[Colloq] PhD Thesis Defense: Therapon Skotiniotis, Wednesday, April 28th at 9:00AM, Room 366

[kirsten at ccs.neu.edu]

Thesis Defense Presentation:
Speaker: Therapon Skotiniotis
Thesis title: Modular Adaptive Programming
Date: April 28, 2010 (Wed.) at 9:00AM
Where: 366 WVH

Abstract

Adaptive Programming (AP) provides advanced code modularization
for traversal related concerns in object-oriented programs.
Computation in AP programs consists of (i) a graph-based model
of a program's class hierarchy, (ii) a navigation specification,
called a strategy, and (iii) a visitor class with specialized
methods executed before and after traversing objects. Strategy
specifications abstract over graph nodes and edges allowing
modifications to the program's class hierarchy without affecting
visitor behavior.  Despite the benefits of AP there are also
limitations; hardcoded name dependencies between strategies and
the class hierarchy as well as non-modular adaptive code (strategies
and visitors). These limitations hamper adaptive code reuse and
make composition and extension of adaptive code difficult.


To address these limitations we define WYSIWYG strategies,
constraints and Demeter Interfaces. WYSIWYG strategies guarantee
the order of strategy nodes in selected paths simplifying the
semantics of strategies and leading to more predictable behavior.
We extend AP systems to enforce the interface between strategy
and visitors limiting visitor operations to strategy nodes making
dependencies between strategies and visitors explicit. Constraints
provide a new mechanism that allows programmers to define invariants
on the graph-based model of a program's hierarchy making programmer's
assumptions explicit and verifiable at compile time. Finally,
Demeter Interfaces provide (i) an interface between the program's
class hierarchy and both strategies and visitors, (ii) constraints
on the structure of a class hierarchy that implements a Demeter
interface and (iii) the ability to parametrize adaptive code.

While our combination of extensions to AP provide modular, reusable
and resilient adaptive programs, our definition of WYSIWYG
strategies also leads to a new simpler algorithm for calculating
all valid paths for a strategy and a straight forward code
generation process that results in a program whose size is
polynomial in the size of the strategy and the input program's
class hierarchy.


Thesis committee:

Prof. Karl Lieberherr (Thesis Advisor)
Prof. Mitchell Wand
Prof. Panagiotis (Pete) Manolios
Prof. Ralf Laemmel (University of Koblenz-Landau, Germany, external member)




Kirsten Anderson
Northeastern University
College of Computer & Information Science
202 West Village H
360 Huntington Ave
Boston, MA 02115
(617) 373-8685
(617) 373-5121