BuckleScript User Manual

BuckleScript is a backend for the OCaml compiler which emits JavaScript. It works with both vanilla OCaml and Reason, the whole compiler is compiled into JS (and ASM) so that you can play with it in the browser.

Note	Documents under bloomberg.github.io match the master branch. If you find errors or omissions in this document, please don’t hesitate to submit an issue, sources are here.

Why BuckleScript

Benefits of JavaScript platform

JavaScript is not just the browser language, it’s also the only existing cross platform language. It is truly everywhere: users don’t need to install binaries or use package managers to access software, just a link will work.

Another important factor is that the JavaScript VM is quite fast and keeps getting faster. The JavaScript platform is therefore increasingly capable of supporting large applications.

Problems of JavaScript && how BuckleScript solves them

BuckleScript is mainly designed to solve the problems of large scale JavaScript programming:

Type-safety

OCaml offers an industrial-strength state-of-the-art type system and provides very strong type inference (i.e. No verbose type annotation required compared with TypeScript), which proves invaluable in managing large projects. OCaml’s type system is not just for tooling, it is a sound type system which means it is guaranteed that there will be no runtime type errors after type checking.

High quality dead code elimination

A large amount of web-development relies on inclusion of code dependencies by copying or referencing CDNs (the very thing that makes JavaScript highly accessible), but this also introduces a lot of dead code. This impacts performance adversely when the JavaScript VM has to interpret code that will never be invoked. BuckleScript provides powerful dead-code elimination at all levels:

Function and module level elimination is facilitated by the sophistication of the type-system of OCaml and purity analysis.
At the global level BuckleScript generates code ready for dead-code elimination done by bundling tools such as the Google closure-compiler.

Offline optimizations

JavaScript is a dynamic language, it takes a performance-hit for the VM to optimize code at runtime. While some JS engines circumvent the problem to some extent by caching, this is not available to all environments, and lack of a strong type system also limits the level of optimizations possible. Again, BuckleScript, using features of the OCaml type-system and compiler implementation is able to provide many optimizations during offline compilation, allowing the runtime code to be extremely fast.

JS platform and Native platform

Run your programs on all platforms, but run your system faster under specific platforms. JavaScript is everywhere but it does not mean we have to run all apps in JS, under several platforms, for example, server side or iOS/Android native apps, when programs are written in OCaml, it can also be compiled to native code for better and reliable performance.

While a strong type-system helps in countering these problems, at the same time we hope to avoid some of the problems faced in using other offline transpilation systems:

Slow compilation: OCaml byte-code compilation is known to be fast (one or two orders of magnitude faster than other similar languages: Scala or Haskell), BuckleScript shares the same property and compiles even faster since it saves the link time. See the speeds at work in the playground, the native backend is one order faster than the JS backend.
Un-readable JS Code and hard to integrate with existing JS libraries: When compiling to JavaScript, many systems generate code, that while syntactically and semantically correct is not human-readable and very difficult to debug and profile. Our BuckleScript implementation and the multi-pass compilation strategy of OCaml, allows us to avoid name-mangling, and produce JavaScript code that is human-readable and easier to debug and maintain. More importantly, this makes integration with existing JS libraries much easier.
Large JS output even for a simple program: In BuckleScript, a Hello world program generates 20 bytes JS code instead of 50K bytes. This is due to BuckleScript having an excellent integration with JS libs in that unlike most JS compilers, all BuckleScript’s runtime is written in OCaml itself so that these runtime libraries are only needed when user actually calls it.
Loss of code-structure: Many systems generate JavaScript code that is essentially a big ball of mud. We try to keep the original structure of the code by mapping one OCaml module to one JS module.

Installation

Below is a list of different ways to install BuckleScript:

Windows Installation

npm install bs-platform

*nix installation

Prerequisites

Standard C toolchain
npm (should be installed with Node)

The standard npm package management tool can be used to install BuckleScript. If you don’t already have npm installed, follow the directions listed here. Once npm is installed, run the following command:

npm install --save bs-platform

or install it globally

npm install -g bs-platform

Recommended installation with OPAM

When working with OCaml we also recommend using opam package manager to install OCaml toolchains, available here. You will benefit from the existing OCaml ecosystem.

Once you have opam installed, ask opam to switch to using our version of the compiler:

opam update
opam switch 4.02.3+buckle-master
eval `opam config env`
npm install bs-platform

Note that using this approach, the user can also install other OCaml tools easily.

Install from source

using NPM

Prerequisites:

Standard C toolchain
npm (should be installed with Node)

Instructions:

git clone https://github.com/bloomberg/bucklescript
cd bucklescript
npm install

Minimal dependencies

Prerequisites:

Standard C toolchain

BuckleScript has very few dependencies and building from source can easily be done.

Build OCaml compiler

git clone --recursive https://github.com/bloomberg/bucklescript
cd bucklescript/ocaml
./configure -prefix `pwd` # put your preferred directory
make world.opt
make install

The patched compiler is installed locally into your $(pwd)/bin directory. To start using it temporarily, check if ocamlc.opt and ocamlopt.opt exist in $(pwd)/bin, and temporarily add the location to your $(PATH) (e.g. PATH=$(pwd)/bin:$PATH).

Building BuckleScript

The following directions assume you already have the correct version of ocamlopt.opt in your $PATH, having followed the process described in the previous section.

make world

At the end, you should have a binary called bsc.exe under jscomp/bin directory, which you can add to your $PATH. You could also set an environment variable pointing to the stdlib, e.g. BSC_LIB=/path/to/jscomp/stdlib for ease of use.

Warning

The built compiler is not relocatable out of box, please don’t move it around unless you know what you are doing

Get Started

Using existing templates (@since 1.7.4 )

BuckleScript provide some project templates, to help people get started on board as fast as possible.

npm install -g bs-platform && bsb -init hello && cd hello && npm run build

First, it installs bs-platform globally, then we use the command line tool bsb to set up a sample project named hello, then we run the build.

The output project layout is similar as below

hello>tree -d .
.
|---.vscode
├── lib
│   ├── bs
│   │   └── src
│   └── js
│       └── src
├── node_modules
│   └── bs-platform -> ...
└── src

To enter watch mode, you can run npm run watch, then you can edit src/demo.ml and see it compiled on the fly, and changes are updated on lib/js/src/demo.js

If you happen to use VSCode as editor, we provide .vscode/tasks.json as well, type Tasks>build, it will enter watch mode, the great thing is that VSCode comes with a Problems panel which has a much nicer UI.

Currently there are not too many project templates, you can contribute more project templates here here,

First example by hand without samples

Create a directory called hello and create bsconfig.json and package.json as below:

bsconfig.json

{
    "name" : "hello",
    "sources" : { "dir" : "src"}
}

package.json

{
    "dependencies": {
        "bs-platform": "1.7.0" (1)
    },
    "scripts" : {
        "build" : "bsb",
        "watch" : "bsb -w" (2)
    }
}

Version should be updated accordingly
Watch mode when developing

Create src/main_entry.ml as below:
src/main_entry.ml
```
let () =
  print_endline "hello world"
```
Build the app
```
npm run build
```

Now you should see a file called lib/js/src/main_entry.js generated as below:

lib/js/src/main_entry.js

// GENERATED CODE BY BUCKLESCRIPT VERSION 1.0.1 , PLEASE EDIT WITH CARE
'use strict';


console.log("hello world");

/*  Not a pure module */ (1)

The compiler detects this module is impure due to the side effect.

User can also run watch mode to see changes instantly while editing.

npm run watch

Tip	The working code is available here:

An example with multiple modules

Now we want to create two modules, one file called fib.ml which exports fib function, the other module called main_entry.ml which will call fib.

Create a directory fib and create files bsconfig.json and package.json

bsconfig.json

{
    "name" : "helo",
    "sources" : { "dir" : "src"}
}

package.json

{
    "dependencies": {
        "bs-platform": "1.7.0"
    },
    "scripts" : {
        "build" : "bsb",
        "watch" : "bsb -w"
    }
}

Create file src/fib.ml and file src/main_entry.ml

src/fib.ml

let fib n =
  let rec aux n a b =
    if n = 0 then a
    else
      aux (n - 1) b (a+b)
  in aux n 1 1

src/main_entry.ml

let () =
  for i = 0 to 10 do
    Js.log (Fib.fib i) (1)
  done

Js module is a built-in module shipped with BuckleScript

Build the app

npm install
npm run build
node lib/js/src/main_entry.js

If everything goes well, you should see the output as below:

Built in NPM support

Build an OCaml library as a npm package

BuckleScript build system has built in support for NPM packages, please checkout the section about the bsb for NPM support.

Note	This section covers some basics of how NPM is supported internally, normal users are safe to skip this section.

BuckleScript extends the OCaml compiler options with several flags to provide a better experience for NPM users.

In general, you are expected to see two kinds of build artifacts, the generated JS files and metadata which your OCaml dependencies rely on.

Since CommonJS has no namespaces, to allow JS files to live in different directories, we have a flag

bsc.exe -bs-package-name $npm_package_name -bs-package-output modulesystem:path/to/your/js/dir -c a.ml

By passing this flag, bsc.exe will store your package_name and relative path to package.json in .cmj files. It will also generate JS files in the directory you specified. You can, and are encouraged to, store JavaScript files in a hierarchical directory.

For the binary artifacts (Note that this is not necessary if you only want your libraries to be consumed by JS developers, and it has benefit since end users don’t need these binary data any more), the convention is to store all *.cm data in a single directory package.json/lib/ocaml and Javascript files in a hierachical directory like package.json/lib/js

To use OCaml library as a npm package

If you follow the layout convention above, using an OCaml package is pretty straightforward:

bsc.exe -I path/to/ocaml/package/installed -c a.ml

Together

Your command line would be like this:

bsc.exe -I path/to/ocaml/package1/installed -I path/to/ocaml/package2/installed  -bs-package-name $npm_package_name -bs-package-output commonjs:path/to/lib/js/ -c a.ml

Examples

Please visit https://github.com/bloomberg/bucklescript-addons for more examples.

JS Calling OCaml

Since BuckleScript guarantees that all OCaml functions are exported as is, no extra work is required to expose OCaml functions to JavaScript.

Caution

external exports are not exported as JS functions, if you really want to export those external functions, please write val instead
operators are escaped, since Javascript does not support user defined operators. For example, instead of calling Pervasives.(^), you have to call Pervasives.$caret from your JavaScript functions

If users want to consume some OCaml features only available in OCaml but not in JS, we recommend users to export it as functions.

