+			+ 'client would block on customConfirm(). Buttons seen: '
+			+ JSON.stringify(buttons.map((b) => b.textContent))
+			+ '. Update the pattern in tests/js/dom.mjs if the button was reworded.'
+		);
+		process.exit(3);
+	}
 	return child;
 };
 

tests/test_batch_404_dialog_guard.py

@@ -59,9 +59,10 @@
 #: (`web.Response(status=404, text=...)` in glob/manager_server.py).
 SERVER_404_BODY = "A security error has occurred. Please check the terminal logs"
 
-#: Fragments of the false message the fix removed. Kept as separate patterns
-#: because a reinstatement is more likely to be a reworded variant than a
-#: byte-identical copy.
+#: Fragments of the specific false message the fix removed. These name the known
+#: regression; they are NOT what makes the guard general — a reworded
+#: substitute would evade them, and is caught by the server-body clause instead
+#: (see _c1_violations).
 _FALSE_LEAD_PATTERNS = (
     re.compile(r"default\s+channel", re.I),
     re.compile(r"security\s+level\s+configuration", re.I),
@@ -137,6 +138,11 @@ def _c1_violations(payload):
 
     Both arms call THIS function, so the RED arm demonstrates the failure of
     the same check the GREEN arm passes — not of a differently-worded cousin.
+
+    The load-bearing clause is the FIRST one: any substitution, however it is
+    worded, displaces the server's own text and is caught. The phrase patterns
+    below it only name the specific regression that already happened; a
+    reworded revival would slip past them alone.
     """
     dialog = payload["dialog"]
     problems = []