Sunday, September 13, 2009

Language wars!

This one gets all the flames when development teams meet to decide on what language to use for the next project. The crusty old Unix guy in the corner says "C, it was good enough for Dennis and Ken, it's good enough for us." The Microsofty says "C++ of course. C is for Neanderthals." The J2E guy says, "Why does anybody want to use those antiquated languages full of stack smash attacks and buffer overflows anyhow? Write once, run everywhere!" And finally, the Python/Ruby guy says, "look, I can write a program all by my lonesome in two weeks that would take months for a whole team to write in Java, why is there even any question?"

And all of them are wrong, and here's why: They're talking about technologies, when they should be talking about the problem that needs solving and the business realities behind that.

I'll put myself pretty firmly in the Python/Ruby camp. My first Python product hit the market in 2000, and is still being sold today. My most recent project was also written primarily in Python. I also had a released product written largely in Ruby, albeit that was supposed to be only a proof-of-concept for the Java-based final version (but the proof of concept shipped as the real product, huh!). Still, none of these products are in Python or Ruby because of language wars, and indeed these products also included major portions written in "C" that did things like encryption, fast network file transfer, fast clustered network database transactions, and so forth. Rather, the language chosen was chosen because it was the right tool for the job. The first product was basically a dozen smaller projects written in "C" with a Python glue layer around them to provide a good user interface and database tracking of items. The second product, the one prototyped in Ruby, was prototyped in Ruby because Ruby and Java are quite similar in many conceptual ways (both allow only single inheritance, both have a similar network object execution model, etc.) and it made sense to prototype a quicky proof-of-concept in a language similar to what the final product would be written in. The last project was written in Python because Python provided easy-to-use XML and XML-RPC libraries that made the project much quicker to get to market, but also included major "C" components written as Unix programs.

So, how do you choose what language to use? Here's how NOT to choose:

  1. The latest fad is XYZ, so we will use XYZ.
  2. ABC programmers are cheap on the market, so we'll use ABC.
  3. DEF is the fastest language and is the native language of our OS, so we'll use DEF.
  4. I just read about this cool language GHI in an airline magazine...
Rather, common sense business logic needs to be used:
  1. We need to get a product out the door as quickly as possible to beat the competition to market.
  2. We need tools that support rapid development of object and component-oriented programs that are easy to modify according to what user feedback says future versions will need .
  3. Performance must be adequate on the most common platforms our customers will be using
  4. Whatever tools we use must allow a reasonable user experience.
The reality is that if you're trying to get a product to market as quickly as possible, you want to use the highest level language that'll support what you're trying to do. Don't write Python code when you can describe something in XML. Don't write Java code where Python will do the job. Don't write C++ code where Java will do the job. Don't write "C" code where C++ will do the job. Don't write assembly code where "C" will do the job. In short, try to push things to the highest level language possible, and don't be ashamed to mix code. I once worked on a product that had Ruby, Java, and "C" components according to what was needed for a particular part of the product set. There were places where Ruby lacked functionality and its performance was too poor to handle the job, for example, but Java would do the job just fine. And there was places where absolute performance was needed or where we were interfacing to low-level features of the Linux operating system where we went straight to "C", either as a network component accepting connections and data in a specified format, or as a JNI or Ruby equivalent.

The whole point is to get product out the door in a timely manner. If you decide, "I will write everything in 'C' because it is the native language of Unix and my product will be smaller and faster", you can get a product out the door... in three years, long after your competition's product hits the market and gains traction. At that point you'll be just an also-ran. What you have to do is get product out the door, get it out the door as quickly as possible with the necessary functionality and features and performance (not theoretical best, but "good enough"), and then work on getting traction in the marketplace. Perfection is the enemy of good enough, and seeking perfection often doesn't even produce a product that's any closer to perfection than the product originally written to be "good enough". That program that took three years to write? That company went bankrupt, and one reason was because the constant search for perfection ended up with a product that was inflexible, difficult to modify to work for different problem sets, and, frankly, poorly architected. Seeking the smallest memory footprint and theoretical best performance resulted in a product that failed in the marketplace because they missed the main reason we're writing programs in the first place: To meet customer needs. A program whose architecture is about memory footprint and performance at all costs is unlikely to have an architecture capable of being changed to meet changing customer needs, so not only were they late to market -- their product sucked. And the hilarious thing is, they didn't even manage to achieve the good performance their head architect claimed would happen with their architecture! So much for the three year "write everything in C" plan... late to market, poor performance, hard to modify, and they claim that those of us who advocate "good enough" rapid application development in the highest-level language feasible are "sloppy" and "advocating non-optimal practices"? Heh!

Next up... I talk about tools, and the odd story of how I got into the Linux industry in the first place. Note - I was the guy against using Linux in our shop. But that's a tale for another post :).


No comments:

Post a Comment