spa: a blank page can never be silent again

boot-guard.js is a classic script loaded before the module graph, so it survives the graph failing to link. If #app is still empty after ten seconds it names what failed and offers a reset of this origin — cache, storage, databases and the service worker, which clearing the cache does not touch. Two real defects found building it: openDB never settled when an upgrade was blocked by another tab, and a connection it gave up on stayed open and squatted the database. Co-Authored-By: Claude Opus 5 <noreply@anthropic.com>
author: Christophe Besson <cbesson@gmail.com> 2026-09-17 03:40:52 +0200
committer: Christophe Besson <cbesson@gmail.com> 2026-09-17 03:40:52 +0200
commit: f4fd6db8faa15bf02f38a14b652cc46936f8d6bf (patch)
tree: 396cbe7fe188fb565231be937ccfa6419384814f /packages/meshbay-hub/tests
parent: 929bfcccf8928a30d86d48a8b0137db0dd4a1ee0 (diff)
download: meshbay-f4fd6db8faa15bf02f38a14b652cc46936f8d6bf.tar.gz
2 files changed, 382 insertions, 0 deletions
diff --git a/packages/meshbay-hub/tests/harness/boot_guard_probe.py b/packages/meshbay-hub/tests/harness/boot_guard_probe.py
new file mode 100644
index 0000000..01751ba
--- /dev/null
+++ b/packages/meshbay-hub/tests/harness/boot_guard_probe.py
@@ -0,0 +1,260 @@
+#!/usr/bin/env python3
+"""
+The shell cannot end up silently blank, and there is a way out of it.
+
+A reader spent an evening on a white page: the shell arrived, every module came
+from the browser's own store so not one request reached the hub, and nothing
+rendered. Clearing the site's data fixed it — clearing the *cache*, three times,
+did not, because the cache is not where a service worker or IndexedDB lives.
+
+What is measured here is the guard, not the cause. Three things have to hold:
+it says nothing when the application mounts; it appears, with the failure named
+on screen, when the module graph does not link; and its reset button genuinely
+empties this origin — a button that claims to and does not would be worse than
+none.
+
+The last case is why the guard is a classic script and not part of the module
+graph: it has to survive the graph failing to link, which is exactly when a
+reader needs it.
+
+    boot_guard_probe.py
+
+Prints JSON: one entry per case.
+"""
+
+import http.server
+import json
+import socketserver
+import subprocess
+import sys
+import tempfile
+import threading
+import time
+from pathlib import Path
+
+STATIC = Path(__file__).resolve().parents[2] / "src" / "meshbay_hub" / "static"
+PORT = 8763
+RECORDS = []
+socketserver.TCPServer.allow_reuse_address = True
+
+# The shell, as `webapp.py` builds it: the guard first, as a classic script,
+# then the module graph. `?fail=1` asks for a module that is not there.
+SHELL = r"""<!doctype html><html><head><meta charset=utf-8></head><body>
+<div id="app"></div>
+<script src="/boot-guard.js"></script>
+<script type="module" src="/__case.js"></script>
+</body></html>"""
+
+MOUNTS = r"""// A successful mount: the guard must stay out of the way.
+document.getElementById('app').appendChild(
+  Object.assign(document.createElement('p'), { textContent: 'the application' }));
+"""
+
+BLOCKED = r"""<!doctype html><html><head><meta charset=utf-8></head><body>
+<script type="module">
+// Version 1, held open, as a second tab of this site running yesterday's build
+// holds it. The shipped `openDB` wants version 2: the upgrade cannot proceed
+// while this connection lives, and `indexedDB.open` then fires neither
+// `success` nor `error`.
+import { openDB } from '/hub-client.js';
+const held = await new Promise((res, rej) => {
+  const r = indexedDB.open('meshbay', 1);
+  r.onupgradeneeded = () => r.result.createObjectStore('group_indexes', { keyPath: 'groupId' });
+  r.onsuccess = () => res(r.result);
+  r.onerror = () => rej(r.error);
+});
+const started = Date.now();
+let outcome = 'never settled';
+try {
+  await openDB();
+  outcome = 'resolved';
+} catch (err) {
+  outcome = 'rejected: ' + (err && err.message);
+}
+window.__result = { outcome, ms: Date.now() - started, held: !!held };
+// Let go: the reset case below needs a database nobody is holding.
