[joaodordio]

📝 Summary

A real switchProject always reported .switchedWithWarnings and logged an error, because the new project's first in-app sync runs before the app has re-identified the user.

🎟️ Jira Ticket: SDK-822

📖 Description

ATG hit this while testing a pre-release build and asked whether the warning needed fixing before go-live (Slack). Nothing was broken, but .switchedCleanly was unreachable after a real switch.

Cause. The switch clears the previous user, then starts the new project. start() returns inAppManager.start(), whose first sync goes through RequestCreator.createGetInAppMessagesRequest, which refuses the request while auth.emailOrUserId is .none. initialize2 turns that refusal into started == false, and the switch folded it into the result. The app's setEmail from the callback schedules the sync again, so nothing is lost.

Fix. startCountsAsClean counts a failed start as clean when no identity is set, and logs at info instead of error. It checks auth, not noUserLoggedIn(): the unknown-user ID also lets the request through, so with it set a failed sync is a genuine failure and still reports warnings and logs an error. The start is evaluated before teardownWasClean &&, so a genuine failure is still logged when the teardown was noisy too.

Two commits, reviewable separately:

1. dfdd53a2 The post-switch path. This is the ticket scope and the log line ATG saw.
2. 4f35ad8c The pre-init path (switch before any initialize). Since SDK-758 it forwards the same start result, so a fresh install reported warnings on iOS while Android reports SWITCHED_CLEANLY, because Android's failure signal is an initialization that times out or throws and a missing user is neither. This narrows what SDK-758 locked in, so it is its own commit and can be dropped. SDK-758's existing failure test still passes: a start that fails with someone identified still warns.

Why the tests missed it. MockDependencyContainer.createInAppFetcher returns a MockInAppFetcher that succeeds without a user, so the real request creation never ran. The new tests subclass it to build the real InAppFetcher. FailingInAppFetcher now sets an unknown-user identity before failing, so it keeps modelling a genuine failure rather than the expected refusal.

🧪 How to test?

• SwitchProjectGateTests.testASwitchWithNoUserYetOnTheNewProjectReportsCleanly: real fetch path, asserts the result is clean, that no in-app request goes out before identification, and that after setEmail from the callback the sync goes to project B with B's key. Fails without commit 1, on the result assertion only.
• SwitchProjectGateTests.testSwitchProjectBeforeInitializeWithNoUserYetReportsCleanly: fresh-install pre-init switch on the real fetch path. Fails without commit 2.
• Existing failing-start tests (post-switch and SDK-758's pre-init one) still report warnings.
• Full suite: unit-tests 739, notification-extension-tests 15, offline-events-tests 103, 0 failures.

Local note: on Xcode 27.2 this repo only runs with IPHONEOS_DEPLOYMENT_TARGET=15.0 passed to xcodebuild and on a pre-iOS-27 simulator (I used iOS 18.5), because Xcode 27 rejects the 12.0 target and the iOS 27 runtime refuses to launch the test host app without UIScene adoption. CI is pinned to an older Xcode so it is unaffected for now. Not changed here.

Manual: on a device, switch from an identified user on A to B and re-identify from the callback. The result should be .switchedCleanly, with no did not start cleanly error, and in-apps should load for B after setEmail.

🧾 Changelog

Updated under [Unreleased]. The warning-cause list in the switchProject entry gains the real fifth cause, says that nobody identified yet on the new project is not a warning, explains the expected Both email and userId are nil log line, and notes that a simulator has no APNs token. A new Fixed entry covers this change. The pre-init Added bullet, left stale by SDK-758, and the SDK-758 Fixed entry now describe the actual behaviour.

📹 Loom recording if applicable

Not recorded.

🐞 Github Issues solved

None.

📚 Docs PR if applicable

switchProject docs are tracked in SDK-676. The result contract described there should match the corrected CHANGELOG wording.

[codecov]

Codecov Report

✅ All modified and coverable lines are covered by tests.
✅ Project coverage is 75.23%. Comparing base (6cfaf84) to head (4f35ad8).

Additional details and impacted files

@@                        Coverage Diff                         @@
##           feature/SDK-674-switch-project    #1095      +/-   ##
==================================================================
+ Coverage                           75.03%   75.23%   +0.20%     
==================================================================
  Files                                 118      118              
  Lines                               10634    10642       +8     
==================================================================
+ Hits                                 7979     8007      +28     
+ Misses                               2655     2635      -20     

☔ View full report in Codecov by Harness.
📢 Have feedback on the report? Share it here.

🚀 New features to boost your workflow:

• ❄️ Test Analytics: Detect flaky tests, report on failures, and find test suite problems.