How to get arround java.lang.verrifyError

Question

I am trying to generate a random number with the instrumented code. For that I have added following expression to generate a random number. But it is throwing a verify Error as given below. I have added a long variable and assign the above generated value. But I am getting following exception during the run time. But if I run

Accepted Answer

It seems you hit a bug in Javassist’s byte code generator. Here is the disassembled code as reported from the VerifyError. Note that due to the absence of the constant pool in the exception message, the actual target methods are guessed from your source code (but it looks plausible): 0 invokestatic [876] 3 lstore_3 4 new [878] guess: new StringBuilder 7 dup 8 invokespecial [879] guess: StringBuilder. 11 invokestatic [876] guess: System.nanoTime() 14 invokestatic [883] guess: String.valueOf(long) 17 invokevirtual [886] guess: StringBuilder.append(String) 20 invokestatic [892] guess: Math.random() 23 invokestatic [896] guess: Math.round(double) 26 invokestatic [883] guess: String.valueOf(long) 29 invokevirtual [886] guess: StringBuilder.append(String) 32 invokevirtual [897] guess: StringBuilder.toString() 35 invokestatic [902] guess: Long.valueOf(String) 38 i2l 39 lstore <5>… rest omittedNote that the method invocations match your source code fragment (being static or virtual as expected and being the same reference where supposed to be the same), whereas the instruction at location 38, the one rejected by the verifier, is a spurious i2l instruction (a conversion from int to long). At this place, an unboxing conversion from Long to long should happen. Since the following lstore instruction will store a long value into a local variable, it seems that Javassist got your local variable declaration right.Note that the first two instructions indicate that there is another instrumentation happening; since it’s invoking the same method as the instruction at location 11, which is supposed to be System.nanoTime, it seems the other instrumentation is intended to measure the method’s overall execution time. But that shouldn’t affect your instrumentation code.You may contact the authors of Javassist to find out whether Long unboxing should work. At the same time, you may workaround the problem by using Long.parseLong instead of Long.valueOf to avoid the boxing/unboxing in the first place. You may also simplify the operation from the generated code’s perspective by using String.concat instead of the + operator as this eliminates the need to deal with StringBuilder internally:Replacemethod.addLocalVariable("instMethod_correlationId", CtClass.longType);beforeBuilder.append("instMethod_correlationId = Long.valueOf(String.valueOf(System.nanoTime()) + String.valueOf(Math.round(Math.random())));");withmethod.addLocalVariable("instMethod_correlationId", CtClass.longType);beforeBuilder.append("instMethod_correlationId = Long.parseLong(String.valueOf(System.nanoTime()).concat(String.valueOf(Math.round(Math.random()))));");This should lower the requirements to Javassist’s compiling capabilities and workaround the bug triggered by your original code fragment.Update: for those, interested in decoding the hex dump of a VerifyError to a readable instruction sequence: Online Decoder using tio.run

Advertisement

Answer