jordan scales home projects twitter github
August 19, 2017

Typesafe JavaScript Chaining with OCaml and BuckleScript

OCaml code: express () |> get "/" index |> get "/about" about |> listen 1337

In my previous article, we explored how BuckleScript allows you to turn OCaml code into readable JavaScript, and how to interface with other modules in the JavaScript ecosystem.

Today I’d like to continue on this path and show you the awesome @@bs.send.pipe binding attribute, which enables us to write concise OCaml code to interface with JavaScript libraries that have a chainable API.


Exhibit A: Express

To interface with the express Node.js web framework, we may write the following bindings in src/FFI/Express.ml. (NOTE: Remember to include src/FFI in the sources field of bsconfig.json!)

type app
external express : unit -> app = "" [@@bs.module]
external listen : app -> int -> unit = "" [@@bs.send]

type req
type res
external get : app -> string -> (req -> res -> res) -> unit = "" [@@bs.send]
external send : res -> string -> res = "" [@@bs.send]

Then, in src/index.ml we could use this code as follows:

open Express

let app = express ();;

get app "/" (fun _ -> fun res ->
    send res "Hello, world! <a href='/page'>Page 2</a>");;

get app "/page" (fun _ -> fun res ->
    send res "Hey <a href='/'>Go back</a>");;

listen app 1337;;

Running bsb results in the following lib/js/src/index.js:

// Generated by BUCKLESCRIPT VERSION 1.8.2, PLEASE EDIT WITH CARE
'use strict';

var Express = require("express");

var app = Express();

app.get("/", (function (_, res) {
        return res.send("Hello, world! <a href='/page'>Page 2</a>");
      }));

app.get("/page", (function (_, res) {
        return res.send("Hey <a href='/'>Go back</a>");
      }));

app.listen(1337);

exports.app = app;
/* app Not a pure module */

Nice! We can run node lib/js/src/index.js and get ourselves a running express server.

The Chaining Express API

Consider the type we wrote for the Express.get function:

external get : app -> string -> (req -> res -> res) -> unit = "" [@@bs.send]

get takes an app representing our express instance, a string for the path, a function (which takes a request and response), and returns a no-op (type unit).

However — did you know we can chain this API like so? In JavaScript:

app
  .get("/", (req, res) => res.send("Hello, world!"))
  .get("/about", (req, res) => res.send("About ..."))
  .listen(1337)

This pattern is very common in JS, and works in the following way: instead of get accepting an app and returning a unit (or no-op), we return another app which we can then use on a subsequent get!

That’s a lot to unpack, so let’s demonstrate how to get from A to B in code.

Step 1: Take an app, return an app

external get : app -> string -> (req -> res -> res) -> app = "" [@@bs.send]

let f: app = get (express ()) "/" index;;
let g: app = get f "/about" about;;
listen g 1337;;

So what’s different here? First, we changed the return type of get from a unit to an app. Next we remove the definition for app and inline express () in f directly.

Then, instead of using app as the first argument for our second call to get, we pass in f. This is type-safe (remember: f, g, and express () all have the same type) and sure enough if we compile this script and run it — we get a working Express app!

In fact, if we wanted to, we could start combining some of these lines by inlining the definition for f entirely like so:

let g: app = get (get (express ()) "/" index) "/about" about;;
listen g 1337;;

Or a step further, inlining g as well:

listen
  (get
    (get (express ()) "/" index)
    "/about"
    about)
  1337

These two examples are identical to the first, but notice that app is only referenced once in our code. Let’s peek at BuckleScript’s output lib/js/src/index.js:

Express().get("/", index).get("/about", about).listen(1337);

🔗🔗🔗🔗🔗🔗🔗🔗!!!

See, once we smush together our get and listen calls, there’s no need for temporary variables like f and g. BuckleScript knows this, and merely puts everything inline for us — in a “chained” manner.

This may start to look a little LISP-y to you, and that’s fair — this syntax is not easier to read than our original example which specifies app multiple times. Let’s move on and see how we can clean up this code a little.