For example, data constructors are not available in JS

  type t =
    | Cons of int * t
    | Nil

Currently, we recommend the user expose the constructor as a function so that it can be constructed from the JS side.

let cons x y = Cons (x,y)
let nil = Nil

Note	In the future, we will derive these functions to automate this process.

BuckleScript annotations for Unicode and JS FFI support

To make OCaml work smoothly with JavaScript, we introduced several extensions to the OCaml language. These BuckleScript extensions facilitate the integration of native JavaScript code and improve the generated code.

Unicode support (@since 1.5.1)

Warning

In this section, we assume the source is encoded using UTF8.

In OCaml, string is an immutable byte sequence (like GoLang), so if the user types some Unicode

Js.log "你好"

It will be translated into

console.log("\xe4\xbd\xa0\xe5\xa5\xbd");

Luckily, OCaml allows customized multiple-line string support, BuckleScript reserves the delimiter js and j (j is unused so far).

Js.log {js|你好，
世界|js}

Output

console.log("你好，\n世界");

Inside the js delimiter, the escape convention is like JavaScript

Js.log {js|\x3f\u003f\b\t\n\v\f\r\0"'|js}

Output

console.log("\x3f\u003f\b\t\n\v\f\r\0\"\'");

Unicode support with string interpolation (@since 1.7.0)

Like {js||js}, {j||j} not only allow unicode point, but also variable interpolation.

For example

let world = {j|世界|j}
let hello_world = {j|你好，$world|j}

Users can parenthesize the interpreted variable as below:

let hello_world = {j|你好，$(world)|j}

Note the syntax of interporated variable is intentionally designed to be simple. Its lexical convention is as below:

identifier := leading_identifier_char identifier_chars
leading_identifier_char := 'a' .. 'z' | '_'
identifier_chars := 'a' .. 'z' | 'A' .. 'Z' | '0' .. '9' | '_' | '\'

FFI

Like TypeScript, when building type-safe bindings from JS to OCaml, users have to write type declarations. In OCaml, unlike TypeScript, users do not need to create a separate .d.ts file, since the type declarations are an integral part of OCaml.

The FFI is divided into several components:

Binding to simple functions and values
Binding to high-order functions
Binding to object literals
Binding to classes
Extensions to the language for debugger, regex, and embedding arbitrary JS code

Binding to simple JS functions values

This part is similar to traditional FFI, with syntax as described below:

external value-name : typexpr = external-declaration attributes
external-declaration := string-literal

Users need to declare types for foreign functions (JS functions for BuckleScript or C functions for native compiler) and provide customized attributes.

Binding to global value: bs.val

external imul : int -> int -> int = "Math.imul" [@@bs.val]
type dom
(* Abstract type for the DOM *)
external dom : dom = "document" [@@bs.val]

bs.val attribute is used to bind to a JavaScript value, it can be a function or plain value.

Note

If external-declaration is the same as value-name, the user can leave external-declaration empty. For example:
```
external document : dom = "" [@@bs.val]
```
If you want to make a single FFI for both C functions and JavaScript functions, you can give the JavaScript foreign function a different name:
```
external imul : int -> int -> int =
  "c_imul" [@@bs.val "Math.imul"]
```

Scoped values: bs.scope (@since 1.7.2)

In JS library, it is quite common to use a name as namespace, for example, if the user want to write a binding to vscode.commands.executeCommand, assume vscode is a module name, the user needs needs to type commands properly before typing executeCommand, and in practice, it is rarely useful to call vscode.commands alone, for this reason, we introduce a convient sugar: bs.scope

Example

type param
external executeCommands : string -> param array -> unit = ""
[@@bs.scope "commands"] [@@bs.module "vscode"][@@bs.splice]

let f a b c  =
  executeCommands "hi"  [|a;b;c|]

Output

var Vscode = require("vscode");
function f(a, b, c) {
  Vscode.commands.executeCommands("hi", a, b, c);
  return process.env;
}

NOTE bs.scope can also be chained as below:

Exmaple

external makeBuffer : int -> buffer = "Buffer"
[@@bs.new] [@@bs.scope "global"]
external hi : string = ""
[@@bs.module "z"] [@@bs.scope "a0", "a1", "a2"]
external ho : string = ""
[@@bs.val] [@@bs.scope "a0","a1","a2"]
external imul : int -> int -> int = ""
[@@bs.val] [@@bs.scope "Math"]
let f2 ()  =
  makeBuffer 20 , hi , ho, imul 1 2

Output

var Z      = require("z");
function f2() {
  return /* tuple */[
          new (global.Buffer)(20),
          Z.a0.a1.a2.hi,
          a0.a1.a2.ho,
          Math.imul(1, 2)
        ];
}

Binding to JavaScript constructor: bs.new

bs.new is used to create a JavaScript object.

external create_date : unit -> t = "Date" [@@bs.new]
let date = create_date ()

Output:

var date = new Date();

Binding to a value from a module: bs.module

Input:

external add : int -> int -> int = "add" [@@bs.module "x"]
external add2 : int -> int -> int = "add2"[@@bs.module "y", "U"] (1)
let f = add 3 4
let g = add2 3 4

"U" will hint the compiler to generate a better name for the module, see output

Output:

var U = require("y");
var X = require("x");
var f = X.add(3, 4);
var g = U.add2(3, 4);

Note	if `external-declaration` is the same as value-name, it can be left empty, for example, `external add : int -> int -> int = "" [@@bs.module "x"]`

Binding the whole module as a value or function

type http
external http : http = "http" [@@bs.module] (1)

external-declaration is the module name

Note	if `external-declaration` is the same as value-name, it can be left empty, for example: `external http : http = "" [@@bs.module]`

Binding to method: bs.send, bs.send.pipe

bs.send helps the user send a message to a JS object.

type id (** Abstract type for id object *)
external get_by_id : dom -> string -> id =
  "getElementById" [@@bs.send]

The object is always the first argument and actual arguments follow.

Input:

get_by_id dom "xx"

Output:

dom.getElementById("xx")

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:

external map : ('a -> 'b [@bs]) -> 'b array =
  "" [@@bs.send.pipe: 'a array] (1)
external forEach: ('a -> unit [@bs]) -> 'a array =
  "" [@@bs.send.pipe: 'a array]
let test arr =
    arr
    |> map (fun [@bs] x -> x + 1)
    |> forEach (fun [@bs] x -> Js.log x)

For the [@bs] attribute in the callback, see Binding to callbacks (high-order function)

Note	if `external-declaration` is the same as value-name, it can be left empty, for example: `external getElementById : dom -> string -> id = "" [@@bs.send]`

Binding to dynamic key access/set: bs.set_index, bs.get_index

This attribute allows dynamic access to a JavaScript property

type t
external create : int -> t = "Int32Array" [@@bs.new]
external get : t -> int -> int = "" [@@bs.get_index]
external set : t -> int -> int -> unit = "" [@@bs.set_index]

Binding to Getter/Setter: bs.get, bs.set

This attribute helps get and set the property of a JavaScript object.

type textarea
external set_name : textarea -> string -> unit = "name" [@@bs.set]
external get_name : textarea -> string = "name" [@@bs.get]

Splice calling convention: bs.splice

In JS, it is quite common to have a function take variadic arguments. BuckleScript supports typing homogeneous variadic arguments. For example,

external join : string array -> string = "" [@@bs.module "path"] [@@bs.splice]
let v = join [| "a"; "b"|]

Output

var Path = require("path")
var v = Path.join("a","b")

Note	For the external call, if the `array` arguments is not a compile time array, the compiler will emit an error message.

Special types on external declarations: bs.string, bs.int, bs.ignore, bs.as

Using polymorphic variant to model enums and string types

There are several patterns heavily used in existing JavaScript codebases, for example, the string type is used a lot. BuckleScript FFI allows the user to model string type in a safe way by using annotated polymorphic variant.

external readFileSync :
  name:string ->
  ([ `utf8
   | `my_name [@bs.as "ascii"] (1)
   ] [@bs.string]) ->
  string = ""
  [@@bs.module "fs"]

let _ =
  readFileSync ~name:"xx.txt" `my_name

Here we intentionally made an example to show how to customize a name

Ouptut:

var Fs = require("fs");
Fs.readFileSync("xx.txt", "ascii");

Polymorphic variants can also be used to model enums.

external test_int_type :
  ([ `on_closed (1)
   | `on_open [@bs.as 3] (2)
   | `in_bin (3)
   ]
   [@bs.int]) -> int =
  "" [@@bs.val]

`on_closed will be encoded as 0
`on_open will be 3 due to the attribute bs.as
`in_bin will be 4

Using polymorphic variant to model event listener

BuckleScript models this in a type-safe way by using annotated polymorphic variants.

type readline
external on :
    (
    [ `close of unit -> unit
    | `line of string -> unit
    ] (1)
    [@bs.string])
    -> readline = "" [@@bs.send.pipe: readline]
let register rl =
  rl
  |> on (`close (fun event -> () ))
  |> on (`line (fun line -> print_endline line))

This is a very powerful typing: each event can have its own different types.

Output:

function register(rl) {
  return rl.on("close", function () {
                return /* () */0;
              })
           .on("line", function (line) {
              console.log(line);
              return /* () */0;
            });
}

Warning

These annotations will only have effect in external declarations.
The runtime encoding of using polymorphic variant is internal to the compiler.
With these annotations mentioned above, BuckleScript will automatically transform the internal encoding to the designated encoding for FFI. BuckleScript will try to do such conversion at compile time if it can, otherwise, it will do such conversion in the runtime, but it should be always correct.

Phantom Arguments and ad-hoc polymorphism

bs.ignore allows arguments to be erased after passing to JS functional call, the side effect will still be recorded.

For example,

external add : (int [@bs.ignore]) -> int -> int = ""
[@@bs.val]
let v = add 0 1 2 (1)

the first argument will be erased

Output:

var v = add (1,2)

This is very useful to combine GADT:

type _ kind =
  | Float : float kind
  | String : string kind
external add : ('a kind [@bs.ignore]) -> 'a -> 'a -> 'a = "" [@@bs.val]

let () =
  Js.log (add Float 3.0 2.0);
  Js.log (add String "x" "y");

User can also have a payload for the GADT:

let string_of_kind (type t) (kind : t kind) =
  match kind with
  | Float -> "float"
  | String -> "string"

external add_dyn : ('a kind [@bs.ignore]) -> string -> 'a -> 'a -> 'a = ""
[@@bs.val]

let add2 k x y =
  add_dyn k (string_of_kind k) x y

Fixed Arguments

Contrary to the Phantom arguments, _[@bs.as] is introduced to attach the const data.