+held.close();
+</script></body></html>"""
+
+
+PAGE = r"""<!doctype html><html><head><meta charset=utf-8></head>
+<body style="margin:0"><div id="frames"></div><script>
+const cases = [];
+const post = (o) => fetch('/log', { method: 'POST', body: JSON.stringify(o) });
+addEventListener('error', (e) => post({ error: 'page error: ' + (e.message || e) }));
+addEventListener('unhandledrejection',
+  (e) => post({ error: 'rejection: ' + ((e.reason && (e.reason.stack || e.reason.message)) || e.reason) }));
+const sleep = (ms) => new Promise((r) => setTimeout(r, ms));
+const add = (src) => {
+  const f = document.createElement('iframe');
+  f.src = src;
+  f.style.cssText = 'width:420px;height:700px;border:0;display:block';
+  document.getElementById('frames').appendChild(f);
+  return new Promise((res) => { f.addEventListener('load', () => res(f)); });
+};
+
+(async () => {
+  try {
+    const ok = await add('/shell?case=mounts');
+    const broken = await add('/shell?case=fails');
+    // Started now, read after the wait: it settles well inside it.
+    const blocked = await add('/blocked');
+
+    const w = broken.contentWindow;
+    w.localStorage.setItem('mb_auth', '{"token":"x"}');
+    await new Promise((res) => {
+      const r = w.indexedDB.open('meshbay', 1);
+      r.onupgradeneeded = () => r.result.createObjectStore('group_indexes', { keyPath: 'groupId' });
+      r.onsuccess = () => { r.result.close(); res(); };
+      r.onerror = () => res();
+    });
+
+    // The guard looks once, late. Waiting it out is the measurement.
+    await sleep(11000);
+
+    const guardIn = (f) => {
+      const h = f.contentDocument.getElementById('app');
+      return { text: h ? h.textContent : '', buttons:
+        [...f.contentDocument.querySelectorAll('#app button')].map((b) => b.textContent) };
+    };
+    const resetButton = (f) => [...f.contentDocument.querySelectorAll('#app button')]
+      .find((x) => /r.initialis|reset/i.test(x.textContent));
+
+    const a = guardIn(ok);
+    cases.push({ case: 'the application mounted', text: a.text.slice(0, 80),
+                 buttons: a.buttons });
+
+    const b = guardIn(broken);
+    cases.push({ case: 'the module graph did not link',
+                 text: b.text.slice(0, 400), buttons: b.buttons });
+
+    cases.push({ case: 'a blocked database upgrade gives up instead of hanging',
+                 result: blocked.contentWindow.__result || 'never settled' });
+
+    // ── a reset another tab is holding open ──────────────────────────────
+    //
+    // `deleteDatabase` waits for every other connection to close, silently: it
+    // neither fails nor completes. A reset that does not watch for that
+    // reloads into the state it has just claimed to clear.
+    // No version: whatever the database is at now. Asking for a particular one
+    // is how this fixture quietly stopped holding anything — the store had
+    // moved to 2 underneath it and the open failed instead of connecting.
+    const holder = await new Promise((res) => {
+      const r = indexedDB.open('meshbay');
+      r.onsuccess = () => res(r.result);
+      r.onerror = () => res(null);
+    });
+    let reloadedAnyway = false;
+    broken.addEventListener('load', () => { reloadedAnyway = true; }, { once: true });
+    const r1 = resetButton(broken);
+    let told = null;
+    if (r1) {
+      r1.click();
+      await sleep(6000);
+      const m = broken.contentDocument.querySelector('#app [data-blocked]');
+      told = m ? m.getAttribute('data-blocked') : null;
+    }
+    cases.push({ case: 'a reset another tab is blocking says so',
+                 clicked: !!r1, told, reloadedAnyway,
+                 stillOffersReset: !!resetButton(broken) });
+
+    // ── and with nothing in its way ──────────────────────────────────────
+    if (holder) holder.close();
+    const r2 = resetButton(broken);
+    let after = null;
+    if (r2) {
+      const reloaded = new Promise((res) => broken.addEventListener('load', res, { once: true }));
+      r2.click();
+      await Promise.race([reloaded, sleep(9000)]);
+      await sleep(1500);
+      // Read from here, not from the frame: the frame is reloading, and a
+      // window mid-navigation answers for whichever document happens to be
+      // current. Same origin, so this is the same store.
