IssueTracker

Some related mappings.

io.fabric.sdk.android.services.common.ExecutorUtils -> d.a.a.a.a.b.x:
    ...
    958:958:java.lang.Object kotlin.jvm.internal.TypeIntrinsics.beforeCheckcastToFunctionOfArity(java.lang.Object,int):341 -> a
    959:959:java.lang.Object kotlin.jvm.internal.TypeIntrinsics.beforeCheckcastToFunctionOfArity(java.lang.Object,int):342:342 -> a
    ...
    20:20:void kotlin.jvm.internal.TypeIntrinsics.throwCce(java.lang.String):26:26 -> b
    20:20:void kotlin.jvm.internal.TypeIntrinsics.throwCce(java.lang.Object,java.lang.String):22 -> b

which means, this line: https://github.com/JetBrains/kotlin/blob/master/libraries/stdlib/jvm/runtime/kotlin/jvm/internal/TypeIntrinsics.java#L342 is inlined to io.fabric's report generating routine.

For the record.

Same result with R8 1.5.33
I tested to fully remove Crashlytics + fabric and same crash occurs but of course with different inlining:

     Caused by: java.lang.ClassCastException: ta cannot be cast to kotlin.jvm.functions.Function1
        at e.f.b.k.b(Intrinsics.java:5)
        at e.f.b.k.a(Intrinsics.java:1034)
        at c.f.a.c.b.a.b.f.a(EventBus.kt:12)
        at k.a.a.a.l.l.onResume(BaseMenuActivity.kt:44)
        at xxxxxxxStartActivity.onResume(StartActivity.kt:1)
        at android.app.Instrumentation.callActivityOnResume(Instrumentation.java:1413)
        at android.app.Activity.performResume(Activity.java:7292)
        at android.app.ActivityThread.performResumeActivity(ActivityThread.java:3803)
        at android.app.ActivityThread.handleResumeActivity(ActivityThread.java:3843) 
        at android.app.servertransaction.ResumeActivityItem.execute(ResumeActivityItem.java:51) 
        at android.app.servertransaction.TransactionExecutor.executeLifecycleState(TransactionExecutor.java:145) 
        at android.app.servertransaction.TransactionExecutor.execute(TransactionExecutor.java:70) 
        at android.app.ActivityThread$H.handleMessage(ActivityThread.java:1823) 
        at android.os.Handler.dispatchMessage(Handler.java:107) 
        at android.os.Looper.loop(Looper.java:198) 
        at android.app.ActivityThread.main(ActivityThread.java:6729) 
        at java.lang.reflect.Method.invoke(Native Method) 
        at com.android.internal.os.RuntimeInit$MethodAndArgsCaller.run(RuntimeInit.java:493) 
        at com.android.internal.os.ZygoteInit.main(ZygoteInit.java:876) 

kotlin.jvm.internal.Intrinsics -> e.f.b.k:
But this one have a million lines after mapping to -> a and -> b

New APK and mapping sent to Jinseong.

As far as I understand the issue so far, when we merge Kotlin's FunctionX instances into a lambda group, we define it as a sub class of kotlin Lambda and call its constructor with a proper arity. For example, that `ta` is collecting mergeable instances of Function1, which literally inputs one argument and has a lambda function, invoke(Object). The issue is, in the generated lambda group <init>, I saw that arity is set as 0, not 1. Then, before hitting the direct type match (https://github.com/JetBrains/kotlin/blob/master/libraries/stdlib/jvm/runtime/kotlin/jvm/internal/TypeIntrinsics.java#L264), it's matched with FunctionBase (a subtype of Lambda) first, and tried to retrieve the arity, where now at runtime that merged group returned a wrong arity. Manually over-writing that arity to the correct one made that type free from casting exceptions. (There are more lambda groups with wrong arities, so the report saw a similar cast exception with a different type.)

It's quite not trivial to reproduce the issue locally. To understand where such arity can go wrong, I add some logging (to R8 1.5). Also, instead of using the size of parameters of the representative method (which could be altered, e.g., via unused argument removal), I change it to use the arity from Kotlin FunctionX interface name directly. See the attached diff.txt.

