feat: host the server-delivered mobile core (JS brain) in a hidden WebView with native fallback

[pandeymangg]

Ref ENG-2912

What & why

Was: The survey display decision (display types, recontact days, segments, display percentage, language resolution) is compiled into the SDK; changing it requires an SDK release plus a customer app store release.

Now: setup() fetches a small JS bundle from the Formbricks server (/js/mobile/v1/core.umd.cjs), evaluates it in a hidden, never-attached WebView (Play-policy-exempt interpreter path), and track() asks it which survey to show. Any failure — no bundle, wrong protocol version, evaluation error, unreadable decision — falls back to the untouched native logic (nativeTrack), so behaviour without a bundle is identical to today. The brain receives the raw originalResponseMap JSON, so no native re-encoding sits in the bridge contract.

First Android PR of the JS-brain epic (targets epic/js-brain, not main). Server side: formbricks/formbricks#9152 · iOS twin: formbricks/ios#55.

Where to look

• android/src/main/java/com/formbricks/android/mobilecore/MobileCoreRuntime.kt — off-screen WebView host, main-thread confined, protocol-version gate
• android/src/main/java/com/formbricks/android/mobilecore/MobileCoreLoader.kt — OkHttp fetch + per-host SharedPreferences cache
• android/src/main/java/com/formbricks/android/manager/SurveyManager.kt — brain-first track(), native path unchanged as fallback

Testing

• 8 new instrumented tests in MobileCoreRuntimeInstrumentedTest (init validation, protocol rejection, state pass-through across the bridge, brain-throws fallback, loader URL shape).
• Full instrumented suite: 132/132 green on a Pixel 9a emulator (API 36).
• End-to-end verified earlier against a live local server: brain fetched, decision logged from server-delivered JS, survey rendered — including multi-language resolution (setLanguage("hi") → survey in Hindi, response persisted as hi-IN).

Draft — remaining scope for ENG-2912

• Opt-out flag on FormbricksConfig (disable remote code for enterprise policies).
• This diff ships no http/cleartext relaxations; local-testing conveniences stay out of the SDK.

---

Note

AI model used — claude-fable-5, reasoning effort unknown.

[sonarqubecloud]

Quality Gate failed

Failed conditions
46.7% Coverage on New Code (required ≥ 80%)

See analysis details on SonarQube Cloud