Bytecode layout deviations from 2.11 related to synchronized

[retronym]

https://gist.github.com/retronym/919cd6279eae31db2958

I haven't benchmarked to see if this makes a difference in C1 / C2 performance, but it might be worth replicating the status quo here and stick the exception handlers at the end of the method.

I noticed the difference in the bytecode of newTermName, which was showing up in my 2.12.x profiling.

[retronym]

Here's what javac does:

⚡ javac -d . sandbox/Test.java && javap -c Test
Compiled from "Test.java"
class Test {
  Test();
    Code:
       0: aload_0
       1: invokespecial #1                  // Method java/lang/Object."<init>":()V
       4: return

  public java.lang.String foo();
    Code:
       0: ldc           #2                  // String
       2: ldc           #2                  // String
       4: invokevirtual #3                  // Method java/lang/String.contains:(Ljava/lang/CharSequence;)Z
       7: ifeq          26
      10: aload_0
      11: dup
      12: astore_1
      13: monitorenter
      14: aload_0
      15: invokevirtual #4                  // Method java/lang/Object.toString:()Ljava/lang/String;
      18: aload_1
      19: monitorexit
      20: areturn
      21: astore_2
      22: aload_1
      23: monitorexit
      24: aload_2
      25: athrow
      26: aload_0
      27: invokevirtual #4                  // Method java/lang/Object.toString:()Ljava/lang/String;
      30: areturn
    Exception table:
       from    to  target type
          14    20    21   any
          21    24    21   any
}

Another deviation is whether or not the exception type is declared as any or java.lang.Throwable.

[retronym]

Merging with #233