For example:

external process_on_exit : (_ [@bs.as "exit"]) -> (int -> unit) -> unit =
  "process.on" [@@bs.val]

let () =
    process_on_exit (fun exit_code ->
        Js.log( "error code: " ^ string_of_int exit_code ))

Output

process.on("exit", function (exit_code) {
      console.log("error code: " + exit_code);
      return /* () */0;
    });

It can also be used in combination with other attributes, for example:

type process

external on_exit : (_ [@bs.as "exit"]) -> (int -> unit) -> unit =
    "on" [@@bs.send.pipe: process]
let register (p : process) =
        p |> on_exit (fun i -> Js.log i)

Output

function register(p) {
  return p.on("exit", function (i) {
              console.log(i);
              return /* () */0;
            });
}

Input

external io_config :
    stdio:(_ [@bs.as "inherit"]) -> cwd:string -> unit -> _ = "" [@@bs.obj]

let config = io_config ~cwd:"." ()

Output

var config = {
  stdio: "inherit",
  cwd: "."
};

Fixed Arguments with arbitrary JSON literal (@since 1.7.0)

So the payload can be more flexiblie with JSON literal support

external on_exit_slice5 :
    int
    -> (_ [@bs.as 3])
    -> (_ [@bs.as {json|true|json}])
    -> (_ [@bs.as {json|false|json}])
    -> (_ [@bs.as {json|"你好"|json}])
    -> (_ [@bs.as {json| ["你好",1,2,3] |json}])
    -> (_ [@bs.as {json| [{ "arr" : ["你好",1,2,3], "encoding" : "utf8"}] |json}])
    -> (_ [@bs.as {json| [{ "arr" : ["你好",1,2,3], "encoding" : "utf8"}] |json}])
    -> (_ [@bs.as "xxx"])
    -> ([`a|`b|`c] [@bs.int])
    -> (_ [@bs.as "yyy"])
    -> ([`a|`b|`c] [@bs.string])
    -> int array
    -> unit
    =
    "xx" [@@bs.send.pipe: t] [@@bs.splice]

let _ = x |> on_exit_slice5 __LINE__ `a `b [|1;2;3;4;5|]

Output

x.xx(114, 3, true, false, ("你好"), ( ["你好",1,2,3] ), ( [{ "arr" : ["你好",1,2,3], "encoding" : "utf8"}] ), ( [{ "arr" : ["你好",1,2,3], "encoding" : "utf8"}] ), "xxx", 0, "yyy", "b", 1, 2, 3, 4, 5)

Binding to NodeJS special variables: bs.node

NodeJS has several file local variables: dirname, filename, _module, and require. Their semantics are more like macros instead of functions.

BuckleScript provides built-in macro support for these variables:

let dirname : string option = [%bs.node __dirname]
let filename : string option = [%bs.node __filename]
let _module : Node.node_module option = [%bs.node _module]
let require : Node.node_require option = [%bs.node require]

Binding to callbacks (high-order function)

High order functions are functions where the callback can be another function. For example, suppose JS has a map function as below:

function map (a, b, f){
  var i = Math.min(a.length, b.length);
  var c = new Array(i);
  for(var j = 0; j < i; ++j){
    c[j] = f(a[i],b[i])
  }
  return c ;
}

A naive external type declaration would be as below:

external map : 'a array -> 'b array -> ('a -> 'b -> 'c) -> 'c array = "" [@@bs.val]

Unfortunately, this is not completely correct. The issue is by reading the type 'a → 'b → 'c, it can be in several cases:

let f x y = x + y

let g x = let z = x + 1 in fun y -> x + z

In OCaml they all have the same type; however, f and g may be compiled into functions with different arities.

A naive compilation will compile f as below:

let f = fun x -> fun y -> x + y

function f(x){
  return function (y){
    return x + y;
  }
}
function g(x){
  var z = x + 1 ;
  return function (y){
    return x + z ;
  }
}

Its arity will be consistent but is 1 (returning another function); however, we expect its arity to be 2.

Bucklescript uses a more complex compilation strategy, compiling f as

function f(x,y){
  return x + y ;
}

No matter which strategy we use, existing typing rules cannot guarantee a function of type 'a → 'b → 'c will have arity 2.

[@bs] for explicit uncurried callback

To solve this problem introduced by OCaml’s curried calling convention, we support a special attribute [@bs] at the type level.

external map : 'a array -> 'b array -> ('a -> 'b -> 'c [@bs]) -> 'c array
= "map" [@@bs.val]

Here ('a → 'b → 'c [@bs]) will always be of arity 2, in general, 'a0 → 'a1 … 'aN → 'b0 [@bs] is the same as 'a0 → 'a1 … 'aN → 'b0 except the former’s arity is guaranteed to be N while the latter is unknown.

To produce a function of type 'a0 → .. 'aN → 'b0 [@bs], as follows:

let f : 'a0 -> 'a1 -> .. 'b0 [@bs] =
  fun [@bs] a0 a1 .. aN -> b0
let b : 'b0 = f a0 a1 a2 .. aN [@bs]

A special case for arity of 0:

let f : unit -> 'b0 [@bs] = fun [@bs] () -> b0
let b : 'b0 = f () [@bs]

Note that this extension to the OCaml language is sound. If you add an attribute in one place but miss it in other place, the type checker will complain.

Another more complex example:

type 'a return = int -> 'a [@bs]
type 'a u0 = int -> string -> 'a return [@bs] (1)
type 'a u1 = int -> string -> int -> 'a [@bs] (2)
type 'a u2 = int -> string -> (int -> 'a [@bs]) [@bs] (3)

u0 has arity of 2, return a function with arity 1
u1 has arity of 3
u2 has arity of 2, return a function with arity 1

[@bs.uncurry] for implicit uncurried callback (@since 1.5.0)

Note the [@bs] annotation already solved the problem completely, but it has a drawback that it requires users to write [@bs] both in definition site and call site.

For example:

external map : 'a array -> ('a -> 'b[@bs]) -> 'b array = "" [@@bs.send] (1)
map [|1;2;3|] (fun [@bs] x -> x + 1) (2)

[@bs] annotation in definition site
[@bs] annotation in call site

This is less convenient for end users, so we introduce another implicit annotation [@bs.uncurry] so that the compiler will automatically wrap the curried callback (from OCaml side) to JS uncurried callback. In this way, the [@bs.uncurry] annotation is defined only once.

external map : 'a array -> ('a -> 'b [@bs.uncurry]) -> 'b array = "" [@@bs.send] (1)
map [|1;2;3|] (fun x -> x+ 1) (2)

[@bs.uncurry] annotation in definition site
Idiomatic OCaml code

Note

In general, bs.uncurry is recommended, and compiler will do lots of optimizations to resolve the curry to uncurry calling convention at compile time. However, there are some cases the compiler optimizer could not do it, in that case, it will be converted runtime.

This means [@bs] are completely static behavior (no any runtime cost), while [@bs.uncurry] is more convenient for end users but in some very rare cases it might be slower than [@bs]

Uncurried calling convention as an optimization

Background:

As we discussed before, we can compile any OCaml function as arity 1 to support OCaml’s curried calling convention.

This model is simple and easy to implement, but the native compilation is very slow and expensive for all functions.

let f x y z = x + y + z
let a = f 1 2 3
let b = f 1 2

can be compiled as

function f(x){
  return function (y){
    return function (z){
      return x + y + z
    }
  }
}
var a = f (1) (2) (3)
var b = f (1) (2)

But as you can see, this is highly inefficient, since the compiler already saw the source definition of f, it can be optimized as below:

function f(x,y,z) {return x + y + z}
var a = f(1,2,3)
var b = function(z){return f(1,2,z)}

BuckleScript does this optimization in the cross module level and tries to infer the arity as much as it can.

Callback optimization

However, such optimization will not work with high-order functions, i.e, callbacks.

For example,

let app f x = f x

Since the arity of f is unknown, the compiler can not do any optimization (unless app gets inlined), so we have to generate code as below:

function app(f,x){
  return Curry._1(f,x);
}

Curry._1 is a function to dynamically support the curried calling convention.

Since we support the uncurried calling convention, you can write app as below

let app f x = f x [@bs]

Now the type system will infer app as type ('a →'b [@bs]) → 'a and compile app as

function app(f,x){
  return f(x)
}

Note	In OCaml the compiler internally uncurries every function declared as `external` and guarantees that it is always fully applied. Therefore, for `external` first-order FFI, its outermost function does not need the `[@bs]` annotation.

Bindings to `this` based callbacks: bs.this

Many JS libraries have callbacks which rely on this (the source), for example:

x.onload = function(v){
  console.log(this.response + v )
}

Here, this would be the same as x (actually depends on how onload is called). It is clear that it is not correct to declare x.onload of type unit → unit [@bs]. Instead, we introduced a special attribute bs.this allowing us to type x as below:

type x
external set_onload : x -> (x -> int -> unit [@bs.this]) -> unit = "onload" [@@bs.set]
external resp : x -> int = "response" [@@bs.get]
set_onload x begin fun [@bs.this] o v ->
  Js.log(resp o + v )
end

Output:

x.onload = function(v){
  var o = this ; (1)
  console.log(o.response + v);
}

The first argument is automatically bound to this

bs.this is the same as bs : except that its first parameter is reserved for this and for arity of 0, there is no need for a redundant unit type:

let f : 'obj -> 'b [@bs.this] =
  fun [@bs.this] obj -> ....
let f1 : 'obj -> 'a0 -> 'b [@bs.this] =
  fun [@bs.this] obj a -> ...

Note

There is no way to consume a function of type 'obj → 'a0 .. → 'aN → 'b0 [@bs.this] on the OCaml side. This is an intentional design choice, we don’t encourage people to write code in this style.

This was introduced mainly to be consumed by existing JS libraries. User can also type x as a JS class too (see later)

Binding to JS objects

Convention:

All JS objects of type 'a are lifted to type 'a Js.t to avoid conflict with OCaml’s native object system (we support both OCaml’s native object system and FFI to JS’s objects), ## is used in JS’s object method dispatch and field access, while # is used in OCaml’s object method dispatch.

