Measure production mail-shell release budgets - #269

apps/unwired-mail/unwired-mailTests/MailboxConnectionAdapterTests.swift (1)

4463-4618: 🩺 Stability & Availability | 🟡 Minor | ⚡ Quick win

Explicitly tear down each hosted launch view

releaseFixtureWindow only makes the window visible, and loop cleanup only sets isHidden. This does not guarantee that AccountView’s .onDisappear runs. Add explicit teardown so MailboxWorkCoordinator.shared does not retain each iteration’s registration and inboxViewModel closure.

🤖 Prompt for AI Agents

Verify each finding against current code. Fix only still-valid issues, skip the
rest with a brief reason, keep changes minimal, and validate.

In `@apps/unwired-mail/unwired-mailTests/MailboxConnectionAdapterTests.swift`
around lines 4463 - 4618, Explicitly tear down each hosted SwiftUI view before
iteration cleanup so AccountView’s onDisappear executes. Update the launch,
body, and draft hosting flow around launchHost, bodyHost, and draftHost to
remove or dismantle their hosting controllers/views and release their windows,
rather than only setting isHidden, ensuring MailboxWorkCoordinator.shared drops
the iteration’s registration and inboxViewModel closure.

Source: Path instructions

apps/unwired-mail/unwired-mail/Features/Account/AccountView.swift (3)

1032-1035: 🚀 Performance & Scalability | 🔵 Trivial | ⚡ Quick win

Use constant-time membership tracking for rendered rows.

recordRenderedItem scans the full renderedItems array for every row appearance. This makes tracking O(n²) on the main actor. The release helper reads only id, but the driver stores each complete MailShellThreadListItem. Track identifiers in a Set and retain only the data required by the measurement.

🤖 Prompt for AI Agents

Verify each finding against current code. Fix only still-valid issues, skip the
rest with a brief reason, keep changes minimal, and validate.

In `@apps/unwired-mail/unwired-mail/Features/Account/AccountView.swift` around
lines 1032 - 1035, Update recordRenderedItem to use a Set of item identifiers
for constant-time duplicate checks instead of scanning renderedItems. Retain
only the measurement-required item data while preserving the existing release
helper’s identifier-based behavior, and update related storage and accessors
consistently.

---

1417-1417: 🎯 Functional Correctness | 🟡 Minor | ⚡ Quick win

Gate launch measurement on shell readiness and row rendering.

initialLaunchDidFinish() runs after refreshPinnedBodyPrefetch(...), but that method starts an unstructured Task and returns immediately. The release test also checks only for an expected thread ID in itemDidRender. Add an explicit NavigationSplitView shell-readiness signal, and wait for both that signal and a rendered row before recording launch time. Add focused success and timeout-path tests for shell readiness and prefetch ordering.

🤖 Prompt for AI Agents

Verify each finding against current code. Fix only still-valid issues, skip the
rest with a brief reason, keep changes minimal, and validate.

In `@apps/unwired-mail/unwired-mail/Features/Account/AccountView.swift` at line
1417, Update the launch measurement flow around itemDidRender and
initialLaunchDidFinish() to require both an explicit NavigationSplitView
shell-readiness signal and a rendered expected row before recording launch time.
Ensure refreshPinnedBodyPrefetch(...)’s asynchronous work cannot allow
measurement to finish prematurely, and add focused tests covering successful
shell/prefetch ordering and the timeout path.

Source: Path instructions

---

1574-1574: 🩺 Stability & Availability | 🟡 Minor | ⚡ Quick win

Guard the launch completion callback against cancellation.

The canceled .task can still reach initialLaunchDidFinish() because synchronization converts cancellation into a normal return. Add guard !Task.isCancelled else { return } before the callback. Add a cancellation/reappearance test and run the specified xcodebuild test command after cleanup.

🤖 Prompt for AI Agents

Verify each finding against current code. Fix only still-valid issues, skip the
rest with a brief reason, keep changes minimal, and validate.

In `@apps/unwired-mail/unwired-mail/Features/Account/AccountView.swift` at line
1574, In the launch task flow immediately before initialLaunchDidFinish(), check
Task.isCancelled and return when cancellation is detected so a canceled task
cannot invoke the completion callback. Add a test covering cancellation followed
by reappearance, then clean up and run the specified xcodebuild test command.

Sources: Coding guidelines, Path instructions

apps/unwired-mail/unwired-mail/Features/Account/AccountView.swift (2)

1576-1583: 🩺 Stability & Availability | 🟠 Major | ⚡ Quick win

Guard the launch completion hook.

