NPE in code path through EmbeddedSocial.initGlobalObjects()

[barbeau]

Summary:

I'm seeing this NPE with 18 occurrences in OneBusAway Android in the Android Developer Console:

Caused by: java.lang.NullPointerException: 
  at com.microsoft.embeddedsocial.ui.c.a.<init> (Unknown Source)
  at com.microsoft.embeddedsocial.sdk.EmbeddedSocial.initGlobalObjects (Unknown Source)
  at com.microsoft.embeddedsocial.sdk.EmbeddedSocial.init (Unknown Source)
  at org.onebusaway.android.app.Application.setUpSocial (Application.java:579)
  at org.onebusaway.android.app.Application.onCreate (Application.java:111)
  at android.app.Instrumentation.callApplicationOnCreate (Instrumentation.java:1032)
  at android.app.ActivityThread.handleBindApplication (ActivityThread.java:5881)

...from Samsung Galaxy S7 devices (Sprint and Verizon) and:

Caused by: java.lang.NullPointerException: 
  at com.microsoft.embeddedsocial.ui.c.a.<init> (Unknown Source)
  at com.microsoft.embeddedsocial.sdk.EmbeddedSocial.initGlobalObjects (Unknown Source)
  at com.microsoft.embeddedsocial.sdk.EmbeddedSocial.init (Unknown Source)
  at org.onebusaway.android.app.Application.setUpSocial (Application.java:579)
  at org.onebusaway.android.app.Application.onCreate (Application.java:111)
  at android.app.Instrumentation.callApplicationOnCreate (Instrumentation.java:1037)
  at android.app.ActivityThread.handleBindApplication (ActivityThread.java:6499)

...from Galaxy J3 and J1 devices.

EDIT May 17 - Unobfuscated stack trace is below:

Caused by: java.lang.NullPointerException: 
  at com.microsoft.embeddedsocial.ui.notification.NotificationController.<init> (NotificationController.java)
  at com.microsoft.embeddedsocial.sdk.EmbeddedSocial.initGlobalObjects (EmbeddedSocial.java)
  at com.microsoft.embeddedsocial.sdk.EmbeddedSocial.init (EmbeddedSocial.java)
  at org.onebusaway.android.app.Application.setUpSocial (Application.java:579)
  at org.onebusaway.android.app.Application.onCreate (Application.java:111)
  at android.app.Instrumentation.callApplicationOnCreate (Instrumentation.java:1037)
  at android.app.ActivityThread.handleBindApplication (ActivityThread.java:6499)
  at android.app.ActivityThread.access$1800 (ActivityThread.java:231)
  at android.app.ActivityThread$H.handleMessage (ActivityThread.java:1889)
  at android.os.Handler.dispatchMessage (Handler.java:102)
  at android.os.Looper.loop (Looper.java:148)
  at android.app.ActivityThread.main (ActivityThread.java:7422)
  at java.lang.reflect.Method.invoke (Method.java)
  at com.android.internal.os.ZygoteInit$MethodAndArgsCaller.run (ZygoteInit.java:1230)
  at com.android.internal.os.ZygoteInit.main (ZygoteInit.java:1120)

Looks to be the same error, but it's in the obfuscated part of the ES SDK. We'll need the Proguard mapping file to debug further to see the method name for ...ui.c.a.

Steps to reproduce:

Unknown (reported via Android Developer Console)

Expected behavior:

Not crash

Observed behavior:

Crash

Device and Android version:

Samsung Galaxy S7, J3, and J1 devices (Android 7.0 for S7 and Android 6.0 for J3 and J1)

[barbeau]

It looks like in NotificationController constructor the most likely contender is this:

Intent intent = context.getPackageManager()
	.getLaunchIntentForPackage(context.getPackageName())
	.setFlags(Intent.FLAG_ACTIVITY_CLEAR_TOP | Intent.FLAG_ACTIVITY_NEW_TASK);

I'm guessing PackageManager.getLaunchIntentForPackage(context.getPackageName()) returns null. From the Android docs, though, it seems like that should never happen. My first thought is that maybe context.getPackageName() is returning the package name of org.onebusaway.android instead of the applicationId (which is com.joulespersecond.seattlebusbot due to legacy version of the app), but the Android docs for application ID say:

Note: The application ID used to be directly tied to your code's package name; so some Android APIs use the term "package name" in their method names and parameter names, but this is actually your application ID. For example, the Context.getPackageName() method returns your application ID. There's no need to ever share your code's true package name outside your app code.

I don't know - maybe Samsung got this wrong on some devices? That doesn't seem likely, and I'd think we'd see more crashes given the popularity of the S7 device.

Another possibility here is that there is a multithreading issue in the Android platform where the user requests to uninstall the app and the uninstallation process removes the application from the PackageManager while OBA Android is still running. Again, doesn't seem likely.

