background image
avatar
James DiGioia
my little web home

Hack Pipe for Functional Programmers: How I learned to stop worrying and love the placeholder

I’ve been involved with the development of the pipe operator (|>) in JavaScript for 4 years, going all the way back to 2017 when it was just a humble pipeline that did basic function application. Over the course of the discussion, the proposal evolved into two competing syntaxes, F# & Smart-Mix, exploring different ways of handling await, arrow functions, and the overall approach to composition. Throughout the process, I was strongly in favor of the F# variant, preferring the simplicity & functional nature of applying values to functions, and I argued strenuously for it over Smart-Mix, which I viewed as overly complicated.

The Smart-Mix syntax has since evolved into its current Hack iteration, dropping the “bare style” (x |> a.b, without a placeholder) and simplifying its syntax accordingly. Around that time, the previous champion of the proposal, Daniel Ehrenberg, stepped down and Tab Atkins-Bittner stepped in to take his place. While Daniel favored the F# pipe, Tab favored the Smart pipe previously and the evolved Hack pipe now and as champion, had been working to bring the TC39 committee (the one that specifies JavaScript) to consensus around the Hack pipe.

With Tab taking on champion responsibilities now, I’ve been discussing the proposal with them and the rest of the champion group. While I originally came into the conversation advocating for the F# pipe, I’m now convinced the Hack pipe is the superior option, and that it would be beneficial to the functional community to embrace it. At its core, Hack pipe is a better bridge between mainstream & functional JavaScript and makes it easier for mainstream JavaScript to adopt functional patterns.

The functional/mainstream split

Right now, the functional community is cut off from the rest of the JavaScript ecosystem. In order to do any of the fun & exciting things functional programming enables, you need to use specially-designed functional libraries. lodash/fp is one well-known example; Ramda is another. While the immutability & side-effect-free behaviors of these libraries are extremely valuable to both functional programmers & the wider JavaScript community, they both also bring in the overhead of one of functional programming’s more complicated concepts: currying.

Traditional currying vs JavaScript currying

Currying traditionally means functions that take a single argument at a time, returning a new function until it’s received all of its arguments, at which point the function is evaluated. To emulate this directly, you’d have to write functions in JavaScript like this:

const add = a => b => a + b;
add(1)(2); // 3

This is great for functional composition. It’s a small, narrowly-focused function you can combine with other small, narrowly-focused functions to solve bigger problems. add can be partially applied to create functions and use it in a larger composition:

const doubleThenAdd2 = pipe(multiply(2), add(2));
doubleThenAdd2(5); // 12

But even with that benefit, this is clearly not the normal way to write functions. If you call them the normal way, they just take the first argument and discard everything after that.

add(1, 2); // b => 1 + b;

Even functional programmers recognize that this is un-Javascipt-y, which is why they invented the curry function.

const add = curry((a, b) => a + b);
add(1); // b => 1 + b;
add(1, 2); // 3
add(1)(2); // 3

curry takes in a normal function and returns a curried version that works in both forms. At core, curry works by looping over the arguments provided and counting them, building up its internal list of arguments. If the function is missing arguments, it returns a new function which takes the remaining arguments; if the number of arguments provided matches the length of the function, the underlying function is called. Care needs to be taken in this implementation to ensure the length of the returned function matches the expected remaining arguments, which is done with this nasty-looking internal helper.

With curry, functions are free to work both the normal way and the partially applied way, enabling functions not written for functional composition to be composable. curry is a bridge between more traditional approaches to functional programming, where currying is built into the language, and JavaScript’s behavior expected behavior.

curry is a suboptimal bridge

While it does provide a bridge, it’s a bridge that actually exacerbates the divide between functional & mainstream JavaScript. It’s imperfect, attempting to bring functionality to JavaScript not native to the language; it’s one-way, sucking functions into the functional world while giving nothing back; and it’s complicated, making functional programming more difficult for mainstream JavaScript developers. Ultimately, this produces an approach to functional programming that is fundamentally un-JavaScript-y.

curry is an imperfect bridge

If you’ve been working in functional JavaScript, you’ve probably come across some variations on curry: curry2 & curryN. curryN takes an additional argument, the length of the function, and returns a function that evaluates when that number of arguments have been provided (with curry2 specifically dealing with functions that take 2 arguments). This is necessary because in JavaScript, using length may not mean what you want.

