19 December 2007

Code Noise Ratio

Using Scott Hickey's article on reducing code noise in Groovy as a starting point, can we measure of the noisiness of a programming language and environment? Let's say that the quietest code where the developer has to add the least amount boilerplate code to implement a particular feature. Just to keep the metric simple, let's just measure the size of the source files for different implementations of the same feature and assume that the developer is trying to write sensible code. We assume that the shortest version is the quietest and calculate the ratio between the shortest version and all other versions.

Let's test this ratio on several versions of Hello World implemented using different scripting languages in previous articles. What is the noisiness of each implementation?

Groovy + SwingBuilderJava2400%

What does this table tell us about reducing code noise for small programs?

  • Script environment should pre-import common classes. Groovy + SwingBuilder and plain Groovy makes it very easy to write a small GUI program because all the Java Swing references are pre-imported. The Jython and IronPython implementations are nearly the same but the IronPython version is longer because it has to load .Net references.
  • Use class aliases. One reason the PowerShell version is very long is you can't add a class name into the current namespace (such as Python's from <library> import <class> or C# using namespace <blah>). You can define class aliases like this: $Form = [System.Windows.Forms.Form] but that only starts reducing noise when you use that class more than once.
  • Define properties in constructors. Another reason the PowerShell version is long is that only the .Net constructors for GUI controls are available through the platform interface, so to define a control, you have to write a sequence of statements starting with creating a new object followed by some SetX() methods. I wonder if PowerShell adaptors can overload constructors?