Programming languages for geophysics students

I was recently asked which programming language a university should require their geophysics graduate students to learn. Here was my reply.

If I had to pick one language, I'd recommend Java. For an excellent appreciation of Java as a teaching language, see Doug Lea's web page at http://g.oswego.edu/dl/html/javaInCS.html

A second-place choice would be C with some exposure to C++. But if students know Java, they can pick up C/C++ on their own.

If I were teaching programming to students that had never programmed before, I'd use Python as a learning language. You won't waste time explaining tricky syntax. Python supports both procedural and object-oriented approaches. If I can get Java to compile, it usually does what I wants. The same is true of Python, but it usually compiles the first time.

Do not require Fortran. I have not written Fortran on the job for ten years now.

If a teacher is using Mathematica in class exercises, then teach what is needed to do the exercises, as you go. Mathematica is a software application, not a language. The same is true of Visual Basic and Perl. (Test: if only one "compiler" exists, then the language probably does not have a specification, or even a grammar.)

If faculty do not consider programming languages worth learning, then students will probably follow their example.

I now consider formal computer science to be essential training for any scientist who expects to program as a part of their career. Make them take courses on programming and read programming books. They will not have time to learn by trial and error, as most of my generation did. Students need courses in style, data structures, algorithms, encapsulation, object-oriented programming, functional/generic programming, and software patterns. Students need concepts more than the mechanics of any single language.

Do not choose a language because it is "realistic" for "real projects" in the "real world." Those criteria change. Teach them principles of programming that are likely to be around for a while. The language is often unimportant.

I personally wish I had been taught programming in Scheme (a Lisp dialect), using the freshman MIT book "Structure and Interpretation of Computer Programs," by Abelson and Sussman. No language has given me more insight into what a compiler really does.

Teaching C++ is tricky because you must identify a safe minimal subset. The best way to improve one's style in C++ is to write more Java.

Java is so easy to debug that I almost never use a debugger. Memory management is the source of most bugs in C/C++. Java prevents you from writing outside the bounds of an array, prevents you from casting a reference to the wrong type, prevents mischief with pointer arithmetic, prevents you from reading uninitialized variables, prevents you from allocating too little space for an object, prevents you from leaking unfreed memory, prevents bad assumptions about the size and byte order of primitive types, and so on. Those are the kind of problems that require a debugger and cost you days. The hardest problems to debug in Java are concurrency and race conditions with threads. Of course, most users of C/C++ never attempt threads.

Performance should be for advanced students, not novices. Two thirds of our new processing system is in Java. We use C++ if we need "machine code," mostly legacy code. There is not a single line of Fortran. We have seen Java code run 20% FASTER than equivalent C code. The latest Java virtual machines from Sun and IBM profile, optimize, and recompile during runtime.

Bill Harlan

June, 2001


Return to parent directory.