Lessons Unlearned

Blog by Nils Mehlhorn

Finding Harmony Between Refactoring and Implementation

January 07, 2019

architectureoop

puddle-piano

I saw this inspiring music video and it gave me the idea for a little exercise. I often struggle between getting things done and finding the perfect abstraction. Some code is already written and now you need to add a certain functionality coming in from a different point of view. Parts of your code shall do things you didn't envision them to be doing when you first designed them. Do you squeeze that new functionality somewhere between the existing lines? Will you really refactor it afterwards? Can you even effectively refactor afterwards?

Clean coders from all over the web would argue: either implement or refactor - which one goes first would be up to style and situation. But in many cases, this seems very counter-intuitive due to the fact that both actions are often inter-dependent. Well, from what I've seen, few good intents ever made it freely into any major codebase. So let's look at an example and try to find some middleground.

Picture this: We've got to bring these two parts together: Piano and Player. It's foreseeable that in the future the player might play something different from a piano - be it a grand piano, a keyboard or even a puddle for our music video. So let's use an interface to extract the contract between these parts, we'll call it Playable. It could look like this:

interface Playable {
    play(key: Key)
}

This interface is associated with the Player as the actions originate from him. He deems what's playable - not the other way around. A player could get an instance of Playable and then may be asked to play a piece of music on it - just like the following class may suggest:

class Player {
    constructor(playable: Playable) {...}
    play(piece: Piece) {...}
}

Alright, that's it you might think, after quickly implementing an algorithm for playing a piece of music. You push the code, it passes review but then it's rejected by QA. They're saying that it sounds off when a piece of music goes on for longer. Your player is unable to keep a balanced tempo because he's playing without audible feedback - what a bummer. The player needs to hear what he's playing. Meanwhile, he's supposed to keep the ability to play a puddle - for the video, you know.

What I like to do in such a situation is write down some ideas that pop into my head, then put the problem on hold and come back later with a new perspective. If the solution doesn't present itself then, start shuffling things around. Try to find out which solution has the greatest return of investment - which way solves the problem with minimal obstruction to your core concepts.

So, the player needs to hear what he's playing. This could be some kind of stream (think ReactiveX or similar). A first intuition could be to work the feedback somehow into the Playable interface. It could be the return value of Playable.play(key) or we could add a second method as some kind of getter. Yet, this would introduce significant unwanted coupling through the Playable interface. The player won't be able to play the puddle anymore as it wouldn't be able to provide feedback - well except for some splash sounds maybe. A puddle would have to return an undefined value or throw an exception - eww that's ugly. It might be made prettier by extending the Playable interface and creating something like the following

interface PlayableWithFeedback extends Playable {
    play(key: Key)
    listen(): AudioStream
}

The piano would be able to implement the more specific interface, while the puddle remains just playable without feedback. But then it would be the player's responsibility to figure out what he's working with. He might do this through some instanceof checking or even pattern matching, if your language is capable of such things. Still, for my taste this would shift responsibility in the wrong direction.

Instead let's extract this new functionality into a different contract Hearable

interface Hearable {
    listen(): AudioStream
}

A player which isn't supposed to be off-tempo at some point should have all the necessary resources at hand for that task. Therefore, let's adjust our player implementation. This way he can adapt his playing according to what he's hearing:

class Player {
    constructor(playable: Playable, hearable: Hearable) {...}
    play(piece: Piece) {...}
}

We can now have a player play a piano or a puddle without getting off-tempo. We might facilitate this by providing something different as the feedback source for objects that are Playable but not Hearable. This way we could combine the puddle with a stereo playing a recording to reconstruct the music video.

const piano: Playable & Hearable = new Piano()
const pianoPlayer: Player = new Player(piano, piano)

const stereo: Hearable = new Stereo(new Recording('recording.mp3'))
const puddle: Playable = new Puddle()
const puddlePlayer: Player = new Player(puddle, stereo)

Bottom line

As is the curse of examples, your current problem might not be as simple as this one. Still, I think it makes the point that it's seldom easy to see the consequences of introducing new functionality to your code. It's hard to draw the line between refactoring and implementation. Sometimes you can't prevent changes to the way your code is coupled. Instead, what you want to do is prevent coupling in ways you don't deem future-proof. Your refactoring will be opinionated towards your implementation. Therefore you'll need a vision which you have to match constantly in congruence to your code. This way the vision is not only in your head and manifested at runtime, but also visible in code.

developer consultant and product developer writing about just that and everything in between