Mirko Stocker

In the last few years, I worked on several Java projects where we transformed and analyzed abstract syntax trees, so when I started learning Scala, pattern-matching quickly became one of my favorite language features. I could never warm up to the visitor pattern, so I was thankful that Scala offered a much more powerful alternative.

What I also liked very much was Scala’s consistent use of the Option type in its standard library. Now, instead of having to read the documentation to find out whether some call could return null, the type checker forced me to handle this where necessary. So a lot of my early Scala code looked as follows:

doSomething() match {
  case Some(value) => Some(doSomethingElse(value))
  case None => None
}

While it’s still possible to get a NPE in Scala, it just doesn’t happen with well-written libraries, simply because there’s no need to ever use null.

(By the way, isn’t it funny that Java forces you to check exceptions but doesn’t help you with the much more common and annoying null problem?)

So yes, in practice, Options do save you from NPEs. Does your code also get smaller (because usually, in Scala it will)? Not if you pattern match on Some/None, all the un-wrapping and lifting is quite verbose.

Nowadays, certainly influenced by all the discussions on monads, I realize that pattern-matching on Option is a very primitive form of abstraction, and instead of the code above I now write:

doSomething() map (value => doSomethingElse(value))

(For those unfamiliar with functional programming, map applies the function to the value inside a Some, and does nothing when called on a None.)

Even better, we can fully automate this refactoring! I’m currently working on a first version in the scala-refactoring library. So far, I’ve implemented the refactoring for map, but there are many more we can do, for example:

• If the Some case does not construct a Some but calls a function that returns an Option, we use flatMap instead of map.
• If the Some case evaluates to Boolean and the None case returns false, we can replace it with exists. If Some returns true and None returns false, we can just replace the whole pattern match with isDefined.
• When the None case is (), we can transform to foreach.

So far we have only looked at Option, but there is more: for example, we could also replace pattern-matching on lists and recursion with folds. I’m sure somebody has already written a paper about such refactorings, but I haven’t found anything yet.

What do you think about eliminating pattern matching? Do you also prefer mapping to explicit pattern-matching?

I’m done! A few minutes ago, I handed in my thesis on Scala Refactoring.

The project also has a new home at scala-refactoring.org, where you can find more about the refactorings I implemented, how they can be used, etc. I also recommend reading my report, but if you just want an overview, take a look at the poster:

The next steps for me now are to go on holidays, so please excuse if I don’t respond for a week.

I haven’t blogged in quite a while now. Actually, the last entry was to announce my term project on Scala Refactoring. My excuse is, I was hard at work! So without further ado:

class Demo1 {
  def demo1(i: Int): Int = {
    val a = i
    val b = a + i
    b
  }
}

Select the line with the assignment to b, murmur the incantation press some keys, et voilà:

class Demo1 {
  def newMethod(i: Int, a: Int): Int = {
    val b = a + i
    b
  }
  def demo1(i: Int): Int = {
    val a = i
    val b = newMethod(i, a)
    b
  }
}

It also works with multiple return values and when passing functions. Now, the code isn’t ready yet, and I have to concentrate on writing my report right now, so it might take a couple more weeks until I can ship something.