IssueTracker

TL;DR: minifier should stick to abstract method's name in a super type.

public class Base implements Runnable {
  public abstract a();
  public void run() {
    try { a(); } ...
  }
}

public class Sub1 extends Base {
  public final b() { ... } // should have been `a`
}

public class Sub2 ... // and more

------

All of affected classes came from "com.google.android.play:core", so I think it's fine to describe the whole details here.

First, here is the mapping of the problematic class:

com.google.android.play.core.internal.r -> b4.h.a.e.a.c.l:
    com.google.android.play.core.tasks.i a -> d
    1:1:boolean com.google.android.play.core.tasks.i.a(java.lang.Exception):0:0 -> run
    1:1:void run():0 -> run

------

The disassembled code of class `b4.h.a.e.a.c.l` is:

.class public abstract Lb4/h/a/e/a/c/l;
.super Ljava/lang/Object;

# interfaces
.implements Ljava/lang/Runnable;


# instance fields
.field public final d:Lb4/h/a/e/a/d/i;
    .annotation system Ldalvik/annotation/Signature;
        value = {
            "Lb4/h/a/e/a/d/i<",
            "*>;"
        }
    .end annotation
.end field


# direct methods
.method public constructor <init>()V
    .locals 1

    invoke-direct {p0}, Ljava/lang/Object;-><init>()V

    const/4 v0, 0x0

    iput-object v0, p0, Lb4/h/a/e/a/c/l;->d:Lb4/h/a/e/a/d/i;

    return-void
.end method

.method public constructor <init>(Lb4/h/a/e/a/d/i;)V
    .locals 0
    .annotation system Ldalvik/annotation/Signature;
        value = {
            "(",
            "Lb4/h/a/e/a/d/i<",
            "*>;)V"
        }
    .end annotation

    invoke-direct {p0}, Ljava/lang/Object;-><init>()V

    iput-object p1, p0, Lb4/h/a/e/a/c/l;->d:Lb4/h/a/e/a/d/i;

    return-void
.end method


# virtual methods
.method public abstract a()V
.end method

.method public final run()V
    .locals 2

    :try_start_0
    invoke-virtual {p0}, Lb4/h/a/e/a/c/l;->a()V
    :try_end_0
    .catch Ljava/lang/Exception; {:try_start_0 .. :try_end_0} :catch_0

    return-void

    :catch_0
    move-exception v0

    iget-object v1, p0, Lb4/h/a/e/a/c/l;->d:Lb4/h/a/e/a/d/i;

    if-eqz v1, :cond_0

    .line 1
    iget-object v1, v1, Lb4/h/a/e/a/d/i;->a:Lb4/h/a/e/a/d/k;

    invoke-virtual {v1, v0}, Lb4/h/a/e/a/d/k;->a(Ljava/lang/Exception;)Z

    :cond_0
    return-void
.end method

------

So, in the very first line of run(), it is calling its own abstract method `a`, which in turn raised AbstractMethodError at runtime. By adding:

    invoke-virtual {p0}, Ljava/lang/Object;->getClass()Ljava/lang/Class;
    move-result-object v1
    invoke-virtual {v1}, Ljava/lang/Class;->getName()Ljava/lang/String;
    move-result-object v1
    const-string v0, "[R8]"
    invoke-static {v0, v1}, Landroid/util/Log;->e(Ljava/lang/String;Ljava/lang/String;)I

as a prologue of run(), I can confirm all sub-types of class `b4.h.a.e.a.c.l` missed to implement `a`. Actually, they implemented, but R8 renamed them to a different name. Here is one example subtype:

com.google.android.play.core.internal.s -> b4.h.a.e.a.c.m:
    com.google.android.play.core.internal.r a -> e
    com.google.android.play.core.internal.q b -> f
    void a() -> b   # <------------- :(

.class public final Lb4/h/a/e/a/c/m;
.super Lb4/h/a/e/a/c/l;


# instance fields
.field public final synthetic e:Lb4/h/a/e/a/c/l;

.field public final synthetic f:Lb4/h/a/e/a/c/k;


# direct methods
.method public constructor <init>(Lb4/h/a/e/a/c/k;Lb4/h/a/e/a/c/l;)V
    .locals 0

    iput-object p1, p0, Lb4/h/a/e/a/c/m;->f:Lb4/h/a/e/a/c/k;

    iput-object p2, p0, Lb4/h/a/e/a/c/m;->e:Lb4/h/a/e/a/c/l;

    invoke-direct {p0}, Lb4/h/a/e/a/c/l;-><init>()V

    return-void
.end method


# virtual methods
.method public final b()V
    .locals 2

    iget-object v0, p0, Lb4/h/a/e/a/c/m;->f:Lb4/h/a/e/a/c/k;

    iget-object v1, p0, Lb4/h/a/e/a/c/m;->e:Lb4/h/a/e/a/c/l;

    invoke-static {v0, v1}, Lb4/h/a/e/a/c/k;->a(Lb4/h/a/e/a/c/k;Lb4/h/a/e/a/c/l;)V

    return-void
.end method

------

After renaming all those `b` in subtypes, I saw the app is working. Just to confirm, will send a manually modified apk. Also, will prepare some logging to see why we chose a different name.

You don't need to wait for logging. I just reproduced the issue on master.

The reproduction and fix is under review: https://r8-review.googlesource.com/c/r8/+/39060

Meanwhile, the attached is a patched R8 on 1.5.41

Thanks a lot I can confirm it works, full PlayCore update process tested (only immediate but I don't see a reason other mode don't work as the issue was at the start)

Thank you for the confirmation! The fix (mentioned at comment #4) is landed on master and 1.5 branch (1.5.42)

Any timeline when this will be released?

It will be included in beta05 when that is released. You can use a version of R8 that includes the fix by adding the following to your build.gradle:

buildscript {

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

    dependencies {
        classpath 'com.android.tools:r8:1.5.42'          // Must be before the Gradle Plugin for Android.
        classpath 'com.android.tools.build:gradle:X.Y.Z' // Your current AGP version.
     }
}

1.5.42 still have some potential major issue.

I'd not recommend using anything under 1.5.49. (Currently in prod with that one)