Why I love learning functional programming

This is the first part of a series on my journey in learning functional programming (FP). In this first part, I'd like to share why I spend time on learning functional programming in the first place.

At work, I mostly write non-functional code and I still haven't written purely functional production software. However, I still spend time learning it every now and then, and here's why.

It brings math to programming

The first reason I like functional programming is that it brings math back to programming.

At the university, I minored in math. I'll probably never have any practical use to the courses in topology, differential geometry or group theory, but none of those courses were a waste of time. They all taught the power of abstraction, how to find and see the big concepts underlying seemingly unrelated problems. Like Eugenia Cheng wrote in Cakes, Custard, and Category Theory:

"The power and beauty [of abstract mathematics] lie not in the answers it provides or the problems it solves, but in the light that it sheds. The light enables us to see clearly, and that is the first step to understanding the world around us."

In functional programming, you encounter abstractions like functors and monads all the time. Functional programming has roots deep in category theory, a branch of mathematics studying objects and their relationships. Category theory tells us, for example, that monad is just a monoid in the category of endofunctors. What the heck do those words even mean? I have no idea, but I must find out!

I've been learning category theory from the wonderful Category Theory for Programmers blog posts. They're are an easy and accessible way to access category theory. Maybe some day I'll be able to pick up a serious textbook on category theory!

It forces you to think differently

My second reason to learn functional programming is that it forces me to think differently.

Putting aside playing with Basic in the 90s, I first learned programming at the university in Java and C. Programs were written using if-clauses and for-loops. Data was modified in-place with functions or method calls returning nothing.

If-clauses, for-loops and in-place mutations are easy for us humans to understand, because that's how we intuitively process data. If you're given a list of N skills that you need to learn unless you already know the skill, here's the algorithm:

1. Set i=1
2. Take the i'th skill from the list
3. Check if you know the skill. If you don't, learn the skill.
4. If i=N, exit. Otherwise, set i = i+1 and go to 1.

This is an imperative program, with one command after another modifying the program state (your skills). To us, world seems to be made of mutable objects. That's how computers also work, one statement after another modifying the program state.

Now, imagine you're told you need to write code for a program without a single if-clause or for-loop. You are also forbidden to mutate objects. What you're allowed to do is create new objects and write pure, referentially transparent functions. Referential transparency means that a function call can be replaced by its return value without any change in the program. For example, this function is not referentially transparent:

def square(x):
    print(f"Computing the square of {x}") 
    return x*x

You can't replace square(x) with x*x and expect the program to remain unchanged. Pure function is a function that's referentially transparent for any referentially transparent arguments.

It goes without saying that such constraints force you to think differently about writing code. To me, that's a very good thing. Recently I've been writing code mostly in Python and JavaScript. While I love both languages for their flexibility and simple syntax, and there's always something new to learn in both of them, I don't think they offer that many chances for learning new concepts. Last time I learned something genuinely new about Python was when we wrote a command-line tool making heavy use of asyncio or when I had to understand generics in the typing module. Most of the time, the code consists of the same if-clauses and for-loops, possibly in some new framework.

With functional programming, programs will inevitably look different. Are they better? That's an ill-posed question, as there's no one best code for a particular task. It depends on factors like with whom you work and who will maintain the code. But I do think writing functional programs teaches you something fundamentally new about computing, and the more you know, the more likely it is that you can pick the best approach when new problems emerge.

Of course, my employer most likely wouldn't appreciate me spending the whole morning figuring out how to make a HTTP call or explain my colleagues how data type Maybe replaces if. That's one reason why FP is mostly a hobby to me at the moment. For me to be truly productive in writing purely functional programs, I would need to be surrounded by colleagues supporting me, with a team where knowledge about solving problems in a functional way would spread. In such a team, the cost of learning new concepts would also be lower as those new concepts might improve everybody's code base.

From the above examples, one may get the idea that imperative programming is "non-functional". To see that's not the case, here's one excerpt of Scala code from Functional Programming in Scala ("the red book"):

val factorialREPL: IO[Unit] = sequence_(
    IO { println(helpstring) },
    doWhile { IO { readline } } { line =>
        when (line != "q") {
            for {
                n <- factorial(line.toInt)
                _ <- IO { println("factorial: " + n) }
            }
        } yield ()
    }
)

That's a purely functional program written in imperative fashion. Why's there a for-loop? It's Scala's syntactic sugar for composing functions such as map, filter and flatMap.

FP is a logical conclusion to many ideas considered good programming style

The last reason to learn FP is that I think it pushes the boundaries of many ideas considered good programming style.

My first touch to functional programming came from attending lectures in functional programming at CMU, when I was a visiting researcher there. I attended maybe six lectures, where the lecturer wrote formal proofs showing that given recursive function calls would terminate with the expected result. It all seemed very theoretical to me and I thought I would not meet FP again.

However, as soon as I started in my first programming job, I was introduced to FP as more experienced programmers told me to avoid writing code with implicit side effects and mutable state where possible. I didn't understand at the time that the ideas had anything to do with FP, but I can see now how many such ideas are built-in to FP.

As an example of how FP can help write cleaner code, let's say you have a function like this:

const containsFinnishLapphund: (jpegBase64: String) => boolean = ...

It checks if an image contains a Finnish lapphund. The signature says the function takes a base64 encoded string and returns a boolean. Based on the signature, I expect this function to not have implicit side effects. Therefore, I can safely call the function for 100 images in parallel without worrying, for example, about race conditions, deadlocks or hitting rate limits of external APIs.

The key here is the word implicit. In the context of my TypeScript codebase, I do not mind if the function prints to console: my code would most likely already be interspersed with such logging statements. However, I would be very surprised if calling the function incremented a database counter or stored the image to Google storage. Such surprises could lead to hard-to-find bugs, let alone make testing a pain.

In non-functional languages, it's the developer's responsibility to write code that is not surprising. In Haskell, however, a type signature such as

containsFinnishLapphund :: String -> Bool

would make it impossible for the implementation to have observable side effects such as storing the image somewhere. If the function insisted on making a network call or logging to console, it would need a type signature

containsFinnishLapphund :: String -> IO Bool

The IO typeclass here makes it explicit that the function is doing something with the external world. What does it do? For that, you'll need to read the code or trust the function docstring saying it doesn't do anything other than print to console. But at least, it's not a surprise anymore.

Another example of an "FP idea" considered good programming style nowadays is declarative style. For example, most programmers would nowadays agree that to remove even elements from an array and square the rest, this

const double = (arr) => 
    arr.filter(v => v % 2 === 0).map(v => v*v);

is preferred to this:

const double = (arr) => {
    const newArr = []; 
    for (const i = 0; i++; i < arr.length) {
        if (arr[i] % 2 === 0) {
            newArr.push(arr[i] * arr[i]);
        }
    }
    return newArr;
}

In functional languages, the former would be the default way of solving the problem. Again, this doesn't mean declarative style is better than imperative, but it does show that declarative style has its pros. In FP, the declarative style can be pushed even further with function composition and point-free style:

square :: Int -> Int
square num = num * num

isEven :: Int -> Bool
isEven n = n `mod` 2 == 0

double :: [Int] -> [Int]
double = map square . filter isEven

To me, code like this is elegant and beautiful. While function composition and point-free style take time to get used to, I find it worth the effort.

Conclusion

That concludes the first part of the series. I love learning functional programming because it gives me reason to read math again, is forces me to think differently, and it pushes the boundaries of good programming style. Thanks for reading, please leave a comment if you have any!