DEV Community

Dmitry Zakharov
Dmitry Zakharov

Posted on

The ultimate answer to Belt vs Js in ReScript

Probably every developer that comes to ReScript stumbles upon a dilemma about which module from the standard library they should use to work with JavaScript API.

  • Should I take the Js module with a familiar design for JavaScript developers and runtime-free bindings? šŸ§
  • Or should I take the more powerful Belt? šŸ¤”
  • Wait, I heard something about rescript-js! Maybe thatā€™s what I need? šŸ˜…

Those are pretty familiar thoughts, right? Iā€™ve seen a lot of discussions about which one you should use, but there were never solid answers, and the result was either ā€œit dependsā€ or ā€œI personally like one over another because of Xā€.

My lovely Carla colleague Daggala has written a very detailed article about the problem, so I wonā€™t repeat her and continue.

Problem 2: Multiple sources of truth

No matter what we choose, sometimes we still need to use another module. Itā€™s because some functions which exist in Js donā€™t exist in Belt and vise-versa.

Multiple sources of truth

But it might happen that a function doesnā€™t exist in both of the modules, eg infamous padStart. At Carla, we used to solve the problem by creating a lot of StringExtra, OptionExtra, and WhateverExtra modules to add missing helper functions.

This leads to another problem when developers donā€™t know where they can find the desired helper:

  • Will it be in Belt that we agreed to use as default, or the helper doesnā€™t exist there, and I should use Js instead, or maybe StringExtra?

A few times, we even end up using Pervasives by mistake.

Problem 3: Creating guidelines

Depending on the project, we may have rules for the team to follow. For example, Belt.Option.getExn raises a ReScript exception which is very difficult to trace, so at Carla we decided to use our own OptionExtra.getExnWithMessage instead. Although there's an agreement, it's very easy to forget about and continue using Belt.Option.getExn. We want a way to prevent the usage of Belt.Option.getExn whatsoever.

Solution

I will not create intrigue and say that the solution I suggest is to create your own vendored standard library and enforce its usage across the codebase. You can reuse existing modules like Js, Belt, or RescriptJs, and adjust them for our needs.

The enforcing is the most crucial part here because if we don't automate it with CI, our colleagues and even ourselves will continue using all different modules instead of the single vendored one.

And to solve the problem, I've created rescript-stdlib-vendorer, an easy-to-use linter to support the usage of a vendored standard library.

Linter

The linter does a straightforward thing - It checks all project files and detects the usage of Js, Belt, and ReScriptJs modules. So you can easily find and replace them with your Stdlib.

To start using the linter in your project, install it as a dev dependency:

npm install -D rescript-stdlib-vendorer
Enter fullscreen mode Exit fullscreen mode

Let's also add an npm run script for convenience and immediately use it:

npm pkg set scripts.lint:stdlib="rescript-stdlib-vendorer lint"
npm run lint:stdlib
Enter fullscreen mode Exit fullscreen mode

So if we have a project created from the official template repository, which has a single ReScript file with Js.log("Hello, World!"), weā€™ll get the following error:

~/rescript-project-template/src/Demo.res:1
  Found "Js" module usage.

Use the vendored standard library instead. Read more at: https://github.com/DZakh/rescript-stdlib-vendorer
Enter fullscreen mode Exit fullscreen mode

To fix it, letā€™s create a directory stdlib and add the first vendored module:

cd src
mkdir stdlib
cd stdlib
echo 'include Js.Console' >> Console.res
Enter fullscreen mode Exit fullscreen mode

We can now update the Demo.res by replacing Js.log with Console.log.

Letā€™s run the linter again and see that thereā€™s no error:

npm run lint:stdlib
Enter fullscreen mode Exit fullscreen mode

In the following way, you can create reexports for other modules and adjust them to make them better suit you.

For example, for the Array module, you can redefine some functions with Beltā€™s implementation to make the code safer:

// src/stdlib/Array.res
include Js.Array
let get = Belt.Array.get
Enter fullscreen mode Exit fullscreen mode

This way, the usage of square brackets to get an array item will return option, whichā€™s more correct:

let array = [1, 2, 3]
let item = array[3]
// The item will have the option<int> type instead of the default int one
Console.log(item)
Enter fullscreen mode Exit fullscreen mode

Returning to the Belt.Option.getExn, mentioned in the third problem, instead of reexporting all functions from Js/Belt, we can explicitly reexport only the functions we need and replace the ones we consider harmful:

// src/stdlib/Option.res
let forEach = Belt.Option.forEach
let mapWithDefault = Belt.Option.mapWithDefault
let map = Belt.Option.map
let flatMap = Belt.Option.flatMap
let isSome = Belt.Option.isSome
let isNone = Belt.Option.isNone

let getExnWithMessage = (x, message) =>
  switch x {
  | Some(x) => x
  | None => Js.Exn.raiseError(message)
  }
Enter fullscreen mode Exit fullscreen mode

Reusing the vendored stdlib

Iā€™ve shown how to start vendoring stdlib in a short way. But having multiple projects following the same guidelines, youā€™d soon want to start reusing the vendored stdlib. Itā€™s very easy to do by moving the code from the stdlib directory featured above to a separate package.

For my personal projects, I copied Gabriel's repository with a proposal for a new ReScript stdlib, which did not work out. Afterward, I modified it to suit my taste better and published it to npm, making it easy to use.

I recommend taking a look at it as a reference @dzakh/rescript-stdlib, but I donā€™t bring it to your own projects. You'll lose a very good part of vendoring - full control over the code.

At Carla, we had a different situation. Having a huge codebase with Js, Belt, and WhateverExtra all over the place, it would be too much work to take some existing customized stdlibs like rescript-js or @dzakh/rescript-stdlib and update old code with them.

First of all, since at Carla we have different guidelines compared to my personal projects, it would be a bad idea to use @dzakh/rescript-stdlib from the get-go. As I said before, it'd lose the whole point of vendoring.

So, to migrate the whole codebase, weā€™ve started with creating a small stdlib package in our pnpm mono repository that simply reexported functions from Belt or Js. And started updating file by file, replacing open Belt with open Stdlib and Js.Array2 with Array, etc.

Also, you can use a tool like Comby to transform the whole codebase in one go.

Before you start the migration to the vendored stdlib, I highly recommend replacing Js.String and Pervasives.String with Js.String2. Since some of the functions have the same API, but different logic, thereā€™s a chance of missing one of them and getting a runtime error.

But we didnā€™t rush with the migration, and to avoid regressions of the process, we ran the linter script in CI with the --ignore-without-stdlib-open flag to skip files not containing open Stdlib.

After every file was updated, we removed open Stdlib from the beginning of the files and opened it globally via bsconfig.json. When it was done, we could finally start gradually adjusting the Stdlib to make the process of writing ReScript code more convenient and reliable.

If you have any questions feel free to write them in the comment section or ping me on Twitter.

Top comments (1)

Collapse
 
hoichi profile image
Sergey Samokhov

Thatā€™s what I do in my pet as well, and thatā€™s what they did at ahrefs back when I worked with them.

It's obviously not perfect (I think one of the reasons people still enjoy developing in Elm is that Elm has a much more complete and cohesive standard library out of the box), but for long-living projects I think itā€™s quite fine.