czwartek, 27 czerwca 2013

A reflection on the future of programming languages

I started my first job as a programmer in 2000, a year before I got Master's degree in Computer Science. At that time, a recurring question that was asked by students and graduates was:

Which language should I learn in order to be successful on the market?

It seems like this question is universal and has always been asked, in some cases inquiring about a programming language, another time on a library or a framework.

So, will learning Java pay off better than C++? Does it even make sense to learn C these days? Or will mastering the Java Servlet Pages position one better on the market than learning PHP?

The simple and provocative answer is: it does not matter.

The longer answer is really the rest of this post, so if you are interested, read on.

An interesting phenomenon in the history of software is that many great libraries and products have been created by individuals or, at most, tiny groups of people. Consider these examples - if any of the products seems unfamiliar, read about it on Wikipedia.

Chipmunk                 ---------          Scott Lembcke
jQuery Form Plugin       ---------          Mike Alsup
Clojure                  ---------          Rich Hickey
FileZilla                ---------          Tim Kosse
Python                   ---------          Guido van Rossum
web2py                   ---------          Massimo Di Pierro
PHP                      ---------          Rasmus Lerdorf
Emacs                    ---------          Richard Stallman and Guy Steele
Gtk                      ---------          Spencer Kimball and Peter Mattis
JUnit                    ---------          Erich Gamma and Kent Beck

We all use these daily to test, develop and deliver software. Sometimes they become part of our products. They have not initially been written by an large team or open source community and that fact is not without meaning in terms of programming language that the authors used. This statement is going to be a bit speculative on my part, but I imagine the authors have not had terrible doubts in terms of which language to choose and why. Working alone or with just two colleagues they could decide on the language based on their own preference. It was only later that developers from open source community joined these projects, also based on their preference of language and domain.

Now, speaking between us, the humble programmers, we have to face the fact that very few of us (< 90%) will be next Rossums and next Becks. We will use the excellent works of others and integrate them into a working products using languages that happen to have bindings to these libraries. Therefore, the choice of the language, in a particular situation, will depend very strongly on the availability of library bindings and then secondly on current fashion, our legacy code base and other factors. This means we must be able to write clean code in at least a handful of languages for which there is really good library choice. We have to acccept this variety and work towards mastering our coding skills, rather than towards mastering one given language.

If there is a very good, stable library already available, written, say, in C, there will likely be at least a couple of bindings created for languages other than C. Re-writing this 10-year old library from scratch in "pure X language" would most often be impractical and too costly to undertake. The library itself may well outlive some languages and the new ones that will pop up in the future will create their own bindings for it. It may even outlive the language it is itself written in, in the sense that the library is widely used, although the language disappeared from the mainstream long ago.

Of course it is not completely true that the widely used libraries are never created from scratch in more than one language. Take SNMP for example - we have at least two different implementations (not bindings), namely net-snmp written in C and snmp4j - in Java. For that matter, Java has demonstrated something that I strongly doubt we will ever witness again - the proliferation of pure Java libraries for almost everything we already had had C/C++ libraries in place. But today, these Java libraries are also being used from within newer languages like Scala or Clojure. These new languages also create their bindings, taking advantage of the JVM, instead of re-implementing all Java heritage from scratch.


With all that said, I think we should not treat the discussions like "Is Scala better than Java" or "Why C++ is not truly object oriented language" too seriously. Some languages are definitely more expressive than others and some support given programming paradigms better than others, but from the point of view of delivering what the customer needs quickly, these features are less important than the availability of libraries and frameworks.

We can hear from time to time the statement that our software industry is young, compared to other industries like automotive, for example. That's true and this is a very important observation, in my opinion. I believe in the next few tens of years, as the industry matures, some of the well-established libraries will settle for good, just like tools and techniques used to produce cars had settled in automotive industry. If we are tasked with developing some new functionality, we will know there are, let's say, two libraries and four languages that we can use exactly for that purpose. And we will use them. For a majority of customer requirements, the library choices will be as obvious, as today is mounting an ABS module in a car in order to provide the required safety level during braking.