CS 609 Diskussionsseminar: Modular Extensibility
|Weekly meeting:||Mondays, 16:30 – 17:45|
|Seminarraum V D5|
|Course web page:||http://www.staff.uni-marburg.de/~walkings/teaching/cs609-su14/|
|(This document, which will be updated throughout the course.)|
A property of good software is the ability to add new features in a modular way. That is, it should be possible to (1) add new features without making invasive changes to the existing code, and (2) keep the code that implements each new feature together. Modular extensibility makes software more maintainable by keeping separate concerns separated in the code; it increases flexibility by making software easier to extend in new and unforeseen ways; and it supports customizability by making it easier to add and remove features according to the needs of different users.
Many programming languages provide good support for some kinds of modular extensions but not others. Most significantly, object-oriented languages and functional languages support modular extensibility in orthogonal dimensions. The challenge of supporting all kinds of modular extensions has been called the “expression problem” or the “tyranny of the dominant decomposition”.
In this seminar we will study modular extensibility, focusing especially on the design patterns and language features that address the expression problem. Each week we will meet to read and discuss one paper. Our working language will be English.
In addition to the general goals of a discussion seminar (exposure to research, practice reading and discussing papers, etc.) the specific goals of this seminar are to:
- Understand the specific concept of modular extensibility, how it can be achieved, and its significance in language design and software engineering.
- Develop some working definitions of more general concepts like modularity and expressiveness that arise in discussions about languages and software.
- Be able to recognize and evaluate the tradeoffs between different solutions to the expression problem.
- Practice using one of these solutions in your own work, in order to develop a deeper understanding of the benefits and limitations of the approach.
Structure of Course
Throughout the semester, the seminar will be conducted as a reading group. Additionally, students taking the seminar for credit will use one of the techniques or languages discussed in the seminar in a small programming project of their choosing. Each student will write a short paper and give a presentation of their project.
Phase 1: Reading Group
Each week I will assign a paper and name a discussion leader for our next meeting. All participants should read the assigned reading each week.
If you are not the discussion leader, you should submit 2-3 discussion topics or questions to the discussion leader and to me. This is intended to encourage you to think critically about the paper before you arrive at the meeting, and so will hopefully lead to more interesting discussions.
If you are the discussion leader, you should use the submitted questions to guide your preparation. Most importantly, you should dig deeper in order to answer questions or provide additional insights. This will usually require looking into related work.
During the discussion, the discussion leader will briefly summarize the main points of each section, pose topics for discussion, attempt to answer questions, offer additional insights, and keep the discussion on track. Everyone else should come prepared to discuss the paper! Not only will this be more lively and fun, it will make the discussion leader’s job much easier.
To summarize, your responsibilities during this phase of the seminar are to:
- Read the assigned reading each week.
- Submit discussion topics/questions to me and the discussion leader 24 hours before each meeting.
- Attend and participate in the discussions.
- Prepare for the discussions that you will lead.
Phase 2: Final Project
The final project is an opportunity to either (1) explore in more depth (and practice using) a language or design pattern from the seminar that you found interesting, or (2) learn independently about some techniques that we didn’t have time to discuss. You will submit a short paper and give a short presentation to share your insights with the rest of the group.
Schedule and Reading Assignments
Parnas, On the Criteria to be Used in Decomposing Systems into Modules, CACM 1972
Argues for “information hiding” as the most important role of modules. Expresses motivations for what would eventually become object-oriented programming.(Discussion leader: Eric)
Wadler, The Expression Problem, email 1998
Coins the name of the expression problem and establishes some criteria for solutions.(Discussion leader: Eric)
Palsberg and Jay, The Essence of the Visitor Pattern, COMPSAC 1998
Discusses the “visitor pattern”, which inverts the expression problem for object-oriented programs, making it easy to add new operations but difficult to add new cases. Introduces the “Walkabout” class as a reusable solution that is extensible in both dimensions but relies on reflection and is not type safe.(Discussion leader: Bastian)
Zenger and Odersky, Extensible Algebraic Datatypes with Defaults, ICFP 2001
Describes the “extensible visitor pattern” as a solution to the expression problem. For background, you may be interested in the original extensible visitor pattern paper by Krishnamurthi, Felleisen, and Friedman: Synthesizing Object-Oriented and Functional Design to Promote Re-Use, ECOOP 1998.(Discussion leader: Felix)
Janzen and De Volder, Programming with Crosscutting Effective Views, ECOOP 2004
Describes a tool that presents two different, editable views of the same program, to support extending with new cases or new operations. Shows how good tool support can provide a partial solution to the expression problem.(Discussion leader: Elena)
Clifton et al., MultiJava: Modular Open Classes and Symmetric Multiple Dispatch for Java, OOPSLA 2000
Extends Java with open classes and multi-methods. Open classes support extending existing classes with new methods, while multi-methods provide a solution to the “binary method problem”.(Discussion leader: Eric)
Kästner, Apel, and Ostermann, The Road to Feature Modularity?, SPLC 2011.
Discusses modularity and extensibility from the perspective of feature-oriented software development. A unique challenge in this context is that it must be possible not only to add new features to a program, but to selectively enable and disable them.(Discussion leader: Iratxe)
Oliveira and Cook, Extensibility for the Masses: Practical Extensibility with Object Algebras, ECOOP 2012
Describes an elegant design pattern for solving the expression problem in OO languages with generics.(Discussion leader: Jona)
Lämmel and Ostermann, Software Extension and Integration with Type Classes, GPCE 2006.
Shows how Haskell’s type classes can be used to (partially) solve many of the modularity challenges we have discussed during this seminar, including the expression problem and the binary method problem.(Discussion leader: Eric)
Löh and Hinze, Open Data Types and Open Functions, PPDP 2006.
Discusses an extension to Haskell that allows data types and functions to be declared open, allowing them to be modularly extended with new cases later. This is analogous in many ways to the work on MultiJava, but starting from the functional perspective.(Discussion leader: Yi)
Final project presentations
A list of other interesting and relevant papers is provided below.
More Solutions to the Expression Problem
Apel, Kästner, Lengauer, Feature Featherweight Java: A Calculus for Feature-Oriented Programming and Stepwise Refinement, GPCE 2008
Carette, Kiselyov, and Shan, Finally Tagless, Partially Evaluated: Tagless Staged Interpreters for Simpler Typed Languages, JFP 2009
Swiestra, Data Types à la Carte, JFP 2008.
Zenger and Odersky, Independently Extensible Solutions to the Expression Problem, FOOL 2005
Cook, On Understanding Data Abstraction, Revisited, OOPSLA 2009
Ostermann et al., Revisiting Information Hiding: Reflections on Classical and Nonclassical Modularity, ECOOP 2011