Self-documenting is a myth, and how to make your code self-documenting

03 Aug 2019 Best practices 8 min read

Self-documenting code is often presented as a programmer's utopia, where you don't need to write comments at all! But code can never be entirely self-documenting. Here are some tips on how to clean up your code and write fewer comments without losing sight of the big picture.

What is self-documenting code

In order to write code, you have to understand the surrounding code. And in order to understand it, you need to read it. Often repeatedly and frequently. So it's in everyone's best interest that this code is clear, concise and properly documented.

Self-documenting code is a goal that a lot of developers (including myself) set for themselves at least once. For most people it means you should write clean, well-structured code that makes it immediately obvious what's going on, so you don't need to write comments explaining it.

Well actually...

You can write the best, cleanest, most well-structured code anyone has ever seen, but here's the truth: You'll still need to write comments and document your code. You won't have to write as many comments, but you can't stop writing comments altogether.

Three questions

When someone reads what you wrote, for them to build the mental modal they need to understand it, your code needs to answer three questions:

  • What is the goal of this code?
  • How is it achieving that goal?
  • Why does it achieve that goal in this way?

Let's take a little piece of less-than-ideal code, examine why it's written like that, and try to improve it so it answers all three questions.

Note: The code samples are written in Javascript/Typescript, but the theory should be applicable for any language

The code

While experimenting with a new feature, you're just trying to get something working. So it can be tempting to think "I'll clean this up later" and write something like this:

function get() {
    const arr = db.getAll();
    return arr.map(i => i.name);
}

Only you don't clean it up because deadlines are looming and you have ten more things to finish before tomorrow's standup. And so that piece of code sits in your codebase for seven months, until another developer — let's call him Steve — needs to update something in that file.

After reading the function for the first time, Steve will probably have several questions:

  • What is this function getting?
  • What is in arr?
  • Why does it map to name?

By parsing what's happening, Steve can deduce a partial answer to these questions:

  • The variable arr must be an array, given its name and the fact that arr.map() is used.
  • The values in arr must have a property called name since it is mapped.
  • The function returns an array.

As you can see, Steve is trying to deduce the what and the why based on the only thing he has: the how. This happens a lot, in all types of codebases, and most developers don't even think twice about it. But in the end, the time and energy you spent parsing this kind of code adds up and takes away from your time working on the code.

So how can we make it easier on Steve, so he can understand what's going on at a single glance?

Add comments

One solution would be to add some comments to explain what's going on.

/**
 * Gets the names of all participants from the database
 */
function get() {
    // Load the participants from the database
    const arr = db.getAll();
    
    // Map to get just the participant's names
    return arr.map(i => i.name);
}

It has gotten a little clearer already, but adding comments is exactly what we wanted to avoid with our self-documenting code.

Name those variables

Instead of adding comments, how about we change the names of the variables to reflect what they mean?

function getParticipantNames() {
    const participants = database.getAllParticipants();
    return participants.map(p => p.name);
}

Now we have communicated essentially the same thing as before, but we didn't need all those comments. Properly naming your variables is one of the cornerstones of self-documenting code, because they communicate exactly what they represent.

Note how I still used a short variable p in the participants.map() function, since it's abundantly clear from the context that it will contain the participant.

So with these changes to the variable names, we've answered our original questions:

  • What is this function getting? It gets the names of the participants.
  • What is in arr? The participant entities.
  • Why does it map to name? Because we only need the name

Steve will find it much easier next time he has to read our code!

Why?

Another question you could ask is a little more far-fetched and doesn't concern this function specifically, but I'll ask it anyway: Why is there no function called database.getAllParticipantNames(), to query just the names from the database (instead of all this other data we don't need)?

There could be a million different technical reasons for this, but for this example let's say that the database queries are cached. This means that when the query runs, the received data is stored in memory for a little while so subsequent calls don't need to make another roundtrip to the database. So using the same query here is actually an optimisation, even if we get too much data from the call.

This optimisation is an example of something you can't possibly communicate using code alone. As it turns out, purely "self-documenting" code is simply insufficient to paint the whole picture. So we will need some comments after all.

function getParticipantNames() {
    // Because queries are cached, using the `allParticipants` query
    // prevents another roundtrip to the database
    const participants = database.getAllParticipants();

    return participants.map(p => p.name);
}

With this, we've made the Why even more complete. We needed comments to fully document the code, but this code can still be considered "self-documenting".

What?

There is one last question remaining, one asked not by Steve who has to look at your function, but rather by Tom who has to use it in another part of the codebase: What is the return type of this function?

The best solution for that is type annotations. Statically typed languages like Java, C# or Rust don't need any additional work since they require explicit type information to work. But dynamically typed languages like Javascript and Python don't have this luxury. Luckily, most of these dynamically typed languages have solutions for (optional) typing. Javascript even has several — I've worked with JSDoc comments, Flow types and Typescript.

We've already tried the full JSDoc comment above, but all that's necessary to annotate the return type is an @returns statement in the comment:

JSDoc
/**
* @returns {string[]}
  */
  function getParticipantNames() {
  // Because queries are cached, using the `allParticipants` query
  // prevents another roundtrip to the database
  const participants = database.getAllParticipants();

  return participants.map(p => p.name);
  }

Typescript and Flowtype use syntactic notations rather than comments. Note the : string[] after the function name:

Typescript/Flowtype
function getParticipantNames() : string[] {
// Because queries are cached, using the `allParticipants` query
// prevents another roundtrip to the database
const participants = db.getAllParticipants();

    return participants.map(p => p.name);
}

My personal favourite is Typescript. It helps you create clean code with strict interfaces and it makes refactoring a whole lot easier when you need to. I use Typescript in nearly all of my projects.

However, it's important to note that adding Typescript to a project is generally not a decision to take lightly — especially if you have an already developed codebase — so be sure to consider the ramifications before you start. JSDoc is almost always the easiest choice to get started typing your code because it's based on comment blocks, which have no impact on the code itself.

Conclusion

Let's extract the three most important rules from what we've seen in this post, based on the three questions:

  • Use clear and consistent naming, types and function signatures to communicate what the goal of each piece of code is.
  • Use well-structured code to show how you're going about achieving the goal.
  • Use comments to explain why you're doing things a certain way, especially if that way may be non-obvious.

That last one is the hardest for a lot of developers because the why is usually obvious while you're writing the code. But take a moment to think of the developers who will need to look at your code months or even years after you wrote it. They'll be grateful.

Make Steve happy. Document your self-documenting code.

Back to blog overview