fix: reap dead push subscriptions from other devices via relay liveness check

src/lib/__tests__/push-notifications.test.ts

@@ -0,0 +1,131 @@
+import { describe, it, expect, vi, beforeEach } from 'vitest';
+
+// Provide the Android native FCM surface setupPushNotifications needs. The
+// global test-setup mocks react-native with an empty NativeModules, so override
+// it here with a BulwarkFcm module and a pre-33 Platform.Version (which skips
+// the runtime permission request).
+vi.mock('react-native', () => {
+  class NativeEventEmitter {
+    addListener() {
+      return { remove: () => undefined };
+    }
+  }
+  return {
+    Platform: { OS: 'android', Version: 30, select: <T,>(s: { default?: T; android?: T }) => s.android ?? s.default },
+    NativeModules: {
+      BulwarkFcm: {
+        getToken: vi.fn(async () => 'fcm-token-xyz'),
+        deleteToken: vi.fn(async () => undefined),
+      },
+    },
+    NativeEventEmitter,
+    PermissionsAndroid: { RESULTS: { GRANTED: 'granted' }, request: vi.fn(async () => 'granted') },
+  };
+});
+
+vi.mock('../../api/jmap-client', () => ({
+  jmapClient: { username: 'user@example.com', serverUrl: 'https://mail.example.com' },
+}));
+
+vi.mock('../../api/push', () => ({
+  listPushSubscriptions: vi.fn(async () => []),
+  createPushSubscription: vi.fn(async () => 'new-server-id'),
+  verifyPushSubscription: vi.fn(async () => undefined),
+  destroyPushSubscription: vi.fn(async () => undefined),
+  updatePushSubscription: vi.fn(async () => true),
+}));
+
+import AsyncStorage from '@react-native-async-storage/async-storage';
+import { setupPushNotifications, deviceClientIdKey } from '../push-notifications';
+import {
+  listPushSubscriptions,
+  destroyPushSubscription,
+} from '../../api/push';
+import { generateAccountId } from '../account-utils';
+
+const OUR_DCID = 'aaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaa';
+const ACCOUNT_ID = generateAccountId('user@example.com', 'https://mail.example.com');
+const RELAY = 'https://relay.example.com';
+
+// State the fake relay reports for each foreign deviceClientId's /active probe.
+type RelayState = 'dead' | 'live' | 'unknown';
+
+function installFetch(states: Record<string, RelayState>): void {
+  global.fetch = vi.fn(async (input: RequestInfo | URL) => {
+    const url = typeof input === 'string' ? input : input.toString();
+    if (url.includes('/api/push/register')) {
+      return { ok: true, status: 200, json: async () => ({ ok: true }) } as Response;
+    }
+    if (url.includes('/api/push/verify/')) {
+      return { ok: true, status: 200, json: async () => ({ verificationCode: 'CODE' }) } as Response;
+    }
+    const active = url.match(/\/api\/push\/active\/([^/?]+)$/);
+    if (active) {
+      const dcid = decodeURIComponent(active[1]);
+      const state = states[dcid] ?? 'unknown';
+      if (state === 'unknown') {
+        return { ok: false, status: 404, json: async () => ({ error: 'Unknown subscription' }) } as Response;
+      }
+      return { ok: true, status: 200, json: async () => ({ active: state === 'live' }) } as Response;
+    }
+    throw new Error(`unexpected fetch: ${url}`);
+  }) as typeof fetch;
+}
+
+const destroyMock = destroyPushSubscription as ReturnType<typeof vi.fn>;
+const listMock = listPushSubscriptions as ReturnType<typeof vi.fn>;
+
+function sub(id: string, deviceClientId: string) {
+  return { id, deviceClientId, expires: new Date(Date.now() + 86400000).toISOString(), types: ['Email'] };
+}
+
+describe('setupPushNotifications leftover reaping', () => {
+  beforeEach(async () => {
+    vi.clearAllMocks();
+    await AsyncStorage.clear();
+    // Pin our deviceClientId so we control which leftovers are "ours".
+    await AsyncStorage.setItem(deviceClientIdKey(ACCOUNT_ID), OUR_DCID);
+  });
+
+  it('reaps our own and relay-confirmed-dead leftovers, keeps live and unverifiable ones', async () => {
+    listMock.mockResolvedValue([
+      sub('own-old', OUR_DCID), // our own previous attempt -> reap
+      sub('foreign-dead', 'deaddeaddeaddeaddeaddeaddeaddead'), // relay: dead -> reap
+      sub('foreign-live', 'livelivelivelivelivelivelivelive'), // relay: live -> keep
+      sub('foreign-unknown', 'unknwunknwunknwunknwunknwunknwun'), // relay: 404 -> keep
+    ]);
+    installFetch({
+      deaddeaddeaddeaddeaddeaddeaddead: 'dead',
+      livelivelivelivelivelivelivelive: 'live',
+      unknwunknwunknwunknwunknwunknwun: 'unknown',
+    });
+
+    const result = await setupPushNotifications({ relayBaseUrl: RELAY });
+
+    expect(result.verified).toBe(true);
+    const reaped = destroyMock.mock.calls.map((c) => c[0]);
+    expect(reaped).toContain('own-old');
+    expect(reaped).toContain('foreign-dead');
+    expect(reaped).not.toContain('foreign-live');
+    expect(reaped).not.toContain('foreign-unknown');
+  });
+
+  it('keeps foreign subs when the relay probe fails (network error)', async () => {
+    listMock.mockResolvedValue([sub('foreign-x', 'xxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxx')]);
+    global.fetch = vi.fn(async (input: RequestInfo | URL) => {
+      const url = typeof input === 'string' ? input : input.toString();
+      if (url.includes('/api/push/register')) {
+        return { ok: true, status: 200, json: async () => ({ ok: true }) } as Response;
+      }
+      if (url.includes('/api/push/verify/')) {
+        return { ok: true, status: 200, json: async () => ({ verificationCode: 'CODE' }) } as Response;
+      }
+      if (url.includes('/api/push/active/')) throw new Error('network down');
+      throw new Error(`unexpected fetch: ${url}`);
+    }) as typeof fetch;
+
+    await setupPushNotifications({ relayBaseUrl: RELAY });
+
+    expect(destroyMock.mock.calls.map((c) => c[0])).not.toContain('foreign-x');
+  });
+});

