Stop identity-manager deinit from dispatching onto its own queue

The previous fix routed forceSave() through queue.async(flags: .barrier),
and deinit called forceSave(). Dispatching onto the private concurrent
queue from deinit is a teardown hazard: the async block resurrects self
and enqueues barrier work that may not drain before process exit, so at
teardown swift-testing/libdispatch waits on that outstanding work and the
process never exits — the CI "Run Swift Tests (app)" job reached
[144/144], then wedged for ~5 minutes and was Killed:9. This surfaced now
because moving cryptographicIdentities into the persisted cache made far
more test-created managers persist on deinit.

Root cause confirmed locally: under LIBDISPATCH_COOPERATIVE_POOL_STRICT=1
(single-worker pool, mimicking a constrained CI runner) the old code
intermittently stalled the whole suite ~15s from deinit-scheduled barrier
work starving the pool; the fix is stable across 14 full coverage+parallel
runs with the process exiting ~3s after the last test.

Fix:
- deinit no longer touches `queue`. Persistence is already durable (every
  mutating API persists inline within its own barrier), so deinit only
  does a queue-free best-effort flush if `pendingSave` is still set — a
  direct read of in-hand state, safe because a deallocating object has no
  other live references mutating `cache`.
- forceSave() (lifecycle/app-termination only, never deinit) now uses a
  synchronous queue.sync(flags: .barrier): race-free `cache` read on the
  barrier context and nothing left scheduled at teardown. No re-entrant
  deadlock risk since it is never called from deinit.

Test: test_manyManagersDeinitDoNotWedgeTeardown churns 200 managers that
mutate then deinit and asserts the workload completes promptly; combined
with the existing concurrent race guard (still TSan-clean) this covers the
deinit path. Verified: swift test --parallel --enable-code-coverage green
14x with prompt process exit, and TSan-clean.

Co-Authored-By: Claude Fable 5 <noreply@anthropic.com>

bitchat/Identity/SecureIdentityStateManager.swift

@@ -217,7 +217,26 @@ final class SecureIdentityStateManager: SecureIdentityStateManagerProtocol {
     }
     
     deinit {
-        forceSave()
+        // Do NOT dispatch onto `queue` here. `deinit` can itself run *on*
+        // `queue` (fire-and-forget saves capture `self`), and the object is
+        // being deallocated: a `queue.sync` would deadlock, and a
+        // `queue.async` schedules work that resurrects `self` and may never
+        // drain before process exit — at teardown libdispatch/swift-testing
+        // then wait on that outstanding barrier work and the process wedges
+        // (observed as a CI teardown hang once crypto identities moved into
+        // the persisted cache, so far more test-created managers persist on
+        // deinit).
+        //
+        // A flush here is redundant anyway: every mutating API already
+        // persists inline within its own barrier, so the keychain is already
+        // up to date. As a queue-free best-effort belt-and-suspenders, only
+        // flush if something is still pending. This is a direct read of
+        // in-hand state — safe because a deallocating object has no other
+        // live references, so nothing can be mutating `cache` concurrently.
+        if pendingSave {
+            pendingSave = false
+            persist(snapshot: cache)
+        }
     }
     
     // MARK: - Secure Loading/Saving
@@ -282,25 +301,25 @@ final class SecureIdentityStateManager: SecureIdentityStateManagerProtocol {
         }
     }
 
-    // Force a flush (for app termination / lifecycle events). Every mutating
+    // Force a flush (for app-termination / lifecycle events — NOT from
+    // `deinit`, which persists inline; see the deinit note). Every mutating
     // API already persists inline inside its own barrier via
     // `saveIdentityCache`, so by the time this is called the keychain is
     // already up to date and this is normally a no-op; it exists as a
     // belt-and-suspenders flush of any `pendingSave` left set.
     //
-    // The snapshot + persist run inside a `queue.async(flags: .barrier)`:
+    // Runs synchronously inside a `queue.sync(flags: .barrier)`: the barrier
-    // reading `cache` on the barrier context is race-free (a plain off-queue
+    // makes the `cache` read race-free (a plain off-queue read races in-flight
-    // read races in-flight barrier writers — JSONEncoder walking a
+    // barrier writers — JSONEncoder walking a concurrently-mutated dictionary
-    // concurrently-mutated dictionary can spin forever, which surfaced as a CI
+    // can spin forever, which surfaced as a CI hang), and being synchronous it
-    // hang). It is deliberately `async`, NOT `sync`: `forceSave` is reachable
+    // leaves nothing scheduled to keep the process alive at teardown. Safe
-    // from `deinit`, and the final release can itself run *on* `queue` (the
+    // against re-entrant deadlock because this is never invoked from `deinit`
-    // fire-and-forget saves capture `self`), so a `queue.sync` here would be a
+    // (the only path that can run *on* `queue`).
-    // re-entrant same-queue wait and deadlock (SIGTRAP).
     func forceSave() {
-        queue.async(flags: .barrier) { [weak self] in
+        queue.sync(flags: .barrier) {
-            guard let self, self.pendingSave else { return }
+            guard pendingSave else { return }
-            self.pendingSave = false
+            pendingSave = false
-            self.persist(snapshot: self.cache)
+            persist(snapshot: cache)
         }
     }
     

bitchatTests/Services/SecureIdentityStateManagerTests.swift

@@ -4,6 +4,33 @@ import BitFoundation
 @testable import bitchat
 
 final class SecureIdentityStateManagerTests: XCTestCase {
+    func test_manyManagersDeinitDoNotWedgeTeardown() {
+        // Teardown-hang guard: create many managers that mutate (scheduling
+        // fire-and-forget barrier saves) and then go out of scope, triggering
+        // deinit. A previous version dispatched forceSave() onto the private
+        // queue from deinit; at process teardown libdispatch/swift-testing
+        // waited on that outstanding barrier work and the process wedged
+        // (CI-only, killed at the watchdog). deinit must not schedule anything
+        // on `queue`. If this leaked scheduled work, a real CI run would hang;
+        // here we at least assert the whole workload completes promptly.
+        let iterations = 200
+        let start = Date()
+        for i in 0..<iterations {
+            autoreleasepool {
+                let manager = SecureIdentityStateManager(MockKeychain())
+                manager.upsertCryptographicIdentity(
+                    fingerprint: Data(repeating: UInt8(i & 0xFF), count: 32).sha256Fingerprint(),
+                    noisePublicKey: Data(repeating: UInt8(i & 0xFF), count: 32),
+                    signingPublicKey: Data(repeating: UInt8((i + 1) & 0xFF), count: 32),
+                    claimedNickname: "peer-\(i)"
+                )
+                // Drop `manager` -> deinit. Must not enqueue work on `queue`.
+            }
+        }
+        let elapsed = Date().timeIntervalSince(start)
+        XCTAssertLessThan(elapsed, 15, "manager churn/teardown took too long (possible deinit dispatch hang)")
+    }
+
     func test_upsertCryptographicIdentity_withoutClaimedNicknameDoesNotCreateSocialIdentity() async {
         let manager = SecureIdentityStateManager(MockKeychain())
         let fingerprint = String(repeating: "aa", count: 32)