Typing JavaScript objects:

OCaml supports object oriented style natively and provides structural type system. OCaml’s object system has different runtime semantics from JS object, but they share the same type system, all JS objects of type 'a are typed as 'a Js.t

OCaml provides two kinds of syntaxes to model structural typing: < p1 : t1 > style and class type style. They are mostly the same except that the latter is more feature rich (supporting inheritance) but more verbose.

Simple object type

Suppose we have a JS file demo.js which exports two properties: height and width:

demo.js

exports.height = 3
exports.width  = 3

There are different ways to writing binding to module demo, here we use OCaml objects to model module demo

external demo : < height : int ; width : int > Js.t = "" [@@bs.module]

There are two kinds of types on the method name:

normal type

< label : int >
< label : int -> int >
< label : int -> int [@bs]>
< label : int -> int [@bs.this]>

method
```
< label : int -> int [@bs.meth] >
```

The difference is that for method, the type system will force users to fulfill its arguments all at the same time, since its semantics depends on this in JavaScript.

For example:

let test f =
  f##hi 1 (1)
let test2 f =
  let u = f##hi in
  u 1
let test3 f =
  let u = f##hi in
  u 1 [@bs]

## is JS object property/method dispatch

The compiler would infer types differently

val test : < hi : int -> 'a [@bs.meth]; .. > -> 'a (1)
val test2 : < hi : int -> 'a ; .. > -> 'a
val test3 : < hi : int -> 'a [@bs]; .. >

.. is a row variable, which means the object can contain more methods.

Complex object type

Below is an example:

class type _rect = object
  method height : int
  method width : int
  method draw : unit -> unit
end [@bs] (1)
type rect = _rect Js.t

class type annotated with [@bs] is treated as a JS class type, it needs to be lifted to Js.t too.

For JS classes, methods with arrow types are treated as real methods (automatically annotated with [@bs.meth]) while methods with non-arrow types are treated as properties.

So the type rect is the same as below:

type rect = < height : int ; wdith : int ; draw : unit -> unit [@bs.meth] > Js.t

How to consume JS property and methods

As we said: ## is used in both object method dispatch and field access.

f##property (1)
f##property #= v
f##js_method args0 args1 args2 (2)

property get should not come with any argument as we discussed above, which will be checked by the compiler.
Here method is of arity 3.

Note

All JS method application is uncurried, JS’s method is not a function, this invariant can be guaranteed by OCaml’s type checker, a classic example shown below:

console.log('fine')
var log = console.log;
log('fine') (1)

May cause exception, implementation dependent, console.log may depend on this

In BuckleScript

let fn = f0##f in
let a = fn 1 2
(* f##field a b would think `field` as a method *)

is different from

let b = f1##f 1 2

The compiler will infer as below:

val f0 : < f : int -> int -> int > Js.t
val f1 : < f : int -> int -> int [@bs.meth] > Js.t

If we type console properly in OCaml, user could only write

console##log "fine"
let u = console##log
let () = u "fine" (1)

OCaml compiler will complain

Note	If a user were to make such a mistake, the type checker would complain by saying it expected `Js.method` but saw a function instead, so it is still sound and type safe.

getter/setter annotation to JS properties

Since OCaml’s object system does not have getters/setters, we introduced two attributes bs.get and bs.set to help inform BuckleScript to compile them as property getters/setters.

type y = <
  height : int [@bs.set {no_get}] (1)
> Js.t
type y0 = <
  height : int [@bs.set] [@bs.get {null}] (2)
> Js.t
type y1 = <
  height : int [@bs.set] [@bs.get {undefined}] (3)
> Js.t
type y2 = <
  height : int [@bs.set] [@bs.get {undefined; null}] (4)
> Js.t
type y3 = <
  height : int [@bs.get {undefined ; null}] (5)
> Js.t

height is setter only
getter return int Js.null
getter return int Js.undefined
getter return int Js.null_undefined
getter only, return int Js.null_undefined

Note	Getter/Setter also applies to class type label

Create JS objects using bs.obj

Not only can we create bindings to JS objects, but also we can create JS objects in a type safe way on the OCaml side:

let u = [%bs.obj { x = { y = { z = 3}}} ] (1)

bs.obj extension is used to mark {} as JS objects

Output:

var u = { x : { y : { z : 3 }}}}

The compiler would infer u as type:

val u : < x : < y : < z : int > Js.t > Js.t > Js.t

To make it more symmetric, extension bs.obj can also be applied into the type level, so you can write:

val u : [%bs.obj: < x : < y : < z : int > > > ]

Users can also write expression and types together as below:

let u = [%bs.obj ( { x = { y = { z = 3 }}} : < x : < y : < z : int > > > ]

Objects in a collection also works:

let xs = [%bs.obj [| { x = 3 } ; { x = 3 } |] : < x : int > array ]
let ys = [%bs.obj [| { x = 3 } ; { x = 4 } |] ]

Output:

var xs = [ { x : 3 } , { x : 3 } ]
var ys = [ { x : 3 } , { x : 4 } ]

Create JS objects using external

bs.obj can also be used as an attribute in external declarations, as below:

external make_config : hi:int -> lo:int -> unit -> t = "" [@@bs.obj]
let v = make_config ~hi:2 ~lo:3

Output:

var v = { hi : 2 , lo : 3 }

Option argument is also supported:

external make_config : hi:int -> ?lo:int -> unit -> t = "" [@@bs.obj] (1)
let u = make_config ~hi:3 ()
let v = make_config ~lo:2 ~hi:3 ()

In OCaml, the order of label does not matter, and the evaluation order of arguments is undefined. Since the order does not matter, to make sure the compiler realize all the arguments are fulfilled (including optional arguments), it is common to have a unit type before the result.

Output:

var u = {hi : 3}
var v = {hi : 3 , lo: 2}

Now, we can write JS style code in OCaml too (in a type safe way):

let u = [%bs.obj {
  x = { y = { z = 3 } };
  fn = fun [@bs] u v -> u + v (1)
  } ]
let h = u##x##y##z
let a = u##fn
let b = a 1 2 [@bs]

fn property is not method, it does not rely on this. We will show how to create JS method in OCaml later.

Output:

var u = { x : { y : { z : 3 } }, fn : function (u, v) {return u + v}}
var h = u.x.y.z
var a = u.fn
var b = a(1,2)

Note	When the field is an uncurried function, a short-hand syntax `#@` is available: `let b x y h = h#@fn x y` `function b (x,y,h){ return h.fn(x,y) }` The compiler will infer the type of `b` as `val b : 'a -> 'b -> < fn : 'a -> 'b -> 'c [@bs] > Js.t -> 'c`

Create JS objects with `this` semantics

The objects created above can not use this in the method, this is supported in BuckleScript too.

let v2 =
  let x = 3. in
  object (self) (1)
    method hi x y = self##say x +. y
    method say x = x *. self##x ()
    method x () = x
  end [@bs] (2)

self is bound to this in generated JS code
[@bs] marks object .. end as a JS object

Output:

var v2 = {
  hi: function (x, y) {
    var self = this ;
    return self.say(x) + y;
  },
  say: function (x) {
    var self = this ;
    return x * self.x();
  },
  x: function () {
    return 3;
  }
};

Compiler infers the type of v2 as below:

val v2 : <
  hi : float -> float -> float [@bs.meth];
  say : float -> float [@bs.meth];
  x : unit -> float [@bs.meth]
> [@bs]

Below is another example to consume a JS object :

let f (u : rect) =
  (* the type annotation is un-necessary,
     but it gives better error message
  *)
   Js.log u##height;
   Js.log u##width;
   u##width #= 30;
   u##height #= 30;
   u##draw ()

Output:

function f(u){
  console.log(u.height);
  console.log(u.width);
  u.width = 30;
  u.height = 30;
  return u.draw()
}

Method chaining

f
##(meth0 ())
##(meth1 a)
##(meth2 a b)

Object label translation convention

There are two cases, where we might want to do name mangling for a JS object method name.

First, in OCaml, some names are keywords, so we want to add an underscore to avoid a syntax error.

Key-word method:

f##_open
f##_MAX_LENGTH

OUTPUT:

f.open
f.MAX_LENGTH

Second, it is common to have several types for a single method. To model this ad-hoc polymorphism, we introduced a small convention when translating object labels, which is occasionally useful as below

Ad-hoc polymorphism

f##draw__cat (x,y)
f##draw__dog (x,y)

OUTPUT:

f.draw(x,y) // f.draw in JS can accept different types
f.draw(x,y)

Note

Rules

If __[rest] appears in the label, index from the right to left.
- If index = 0, nothing mangled
- If index > 0, __[rest] is dropped
Else if _ is the first char
- If the following char is not 'a' .. 'z', drop the first '_'
- Else if the rest happens to be a keyword, drop the first '_'
- Else, nothing mangled

Return value checking (@since 1.5.1)

In general, the FFI code is error prone, and potentially will leak in undefined or null values.

So we introduced auto coercion for return values to gain two benefits:

More safety for FFI code without performance cost (explained later).
More idiomatic OCaml code for users to consume the FFI.

Below is a contrived core example:

type element
type dom
external getElementById : string -> element option = ""
[@@bs.send.pipe:dom] [@@bs.return null_to_opt] (1)

let test dom =
    let elem = dom |> getElementById "haha" in
    match elem with
    | None -> 1
    | Some ui -> Js.log ui ; 2

null_to_opt attribute will automatically convert null to option

Output

function test(dom) {
  var elem = dom.getElementById("haha");
  if (elem !== null) { (1)
    console.log(elem);
    return 2;
  }
  else {
    return 1;
  }
}

nullable checking without boxing due to compiler optimizations

Currently 4 directives are supported: null_to_opt, undefined_to_opt, null_undefined_to_opt and identity.

Note

null_to_opt, undefined_to_opt and null_undefined_to_opt will semantically convert a nullable value to option which is a boxed value, but the compiler will do smart optimizations to remove such boxing overhead when the returned value is destructed in the same routine.

The three directives above require users to write literally _ option. It is in theory not necessary, but it is required to reduce user errors.

When the return type is unit: the compiler will append its return value with an OCaml unit literal to make sure it does return unit. Its main purpose is to make the user consume FFI in idiomatic OCaml code, the cost is very very small and the compiler will do smart optimizations to remove it when the returned value is not used (mostly likely).

