java.lang.ClassFormatError: StackMapTable format error: bad offset for Uninitialized in method Test.run()V

[laurit]

ClassFormatError happens when advice is applied to a method that

• has a leading stack frame
• creates a new value with the first instruction
• leaves the newly create value uninitialized and performs a jump
  Seems like bytebuddy inserts an extra nop before original method code and the uninitialized value in stack frame points to that nop instead of the following new instruction.

import static net.bytebuddy.matcher.ElementMatchers.named;

import java.lang.invoke.MethodHandles;
import net.bytebuddy.agent.ByteBuddyAgent;
import net.bytebuddy.agent.builder.AgentBuilder;
import net.bytebuddy.asm.Advice;
import net.bytebuddy.jar.asm.ClassVisitor;
import net.bytebuddy.jar.asm.ClassWriter;
import net.bytebuddy.jar.asm.Label;
import net.bytebuddy.jar.asm.MethodVisitor;
import net.bytebuddy.jar.asm.Opcodes;

public class Main implements Opcodes {

  public static void main(String... args) throws Exception {
    new AgentBuilder.Default().type(named("Test"))
            .transform(new AgentBuilder.Transformer.ForAdvice().advice(named("run"), RunnableAdvice.class.getName()))
            .asTerminalTransformation()
            .installOn(ByteBuddyAgent.install());

    ClassWriter cw = new ClassWriter(ClassWriter.COMPUTE_MAXS);
    ClassVisitor cv = cw;
    cv.visit(
            V9,
            ACC_PUBLIC | ACC_SUPER,
            "Test",
            null,
            "java/lang/Object",
            new String[] { "java/lang/Runnable" });

    {
      MethodVisitor mv = cv.visitMethod(ACC_PUBLIC, "<init>", "()V", null, null);
      mv.visitCode();
      mv.visitVarInsn(ALOAD, 0);
      mv.visitMethodInsn(INVOKESPECIAL, "java/lang/Object", "<init>", "()V", false);
      mv.visitInsn(RETURN);
      mv.visitMaxs(0, 0);
      mv.visitEnd();
    }

    {
      MethodVisitor mv = cv.visitMethod(ACC_PUBLIC, "run", "()V", null, null);
      mv.visitCode();
      Label newInsn = new Label();
      //XXX ClassFormatError does not happen without this line
      mv.visitFrame(F_SAME, 0, null, 0, null);

      mv.visitTypeInsn(NEW, "java/util/ArrayList");
      Label jump = new Label();
      mv.visitJumpInsn(GOTO, jump);
      mv.visitFrame(F_FULL, 0, null, 1, new Object[] { "java/lang/Throwable" });
      mv.visitInsn(NOP);
      mv.visitInsn(NOP);
      mv.visitInsn(NOP);
      mv.visitInsn(ATHROW);
      mv.visitLabel(jump);
      mv.visitFrame(F_FULL, 0, null, 1, new Object[] { newInsn });
      mv.visitInsn(DUP);
      mv.visitMethodInsn(INVOKESPECIAL, "java/util/ArrayList", "<init>", "()V", false);
      mv.visitVarInsn(ASTORE, 1);
      mv.visitInsn(RETURN);
      mv.visitMaxs(0, 0);
      mv.visitEnd();
    }

    cv.visitEnd();
    byte[] bytes = cw.toByteArray();
    Class<?> c = MethodHandles.lookup().defineClass(bytes);
    Runnable r = (Runnable) c.getDeclaredConstructor().newInstance();
    r.run();
  }

  public static class RunnableAdvice {

    @Advice.OnMethodEnter(suppress = Throwable.class)
    public static void enter() {
      System.err.println("enter");
    }

    @Advice.OnMethodExit(onThrowable = Throwable.class, suppress = Throwable.class)
    public static void exit() {
      System.err.println("exit");
    }
  }
}

[raphw]

Your example is running fine on the latest version of Byte Buddy. What version did you use?

[laurit]

1.11.16 with jdk-11.0.9

[raphw]

I just executed the example with 1.0.11 from Adopt OpenJdk and the problem does not occur.

Are you using the Eclipse compiler?

[raphw]

My bad, I put the test in a package that Byte Buddy ignores by default.

[raphw]

I see the problem. With these frames without instructions following, Byte Buddy currently adds a NOP instruction to allow adding a frame after another frame. It's unfortunately not allowed to add two frames without an instruction in between. This instruction does however come before the label which is otherwise supposed to come right before the instruction that is referenced. As a consequence, the label now points to the NOP instruction what is not legal.

I will need to see if I can avoid the NOP instruction and rather merge two subsequent frames. Alternatively, I would need to rewrite labels that are placed between two such padded frames.

[raphw]

I am wondering, though. Should the label newInsn not be visited? How does this even work?

[laurit]

It should, that is a bug in the sample. I guess it doesn't matter because asm reads bytecode offset from the label which for unvisited label is 0 which in this case would also be the value when this label was visited.

[raphw]

That's what I figured out and adjusted for a test case in BB. I already fixed the issue but will add some additional testing before merging to master.

Thanks for the report.

[raphw]

Fixed on master.