+      let dbs = null;
+      try { dbs = (await indexedDB.databases()).map((d) => d.name); } catch (e) { dbs = 'n/a'; }
+      after = { auth: localStorage.getItem('mb_auth'), dbs };
+    }
+    cases.push({ case: 'the reset button empties this origin', clicked: !!r2, after });
+
+    fetch('/log', { method: 'POST', body: JSON.stringify({ cases }) });
+  } catch (err) {
+    fetch('/log', { method: 'POST',
+      body: JSON.stringify({ error: String((err && err.stack) || err) }) });
+  }
+})();
+</script></body></html>"""
+
+
+class H(http.server.BaseHTTPRequestHandler):
+    def log_message(self, *a):
+        pass
+
+    def do_POST(self):
+        length = int(self.headers.get("Content-Length") or 0)
+        if self.path == "/log":
+            RECORDS.append(json.loads(self.rfile.read(length).decode()))
+        else:
+            self.rfile.read(length)
+        self.send_response(204)
+        self.end_headers()
+
+    def _send(self, body: bytes, ctype: str, code: int = 200) -> None:
+        self.send_response(code)
+        self.send_header("Content-Type", ctype)
+        self.send_header("Content-Length", str(len(body)))
+        self.end_headers()
+        self.wfile.write(body)
+
+    def do_GET(self):
+        path, _, query = self.path.partition("?")
+        if path == "/":
+            self._send(PAGE.encode(), "text/html; charset=utf-8")
+        elif path == "/blocked":
+            self._send(BLOCKED.encode(), "text/html; charset=utf-8")
+        elif path == "/shell":
+            self._send(SHELL.encode(), "text/html; charset=utf-8")
+        elif path == "/__case.js":
+            # The referring frame decides: one mounts, the other is missing —
+            # a 404 on a module, which is what a graph that will not link looks
+            # like from the outside.
+            if "case=fails" in (self.headers.get("Referer") or ""):
+                self._send(b"not here", "text/plain", 404)
+            else:
+                self._send(MOUNTS.encode(), "text/javascript")
+        else:
+            asset = (STATIC / path.lstrip("/")).resolve()
+            if not str(asset).startswith(str(STATIC)) or not asset.is_file():
+                self.send_response(404)
+                self.end_headers()
+                return
+            self._send(asset.read_bytes(), "text/javascript")
+
+
+def main() -> int:
+    with socketserver.TCPServer(("127.0.0.1", PORT), H) as srv:
+        threading.Thread(target=srv.serve_forever, daemon=True).start()
+        with tempfile.TemporaryDirectory(ignore_cleanup_errors=True) as profile:
+            proc = subprocess.Popen(
+                ["google-chrome", "--headless=new", "--disable-gpu", "--no-sandbox",
+                 f"--user-data-dir={profile}", "--window-size=900,900",
+                 f"http://127.0.0.1:{PORT}/"],
+                stdout=subprocess.DEVNULL, stderr=subprocess.DEVNULL)
+            for _ in range(600):
+                if RECORDS:
+                    break
+                time.sleep(0.1)
+            proc.terminate()
+            try:
+                proc.wait(timeout=10)
+            except subprocess.TimeoutExpired:
+                proc.kill()
+                proc.wait()
+    if not RECORDS:
+        print(json.dumps({"error": "no measurement"}), file=sys.stderr)
+        return 1
+    print(json.dumps(RECORDS[0], indent=1, ensure_ascii=False))
+    return 0
+
+
+if __name__ == "__main__":
+    raise SystemExit(main())
diff --git a/packages/meshbay-hub/tests/test_boot_guard.py b/packages/meshbay-hub/tests/test_boot_guard.py
new file mode 100644
index 0000000..d802a03
--- /dev/null
+++ b/packages/meshbay-hub/tests/test_boot_guard.py
@@ -0,0 +1,122 @@
+"""
+The page cannot end up silently blank, and there is a way out of it.
+
+A reader spent an evening on a white screen. The shell arrived on every reload,
+every module was served from the browser's own store so not one request reached
+the hub, and nothing rendered: no message, no error, nothing in the server log
+to look at. The same account worked in a private window and in another browser
+— the signature of something wrong in this origin's stored state rather than in
+what was deployed. Clearing the site's data fixed it; clearing the *cache*,
+three times, had not, because a service worker and IndexedDB are not the cache.
+
+What is measured here is the silence, not its cause — that evidence was
+destroyed by the fix, necessarily. A blank page is a bug report nobody can
+write, and on a phone there is no console to open.
+
+`boot-guard.js` is a classic script, loaded before the module graph, because the
+failure it guards against includes the graph never linking: one bad module and
+no module code runs at all, so a guard inside `app.js` would be part of what
+failed.
+"""
+
+import json
+import shutil
+import subprocess
+import sys
+from pathlib import Path
+
+import pytest
+
+HARNESS = Path(__file__).parent / "harness" / "boot_guard_probe.py"
+
+
+@pytest.fixture(scope="module")
+def cases():
+    if shutil.which("google-chrome") is None:
+        pytest.skip("Chrome is not available")
+    proc = subprocess.run([sys.executable, str(HARNESS)],
+                          capture_output=True, text=True, timeout=300)
+    data = json.loads(proc.stdout)
+    assert "error" not in data, f"probe failed: {proc.stdout}{proc.stderr}"
+    return {c["case"]: c for c in data["cases"]}
+
+
+def test_the_guard_says_nothing_when_the_application_mounts(cases):
+    # The ordinary case, and the one that matters most: a guard that draws over
+    # a working application would be a worse bug than the one it is for.
