Traversal code is ubiquituous in software, polluting the real intent of programs with accidental details of the current structure in which the intent is expressed.
We present a novel model, called Functional Adaptive Programming (AP-F), for processing objects that supports improved software variability. The software variability is improved by using three approaches: (1) a novel separation of generic concerns (i.e., a functional model for processing objects), (2) constraint-controlled automatic adaptation, and (3) programmer-controlled, crosscutting adaptation using two aspect mechanisms (multi-methods and strategies). Regarding (1), object processing is broken down into three major concerns as in Adaptive Programming (AP): ClassGraph, WhereToGo, WhatToDo. The ClassGraph defines the structural relationships between the classes. WhereToGo uses a high-level traversal language a la AP. WhatToDo is different than in AP and has three subconcerns: Down, Leaf, Up. WhatToDo uses functional-style advice on a general traversal with strategies, inspired by, but different from a visitor. While a visitor uses the traditional Law of Demeter: "Talk only to your friends", WhatToDo uses the dual: "Listen only to your friends". This creates a flow of information through the traversal, separated into what information flows Down, is created at Leafs and flows Up.
Regarding (2), a type checker checks constraints that the multi-methods must satisfy, using strategies to constrain applicable ClassGraphs. Regarding (3), multi-methods and strategies are used to adapt generic WhatToDos in a powerful way: a small change to the program may have a broad effect cutting across the system being built.
A system that supports software variability should have two properties: easy creation of class-graph-generic (or datatype-generic) programs and powerful adaptation and composition capabilities to specialize generic programs. AP-F has both, with advantages over other class-graph-generic programming approaches such as SYB (Scrap Your Boilerplate), Strategic Programming, Polytypic Programming and Generalized Folds. A key feature of AP-F is that it smoothly integrates high-level traversal control with the other desirable features leading to a superior way to abstract a large variety of traversals. This simplifies programs because all traversal-related calls become implicit and they are no longer polluting the real intent of programs.
Examples will be shown in two implementations of AP-F: A Java version, called DemeterF and a Scheme version, called apf-lib.
Joint work with Bryan Chadwick, Ahmed Abdelmeged, and Therapon Skotiniotis.