Line 1583 invokes initialLaunchDidFinish() after multiple awaits without checking cancellation or session freshness. SwiftUI can cancel this task when the view disappears or when the signed-in snapshot changes. Non-throwing operations can still return after cancellation. A stale shell can then satisfy the release measurement.

       inboxViewModel.refreshPinnedBodyPrefetch(connections: gmailViewModel.connections)
+      guard !Task.isCancelled, session.isCurrent(snapshot) else { return }
       initialLaunchDidFinish()

As per path instructions, check cancellation and lifecycle transitions in SwiftUI mail-shell code.

🤖 Prompt for AI Agents

Verify each finding against current code. Fix only still-valid issues, skip the
rest with a brief reason, keep changes minimal, and validate.

In `@apps/unwired-mail/unwired-mail/Features/Account/AccountView.swift` around
lines 1576 - 1583, Guard the launch completion flow around
initialLaunchDidFinish() by checking task cancellation and current
session/lifecycle freshness after the awaited mailbox synchronization and reload
operations. Return or skip the completion hook when the SwiftUI task was
cancelled or the signed-in snapshot has changed, while preserving the existing
completion behavior for the active session.

Source: Path instructions

---

347-376: 🩺 Stability & Availability | 🔵 Trivial | ⚡ Quick win

Add an identity-token polling regression test.

Change only identityToken and assert that pollWhileActive continues polling.

🤖 Prompt for AI Agents

Verify each finding against current code. Fix only still-valid issues, skip the
rest with a brief reason, keep changes minimal, and validate.

In `@apps/unwired-mail/unwired-mail/Features/Account/AccountView.swift` around
lines 347 - 376, Add a regression test for pollWhileActive that mutates only the
session’s identityToken while keeping all other session fields unchanged, then
assert polling continues rather than stopping. Reuse the existing polling test
setup and counters around pollWhileActive; do not alter production behavior or
introduce unrelated session changes.

Source: Path instructions

apps/unwired-mail/unwired-mailTests/MailboxConnectionAdapterTests.swift (2)

4532-4550: 🗄️ Data Integrity & Integration | 🟠 Major | ⚡ Quick win

Assert the cached body content before recording the sample.

The bodyLoaded expectation only observes onLoaded. It does not verify that the selected message returned "Cached body". A stale cache entry or incorrect message identity can therefore pass the performance gate.

Expose the loaded body through the test harness and assert its text and message identity before appending bodyOpenSamples.

As per path instructions, verify externally observable behavior and privacy invariants in tests.

🤖 Prompt for AI Agents

Verify each finding against current code. Fix only still-valid issues, skip the
rest with a brief reason, keep changes minimal, and validate.

In `@apps/unwired-mail/unwired-mailTests/MailboxConnectionAdapterTests.swift`
around lines 4532 - 4550, Update the body-loading test around
MailShellMessageBody and the bodyLoaded expectation to expose the rendered body
and associated message identity through the test harness. Before appending to
bodyOpenSamples, assert that the loaded content is exactly “Cached body” and
belongs to the selected latestMessage, preserving the existing timing
measurement only after these checks pass.

Source: Path instructions

---

7571-7580: 🎯 Functional Correctness | 🟠 Major | ⚡ Quick win

Require the rendered set to contain only current-mailbox IDs.

The helper returns true when any expected ID appears. It does not reject stale IDs or prove that the rendered dataset changed. A stale previous row plus one new row can satisfy this condition.

Require a non-empty rendered set that intersects the expected IDs and is a subset of the expected set. Add a stable-frame or completeness check appropriate for the lazy list.

Suggested predicate

   for _ in 0..<100 {
     await releaseRenderFrame(view)
-    if !driver.renderedItemIds.isDisjoint(with: expectedIdSet) {
-      return true
+    let renderedItemIds = driver.renderedItemIds
+    if !renderedItemIds.isDisjoint(with: expectedIdSet),
+      renderedItemIds.isSubset(of: expectedIdSet)
+    {
+      await releaseRenderFrame(view)
+      return driver.renderedItemIds.isSubset(of: expectedIdSet)
     }
   }

As per path instructions, verify externally observable behavior and deterministic isolation in tests.

🤖 Prompt for AI Agents

Verify each finding against current code. Fix only still-valid issues, skip the
rest with a brief reason, keep changes minimal, and validate.

In `@apps/unwired-mail/unwired-mailTests/MailboxConnectionAdapterTests.swift`
around lines 7571 - 7580, Update releaseWaitForRenderedThreads to require a
non-empty renderedItemIds set that intersects expectedIdSet and is entirely
contained within it, rejecting stale IDs from other mailboxes. Also add a
stable-frame or completeness check appropriate for the lazy list so the helper
only returns true after the current rendered dataset has settled, and verify the
observable behavior with deterministic test isolation.

Source: Path instructions