fix: wait for pending user updates before evaluating targeting [ENG-3153]

[pandeymangg]

Ref ENG-3153

What & why

Was: setUserId() / setAttribute() queue a debounced POST /user, but track() picked a survey straight away. Identify-then-track on login rendered the survey before the contact existed: segment-targeted surveys were dropped (the filter still saw an anonymous user), and an unsegmented survey rendered with no contactId, so its response was recorded as Anonymous.

Now: track() resolves the action class, then waits for the queued update to land (flushing it at once instead of sitting out the 500ms debounce), and only then picks a survey. If the update fails or times out (5s), a survey with segment filters is skipped rather than shown on stale membership; one without filters still shows.

Reported in #60. Same fix as the JS SDK (formbricks/formbricks#7498) and iOS (formbricks/ios#56).

Also unbreaks main. Every CI run since the 2026-09-23 dependabot merges has been red: kotlinOptions { jvmTarget } is a script compilation error on Kotlin 2.4.20 (#69), and the test APK's AAR-metadata check rejects androidx.core 1.19 (#67, needs AGP ≥ 9.1) and okhttp 5.5 (#71, needs compileSdk 37) on AGP 8.9.2. The second commit migrates the script to compilerOptions, pins those two back to 1.16.0 / 4.12.0, and tells dependabot to leave them alone until the AGP 9 migration (#80) lands. appcompat 1.8, fragment 1.9 and the Kotlin bump stay.

Where to look

• UpdateQueue.waitForPendingWork / syncDidFinish: the wait, its two short circuits (nothing queued, anonymous user), the timeout, and the hand-back on failure.
• UserManager.syncUser: filterSurveys() now runs before waiters are released. That order is load-bearing.
• SurveyManager.track / evaluate: the split and the fail-closed branch.

Two bugs this had to fix first

• A write during an in-flight request was silently lost: values stayed queued during the request and the success path's reset() cleared them wholesale. commit() now moves its values out; a failure hands them back.
• commit() had an in-flight flag but no guard, so a debounce firing mid-request sent a second concurrent POST /user, whose response could overwrite segments / displays / responses. Now deferred and re-armed by syncDidFinish.

Coverage

Behaviour	Level
Identify + setAttribute(tier=pro) + track() back-to-back shows the segment-targeted survey; display and response carry the contact	manual — emulator against local web, evidence in the fold below
Identify + track() with no attributes: unsegmented survey shows, attributed	manual
Update times out → segment-targeted survey skipped with a warning, nothing shown	manual — 8s latency injected into POST /user, timeout fired at 5.0s
Wait short-circuits when nothing is queued and for an anonymous user; flushes now and coalesces; reports failure; times out; failed values ride along with the next commit	unit (guard) — UpdateQueueInstrumentedTest
Mid-flight write survives and gets its own request	unit (mutation) — drop the else if (hasNewWork && userId == null) branch in UpdateQueue.syncDidFinish
No second concurrent sync while one is in flight	unit (mutation) — remove the if (isSyncInFlight) guard in UpdateQueue.commit

Rerun: ./gradlew :android:connectedDebugAndroidTest -Pandroid.testInstrumentationRunnerArguments.class=com.formbricks.android.network.queue.UpdateQueueInstrumentedTest

Full suite: 133 tests, 0 failures (Pixel 9a, API 36).

Manual QA evidence (Pixel 9a emulator, local web, a segment-targeted copy of the app survey seeded on the same code action)

Before (38ba734): tap at 07:18:04.4 UTC → WebView loading at +0.6s → contact created 07:18:14 → Display 07:18:20 with contactId = null → Response 07:19:49 with contactId = null. The survey shown was the unsegmented one; the segment-targeted copy had been dropped by the anonymous filter.

After: tap at 07:21:29.3 → UpdateQueue - commit() called in the same millisecond (flushed, no debounce) → contact created 07:21:32.1 → Display 07:21:39 on the segment-targeted survey with contactId = cmukx5lg9… → Response 07:22:19 attributed. Identify without attributes: unsegmented survey shown, Display attributed.

Timeout: 8s setTimeout injected into the local POST /user, then setAttribute(email) + track() back-to-back:

12:53:17.957 D FormbricksSDK: UpdateQueue - commit() called on UpdateQueue with user-04e8a11c and {email=…}
12:53:22.960 D FormbricksSDK: UpdateQueue - a queued user update did not land within 5000ms
12:53:22.961 W FormbricksSDK: Skipping survey "eng3153segmentedtierpro1": the pending user update did not land, so segment membership is stale.

No survey was shown.

Open gaps

• No unit (red on main) row: waitForPendingWork and syncDidFinish(success) do not exist on main, so the tests cannot compile against it. The red-on-main proof is the manual before run: the unsegmented survey rendered 0.6s after track(), the contact was created 9s later, and both the display and the response have contactId = null.
• isShowingSurvey is write-only on Android (never read or reset), so iOS's "claim the latch before the wait" has nothing to port onto. Left as is.

Breaking changes

• This PR contains a breaking change

track() already returned Unit; it now resolves asynchronously. No public signature changes.

[coderabbitai]

Important

• 🔍 Trigger review

This repository does not receive automatic reviews because it has fewer than 10 stars.

⚙️ Run configuration

Configuration used: Organization UI

Review profile: ASSERTIVE

Plan: Advanced

Run ID: b325abb6-7463-43d6-bddd-4b662bac96ce

---

Thanks for using CodeRabbit! It's free for OSS, and your support helps us grow. If you like it, consider giving us a shout-out.

❤️ Share

• X
• Mastodon
• Reddit
• LinkedIn

_{Comment @coderabbitai help to get the list of available commands.}

[sonarqubecloud]

Quality Gate passed

Issues
1 New issue
0 Accepted issues

Measures
0 Security Hotspots
88.8% Coverage on New Code
0.0% Duplication on New Code

See analysis details on SonarQube Cloud