When the return type is bool, the compiler will coerce its return value from JS boolean to OCaml boolean. The cost is also very small and compiler will remove such coercion when it is not needed. Note even if your external FFI does return OCaml bool or unit, such implicit coercion will cause no harm.

identity will make sure that compiler will do nothing about the returned value. It is rarely used, but introduced here for debugging purpose.

Embedding untyped Javascript code

Warning

This is not encouraged. The user should minimize and localize use cases of embedding raw JavaScript code, however, sometimes it’s necessary to get the job done.

Detect global variable existence `bs.external` (@since 1.5.1)

Before we dive into embedding arbitrary JS code, a quite common use case of embedding untyped JS code is detect a global variable (feature detection), Bucklescript provides a relatively type safe approach for such use case: bs.external (or extenral), [%bs.external a_single_identifier] is a value of _ option type, see examples below

let test () =
  match [%external __DEV__] with
  | Some _ -> Js.log "dev mode"
  | None -> Js.log "producton mode"

Output

function test() {
  var match = typeof (__DEV__) === "undefined" ? undefined : (__DEV__);
  if (match !== undefined) {
    console.log("dev mode");
    return /* () */0;
  }
  else {
    console.log("producton mode");
    return /* () */0;
  }
}

let test2 () =
  match [%external __filename] with
  | Some f -> Js.log f
  | None -> Js.log "non node environment"

Output

function test2() {
  var match = typeof (__filename) === "undefined" ? undefined : (__filename);
  if (match !== undefined) {
    console.log(match);
    return /* () */0;
  }
  else {
    console.log("non node environment");
    return /* () */0;
  }
}

Embedding arbitrary JS code as an expression

let keys : t -> string array [@bs] = [%bs.raw "Object.keys" ]
let unsafe_lt : 'a -> 'a -> Js.boolean [@bs] = [%bs.raw{|function(x,y){return x < y}|}]

We highly recommend writing type annotations for such unsafe code. It is unsafe to refer to external OCaml symbols in raw JS code.

Embedding raw JS code as statements

[%%bs.raw{|
  console.log ("hey");
|}]

Other examples:

let x : string = [%bs.raw{|"\x01\x02"|}]

It will be compiled into:

var x = "\x01\x02"

Polyfill of Math.imul

   [%%bs.raw{|
   // Math.imul polyfill
   if (!Math.imul){
       Math.imul = function (..) {..}
    }
   |}]

Warning

So far we don’t perform any sanity checks in the quoted text (syntax checking is a long-term goal).
Users should not refer to symbols in OCaml code. It is not guaranteed that the order is correct.

Debugger support

We introduced the extension bs.debugger, for example:

  let f x y =
    [%bs.debugger];
    x + y

which will be compiled into:

  function f (x,y) {
     debugger; // JavaScript developer tools will set an breakpoint and stop here
     x + y;
  }

Regex support

We introduced bs.re for Javascript regex expressions:

let f = [%bs.re "/b/g"]

The compiler will infer f has type Js.Re.t and generate code as below:

var f = /b/g

Note	`Js.Re.t` can be accessed and manipulated using the functions available in the `Js.Re` module.

Examples

Below is a simple example for the mocha library. For more examples, please visit https://github.com/bloomberg/bucklescript-addons

A simple example: binding to mocha unit test library

This is an example showing how to provide bindings to the mochajs unit test framework.

external describe : string -> (unit -> unit [@bs]) -> unit = "" [@@bs.val]
external it : string -> (unit -> unit [@bs]) -> unit = "" [@@bs.val]

Since, mochajs is a test framework, we also need some assertion tests. We can also describe the bindings to assert.deepEqual from the nodejs assert library:

external eq : 'a -> 'a -> unit = "deepEqual" [@@bs.module "assert"]

On top of this we can write normal OCaml functions, for example:

let assert_equal = eq
let from_suites name suite =
    describe name (fun [@bs] () ->
         List.iter (fun (name, code) -> it name code) suite
         )

The compiler would generate code as below:

 var Assert = require("assert");
 var List = require("bs-platform/lib/js/list");

function assert_equal(prim, prim$1) {
 return Assert.deepEqual(prim, prim$1);
 }

function from_suites(name, suite) {
 return describe(name, function () {
   return List.iter(function (param) {
    return it(param[0], param[1]);
      }, suite);
  });
 }

Exception handling between OCaml and JS (@since 1.7.0)

In Js world, exception could be any data, while OCaml exception is structured data format and supports pattern match, catch OCaml exception on JS side is no-op.

Catch JS exception

To catch Js exception on OCaml side, we categorize all JS exceptions to belong to Js.Exn.Error.

let example1 () =
    match Js.Json.exnParse {| {"x" }|} with
    | exception Js.Exn.Error err ->
        Js.log @@ Js.Exn.stack err;
        None
    | v -> Some v

let example2 () =
    try Some (Js.Json.exnParse {| {"x"}|}) with
    Js.Exn.Error _ -> None

The exception definition of Js.Exn.Error is as below:

type t =
  < stack : string Js.undefined ;
    message : string Js.undefined ;
    name : string Js.undefined;
    fileName : string Js.undefined
  > Js.t

exception Error of t

Raise JS style exception

We provide such functions

(** Raise Js exception Error object with stacktrace *)
val error : string -> 'a
val evalError : string -> 'a
val rangeError : string -> 'a
val referenceError : string -> 'a
val syntaxError : string -> 'a
val typeError : string -> 'a
val uriError : string -> 'a

Please consult module Js.Error for more details

`bs.open`: Type safe external data-source handling (@@since 1.7.0)

There are some cases, the data-source could either come from JS land or OCaml land, it is very hard to give precise type information. For example, for an extenral promise whose creation could come from JS API, its failed value caused by Promise.reject could be in any shape.

BuckleScript provides a solution to filter out OCaml structured exception data from the mixed data source, it preserves the type safety while allow uesrs to deal with mixed source.

It makes use of OCaml’s extensible variant, so that uers can pack mix value of type exn with JS data-source

Example

let handleData = function [@bs.open]
   | Invalid_argument _ -> 0
   | Not_found -> 1
   | Sys_error _ -> 2

val handleData : 'a -> int option (1)

For any input source, as long as it matches the exception pattern (nested pattern match supported), the matched value is returned, otherwise return None.

Use cases

Take promise for example:

let v = Promise.reject Not_found
let handlePromiseFaiure = function [@bs.open]
   | Not_found -> Js.log "Not found"; (Promise.resolve ())

let () =
   v
   |> Promise.catch (fun error ->
        match handlePromiseFaiure error with
        | Some x -> x
        | None -> raise UnhandledPromise
    )

Js module

API reference

Js module is shipped with BuckleScript, both the namespace Js and Node are preserved.

Null and Undefined

BuckleScript does not deal with null and undefined on the language level. Instead they are defined on the library level, represented by Js.Null.empty and Js.Undefined.empty (and Js.Null_undefined.empty) types. But due to this being Ocaml and all, we can’t just pass these values off as if it was any type we’d like it to be, so in order to actually use them we need to lift the type we want to use them with. For this purpose BuckleScript defines the type 'a Js.null, 'a Js.undefined' and ’a Js.null_undefined (for values that can be both null, undefined and 'a). As well as the corresponding modules Js.Null, Js.Undefined and Js.Null_undefined for working with these tpyes.

Here’s an example showing Js.Null used directly:

external get_prop_name_maybe : unit -> string Js.null = "" [@@bs.val]
external set_or_unset_prop : string -> int Js.null -> unit = "" [@@bs.val]

(* unset property if fond *)
let s = get_prop_name_maybe ()
let _ = Js.Null.iter s (function name -> set_or_unset_prop name Js.Null.empty)

(* set some other property *)
let _ = set_or_unset_prop "some_other_property" (Js.Null.return "")

You might also want to map these types to option for a friendlier and more idiomatic API, if the distinction between null and undefined isn’t important to maintain:

external try_some_thing : string Js.null -> int Js.null = ...

let try_some_thing_maybe ( v : string option) : int option =
  Js.Null.to_opt (try_some_thing (Js.Null.from_opt v))

Boolean

"Native" bool is for Bob-ish reasons not represented as true and false literals in JavaScript. For interop you therefore need to use the Js.boolean and its Js.true_ and Js.false_ values. There are also convenience functions for coercing to and from bool: Js.to_bool and Js.Boolean.to_js_boolean.

Stable-ish submodules

These APIs might change, but is considered mostly complete

Experimental/incomplete submodules

Expect these APIs to change drastically, use only if you accept breakage.

Very experimental/internal submodules

Do not use these APIs unless absolutely necessary, they’re mostly internal

Node module

Node

Submodules

Extended compiler options

This section is only for people who want roll their own build system instead of using the recommended build system: BuckleScript build system: bsb, in general, it is safe to skip this section. It also adds some tips for people who debug the compiler.

BuckleScript inherits the command line arguments of the OCaml compiler. It also adds several flags:

-bs-main (single directory build)

bsc.exe -bs-main Main

bsc.exe will build module Main and all its dependencies and when it finishes, it will run node main.js.

bsc.exe -c -bs-main Main

The same as above, but will not run node.

-bs-files

So that you can do

bsc.exe -c -bs-files *.ml *.mli

the compiler will sort the order of input files before starting compilation.

BuckleScript supports two compilation modes: script mode and package mode. In package mode, you have to provide package.json on top and set the options -bs-package-name, -bs-package-output. In script mode, such flags are not needed.

-bs-package-name

The project name of your project. The user is suggested to make it consistent with the name field in package.json

-bs-packge-output

The format is module_system:oupt/path/relative/to/package.json Currently supported module systesms are: commonjs, amdjs and goog:<namespace>.

For example, when you want to use the goog module system, you can do things like this:

bsc.exe -bs-package-name your_package -bs-package-output goog:lib/goog -c xx.ml

Note	The user can supply multiple `-bs-package-output` at the same time.

For example:

bsc.exe -bs-package-name name -bs-package-output commonjs:lib/js -bs-package-output goog:lib/goog -bs-package-output amdjs:lib/amdjs -c x.ml

It will generate x.js in lib/js as a commonjs module, lib/goog as a google module and lib/amdjs as an amdjs module at the same time.

You would then need a bundler for the different module systems: webpack supports commonjs and amdjs while google closure compiler supports all.

