#752 "Inconsistent stackmap frames at branch target" in funky constructor

[Nonya Business]

Hello. I encounter the following exception in my application (a media player) after running ProGuard's optimizer on it:

Exception in thread "main" java.lang.VerifyError: Inconsistent stackmap frames at branch target 23
Exception Details:
  Location:
    ProGuardBug$Effect$Sepia.<init>(F)V @23: new
  Reason:
    Current frame's flags are not assignable to stack map frame's.
  Current Frame:
    bci: @8
    flags: { flagThisUninit }
    locals: { uninitializedThis, float }
    stack: { uninitializedThis, integer }
  Stackmap Frame:
    bci: @23
    flags: { }
    locals: { }
    stack: { uninitializedThis }
  Bytecode:
    0000000: 2a12 0159 44b8 0009 9a00 0f23 0b96 9b00
    0000010: 0923 0c95 9e00 0bbb 0007 59b7 000a bf12
    0000020: 02b7 0008 b1                           
  Stackmap Table:
    full_frame(@23,{},{UninitializedThis})
    full_frame(@31,{UninitializedThis},{UninitializedThis})

I reduced the problem to a short test case:

public final class ProGuardBug {
    public static void main(String[] args) {
        Effect effect;

        effect = new Effect("invert");
        System.out.println(effect.op);

        effect = new Effect.Sepia(0.5f);
        System.out.println(effect.op);
    }

    static class Effect {
        final String op;

        Effect(String op) {
            this.op = op;
        }

        static final class Sepia extends Effect {
            Sepia(float intensity) {
                super(createOp(intensity));
            }

            private static String createOp(float intensity) {
                if (Float.isNaN(intensity) || intensity < 0 || intensity > 1)
                    throw new IllegalArgumentException();
                return "sepia";
            }
        }
    }
}

The problem occurs in the "Sepia" constructor that calls a static helper method to do some processing, before invoking the superconstructor. If the optimization method/inlining/unique is turned off, the problem goes away.

I can work around it, so it's not serious, but I thought I ought to report it.

[Nonya Business]

I forgot to specify, this is Java 9u181 and ProGuard 6.1.0.

[Eric Lafortune]

Thanks for your detailed report. It turns out to be an edge case for the Java bytecode verifier. We've now fixed it for the upcoming ProGuard 6.1.1.

[Nonya Business]

The fix works. Thank you for the very fast service.