The first example is how default parameters are handled in JavaScript. In the below example, sort has a length of 1, not 2.

const sort = (arr, algo = 'bubble') => // ... implementation

The default parameter doesn’t count towards the length of the function, so if you want this function to take two arguments before being called, you need to tell it explicitly because JavaScript considers this a function with one argument.

Another pattern this doesn’t work well with is options-bag parameters. The below example isn’t curry-able at all without some significant hoop-jumping:

const api = ({ url, method = 'GET', body = {}, headers = {} }) => // ... implementation

In order to turn these into composable functions, you have to wrap them in functions that take these options one at a time, which can be awkward at best. This gets worse if the second argument is the options-bag argument, with a handful of default arguments applied. None of this is curry-able in the traditional sense, and for the most part, is solved by wrapping api in an arrow function to make it do what you want. curry just isn’t helpful to make this composable.

curry is a one-way bridge

While curry functions as a bridge between the non-functional world and the functional world, it’s really a one-way bridge. Functions that aren’t as easily composable need to be made composable in order to interface with the wider functional ecosystem, but there is no reciprocity.

Taking the above sort function as an example, that’s written in an mainstream idiomatic fashion, with the data coming first & the optional, defaulted argument coming second, but to make this composable, you need to make the first argument the algorithm:

const sort = curry((algo, arr) => // ... implementation );
sort('bubble', [5, 214, 23, 6]); // [5, 6, 23, 213]
const sortBubble = sort('bubble');

This is weird! The default is now required, and the data comes second, which is often unexpected. In this case, you could feasibly create a separate function for each algorithm (this is a pretty contrived example, for sure), but for other functions with default values, that might not make sense. You really want to just leave this value out, but making it curried makes writing a mainstream function more difficult. In this case, the default argument is just a string, but what if it’s an options-bag? This gets even stranger and more complex.

All of this means you have to write functions specifically for curried functional composition, and those functions are often unidiomatic for mainstream JavaScript. The curry bridge only brings functions into functional JavaScript style; it doesn’t send functions back into mainstream style, and writing mainstream JavaScript often makes it difficult to adopt functional patterns because of this mismatch.

curry is a complicated bridge

In my experience, while these curried functions & compositions can make for really elegant code once written, debugging them when they go wrong can be… challenging. You lose visibility onto where arguments are actually coming from, pipelines can be difficult to step through in a debugger, and the overall behavior of the system can become opaque without a strong sense of what each underlying function does (do you know what converge does without looking it up? Do most JavaScript programmers?).

All of this makes functional programming in JavaScript hard for beginners. Even the basic add function, while appearing simple, can take a second for new programmers to realize it’s a function that returns a function. Functional composition is an incredibly powerful way of thinking through problems, and functional idioms make bringing that to non-functional programmers more difficult by requiring an understanding of several concepts before they can take advantage of it. There are tools like ramda-debug that make debugging these pipelines easier, but exacerbates the issue of needing to opt into a whole new ecosystem in order to use these tools and deepens the divide between functional & mainstream JavaScript.

Curried (& point-free) code is un-JavaScript-y

I’m not the first person to point this out. Composing functions in the way described is used widely in functional JavaScript, and (in my experience) rarely outside of it. Every time a beginner runs into code that looks like this:

const doubleThenAdd2 = pipe(multiply(2), add(2));
doubleThenAdd2(5); // 12

there’s like 4+ concepts to learn in order to understand what’s going on here, plus new tools & techniques to verify & debug these compositions. Developers find it “difficult to grasp and hard to read”, and it takes some time & experience with the style for a function like the above to become intuitive.

Early in my career, I used Ramda extensively and tried to introduce it & its associated concepts to my team. Despite several of them being senior engineers, this took a lot of work to explain and it never became a natural approach for them. I also struggled with debugging my own pipelines in the process and have since abandoned Ramda and its associated technqiues as my utility library of choice.

The overhead of these concepts makes adopting other functional concepts in the wider JavaScript community more difficult because they don’t integrate well into non-functional code without adopting an overall functional style. Both mainstream JavaScript and functional JavaScript in particular suffer as a result, as mainstream JavaScript fails to benefit from functional ideas while functional programming is relegated to playing with its own toys instead of sharing everyone else’s.

