• Static variables and binding refs can employ qualified names to refer to variables that otherwise would be out of scope. When refactorings are moving/creating code, the qualification of names may need to be adjusted in order to bind variables, or may benefit from shortening.
  • Inlined or moving (pushing up or down) methods may need to change static variable qualification
  • Expanding pools/imports is not required when pushing down methods, as pools (and now imports) have always been inherited in Dolphin.
  • Methods can be extracted to component with class var refs if qualified without needing to abstract or add imports
  • Offer the user the option to import or use qualified names when pushing methods up.
  • Abstracting class variables needs to be performed globally, not just in the hierarchy
  • Pulling-up/pushing-down class variables may affect qualified variable references and require global method rewrites
  • When static access is needed because of code movement, abstracting is only needed for private variables. Other access can be qualified.
  • Don't want to generate setters for class constants - writes to these are, of course, compilation errors
• ChildrenToSiblingsRefactoring, etc, fails to push up methods with custom namespace: Refactorings that move methods need to preserve the package of the original method in case it is compiled into a package default namespace.
• Fixed a disastrous performance problem in PushUpMethodRefactoring. The algorithm to check sibling supersends is needlessly quadratic in relation to the total number of classes in the superclass hierarchy, so attempting to push up a method to Object takes literally minutes to check the preconditions. This is due to a recursive function using allSubclasses where it should be using subclasses, as it already recurses down the hierarchy.
• Noticed that the supposed breadth-first traversal of all subclasses is not actually a breadth-first traversal. It visits all the immediate subclasses, but then visits all the subclasses of the first subclass before visiting any of the second subclass. This is a kind of hybrid with depth first. There are consequences in fixing this that need to be worked through, so for the moment just fixing in RBAbstractClass by reverting it to its original allSubclasses implementation (which is actually breadth-first).

Builds on work in 77647b5