Tolriq, could you try the attached r8.jar? For logging messages, you can email it to me. Also, hope the resulting apk doesn't crash at runtime. And if so, that would be the potential fix, but I still want to reproduce the issue locally. Thanks!

buildscript {

    dependencies {
        classpath files($PATH_TO_R8_JAR)                 // Must be before the Gradle Plugin for Android.
        classpath 'com.android.tools.build:gradle:X.Y.Z' // Your current AGP version.
     }
}

Oops, the jar attached at comment #4 has been built from ToT (1.6.2-dev). Please try this one (or both :D)

I doubt anyone work on Week-End but quickly tested with the 1.5 version and now have strange crash instead of previous one:)

2019-05-25 11:14:11.115 29479-29577/? E/AndroidRuntime: FATAL EXCEPTION: AppUpdateService
    java.lang.AbstractMethodError: abstract method "void b4.h.a.e.a.c.l.a()"
        at b4.h.a.e.a.c.l.run(Unknown Source:0)
        at android.os.Handler.handleCallback(Handler.java:874)
        at android.os.Handler.dispatchMessage(Handler.java:100)
        at android.os.Looper.loop(Looper.java:198)
        at android.os.HandlerThread.run(HandlerThread.java:65)

Will do complete tests and send logs later today.

After reading correctly the error I tried again at same perimeter as before and it actually fix the issue.

This last crash is a new R8 issue tied to implementation("com.google.android.play:core:1.6.0") I have not yet fully implemented it and have not tested with r8 1.4 so will open another issue when I have more details about it.

Thank you for logging as well as code snippet. It's really helpful to reproduce the issue (in a different way): https://r8-review.googlesource.com/c/r8/+/38840 (I'm glad I didn't reveal any code patterns from your app!)

Also, thank you for filing a separate issue (issue 133686361) for comment #6 (and comment #7). Will take a look at that too.

According to my repro, the issue still exists on master. Could you confirm that it's indeed the case? Here is the build.gradle setting to use the current ToT:

buildscript {

    repositories {
        maven {
            url "http://storage.googleapis.com/r8-releases/raw/master"
        }
    }

    dependencies {
        classpath 'com.android.tools:r8:3682ef4d0842a4963e703f8a81035fb67866bdd6'  // Must be before the Gradle Plugin for Android.
        classpath 'com.android.tools.build:gradle:X.Y.Z'     // Your current AGP version.
     }
}

------

Regarding the fix, it's also glad to hear that my temporary workaround, included in the patch at comment #4, works for you. Later, however, I realized that such pattern matching might be prone to several other issues, like literally missed pattern matching, or arithmetic conversion error. So, I came up with another solution: https://r8-review.googlesource.com/c/r8/+/38900 Could you try it while it's under review? Here is another build.gradle setting to use that tentative fix:

buildscript {

    repositories {
        maven {
            url "http://storage.googleapis.com/r8-releases/raw/master"
        }
    }

    dependencies {
        classpath 'com.android.tools:r8:cef9babcb08fc6281bfb115836e7f4c7391fc878'  // Must be before the Gradle Plugin for Android.
        classpath 'com.android.tools.build:gradle:X.Y.Z'     // Your current AGP version.
     }
}

(Of course, you may still see issue 133686361.)

In case you do not already sleep :)

Tested very quickly 3682ef4d0842a4963e703f8a81035fb67866bdd6

And new fix causes new error

    java.lang.VerifyError: Verifier rejected class z3.c.b.a.a: void z3.c.b.a.a.a(java.lang.StringBuilder, java.lang.String, java.lang.String, z3.f.a.c.c.o, java.lang.Object) failed to verify: void z3.c.b.a.a.a(java.lang.StringBuilder, java.lang.String, java.lang.String, z3.f.a.c.c.o, java.lang.Object): [0xA] register v4 has type Reference: java.lang.Object but expected Precise Reference: java.lang.String (declaration of 'z3.c.b.a.a' appears in base.apk)
        at z3.c.b.a.a.a(Unknown Source:0)
        at z3.d.a.w.o.t.<init>(DecodePath.java:6)
        at z3.d.a.z.d.<clinit>(LoadPathCache.java:2)
        at z3.d.a.n.<init>(Registry.java:3)
        at z3.d.a.c.<init>(Glide.java:14)
        at z3.d.a.c.b(Glide.java:82)
        at z3.d.a.c.a(Glide.java:13)
        at org.leetzone.android.yatsewidget.YatseApplication.onCreate(YatseApplication.kt:18)
        at android.app.Instrumentation.callApplicationOnCreate(Instrumentation.java:1155)
        at android.app.ActivityThread.handleBindApplication(ActivityThread.java:5918)
        at android.app.ActivityThread.access$1100(ActivityThread.java:202)
        at android.app.ActivityThread$H.handleMessage(ActivityThread.java:1665)
        at android.os.Handler.dispatchMessage(Handler.java:107)
        at android.os.Looper.loop(Looper.java:198)
        at android.app.ActivityThread.main(ActivityThread.java:6729)
        at java.lang.reflect.Method.invoke(Native Method)
        at com.android.internal.os.RuntimeInit$MethodAndArgsCaller.run(RuntimeInit.java:493)
        at com.android.internal.os.ZygoteInit.main(ZygoteInit.java:876)

Have to move but will send all APK / mapping for both later today.