-bs-gen-tds

Trigger the generation of TypeScript .d.ts files. bsc.exe has the ability to also emit .d.ts for better interaction with TypeScript. This is still experimental.

For more options, please see the documentation of bsc.exe -help.

-bs-no-warn-ffi-type

Turn off warnings on FFI type declarations.

-bs-eval

Example

bsc.exe -dparsetree -drawlambda -bs-eval 'Js.log "hello"'

[ (1)
  structure_item (//toplevel//[1,0+0]..[1,0+14])
    Pstr_eval
    expression (//toplevel//[1,0+0]..[1,0+14])
      Pexp_apply
      expression (//toplevel//[1,0+0]..[1,0+6])
        Pexp_ident "Js.log" (//toplevel//[1,0+0]..[1,0+6])
      [
        <label> ""
          expression (//toplevel//[1,0+7]..[1,0+14])
            Pexp_constant Const_string("hello",None)
      ]
]
(2)
(setglobal Bs_internal_eval! (seq (js_dump "hello") (makeblock 0)))
// Generated by BUCKLESCRIPT VERSION 1.0.2 , PLEASE EDIT WITH CARE
'use strict';


console.log("hello");

/*  Not a pure module */

Output by flag -dparsetree
Output by flag -drawlambda

For this flag, it will not create any intermediate file, which is useful for learning or troubleshooting.

-bs-no-builtin-ppx-ml, -bs-no-builtin-ppx-mli

If the user doesn’t use any bs specific annotations, the user can explicitly turn it off. Another use case is that users can use -ppx explicitly as below:

bsc.exe -c -ppx bsppx.exe -bs-no-builtin-ppx-ml c.ml

-bs-no-version-header

Don’t print version header.

Semantic differences from other backends

This is particularly important when porting an existing OCaml application to JavaScript.

Custom data type

In OCaml, the C FFI allows the user to define a custom data type and customize caml_compare, caml_hash behavior, etc. This is not available in our backend (since we have no C FFI).

Physical (in)equality

In general, users should only use physical equality as an optimization technique, but not rely on its correctness, since it is tightly coupled with the runtime.

String char range

Currently, BuckleScript assumes that the char range is 0-255. The user should be careful when they pass a JavaScript string to the OCaml side. Note that we are working on a solution for this problem.

Weak map

OCaml’s weak map is not available in BuckleScript. The weak pointer is replaced by a strict pointer.

Integers

OCaml has int, int32, nativeint and int64 types. - Both int32 and int64 in BuckleScript have the exact same semantics as OCaml. - int in BuckleScript is the same as int32 while in OCaml it’s platform dependent. - nativeint is treated as JavaScript float, except for division. For example, Nativeint.div a b will be translated into a /b | 0.

Warning

Nativeint.shift_right_logical x 0 is different from Int32.shift_right_local x 0. The former is literally translated into x >>> 0 (translated into an unsigned int), while the latter is x | 0.

Printf.printf

The Printf.print implementation in BuckleScript requires a newline (\n) to trigger the printing. This behavior is not consistent with the buffered behavior of native OCaml. The only potential problem we foresee is that if the program terminates with no newline character, the text will never be printed.

Obj module

We do our best to mimic the native compiler, but we have no guarantee and there are differences.

Hashtbl hash algorithm

BuckleScript uses the same algorithm as native OCaml but the output is different due to the runtime representation of int/int64/int32 and float.

Marshall

Marshall module is not supported yet.

Sys.argv, Sys.max_array_length, Sys.max_string_length

Command line arguments are always empty. This might be fixed in the near future. Sys.max_array_length and Sys.max_string_length will be the same as max_int, but it might be respected.

Unsupported IO primitives

Because of the JavaScript environment limitation, Pervasives.stdin is not supported but both Pervasives.stdout and Pervasives.stderr are.

Conditional compilation support - static if

It is quite common that people want to write code that works with different versions of compilers and libraries.

People used to use preprocessors like C preprocessor for the C family languages. In the OCaml community there are several preprocessors: cppo, ocp-pp, camlp4 IFDEF macros, optcomp and ppx optcomp.

Instead of using a preprocessor, BuckleScript adds language level static if compilation to the language. It is less powerful than other preprocessors since it only supports static if (no #define, #undefine, #include) but there are several advantages.

It’s very small (only around 500 LOC) and highly efficient. There is no added pass, allowing everything to be done in a single pass. It is easy to rebuild the pre-processor in a stand alone file, with no dependencies on compiler libs to back-port it to old OCaml compilers.
It’s purely functional and type safe, easy to work with IDEs like Merlin.

Concrete syntax

static-if
| HASH-IF-BOL conditional-expression THEN (1)
   tokens
(HASH-ELIF-BOL conditional-expression THEN) *
(ELSE-BOL tokens)?
HASH-END-BOL

conditional-expression
| conditional-expression && conditional-expression
| conditional-expression || conditional-expression
| atom-predicate

atom-predicate
| atom operator atom
| defined UIDENT
| undefined UIDENT

operator
| (= | < | > | <= | >= | =~ )

atom
| UIDENT | INT | STRING | FLOAT

IF-BOL means #IF should be in the beginning of a line

Typing rules

type of INT is int
type of STRING is string
type of FLOAT is float
value of UIDENT comes from either built-in values (with documented types) or an environment variable, if it is literally true, false then it is bool, else if it is parsable by int_of_string then it is of type int, else if it is parsable by float_of_string then it is float, otherwise it would be string
In lhs operator rhs, lhs and rhs are always the same type and return boolean. =~ is a semantic version operator which requires both sides to be string.

Evaluation rules are obvious. =~ respect semantic version, for example, the underlying engine

semver Location.none "1.2.3" "~1.3.0" = false;;
semver Location.none "1.2.3" "^1.3.0" = true ;;
semver Location.none "1.2.3" ">1.3.0" = false ;;
semver Location.none "1.2.3" ">=1.3.0" = false ;;
semver Location.none "1.2.3" "<1.3.0" = true ;;
semver Location.none "1.2.3" "<=1.3.0" = true ;;
semver Location.none "1.2.3" "1.2.3" = true;;

Examples

lwt_unix.mli

type open_flag =
    Unix.open_flag =
  | O_RDONLY
  | O_WRONLY
  | O_RDWR
  | O_NONBLOCK
  | O_APPEND
  | O_CREAT
  | O_TRUNC
  | O_EXCL
  | O_NOCTTY
  | O_DSYNC
  | O_SYNC
  | O_RSYNC
#if OCAML_VERSION =~ ">=3.13" then
  | O_SHARE_DELETE
#end
#if OCAML_VERSION =~ ">=4.01" then
  | O_CLOEXEC
#end

Built in variables and custom variables

ocamlscript>bsc.exe -bs-D CUSTOM_A="ghsigh" -bs-list-conditionals
OCAML_PATCH "BS"
BS_VERSION "1.2.1"
OS_TYPE "Unix"
BS true
CUSTOM_A "ghsigh"
WORD_SIZE 64
OCAML_VERSION "4.02.3+BS"
BIG_ENDIAN false

Changes to command line options

For BuckleScript users, nothing needs to be done (it is baked in the language level). For non BuckleScript users, we provide an external pre-processor, so it will work with other OCaml compilers too. Note that the bspp.ml is a stand alone file, so that it even works without compilation.

Example

bsc.exe -c lwt_unix.mli
ocamlc -pp 'bspp.exe' -c lwt_unix.mli
ocamlc -pp 'ocaml -w -a bspp.ml' -c lwt_unix.mli

Warning

This is a very small extension to the OCaml language, it is backward compatible with OCaml language with such exceptions.

let f x =
  x
#elif (1)

#elif at the beginning of a line is interpreted as static if, there is no issue with #if or #end, since they are already keywords.

Build system support

The BuckleScript compilation model is similar to OCaml native compiler. If b.ml depends on a.ml, you have to compile a.ml and a.mli first.

Note	The technical reason is that BuckleScript will generate intermediate files with the extension `.cmj` which are later used for cross module inlining, arity inference and other information.

BuckleScript build system: `bsb`

BuckleScript proivdes a native build tool on top of Google’s ninja-build. It is designed for a fast feedback loop (typically 100ms feedback loop) and works cross platform.

bsb can be running in any subdirectory. It is a schema based build tool and the schema is available. It is strongly recommended to check out the schema after you finish reading the tutorials below.

If your editor supports JSON schema validation and auto-completion like VSCode, below is a configuration to help you get auto-completion and validation for free:

settings.json

{
    "json.schemas": [
        {
            "fileMatch": [
                "bsconfig.json"
            ],
            "url" : "file:///path/to/bucklescript/docs/docson/build-schema.json" (1)
        }
    ],
    // ....
}

The build system is installed as bsb.exe in bs-platform/bin/bsb.exe. Due to a known issue in npm, we create a JS wrapper (which is accessible in .bin too) so the user can call either bsb (slightly higher latency but symlinked into .bin) or bsb.exe

Here is a minimal configuration:

bsconfig.json

{
  "name": "test", // package name, required (1)
  "sources": "src" (2)
}

It is an extension to JSON with comment support
Here we did not list files, so all .ml, .mli, .re, .rei will be considered as source files

The entry point is bsb. It will check if there is already build.ninja in the build directory, and if not or it needs to be regenerated it will generate the file build.ninja and delegate the hard work to ninja.

The directory layout (after building) would be

.
├── lib
│   ├── bs
│   │   ├── src
│   │   └── test
│   ├── js
│   │   ├── src
│   │   └── test
│   ├── amdjs (1)
│   │   ├── src
│   │   └── test
│   ├── goog  (2)
│   │   ├── src
│   │   └── test
│   └── ocaml
├── scripts
├── src
└── test

Will generate amdjs modules if flags are turned on
Will generate goog modules if flags are turned on

We wrap bsb.exe as bsb so that it will work across different platforms.