[barbeau]

@acrown-msft The unobfuscated stack trace (obtained from the proguard mapping file) confirms that the NPE came from NotificationController, but there are no line numbers because the code was obfuscated:

Caused by: java.lang.NullPointerException: 
  at com.microsoft.embeddedsocial.ui.notification.NotificationController.<init> (NotificationController.java)
  at com.microsoft.embeddedsocial.sdk.EmbeddedSocial.initGlobalObjects (EmbeddedSocial.java)
  at com.microsoft.embeddedsocial.sdk.EmbeddedSocial.init (EmbeddedSocial.java)
  at org.onebusaway.android.app.Application.setUpSocial (Application.java:579)
  at org.onebusaway.android.app.Application.onCreate (Application.java:111)
  at android.app.Instrumentation.callApplicationOnCreate (Instrumentation.java:1037)
  at android.app.ActivityThread.handleBindApplication (ActivityThread.java:6499)
  at android.app.ActivityThread.access$1800 (ActivityThread.java:231)
  at android.app.ActivityThread$H.handleMessage (ActivityThread.java:1889)
  at android.os.Handler.dispatchMessage (Handler.java:102)
  at android.os.Looper.loop (Looper.java:148)
  at android.app.ActivityThread.main (ActivityThread.java:7422)
  at java.lang.reflect.Method.invoke (Method.java)
  at com.android.internal.os.ZygoteInit$MethodAndArgsCaller.run (ZygoteInit.java:1230)
  at com.android.internal.os.ZygoteInit.main (ZygoteInit.java:1120)

We could ship an update soon with the ES SDK library unobfuscated to see if we can nail down a line number. There are around 44 crashes from this which isn't huge for the number of installs currently (106,725) so it doesn't warrant a hot fix specifically for this.

[barbeau]

Finally got a stack trace from this with the new ES SDK release without obfuscation, so we now have a line number (from a Samsung Galaxy S5 (klteatt), 2048MB RAM, Android 6.0):

java.lang.RuntimeException: 
  at android.app.ActivityThread.handleBindApplication (ActivityThread.java:6319)
  at android.app.ActivityThread.access$1800 (ActivityThread.java:221)
  at android.app.ActivityThread$H.handleMessage (ActivityThread.java:1860)
  at android.os.Handler.dispatchMessage (Handler.java:102)
  at android.os.Looper.loop (Looper.java:158)
  at android.app.ActivityThread.main (ActivityThread.java:7224)
  at java.lang.reflect.Method.invoke (Native Method)
  at com.android.internal.os.ZygoteInit$MethodAndArgsCaller.run (ZygoteInit.java:1230)
  at com.android.internal.os.ZygoteInit.main (ZygoteInit.java:1120)
Caused by: java.lang.NullPointerException: 
  at com.microsoft.embeddedsocial.ui.notification.NotificationController.<init> (NotificationController.java:89)
  at com.microsoft.embeddedsocial.sdk.EmbeddedSocial.initGlobalObjects (EmbeddedSocial.java:147)
  at com.microsoft.embeddedsocial.sdk.EmbeddedSocial.init (EmbeddedSocial.java:101)
  at org.onebusaway.android.app.Application.setUpSocial (Application.java:579)
  at org.onebusaway.android.app.Application.onCreate (Application.java:111)
  at android.app.Instrumentation.callApplicationOnCreate (Instrumentation.java:1036)
  at android.app.ActivityThread.handleBindApplication (ActivityThread.java:6316)

Here's line 89 in NotificationController:
https://github.com/Microsoft/EmbeddedSocial-Android-SDK/blob/master/EmbeddedSocialClient/sdk/src/main/java/com/microsoft/embeddedsocial/ui/notification/NotificationController.java#L89

So it looks like in:

        Intent intent = context.getPackageManager()
            .getLaunchIntentForPackage(context.getPackageName())
            .setFlags(Intent.FLAG_ACTIVITY_CLEAR_TOP | Intent.FLAG_ACTIVITY_NEW_TASK);  // <---Line 89

..PackageManager.getLaunchIntentForPackage(context.getPackageName()) does indeed return null, which matches my guess in a previous post, although as I said earlier this should never actually happen on a device. See #91 (comment) for some explanations of how this could theoretically happen.

Also, Google Play Console reporting on this issue is a bit strange. Here's the occurences by Android version and device for the previous OBA release (when ES SDK was still obfuscated), which are all on Android 7.0:

...and here's the report for the newest unobfuscated version where I got the line number from, but which is also grouped with some obfuscated reports from the previous app version:

...which are all Android 6.0 and below.

Note that in both cases, reports are largely dominated by Samsung devices, so it's likely a Samsung issue in their Android implementation.

I'm not sure what the best resolution to this issue is. Would it cause any damage to catch and log/ignore the notification controller instantiation/registration, and proceed with the rest of the EmbeddedSocial.init() code path for this execution? I don't think this is the worst thing if it only fails once, likely the first execution after install. In theory future executions should then work fine (?).