Step 2: Some light plumbing, and a leak

As we start composing functions (like we did by inlining f and g in the previous section), we’ll start to see quite a bit of parentheses. Consider the following bit of code:

apply_discount(
  (get_age_group(get_age(user_from_id(id))))
  price)

Sure we can dress this up with further indentation, but developers reading this code will still construct a sort of “stack” in their head as they read the subsequent functions from left to right (“Okay apply discount of the age group of the age of the…”)

To remedy this, OCaml provides the infix |> (or “pipe”) operator. We can inspect its type via utop :

utop # (|>);;
- : 'a -> ('a -> 'b) -> 'b = <fun>

We see that we take an item of type a, a function from a to b and return an item of type b. *Exhale *In code:

f(x) === x |> f

And if we were to use this pipe multiple times:

f(g(x)) === x |> g |> f

We can see here how the pipe operator (|>) allows us to unfold various layers of function composition. It’s quite neat, and leads to some very readable code. Let’s use it with our example above:

apply_discount(
  (get_age_group(get_age(user_from_id(id))))
  price)

(* turns into... *)

apply_discount(
  (id |> user_from_id |> get_age |> get_age_group)
  price)

How about that last layer? What if we wanted to unfold apply_discount as well?

let f = id |> user_from_id |> get_age |> get_age_group |> apply_discount;;

f price;;

Decent! However we hit a snag. apply_discount takes two arguments, the user’s age group, and a price (group -> price -> total). If we were to write our code like so:

... |> get_age_group |> apply_discount price

We would receive a type error because price would be used as the first argument to apply_discount. This means we need some parentheses (technically you could use OCaml’s @@, but hold your horses), which we are trying to avoid!

(... |> get_age_group |> apply_discount) price

One way to fix this? Just make price the first argument!

Step 3: Save the app for last

If we were to redefine apply_discount from group -> price -> total to price -> group -> total, we could then remove our parentheses entirely:

... |> get_age_group |> apply_discount price

Now price is used as the first argument, and second argument (the age group) makes its way to apply_discount from the pipeline.

“Jordan this is great but I don’t really care about discounts and age groups, I’m trying to write a web server before my startup goes under.”

Well fear no more, let’s return to our express example from earlier.

listen
  (get
    (get (express ()) "/" index)
    "/about"
    about)
  1337

If we were to swap in some |> operators, we’ll quickly run into the same exact problem we had with apply_discount:

(((express () |> get) "/" index |> get) "/about" about |> listen) 1337

Notice how |> doesn’t really buy us much. Since an app type must be the first argument to get and listen, we’re left with a confusing mix of parentheses and |> operators.

As we learned in the previous section, our solution is to move this argument to the end. Let’s try it with some helper functions:

let get_ route handler app = get app route handler
let listen_ port app = listen app port

And use ’em like so:

express () |>

get_ "/" index |>
get_ "/about" about |>

listen_ 1337

And voila! An app type makes it way from express (), through the pipe and onto the end of get_ “/" index. That method also returns an app type, which finds its way at the end of get_ “/about" about, and so on and so forth. We now have ourselves a beautiful, type-safe chain of functions that map to the chainable express API.

Express().get("/", index).get("/about", about).listen(1337);

Step 4: BuckleScript can do this for us

Defining a function_ for every function you bind to JavaScript-land doesn’t sound all that exciting, though. Wouldn’t it be great if get and listen could work like that for us? Well they can!

The current bindings for get and listen are defined using the @@bs.send attribute as follows:

external listen : app -> int -> unit = "" [@@bs.send]
external get : app -> string -> (req -> res -> res) -> app = "" [@@bs.send]

However, BuckleScript also provides us with a @@bs.send.pipe which, you guessed it, allows us to define functions that work well with the |> operator. From the docs:

bs.send.pipe is similar to bs.send except that the first argument, i.e, the object, is put in the position of last argument to help user write in a chaining style:

Here’s a modified binding for get:

external get : string -> (req -> res -> res) -> app = "" [@@bs.send.pipe: app]

The difference here is that the first app in the type definition has been moved into the attribute, right after @@bs.send.pipe: . Here’s our new definition for listen:

external listen : int -> unit = "" [@@bs.send.pipe: app]

Now, we can swap out get_ and listen_ in favor of their original counterparts.

express () |>

get "/" index |>
get "/about" about |>

listen 1337

🎉🎉🎉🎉🎉🎉


Closing Thoughts

Okay so that was a lot of words to tell you how @@bs.send.pipe works, but I hope this post gave you a bit of intuition for why it exists and why you may want to use it. With that, here a few more questions to ponder on:

  • You may have noticed that the type of the callback for get is req -> res -> res. Why the second res? Well, express has operations on res like send, status, and cookie which are also chainable (they return a res type). Write chainable bindings for these methods.
  • Imagine @@bs.send.pipe did not exist and we were stuck with our old definitions of get and listen: could we create a function called make_chainable where make_chainable get === get_ and make_chainable listen === listen_? Why or why not? (As a hint: what if get and listen both had three arguments, could we do it then?)