Watch mode

bsb (1)
bsb -w (2)

Do the plain build
Do the plain build and watch

Build with other BuckleScript dependencies

List your dependency in bs-dependencies and install it via npm install as below:

bsconfig.json

{
    "name": "bs-string",
    "version": "0.1.3",
    "bs-dependencies": [
        "bs-mocha" (1)
    ],
    "sources": [
       .. .
    ],
    "generate-merlin" : false (2)
}

Yet another BuckleScript dependency
If true (default) bsb will generate merlin file for you

package.json

{
 "dependencies": {
    "bs-mocha": "0.1.5"
    },
  ...
}

After your npm install,

bsb -clean-world (1)
bsb -make-world (2)
bsb -w (3)

Clean the binary artifact of current build and your dependency
Build dependencies and lib itself
Start watching the project and whenever your changes are made, bsb will rebuild incrementally

You can also streamline the three commands as below:

bsb -clean-world -make-world -w

Mark your directory as dev only

Note sometimes, you have directories which are just tests that you don’t need your dependency to build. In that case you can mark it as dev only:

bsconfig.json

{
        "sources" : {
                "dir" : "test",
                "type" : "dev" (1)
        }
}

directory test is in dev mode, it will not be built when used as a dependency

A real world example of using `bsb`

Below is a json configuration for the bucklescript-tea: the Elm artchitecture in BuckleScript

bsconfig.json

{
  "name": "bucklescript-tea",
  "version": "0.1.3",
  "sources": [
   "src", (1)
    {
      "dir": "test",
      "type": "dev" (2)
    }
  ]
}

Source directory, by default it will export all units of this directory to users.
Dev directory, which will only be useful for developers of this project.

package.json

{
  "name": "bucklescript-tea",
  "version": "0.1.3",
  "description": "TEA for Bucklescript",
  "scripts": {
    "build": "bsb",
    "watch": "bsb -w",
    "test": "echo \"Error: no test specified\" && exit 1"
  },
  "peerDependencies": {
    "bs-platform": "^1.7.0" (1)
  }
}

Here we list bs-platform as a peer dependency so that different repos shares the same compiler.

Now, we have a repo bucklescript-have-tea to depend on bucklescript-tea, its configurations are as below:

bsconfig.json

{
    "name" : "bucklescript-have-tea",
    "sources" : "src",
    "bs-dependencies": [
      "bucklescript-tea"
    ]
}

package.json

{
    "name" : "bucklescript-have-tea",
    "version" : "0.1.0",
    "dependencies" : { "bucklescript-tea" : "^0.1.2" }, (1)
    "peerDependencies" : { "bs-platform" : "^1.7.0" } (2)
}

List bucklescript-tea as dependency
List bs-platform as peer dependency

Suppose you are in bucklescript-have-tea top directory,

npm install (1)
npm install bs-platform (2)
./node_modules/.bin/bsb -clean-world -make-world -w (3)

Install the dependencies
Install peer dependencies
On Windows, it would be .\node_modules\.bin\bsb -clean-world -make-world -w

You can also change the package-specs to have another module format, for example, tweak your bsconfig.json:

{
  ... ,
  "package-specs" : ["amdjs", "commonjs"],
  ...
}

Rerun the command

bsb -clean-world -make-world

You will get both commonjs and amdjs support. In the end, we suggest you check out the schema and enjoy the build!

Build using Make

Warning

bsb is the officially recommended build system. This section is included here only for people who are curious about how the build works.

BuckleScript distribution has bsdep.exe which has the same interface as ocamldep

Here is a simple Makefile to get started:

Makefile

OCAMLC=bsc.exe (1)
OCAMLDEP=bsdep.exe (2)
SOURCE_LIST := src_a src_b
SOURCE_MLI = $(addsuffix .mli, $(SOURCE_LIST))
SOURCE_ML  = $(addsuffix .ml, $(SOURCE_LIST))
TARGETS := $(addsuffix .cmj, $(SOURCE_LIST))
INCLUDES=
all: $(TARGETS)
.mli:.cmi
        $(OCAMLC) $(INCLUDES) $(COMPFLAGS) -c $<
.ml:.cmj:
        $(OCAMLC) $(INCLUDES) $(COMPFLAGS) -c $<
-include .depend
depend:
        $(OCAMLDEP) $(INCLUDES) $(SOURCE_ML) $(SOURCE_MLI) > .depend

bsc.exe is the BuckleScript compiler
ocamldep executable is part of the OCaml compiler installation

In theory, people need run make depend && make all. make depend will calculate dependencies while make all will do the job.

However in practice, people used to a file watch service, such as watchman for example, will need the JSON configuration:

build.json

[
    "trigger", ".", {
        "name": "build",
        "expression": ["pcre", "(\\.(ml|mll|mly|mli|sh|sh)$|Makefile)"], (1)
        "command": ["./build.sh"],
        "append_files" : true
    }
]

whenever such files changed, it will trigger command field to be run

build.sh

make -r -j8 all (1)
make depend (2)

build
update the dependency

Now in your working directory, type watchman -j < build.json and enjoy the lightning build speed.

FAQ

How does IO work in the browser?

In general, it is very hard to simulate IO in the browser, we recommend users write bindings to NodeJS directly for server side, or use Js.log in client side, see disucssions in #748
The compiler does not build?

In production mode, the compiler is a single file in jscomp/bin/compiler.ml. If it is not compiling, make sure you have the right OCaml compiler version. Currently the OCaml compiler is a submodule of BuckleScript. Make sure the exact commit hash matches (we only update the compiler occasionally).
Which version of JavaScript syntax does BuckleScript target?

BuckleScript targets ES5.
Does BuckleScript work with merlin?

Yes, you need edit your .merlin file:
```
B node_modules/bs-platform/lib/ocaml
S node_modules/bs-platform/lib/ocaml
FLG -ppx node_modules/bs-platform/bin/bsppx.exe
```
Note there is a upstream fix in Merlin, make sure your merlin is updated.

What polyfills does BuckleScript need?

Math.imul: This polyfill is needed for int32 multiplication. BuckleScript provides this by default (when feature detection returns false), no action is required from the user.

TypedArray: The TypedArray polyfill is not provided by BuckleScript and it’s the responsibility of the user to bundle the desired polyfill implementation with the BuckleScript generated code.

The following functions from OCaml stdlib
require the TypedArray polyfill:

Int64.float_of_bits
Int64.bits_of_float
Int32.float_of_bits

Int32.bits_of_float

Warning

For the current BuckleScript version, if the user does not bundle the TypedArray polyfill, the JavaScript engine does not support it and user used functions mentioned above, the code will fail at runtime.

Uncurried functions cannot be polymorphic?

E.g. if you try to do this at toplevel:

let id : ('a -> 'a [@bs]) = ((fun v -> v) [@bs])

You’ll get this dreaded error message

Error: The type of this expression, ([ `Arity_1 of '_a ], '_a) Js.fn,
       contains type variables that cannot be generalized

The issue here isn’t that the function is polymorphic. You can use (and probably have used) polymorphic uncurried functions without any problem as inline callbacks. But you can’t export them. The issue here is the combination of using the uncurried calling convention, polymorphism and exporting (by default). It’s an unfortunate limitation partly due to how OCaml’s type system incorporates side-effects, and partly due to how BuckleScript handles uncurrying. The simplest solution is in most cases to just not export it, by adding an interface to the module. Alternatively, if you really, really need to export it, you can do so in its curried form, and then wrap it in an uncurried lanbda at the call site. E.g.:

lat _ = map (fun v -> id v [@bs])

High Level compiler workflow

The high level architecture is illustrated below:

Source Language
  |
  | (Reuse OCaml Parser)
  v
Surface Syntax Tree
  |
  | (built in Syntax tree transformation)
  v
Surface Syntax Tree
  |
  | (Reuse OCaml Type checker)
  v
Typedtree
  |
  | (Reuse OCaml pattern match compiler and erase types)
  | (Patches to pass more information down to Lambda )
  |
OCaml Lambda IR
  |
  |
  v
Buckle Lambda IR ------------------+
  |   ^                            |
  |   |                     6 Lambda Passes (lam_* files)
  |   |             Optimization/inlining/dead code elimination
  |   \                            |
  |    \ --------------------------+
  |
  |  Self tail call elimination
  |  Constant folding + propagation
  V
JS IR (J.ml)  ---------------------+
  |   ^                            |
  |   |                     6 JS Passes (js_* files)
  |   |            Optimization/inlining/dead code elimination
  |   \                            |
  |    \  -------------------------+
  |
  |  Smart printer includes scope analysis
  |
  V
Javascript Code

Design Principles

The current design of BuckleScript follows several high level principles. While those principles might change in the future, they are enforced today and can explain certain technical limitations BuckleScript has.

Lambda Representation

As pictured in the diagram above, BuckleScript is primarily based on the Lambda representation of the OCaml compiler. While this representation is quite rich, some information is lost from the upstream representation. The patch to the OCaml compiler tries to enrich this representation in a non-intrusive way (see next section).

Minimal Patch to the OCaml compiler

BuckleScript requires patches to the OCaml compiler. One of the main reasons is to enrich the Lambda representation so that the generated code is as nice as possible. A design goal is to keep those patches minimal and useful for the OCaml compiler in general so that they can later be integrated.

Note

A common question is to wonder why BuckleScript transpiles an OCaml record value to a JavaScript array while a more intuitive representation would be a JavaScript object. This technical decision is a direct consequence of the above 2 design principles: the Lambda layer assumes in a lot of places that a record value is an array and such modification would be too large of a change to OCaml compiler.

Soundness

BuckleScript preserves the soundness of the OCaml language. Assuming the FFI is correctly implemented, the type safety is preserved.

Minimal new symbol creation

In order to make the JavaScript generated code as close as possible to the original OCaml core we thrive to introduce as few new symbols as possible.

Runtime representation

Below is a description of how OCaml values are encoded in JavaScript, the internal description means users should not rely on its actual encoding (and it is subject to change). We recommend that you write setter/getter functions to manipulate safely OCaml values from JavaScript.

For example, users should not rely on how OCaml list is encoded in JavaScript; instead, the OCaml stdlib provides three functions: List.cons, List.hd and List.tl. JavaScript code should only rely on those three functions.

Simple OCaml type

ocaml type JavaScript type

int

number

nativeint

number

int32

number

float

number

bool

number

true → 1
false → 0

int64

Array of size two numbers [hi,lo]. hi is signed while lo is unsigned

char

number

for example:

'a' → 97

string

bytes

number array

Note	We might encode it as buffer in NodeJS.

