Why not Python

Tue 15 May 2007 Apalala  | : caudex

Tags :


2010-03-12 Amendment: After doing the research right, my conclusion is that the tools*are*available for Python, that Python is an excellent software development platform, and one that allows more productivity than, say, Java.

This article is about programming languages in the context of professional software work. While considering several programming languages with very attractive features for my next project I had to take a pause and think about what the objective choice criteria should be. My conclusion, as argued below is that programming language choice should be dictated by programming pragmatics.

I've been programming for some thirty years now, twenty of them doing it for a living. In my work I've come well acquainted with several programming languages, including Fortran, Basic and Visual Basic, Pascal and Object Pascal, C, C++, Java, and C#. During my university training I had to deliver assignments in Concurrent C, Simula, LISP, and Prolog, among others. I specialized in Computer Languages (compilers) and became a professor of the subject later. I have studied the innards of programming language implementation and have implemented several programming languages from scratch. I've studied every promising programming language I have come upon, including AWK, Perl, Python, CAML, Haskell, Nemerle, Groovy, and Boo, among others.

My experience with compilers in particular has brought me some understanding into what some call "programming language beauty", but that subject is better dealt with elsewhere. Here I want to talk about what makes a programming language work.

By "makes it work" I mean which features of a programming language let you get the job done in an efficient way. By "in an efficient way" I mean in a way that lets you do better than just survive in the ever more competitive field of software development.

It turns out that what makes a programming language work has less to do with the idiosyncrasies of the language and much more to do with how the language allows and provides for a good software writing environment. The first breakthroughs in programming language, in Fortran, LISP, and Algol68, consisted in raising the level of abstraction so programs could be more about the solution space than the machine. But the first breakthrough I remember, at least in the context of programming languages that work, came with the Integrated Development Environment (IDE), the first comercialization of which was provided by Borland for its Turbo Pascal programming language.

Turbo Pascal came with an IDE that included an interactive, full-screen editor, an incredibly fast compiler, and a debugger. It was the first time that a generally-available tool reduced the edit-compile-run-debug cycle to a matter of seconds. Today, IDEs are common-place, and few professional programmers would think of doing any significant work without the aid of one.

But what do IDEs have to do with specific programming languages? Turbo Pascal was a simple, structured, and well defined language, and that allowed Borland to build a very fast compiler that fitted in the few kilobytes of RAM remaining in the machines of the time after having loaded an editor.

By the end of the 1980's Apple's MacIntosh had made the advantages of Graphical User Interfaces (GUIs) for users obvious, and GUIs became not just desireable but a necessity. The common language of choice at the time was C. Programmers spent hours dealing with single event-handling functions (like WindProc) and trying to manage modal contexts through innumerable heap structures and casts, while loading and exiting the GUI environment to run a command-line compiler. The next breakthrough came again in the IDE. GUI designers had been around for a while, but the linking between UI elements and event-handling code had to be handcrafted. In 1991 Microsoft launched Visual Basic. Visual Basic was built around an IDE that was itself graphical and that, besides edit-compile-run-debug allowed hooking UI elements and events to individual event handlers with a few clicks. Writing GUIs suddenly became easy. The language feature that allowed for it all was Visual Basic's component model, a form of object orientation.

As computers became more powerful, the size and complexity of the programs we attempted to write increased. Programmers spent an important part of their time studying and trying to memorize Application Programming Interfaces (APIs) and third party libraries. The next breakthrough came again in the IDE in the form of context-sensitive help, and later in coding assistance. Instead of having to browse over a huge API reference, the press of a key would make the IDE produce the exact piece of documentation pertinent to the context in the editor, or provide assistance in filling out a method call.

The language features that enabled the new IDE functionality were static-typing and object- orientation. They are the same features that today enable automated refactorings, a functionality that is indispensable while we try to tackle even more complex programs with iterative and agile methods.

So, Which are the features a working programmer should look for in a programming language? They are the ones that enable good tool support for the act of programming: static-typing, and object-orientation. But wait... Shouldn't the tools actually be available? Of course! This means that the primary criteria for choosing a programming language is the availability of the tools that make us efficient at writing the programs we have to write. The breakthroughs in programming of today are still happening in the tooling.

Fans of programming languages without those features or without the tools available say that their language of choice allows them to think better and write less code. That criteria is completely a non-issue for the simple reason that modern programming language are converging to include every characteristic that programming experience has identified as useful: first-order and anonymous functions, closures, list and map literals, partial classes and AOP features, type inference, and duck typing. You name it. None of then offset the pragmatic need for a complete development environment that makes us competitive at what we do.