Wednesday 25 January 2012

We need more than one programming language

I teach Haskell as a programming language to our undergraduates. I'm sure it's a continual subject of debate in computer science department coffee rooms up and down the country: "Which programming languages should we teach in our CS degrees?" The module that I teach is called "Concepts in Programming", and the idea is that there are indeed concepts in programming that are fundamental to many languages, that you can articulate the differences between programming languages and that these differences give each language different characteristics. Differences such as the type system, the order of evaluation, the options for abstraction, the separation of data and functions.

"We need more than one" is the title of a paper by Kathleen Fisher, a Professor in Computer Science at Tufts University. Her short, eloquent paper describes why we will never have just one programming language ("because a single language cannot be well-suited to all programming tasks"). She has had a career spanning industry and academia, has been the chair of the top programming language conferences (OOPSLA, ICFP), and has been the chair of the ACM Special Interest Group on Programming Languages. Her paper On the Relationship between Classes, Objects and Data Abstraction tells you everything you ever needed to know about objects. She knows about programming.

Her recent work has been looking at that problem of how to read in data files when the data is in some ad-hoc non-standard representation (i.e. not XML with a schema, or CSV or anything obvious). So we all have to write parsers/data structures to fit these data files. We write a program in a language such as Perl. She says "The program itself is often unreadable by anyone other than the original authors (and usually not even them in a month or two)". I've been there and done that.

And when we've written another new parser like this for the umpteenth time (especially in the field of bioinformatics) we begin to wonder "How many programs like this will I have to write?" and "Are they all the same in the end?" and Kathleen's paper The Next 700 Data Description Languages looks at just that question. What is the family of languages for processing data and what properties do they have? I love the title of this paper because it instantly intrigues by its homage to the classic 1966 paper The Next 700 Programming Languages by Peter Landin. (To see why Peter Landin's work on programming languages was so important, the in memorium speech for Peter Landin given at ICFP 2009 is well worth a listen, and also the last 3 mins of Rod Burstall's speech discusses this paper in particular).

So perhaps, in computer science, we're doomed to keep inventing new programming languages, which wax and wane in popularity over time: Lisp, Fortran, C, C++, Java, Ruby, Haskell, F#, Javascript and so on. But as computer scientists we should be able to understand why this happens. They're all necessary, all useful and all part of a bigger theoretical picture. We need more than one.

This is my entry for the BCSWomen Blog Carnival.

No comments:

Post a Comment