Reject currying; embrace composition

Bringing a native pipe operator into JavaScript, on its own, will address a lot of these issues. There will no longer be a need for a pipe or compose function, and the behavior of the pipe is more intuitive than either of those functions could ever be. It literally looks like an arrow! x |> func looks like “x going into func“! Right off the bat, we make composition more accessible to programmers new to functional programming.

Additionally, it makes debugging easier. With a pipe integrated into JavaScript, the devtools can more easily add breakpoints in between pipeline steps, inspecting the output of any given step without additional tooling required. This is similar to how you can add a breakpoint to a one-line arrow function right after the arrow, inspecting the callstack & arguments inline. These two features alone make the value of the pipe worth it for the language.

By and large, between the two proposals, most functional programmers prefer the F# proposal. Looking at doubleThenAdd2 with F# pipe, we see effectively a straight translation of the current usage:

const doubleThenAdd2 = x => x |> multiply(2) |> add(2);

Functional programmers have a pretty straightforward translation from their current pipe to |> because to works like what they’re already used to doing in functional JavaScript. Currying & point-free programming is such a significant part of how functional programmers in JavaScript code that they want a pipe operator that allows them to continue doing that.

But… what if it wasn’t? The Hack pipe version looks like this:

const doubleThenAdd2 = x => x |> multiply(2, ^) |> add(2, ^);

Now, we no longer even need currying; we can use our base, uncurried add function without modification. Because Hack allows arbitrary expressions, rather than functions, this can actually be even simpler:

const doubleThenAdd2 = x => x |> 2 * ^ |> 2 + ^;

We don’t even need functional versions of these operators! We can compose operators, not just functions, without needing to introduce or write a whole library for them.

The entire world of non-curried libraries get opened up to functional composition. We no longer need lodash/fp as a curried alternative to the base lodash functions because the base lodash functions work just fine in the pipe. The Web platform as a whole is now available to functional programmers directly. fetch doesn’t need a special wrapper to handle its options bag; you don’t need to dive into date-fns/fp when Temporal can compose; the entire DOM can be manipulated in a pipeline just like everything else, without wrapping every step in an arrow function! Even future proposals, like Records & Tuples, could possibly be getting syntax for deep path property setting, which works better with Hack’s expression-based operator.

What’s more, this operator is much easier to understand than curry. Beginning programmers understand expressions; they may not call them that, but they know how x + 1 and await fetch() work & behave. With Hack pipe, there are a bunch of concepts that are no longer necessary to learn in order to start taking advantage of composition.

Functional programmers can stop explaining currying & point-free programming and focus on bring other functional features into the code people write every day. Mainstream JavaScript can start to use constructs like Maybe & Either, both of which I’ve found to be quite useful and introduce regularly in non-functional JavaScript, especially for wrapping promises. Being able to buy into these concepts without the overhead of currying & point-free programming would be a great benefit to the wider JavaScript community and much easier for beginner programmers to understand & use on their own.

The Hack pipe is thus good for functional programming: it bridges the divide between the functional programming community & the wider JavaScript community. A whole category of interop tooling goes away in favor of a native operator that both mainstream & functional JavaScript can take advantage of.

The future of functional programming is Hack pipe

With the most recent TC39 meeting, the Hack pipe advanced to Stage 2. I expect this news, especially given my long-standing preference for F#, to cause some consternation amongst the F# adherents and their functional preferences. However, Hack is a win for functional programming in JavaScript, even if it means giving up some time-honored functional traditions.

It will definitely take some time for the ecosystem to adapt to the introduction of Hack pipe, although curried functions can take advantage of the pipe operator as-is.

// Without `curry`
1 |> add(1)(%);
// With `curry`
1 |> add(1, %);

Some algebraic data types that work well with the Hack pipe operator, support tree-shaking, & are comprehensible to beginners would go a long way towards making functional programming more idiomatic in non-functional contexts, decoupling them from more complicated functional approaches to solving problems.

Functional programming in JavaScript has long adopted techniques and tools from other functional languages; this is why currying & point-free programming have become so popular in functional JavaScript. But JavaScript isn’t Haskell, and currying isn’t built-in to JavaScript, so its adoption will always be a bit unnatural by comparison. We should adopt a pipe operator that feels native to JavaScript, and adapt our techniques for more natural functional JavaScript for the masses.