Wednesday, 29 September 2010

Simple and Elegant

One of the most common traits that I come across in us software developers is the tendency to over-engineer our code. We try to future-proof our code and we try to make it deal with scenarios  that may never even be a requirement for the system we are building. Recognising when this is happening and being able to create a simpler and more elegant solution is something we should all be striving towards.

As an example, I'm currently using some of my personal time to build a solution for classifying large amounts of information (more about this in a future post). I'm currently undertaking a TDD exploration of the core domain model. One of the requirements that I have identified is the ability for a user to define a set of properties that can be held against each piece of information. Each property has a name that maps to a value - a fairly simple concept.

As I was writing the test cases and domain classes for this requirement I drifted down a route of allowing typed property values. Thus, rather than just String values, they could be Int, Boolean, ThingaMaBob or whatever you want them to be. However, definitions of properties and assignment of values are managed via a user interface rather than in compiled code so this solution started to grow very complex as it dropped into a world of runtime type validations and conversions and so forth. 

As I wrote more and more code the solution began to smell more and more like over engineering (when you start needing Scala Manifests in a domain model you know something isn't quite right). I could see myself getting bogged down in getting this working, so I decided to take a step back and look at the requirement again. Was I sure that I would actually need typed property values? No, maybe in future but not for any specific cases I could currently identify. Was having typed property values allowing me to get a greater understanding of my domain model? No. Was the requirement for typed property values a valid one, or just an excuse for over engineering? The ultimate conclusion was that I could actually do everything I wanted for the foreseeable future with simple String values and some basic validation rules.

With my new found simplification in hand I dumped a whole load of complex code and ended up with a very simple, very elegant solution that works perfectly for my current needs.

Sadly, I all too often see developers go down this same route but fail to realise that they are over engineering. They then spend an excessive amount of time building a solution that is fragile, difficult to test and unresponsive to changes in requirements that might be different to those they originally imagined when building their solution. These over engineered solutions just contribute to the technical debt of projects that allow them to exist.

Always aim for the simplest and most elegant solution that meets the current set of requirements - nothing more, nothing less. Don't over engineer and don't try to predict future requirements. Note the emphasis on elegant: just because something is very simple doesn't mean it is a good solution. A simple solution can still be built from messy, poorly structured code that is just as difficult to maintain and enhance as its complex counterpart.

If we all focus on this goal of simple, elegant code then software becomes simpler, easier to understand and much easier to maintain and enhance. This then benefits us all through increased productivity and improved ability to deliver.

No comments:

Post a Comment