Nah, it turned out ToT has a regression, which has been reverted. Let me quickly prepare a patch on R8 1.5 instead.

Please try the attached r8.jar on R8 1.5

I pasted the wrong one I tested cef9babcb08fc6281bfb115836e7f4c7391fc878 ;(

I'm leaving now so won't be able to report more before couple hours sorry.

No worries. Both SHAs I posted included that regression. :o

The fix is landed on master (https://r8-review.googlesource.com/c/r8/+/38900) and R8 1.5 (https://r8-review.googlesource.com/c/r8/+/38920). Could you try R8 1.5.38? Thanks!

1.5.38 does not work sending apk and mapping by mail.

2019-05-29 09:22:55.738 20381-20381/? E/AndroidRuntime: FATAL EXCEPTION: main
    java.lang.RuntimeException: Unable to resume activity {xxxx/xxx.StartActivity}: java.lang.ClassCastException: s1 cannot be cast to kotlin.jvm.functions.Function1
        at android.app.ActivityThread.performResumeActivity(ActivityThread.java:3811)
        at android.app.ActivityThread.handleResumeActivity(ActivityThread.java:3843)
        at android.app.servertransaction.ResumeActivityItem.execute(ResumeActivityItem.java:51)
        at android.app.servertransaction.TransactionExecutor.executeLifecycleState(TransactionExecutor.java:145)
        at android.app.servertransaction.TransactionExecutor.execute(TransactionExecutor.java:70)
        at android.app.ActivityThread$H.handleMessage(ActivityThread.java:1823)
        at android.os.Handler.dispatchMessage(Handler.java:107)
        at android.os.Looper.loop(Looper.java:198)
        at android.app.ActivityThread.main(ActivityThread.java:6729)
        at java.lang.reflect.Method.invoke(Native Method)
        at com.android.internal.os.RuntimeInit$MethodAndArgsCaller.run(RuntimeInit.java:493)
        at com.android.internal.os.ZygoteInit.main(ZygoteInit.java:876)
     Caused by: java.lang.ClassCastException: s1 cannot be cast to kotlin.jvm.functions.Function1
        at d4.w.c.k.b(Intrinsics.java:5)
        at d4.w.c.k.a(Intrinsics.java:1034)
        at b4.f.a.c.b.a.b.f.a(EventBus.kt:12)
        at j4.a.a.a.u.l.onResume(BaseMenuActivity.kt:44)
        at xxx.StartActivity.onResume(StartActivity.kt:1)
        at android.app.Instrumentation.callActivityOnResume(Instrumentation.java:1413)
        at android.app.Activity.performResume(Activity.java:7292)
        at android.app.ActivityThread.performResumeActivity(ActivityThread.java:3803)
        at android.app.ActivityThread.handleResumeActivity(ActivityThread.java:3843) 
        at android.app.servertransaction.ResumeActivityItem.execute(ResumeActivityItem.java:51) 
        at android.app.servertransaction.TransactionExecutor.executeLifecycleState(TransactionExecutor.java:145) 
        at android.app.servertransaction.TransactionExecutor.execute(TransactionExecutor.java:70) 
        at android.app.ActivityThread$H.handleMessage(ActivityThread.java:1823) 
        at android.os.Handler.dispatchMessage(Handler.java:107) 
        at android.os.Looper.loop(Looper.java:198) 
        at android.app.ActivityThread.main(ActivityThread.java:6729) 
        at java.lang.reflect.Method.invoke(Native Method) 
        at com.android.internal.os.RuntimeInit$MethodAndArgsCaller.run(RuntimeInit.java:493) 
        at com.android.internal.os.ZygoteInit.main(ZygoteInit.java:876)

Ignore previous comment. 1.5.38 was not yet available and it somehow built with 1.5.37 instead of stopping compilation. I should check my build scripts. Will report back correct results in a few minutes.

Oh, no, that's a bad news. :( Perhaps, we may need to resort to invalidate lambda merging candidates if arity mismatches.

Sorry again for the false alarm. My build script does not properly handle dependencies errors, missing R8 did prevent the clean to run, so archiving task at the end did success and I did not see the failure in the middle.

1.5.38 is working for this issue.

Should I also test the other issue? Or wait for a future patch? Will do more pushed tests to ensure 100% of the app is working.

Great! You don't need to test for the other issue. Will keep you posted via that issue. Thanks!