The pipe()
function that I'm talking about is the one that lets you replace b(a(x))
with pipe(x, a, b)
. Yes, that's how many times I've used it over the last few years, and looking back at those usages, I'd like to tell you the reasons you might find it useful too, even when you work with a codebase that doesn't stray from mainstream patterns into functional programming.
Where it's coming from
pipe
takes the first argument and pipes it though each of the functions that you provide as the remaining arguments, and can be implemented as follows:
const pipe = (x, ...fns) =>
fns.reduce((acc, el) => el(acc), x);
You can type it in TypeScript using overloads, and since as far back as TypeScript 3.4, type inference works perfectly:
One way to look at this function is to see it as a fill-in for the proposed pipeline operator (x |> a |> b
). That proposal has been at stage 1 for years, but the good news is that pipe
is not far worse — curiously, it's even better than some of the discussed flavors of the operator in one sense, namely that you don't have to enclose arrow functions in parens. If one of the flavors of the pipeline operator does reach stage 3, you won't be left out in the cold: with AST tools and Prettier, it would be easy to build a codemod that replaces pipe
with the operator.
Putting aside the pipeline operator, pipe
can be just seen as the first choice among different ways to do function composition. Another notable contender is a function that composes functions without applying them,
const ltrCompose = (...fns) => (x) =>
fns.reduce((acc, el) => fn(acc), x);
so b(a(x))
is equivalent to ltrCompose(a, b)(x)
. It's a higher-order function though, and that's where pipe
beats it: pipe
is easier to read because it lets you achieve the same ends without thinking in terms of transforming functions to other functions. At first I tried using both utilities depending on the context, but I found this to be a bad violation of "only one way to do it".
It's like dot-chaining
Now to reasons for using pipe
. The first thing to notice is that rather than introducing a new pattern, pipe
lets you use essentially the same pattern as dot-chaining,
yourArray.filter(predicate).map(project);
yourString.trim().toLowerCase();
only without being constrained to the collection of methods defined for native objects.
One group of use-cases center around the fact that native JavaScript APIs were not designed with an eye to immutable updates that we often use today. sort
method of Array
and add
method of Set
are mutating, but with pipe
, we can define their non-mutating counterparts
const sort = (compare) => (array) =>
[...array].sort(compare);
const add = (value) => (set) =>
new Set(set).add(value);
and use them like we use dot-chained methods:
const newArray = pipe(array, sort(compare));
const newSet = pipe(set, add(value));
Another common use-case is iterables. To take one example, if you need to filter values of a Map
, you would have to write [...yourMap.values()].filter(predicate)
, in other words, you have to convert the iterable returned by yourMap.values
to an array just to get at the filter
method. It wouldn't matter that much if it was just a question of performance, but it's both inefficient and clutters up the code. pipe
gives you an alternative of working with iterables in the same way that you work with arrays:
const filter = (predicate) =>
function* (iterable) {
for (const el of iterable) {
if (predicate(el)) {
yield el;
}
}
};
const filteredValuesIterable = pipe(
yourMap.values(),
filter(predicate)
);
It lets you create locals with expressions
Here's another reason for using pipe
— and this time we're not even going to need any utility functions other than pipe
itself.
Imagine that in an if
clause, you need to convert a string to a number and check if that number is greater than 0.
if (parseFloat(str) > 0) {
// ...
}
Now suppose that we also need to check that the number is less than 1. Unless we want to duplicate parseFloat
calls, we have to define a new constant in the outer scope:
const num = parseFloat(str);
if (num > 0 && num < 1) {
// ...
}
Wouldn't it be better if num
was scoped to the expression in the if
clause, which is the only place where we need it? This can be accomplished with an IIFE, but it's not pretty:
if ((() => {
const num = parseFloat(str);
return num > 0 && num < 1;
})()) {
// ...
}
pipe
solves the problem:
if (pipe(str, parseFloat, (num) => num > 0 && num < 1)) {
// ...
}
Generally speaking, in any context where an expression is expected, whether it's a function argument, an element in an array/object literal, or an operand of a ternary operator, pipe
lets you create a local without resorting to IIFE. This tends to make you rely more on expressions,
const reducer = (state, action) =>
action.type === `incrementA`
? pipe(state, ({ a, ...rest }) => ({ ...rest, a: a + 1 }))
: action.type === `incrementB`
? pipe(state, ({ b, ...rest }) => ({ ...rest, b: b + 1 }))
: state;
but you don't have to use expressions all the time — pipe
just lets you make the choice between expressions and statements not based on syntax limitations, but based on what's more readable in a specific situation.
The pipe
function as defined here is available in fp-ts. If like me you don't need a full-blown functional programming library, you can get pipe
in my own library Antiutils:
ivan7237d / pipe-function
A function to pipe a value through a number of transforms
pipe
function
A function to pipe a value through a number of transforms.
Installation
npm install pipe-function
or
yarn add pipe-function
or
pnpm add pipe-function
Usage
import { pipe } from "pipe-function";
Takes between 2 and 20 arguments. pipe(x, a, b)
is equivalent to b(a(x))
, in other words, this function pipes a value through a number of functions in the order that they appear. This article talks about why this function is useful.
When you have a single argument, like const y = pipe(x)
, pipe
is redundant, so you will get a type error, but the code will run and return x
. Despite the type error, the type of y
will be inferred correctly as type of x
.
Top comments (4)
The pipe() function that I'm talking about is the one that lets you replace b(a(x)) with pipe(x, a, b)
I don't think that is exact to pipe, If you just want to replace the b(a(x)), I think
compose pattern
is closer and exacter for it.Could you tell a bit more about what you mean by "compose pattern"?
Compose is similar to pipe but you don't pass a value at the front so the result is another function, not a value. It's for combining small functions into more complicated functions that you can call later. Check out flow in fp-ts for some examples.
Oh I see - I talk a little bit about this function in the first section where I call it
ltrCompose
. It might well be a good option for folks who use fp-ts or Ramda, but personally I've moved away from it.