Thinking in Functions, Part I: The Input/Output pattern.
Practical advice on how to design better functions.
I was lucky to work with some great minds in the industry. Once I’ve befriended a brilliant Java developer and we spent a decent amount of lunches chatting over patterns and approaches in programming. At that time I was working on a schema design for the forms validation in my project, which involved a lot of functional heavy-lifting. Each time we discussed a particular problem I was fascinated by the way he approached functional design. I’ve adopted one of his thinking patterns during the process, and it has changed the way I comprehend functions since then.
If you are coming from a functional programming background the things I’m about to say may bear no novelty for you. However, I know there are developers who are not accustomed to strong types and Hindley-Milner’s notation isn’t exactly the first thing that pops up in their heads when they think of a function. I address this article to those people.
The Input/Output pattern
We know that a function accepts arguments and returns a value. What we sometimes overlook is that this characteristic is an incredibly powerful design pattern we can use to write better functions.
Try to write any function by answering these two questions first:
- What is this function’s input (what arguments it accepts)?
- What is this function’s output (what data it returns)?
Doing so allows you to put technical details aside and focus on a function as an input/output operation, which, in fact, it is. The answers you give may hint certain implementational details, but, most importantly, they define clear constrains over a function’s responsibilities long before any actual code is being written.
You may use abstract types for your answers. For example, a function may accept a list of apples and return a happy fox. Such type abstractions detach the call signature from the implementation even further.
Let’s put this into practice. Say, you need to write a function that validates a form field. There are myriad of things that can affect a field’s validation, but you can leave those aside and answer the Input/Output questions first:
- My function accepts a field;
- My function returns a validation verdict (
When written down, these answers represent the function’s call signature:
You may have no idea what the
Field type may be at this point, but you know what it represents. The same can be said about the
validate function in total: no matter which factors affect a field’s validity, you have to resolve to the boolean verdict in the end. Defined input and output act as a restriction that prevents our function from becoming too smart along the deadline-driven development. That ensures that the logic we write lies within the function’s responsibility and remains simple, satisfying both single responsibility and KISS principles.
Adopting this pattern doesn’t mean you should immediately rewrite your code in TypeScript or any other strongly-typed language. First of all, it is the way of thinking about your functions. It’s fine to keep the input and output noted in a JSDoc block or in your sketchbook. Start from changing the way you think, and the tools will follow.
Similar to how you put the user’s needs first before making proper UX decisions, you think about what data your function accepts and returns in order to establish the boundaries of its future implementation.
Pattern at scale
Thinking of functions using this pattern is great, but what about real-world operations that often consist of multiple functions and represent more complex logic?
The truth is, even the most complex function can be written as a set of consecutively executing smaller functions. When you approach the task this way you can focus on designing each individual function at a time. However, keeping functions in isolation is dangerous, as you may end up with multiple puzzle pieces that don’t fit together. There is a rule of functional composition to avoid that problem:
Two functions are composable if the output of one can serve as the input to another.
Knowing that, let’s implement a fairly complex operation that accepts a user and returns the amount of likes under all of their posts. To prevent the complexity, we can describe this operation as a sequence of steps:
Get a user → get user’s posts → get the amount of post’s likes
Each of these steps is a function, and we can apply the Input/Output pattern to design its call signature, keeping in mind the composition principle.
Such high-level overview of a functional chain allows you to follow the data flow without distractions and highlight potential problems in the logic. Moreover, it is just a fun exercise to do.
In the end, functions are about transforming the data, so use all means available to ensure that transformation is coherent and efficient.
One more thing!
There is one more hidden gem about the Input/Output approach. Let’s say you answer those primary questions with “my function accepts a list of strings and returns a number”. Congratulations, you have just written a unit test for your function!
The result of this pattern may be reflected in a test scenario, backing up function design decisions with an actual unit test. This encourages TDD (test-driven development) and BDD (behavior-driven development), since we express our function’s intent by describing its input and output.
Focusing on a function’s input and output types defines a concise specification of that function:
- A function’s call signature;
- A minimal unit test for that function.
Implementing things according to the specification is a pleasant and safe experience I absolutely recommend you to get accustomed to.
I hope this pattern helps you next time you decide to write a function. Let me know if you would like to know more practical patterns and approaches to functional design by liking this post and letting me know in the comments.
Follow me on Twitter to have a chat about functions, or programming in general. I’d be glad to hear your opinions and answer your questions.