src/lib/push-notifications.ts

@@ -217,6 +217,33 @@ async function registerWithRelay(params: {
   }
 }
 
+/**
+ * Ask the relay whether a leftover subscription is dead. Returns true ONLY when
+ * the relay positively reports it inactive - a record it knows about that has
+ * never forwarded a push and isn't freshly registered. Every other outcome (the
+ * relay doesn't recognise the id, an older relay without this endpoint, a
+ * network blip, or a live subscription) returns false, so we never reap
+ * anything we can't confirm is dead. This is what lets setup clear its own
+ * abandoned attempts - and dead siblings left by reinstalls that regenerated
+ * the deviceClientId - without disturbing another live device or the PWA that
+ * shares the account.
+ */
+async function relayReportsDead(
+  relayBaseUrl: string,
+  subscriptionId: string,
+): Promise<boolean> {
+  try {
+    const res = await fetch(
+      buildRelayUrl(relayBaseUrl, `/api/push/active/${encodeURIComponent(subscriptionId)}`),
+    );
+    if (!res.ok) return false;
+    const body = (await res.json()) as { active?: unknown };
+    return body.active === false;
+  } catch {
+    return false;
+  }
+}
+
 async function pollVerificationCode(
   relayBaseUrl: string,
   subscriptionId: string,
@@ -306,19 +333,29 @@ export async function setupPushNotifications(
     await AsyncStorage.removeItem(subKey);
   }
 
-  // Reap any leftover Stalwart subscriptions still bound to this account's
-  // deviceClientId. These pile up when a previous enable attempt failed
-  // mid-flow (verify race, network blip, app killed). Stalwart per-account
-  // rate-limits PushVerification posts, so leaving stragglers around blocks
-  // the new one's verification - the symptom is a confusing "code does not
-  // match". Only reap matches for THIS account's deviceClientId; other
-  // accounts' subscriptions on this server (rare but possible if the same
-  // JMAP server backs multiple accounts) must be left alone.
-  const stragglers = existingSubs.filter(
-    (s) => s.deviceClientId === deviceClientId && s.id !== storedServerId,
-  );
-  for (const s of stragglers) {
-    await destroyPushSubscription(s.id).catch(() => undefined);
+  // Reap leftover Stalwart subscriptions that would otherwise starve the new
+  // one's verification. Stalwart emits only one PushVerification per account
+  // per ~60s and picks the oldest unverified subscription, so a single stale
+  // straggler blocks every fresh attempt - the symptom is the confusing
+  // "Timed out waiting for PushVerification" error. We can't read a
+  // subscription's verified state or URL over JMAP (Stalwart hides both), only
+  // its deviceClientId, so we decide what's safe to remove like this:
+  //   - same deviceClientId as ours: a previous attempt from THIS device,
+  //     always safe to reap.
+  //   - a different deviceClientId: could be another live device or the PWA on
+  //     this account. Ask the relay whether it's still alive and only reap the
+  //     ones it confirms are dead. Anything live - or anything the relay can't
+  //     vouch for (a different relay, a non-Bulwark client, a network blip) -
+  //     is left untouched.
+  for (const s of existingSubs) {
+    if (s.id === storedServerId) continue;
+    if (s.deviceClientId === deviceClientId) {
+      await destroyPushSubscription(s.id).catch(() => undefined);
+      continue;
+    }
+    if (await relayReportsDead(relayBaseUrl, s.deviceClientId)) {
+      await destroyPushSubscription(s.id).catch(() => undefined);
+    }
   }
 
   const serverAssignedId = await createPushSubscription({