Rx WorkflowHost adapters swallow "Fatal" (as def'd by Rx2) exceptions from subscribers.

[rjrjr]

• Build the Tic Tac Toe app from branch ray/DNM-swallowed-exception
• Launch the app and login with a bad password, e.g. foo@bar / pass
• Notice that you hang on the Loading screen when you should pop back to login.

Now uncomment the block at line 47 of MainActivity, to make another subscription to workflowRunner.renderings. Repeat the recipe, and crash (stack below).

I suspect that in the hang recipe this exception is being swallowed, but I haven't dug further yet.

Caused by: java.lang.ClassNotFoundException: Didn't find class "com.squareup.workflow.ui.backstack.ViewStateCache$prepareToUpdate$2$1" on path: DexPathList[[zip file "/data/app/com.squareup.sample.tictacworkflow-emIWXcG5jN7vKiZo1257Gg==/base.apk"],nativeLibraryDirectories=[/data/app/com.squareup.sample.tictacworkflow-emIWXcG5jN7vKiZo1257Gg==/lib/x86, /system/lib]]
   at dalvik.system.BaseDexClassLoader.findClass(BaseDexClassLoader.java:134)
   at java.lang.ClassLoader.loadClass(ClassLoader.java:379)
   at java.lang.ClassLoader.loadClass(ClassLoader.java:312)

[rjrjr]

The exception is pretty clearly a compiler bug, I'll get that reported. Goes away if I change the UpdateTools interface in ViewStateCache to a data class -- probably a problem w/anonymous objects in lambdas.

[zach-klippenstein]

If I explicitly throw a RuntimeException just before running the lambda that triggers this exception, it crashes as expected.

IntelliJ will occasionally suck up all my CPU and hang while stepping around after the ClassNotFoundException gets thrown, so that's a good sign.

RxJava is seeing the error as a NoClassDefFoundError, which is a LinkageError, which means that Rx rethrows it directly and bypasses all the usual exception processing logic. Wondering if that might have something to do with it.

[zach-klippenstein]

I can't repro in a clean room project, and the code that it is occuring in (flatMapWorkflow) has been deleted, so I'm going to close this issue.

[zach-klippenstein]

I reproduced in unit tests: 86b82bc

Going to replicate these tests for (our internal-only, for now) RxWorkflowHost.

[zach-klippenstein]

Working around this internally by checking for this condition and manually passing to the thread's UncaughtExceptionHandler if it will be swallowed.

I've also reproduced the bug in an isolated project and filed as Kotlin/kotlinx.coroutines#1297.

[rjrjr]

Finally got around to reporting the swallowed exception to JetBrains, https://youtrack.jetbrains.com/issue/KT-32351

[rjrjr]

@zach-klippenstein we can close this now, yeah?

[zach-klippenstein]

I want to leave this open until I'm able to remove our workarounds (I think using Flow should fix).