+    c = cases["the application mounted"]
+    assert c["buttons"] == [], f"the guard drew over a mounted application: {c}"
+    assert "application" in c["text"]
+
+
+def test_a_graph_that_does_not_link_shows_a_message_not_a_blank_page(cases):
+    c = cases["the module graph did not link"]
+    assert len(c["text"]) > 40, f"still effectively blank: {c['text']!r}"
+
+
+def test_the_failure_is_named_on_screen(cases):
+    # Verbatim, on screen, because a phone has no console and this is the first
+    # thing anybody diagnosing it will ask for.
+    c = cases["the module graph did not link"]
+    assert "__case.js" in c["text"], (
+        f"the module that failed is not named anywhere: {c['text']!r}")
+
+
+def test_a_way_out_is_offered(cases):
+    c = cases["the module graph did not link"]
+    assert len(c["buttons"]) == 2, c["buttons"]
+    joined = " ".join(c["buttons"]).lower()
+    assert "essayer" in joined or "try" in joined
+    assert "initialis" in joined or "reset" in joined
+
+
+def test_the_reset_button_really_empties_this_origin(cases):
+    """A button that claims to clear and does not would be worse than none.
+
+    So it is clicked, for real, against a seeded `localStorage` and a seeded
+    database — and what it leaves behind is what is read back.
+    """
+    c = cases["the reset button empties this origin"]
+    assert c["clicked"], "no reset button to click"
+    assert c["after"]["auth"] is None, (
+        f"localStorage survived the reset: {c['after']}")
+    assert c["after"]["dbs"] == [], (
+        f"a database survived the reset: {c['after']}")
+
+
+def test_a_reset_another_tab_is_blocking_says_so(cases):
+    """Deleting a database waits for every other connection to close — silently.
+
+    Unwatched, `deleteDatabase` neither fails nor completes, so a reset that
+    reloads regardless comes back to exactly the state it claimed to clear. The
+    reader would then have tried the one thing that works, watched it appear to
+    work, and still be staring at the same page.
+    """
+    c = cases["a reset another tab is blocking says so"]
+    assert c["clicked"], "no reset button to click"
+    assert c["told"] == "meshbay", (
+        f"the blocked database was not named on screen: {c}")
+    assert not c["reloadedAnyway"], "it reloaded into the state it had not cleared"
+    assert c["stillOffersReset"], "the reader is left with no way to try again"
+
+
+def test_a_blocked_database_upgrade_gives_up_instead_of_hanging(cases):
+    """`indexedDB.open` fires neither `success` nor `error` when an upgrade is
+    blocked by another connection — it fires `blocked`, and unhandled that
+    leaves the promise unsettled for ever.
+
+    Version 2 arrived with the playlists, so every browser that had used this
+    site before it has a version 1 to upgrade, and a second tab holding one open
+    is all it takes.
+    """
+    c = cases["a blocked database upgrade gives up instead of hanging"]
+    assert c["result"] != "never settled", c
+    assert c["result"]["held"], "the fixture did not hold a connection open"
+    assert c["result"]["outcome"].startswith("rejected:"), c["result"]
+    # `blocked` fires the moment the upgrade is attempted, so this is immediate.
+    # The deadline behind it is a backstop for the case where even `blocked`
+    # never arrives — tolerating five seconds here would let the backstop stand
+    # in for the handler and the handler could be deleted unnoticed.
+    assert "blocked" in c["result"]["outcome"], (
+        f"gave up on a timeout rather than on the blocked event: {c['result']}")
+    assert c["result"]["ms"] < 1000, f"it took {c['result']['ms']}ms to give up"
author	Christophe Besson <cbesson@gmail.com>	2026-09-17 03:40:52 +0200
committer	Christophe Besson <cbesson@gmail.com>	2026-09-17 03:40:52 +0200
commit	f4fd6db8faa15bf02f38a14b652cc46936f8d6bf (patch)
tree	396cbe7fe188fb565231be937ccfa6419384814f /packages/meshbay-hub/tests
parent	929bfcccf8928a30d86d48a8b0137db0dd4a1ee0 (diff)
download	meshbay-f4fd6db8faa15bf02f38a14b652cc46936f8d6bf.tar.gz