Bazel Binary APK frozen on Splash Screen

[anandwana001]

Describe the bug
Launching the testing APK gets frozen on SplashActivity.

To Reproduce
Internal Testing

Expected behavior
Work smoothly on all devices.

Screenshots
If applicable, add screenshots or videos to help explain your problem.

Device

• Device/emulator being used - Pixel XL 3a (not working)/ Redmi (Working)
• SDK version - API 28 and Android 11

Additional context

2021-09-16 23:30:37.740 I/ActivityTaskManager: START u0 {act=android.intent.action.MAIN cat=[android.intent.category.LAUNCHER] flg=0x10200000 cmp=org.oppia.android/.app.splash.SplashActivity bnds=[439,80][641,379]} from uid 10064
2021-09-16 23:30:37.747 E/statsd: Predicate 5980654721335871649 dropping data for dimension key (10)0x2010101->10437[I] (10)0x30000->*launch*[S] 
2021-09-16 23:30:37.767 I/ActivityManager: Start proc 3893:org.oppia.android/u0a437 for pre-top-activity {org.oppia.android/org.oppia.android.app.splash.SplashActivity}
2021-09-16 23:30:37.778 E/g.oppia.androi: Not starting debugger since process cannot load the jdwp agent.
2021-09-16 23:30:37.813 W/g.oppia.androi: JIT profile information will not be recorded: profile file does not exist.
2021-09-16 23:30:37.826 I/MultiDex: VM with version 2.1.0 has multidex support
2021-09-16 23:30:37.826 I/MultiDex: Installing application
2021-09-16 23:30:37.826 I/MultiDex: VM has multidex support, MultiDex support library is disabled.
2021-09-16 23:30:37.841 W/DynamiteModule: Local module descriptor class for com.google.android.gms.measurement.dynamite not found.
2021-09-16 23:30:37.841 W/DynamiteModule: Failed to load module via V2: java.lang.NoSuchFieldException: No field sClassLoader in class Lbg/b; (declaration of 'bg.b' appears in /data/app/~~1o1OMvmxLfLRH6EbQQ7Ovg==/org.oppia.android-eHdGVSc-FveRZnY884YTBg==/base.apk)
2021-09-16 23:30:37.853 I/FirebaseCrashlytics: Initializing Crashlytics 17.1.1
2021-09-16 23:30:37.857 I/g.oppia.androi: The ClassLoaderContext is a special shared library.
2021-09-16 23:30:37.864 I/g.oppia.androi: The ClassLoaderContext is a special shared library.
2021-09-16 23:30:37.870 I/FirebaseInitProvider: FirebaseApp initialization successful
2021-09-16 23:30:37.887 I/TetheringManager: registerTetheringEventCallback:org.oppia.android
2021-09-16 23:30:37.902 I/DynamiteModule: Considering local module com.google.android.gms.measurement.dynamite:0 and remote module com.google.android.gms.measurement.dynamite:51
2021-09-16 23:30:37.902 I/DynamiteModule: Selected remote version of com.google.android.gms.measurement.dynamite, version >= 51

[anandwana001]

Removing the livedata observer on SplashActivityPrsenter fix the issue.

[anandwana001]

@BenHenning still issue on latest internal testing live app on Android 11.

[BenHenning]

@srushtirk also hit this, and I can now repro on an Android 11 device. I was hoping that #3795 would fix this, but I think it primarily fixed #3817. I found that this seems to only occur in the Proguard-optimized version of the app, and the logs are still unhelpful so I'm iterating locally with extra logs in the app to try and investigate. My best guess is that Android 11 (& 12) might be relying on some new reflection that is being stripped out by Proguard (though I'd expect that to trigger a crash, not a hang).

Interestingly, @srushtirk mentioned that MR2 works fine on her Android 11 device so it does seem like this is somehow tied to the splash screen changes (namely that we're now blocking on both language retrieval & onboarding state to proceed). Will keep this thread updated with my findings.

[BenHenning]

Some new notes:

• Removing the combineWith() & just defaulting the locale doesn't fix anything
• 1 optimization pass is sufficient to hit the issue (which is nice since it cuts the build time in about 1/3)
• The app seems to hang immediately after subscribing to the first data provider (interestingly, the LiveData never becomes active nor are any of the constituent data providers processed)

[BenHenning]

FWIW, the comment above regarding Proguard might actually be more likely than at first glance since MR3 is using custom Proguard rules vs. MR2's Gradle (AGP)-generated ones.

Further, Jetpack's Lifecycle relies on some reflection for certain observers (that leverage the annotations). LiveData isn't one of them, but I wonder if something in the Jetpack components library that ties into LiveData does (which could cause a silent failure rather than a crash depending on how it was implemented). Really weird that this wouldn't repro on other versions of Android, though, given AndroidX isn't part of Android.

I'm working to confirm this one way or another now.

Edit: adding exemptions along the lines of https://stackoverflow.com/a/44688917 (but for AndroidX versions) didn't work, so it's probably not the LiveData<->Proguard situation.

[BenHenning]

Spoke too soon. https://stackoverflow.com/a/59162111 fixes the issue, and I suspect the finding is the actual issue. It seems LifecycleObserver might depend on reflection to verify subscriptions are valid, but it's not clear to me at all why this is an Android 11-specific issue.

[BenHenning]

Aha! This might repro on 28+ per the comment here: https://android.googlesource.com/platform/frameworks/support/+/refs/heads/androidx-master-release/lifecycle/lifecycle-runtime/proguard-rules.pro#18. Just confirming...

Edit: confirming that exact line is the issue (it fixes exactly with that line being present). Simple PR & rebuild coming.

Edit2: for posterity (in case the link above ever breaks), the comment & line of code that fixes this:

# Also this rule prevents registerIn from being inlined.
-keepclassmembers class androidx.lifecycle.ReportFragment$LifecycleCallbacks { *; }```

[BenHenning]

For searching context, this issue was found in 0.6-alpha (MR3).