Rework the handling of inner classes.

[SpaceWalkerRS]

By convention inner class names are of the form com/example/Example$InnerName, which encodes both the outer class' name, as well as the inner class' inner name, but this is not required by the JVM spec. However, Stitch fully relies on classes' names following this convention, and if they do not, Stitch treats them as top level classes, even if the appropriate attributes are present. Similarly, if the inner and outer class attributes are not present, but the classes' names suggest the classes are nested, Stitch will treat them as such.

This PR fixes these issues by looking for the inner and outer class attributes to retrieve the enclosing class and inner name of a class, instead of relying on class names following the standard $ convention.

I ended up removing a significant part of JarReader, as it appeared to be dead code and it made it easier to add the new necessary parts. I'm just mentioning it here 'cause it's a fairly large change that isn't strictly related to the PR.

[liach]

I don't think you should track the declaring and enclosing class both with the same variable. Consider:

class TopLevel {
  Comparator<Integer> comparator = new Comparator() {...};
  class Inner {}
}

your model cannot distinguish between the two classes here; Inner has a non-null "outer class" (declaring class), while the anonymous comparator implementation class has an EnclosingMethod with a class reference but no method name and type reference.

[SpaceWalkerRS]

The distinction is that Inner will have a non-null "inner name", while the anonymous class will have its inner name set to null.

I do realize that I only look at the outer class attribute for the enclosing class' name, while for inner classes that attribute isn't present so I need to get it from the inner classes attribute.

[liach]

No:

class TopLevel {
  static {
    class Local {}
  }
  class Inner {}
}

Should've given this example first and foremost

[SpaceWalkerRS]

Huh, I assumed that would give <clinit> as the enclosing method, but that is not the case. You're right, I should track the enclosing/declaring class separately to properly be able to distinguish between inner and local classes.

[SpaceWalkerRS]

I think that's fixed now

[liach]

Suggested change

	this.allClasses = new TreeMap<>(Comparator.naturalOrder());
	this.allClasses = new TreeMap<>();

[liach]

This will return true for this Local:

class TopLevel {
  static { class Local {} }
}

and since this method isn't used and stitch is not an api, I recommend removing this method.