'a array

Array

record

Array internal

For instance:

type t = { x : int; y : int }
let v = {x = 1; y = 2}

Output:

var v = [1,2]

tuple

Array

For example:

(3,4) → [3,4]

’a option

internal

For example:

None → 0
Some a → [a]

list

internal

For example:

[] → 0
x::y → [x,y]
1::2::[3] → [ 1, [ 2, [ 3, 0 ] ] ]

Variant

internal (subject to change)

Simple Variants: (Variants with only one non-nullary constructor)

type tree =
  | Leaf
  | Node of int * tree * tree
(* Leaf --> 0 *)
(* Node(a,b,c) --> [a,b,c]*)

Complex Variants: (Variants with more than one non-nullary constructor)

type u =
     | A of string
     | B of int
(* A a --> [a].tag=0 -- tag 0 assignment is optional *)
(* B b --> [b].tag=1 *)

Polymorphic variant

internal

`a (* 97 *)
`a 1 2 (* [97, [1,2] ]*)

exception

internal

extension

internal

object

internal

Js.boolean

boolean

For example:

Js.true_ → true
Js.false_ → false

Js module

type boolean
val to_bool: boolean -> bool

Js.Boolean module

val to_js_boolean: bool -> Js.boolean

'a Js.Null.t

Either 'a or null. Js.Null.empty represents null too.

Js.Null module

val to_opt : 'a t -> 'a option
val from_opt : 'a option -> 'a t
val return : 'a -> 'a t
val test : 'a t -> bool

'a Js.Undefined.t

Either 'a or undefined. Same operations as 'a Js.Null.t. Js.Undefined.empty represents undefined too.

'a Js.Null_undefined.t

Either 'a, null or undefined. Same operations as 'a Js.Null.t.

Js.Null_undefined.undefined represents undefined, Js.Null_undefined.null represents null.

This module’s null tests check for both null and undefined; if you know the value’s only ever going to be null and not undefined, use Js.Null instead. Likewise for Js.Undefined.

Note	`Js.to_opt` is optimized when the `option` is not escaped

Note	In the future, we will have a debug mode, in which the corresponding js encoding will be instrumented with more information

As we clarified before, the internal representation should not be relied upon. We are working to provide a ppx extension as below:

type t =
  | A
  | B of int * int
  | C of int * int
  | D [@@bs.deriving{export}]

So that it will a automatically provide constructing and destructing functions:

val a : t
val b : int -> int -> t
val c : int -> int -> t
val d : int

val a_of_t : t -> bool
val d_of_t : t -> bool
val b_of_t : t -> (int * int ) Js.Null.t
val c_of_t : t -> (int * int ) Js.Null.t

Integration with Reason

You can play with Reason using the playground Facebook Reason

Note	The playgrounds are only for demos and might not be the latest You should always use the command line as your production tool.

There is a stand alone example here.

Contributions

First, thanks for your interest in contributing! There are plenty of ways to make contributions, like blogging, sharing your experience, open sourcing your libraries using BuckleScript. They are all deeply appreciated.

This section will focus on how to contribute to this repo.

Development set-up

Having opam installed
```
opam switch 4.02.3+buckle-master # use our OCaml compiler
opam install camlp4  <1>
```
Camlp4 is used to generate OCaml code for processing large AST. (j.ml file), if you don’t change j.ml (most likely you won’t), so you probably don’t need it
Having NodeJS installed
Having Make installed
OS: Mac/Linux (Note BuckleScript works on Windows, but the dev mode is not tested)

Below assume that you are working in jscomp directory, and that you’ve followed instructions from Minimal dependencies to build OCaml.

Contributing to `bsb.exe`

The build target is

make -C bin bsb.exe

So whenever you change files relevant to the build tool bsb.exe, try it and do some test. If it works, send a pull request!

Note that for most binaries in BuckleScript, we also have a release mode, which will pack all relevant files into a single file. This is important, since it will cut all our dev-dependencies, so the user does not need install those tools.

You can verify it by typing

make snapshotml # see the diffs in jscomp/bin

But please don’t commit those changes in PR, it will cause painful merge conflicts.

Contributing to `bsc.exe`

make -C bin bsc.exe # build the compiler in dev mode

make libs # build all libs using the dev compiler

There is also a snapshot mode,

make snapshotml

This will actually snapshot your changes into a single ml file and used in npm distribution. But please don’t commit those changes in PR, it will cause painful merge conflicts.

Contributing to the runtime

BuckleScript runtime implementation is currently a mix of OCaml and JavaScript. (jscomp/runtime directory). The JavaScript code is defined in the .ml file using the bs.raw syntax extension.

The goal is to implement the runtime purely in OCaml and you can help contribute.

Each new PR should include appropriate testing.

Currently all tests are in jscomp/test directory and you should either add a new test file or modify an existing test which covers the part of the compiler you modified.

Add the filename in jscomp/test/test.mllib
Add a suite test

The specification is in jscomp/test/mt.ml

For example some simple tests would be like:

let suites : _ Mt.pair_suites =
   ["hey", (fun _ -> Eq(true, 3 > 2));
       "hi", (fun _ -> Neq(2,3));
       "hello", (fun _ -> Approx(3.0, 3.0));
       "throw", (fun _ -> ThrowAny(fun _ -> raise 3))
       ]
let () = Mt.from_pair_suites __FILE__ suites

Run the tests

Suppose you have mocha installed, if not, try npm install mocha

mocha -R list jscomp/test/your_test_file.js

To build libs, tests and run all tests:

make libs && make -C test all && npm test

See the coverage

npm run cover

Contributing to the documentation

You’ll need Asciidoctor installed. Modify the section you want in site/docsource/, then run site/docsource/build.sh.

Contributing to the API reference

The API reference is generated from doc comments in the source code. Here’s a good example: https://github.com/bloomberg/bucklescript/blob/master/jscomp/others/js_re.mli#L146-L161

Some tips and guidelines:

The first sentence or line should be a very short summary. This is used in indexes and by tools like merlin.
Ideally, every function should have at least one @example
Cross-reference another definition with {! identifier}. But use them sparingly, they’re a bit verbose (currently, at least).
Wrap non-cross-referenced identifiers and other code in [ … ]
Escape {, }, [, ] and @ using \
It’s possible to use {%html …} to generate custom html, but use this very, very sparingly.
A number of "documentation tags" are provided that would be nice to use, but unfortunately they’re often not supported for `external`s. Which is of course most of the API.
@param usually doesn’t work. Use {b <param>} … instead
@returns usually doesn’t work. Use {b returns} … instead.
Always use @deprecated when applicable.
Always use @raise when applicable.
Always provide a @see tag pointing to MDN for more information when available.

See Ocamldoc documentation for a lot more details.

To generate the html, run make docs in jscomp/.

Html generation uses a custom generator located in odoc_gen/ and custom styles located in docs/api_static.

Comparisons

Difference from js_of_ocaml

Js_of_ocaml is a popular compiler which compiles OCaml’s bytecode into JavaScript. It is the inspiration for this project, and has already been under development for several years and is ready for production. In comparison, BuckleScript, while moving fast, is still a very young project. BuckleScript’s motivation, like js_of_ocaml, is to unify the ubiquity of the JavaScript platform and the truly sophisticated type system of OCaml, however, there are some areas where we view things differently from js_of_ocaml. We describe below, some of these differences, and also refer readers to some of the original informal discussions.

Js_of_ocaml takes lowlevel bytecode from OCaml compiler, BuckleScript takes the highlevel rawlambda representation from OCaml compiler
Js_of_ocaml focuses more on existing OCaml eco-system(opam) while BuckleScript’s major goal is to target npm
Js_of_ocaml and BuckleScript have slightly different runtime encoding in several places, for example, BuckleScript encodes OCaml Array as JS Array while js_of_ocaml requires its index 0 to be of value 0.

Both projects are improving quickly, so this can change in the future!

Appendix A: Library API documentation

There is a small Library comes with bs-platform, its documentation is available.

Appendix B: CHANGES

1.3.2

Features
- Significantly improve bsb experience (TODO: install instruction)

1.2.1 + dev

Features
- add -bs-assume-no-mli, -bs-no-implicit-include #861 it’s for deterministic build
- add -bs-D -bs-list-conditionals flags #851
- add -bs-syntax-only
- add -bs-binary-ast #854

1.1.2

Fixes
- Bug fix with opam issues #831
Features
- Provide bspp.exe for official compiler

1.1.1

Features
- Add bsdep.exe #822
- Conditional compilation support #820
- Relax syntactic restrictions for all extension point #793 so that bs.obj, obj, bs.raw, raw, etc will both work. Note that all attributes will still be qualified
- Suport bs.splice in bs.new #793
- Complete `bs.splice ` support and documentation #798

1.03

Features
- Add an option -bs-no-warn-unused-bs-attribute #787

Incompatible changes (due to proper Windows support):

bsc, bspack and bsppx are renamed into bsc.exe, bspack.exe and bsppx.exe
no symlink from .bin any more.

Old symlinks

  tmp>ls -al node_modules/.bin/
  total 96
  drwxr-xr-x  14 hzhang295  staff  476 Sep 20 17:26 .
  drwxr-xr-x   4 hzhang295  staff  136 Sep 20 17:27 ..
  lrwxr-xr-x   1 hzhang295  staff   22 Sep 20 17:26 bsc -> ../bs-platform/bin/bsc
  lrwxr-xr-x   1 hzhang295  staff   25 Sep 20 17:26 bspack -> ../bs-platform/bin/bspack
  lrwxr-xr-x   1 hzhang295  staff   24 Sep 20 17:26 bsppx -> ../bs-platform/bin/bsppx

Now these symlinks are removed, you have to refer to bs-platform/bin/bsc.exe

1.02

Bug fixes and enhancement
- Fix Bytes.blit when src==dst #743
Features
- Add an option -bs-no-warn-ffi-type #783 By default, bsc.exe will warn when it detect some ocaml datatype is passed from/to external FFi
- Add an option -bs-eval 784

1.01

FFI
- support fields and mutable fields in JS object creation and private method #694
- Introduce phantom arguments (bs.ignore) for ad-hoc polymorphism #686
Bug fixes and enhancement
- Enforce #= always return unit #718 for better error messages

1.0

Initial release

BuckleScript User Manual

Why BuckleScript

Benefits of JavaScript platform

Problems of JavaScript && how BuckleScript solves them

Installation

Windows Installation

*nix installation

Recommended installation with OPAM

Install from source

using NPM

Minimal dependencies

Get Started

Using existing templates (@since 1.7.4 )

First example by hand without samples

An example with multiple modules

Built in NPM support

Build an OCaml library as a npm package

To use OCaml library as a npm package

Together

Examples

JS Calling OCaml

BuckleScript annotations for Unicode and JS FFI support

Unicode support (@since 1.5.1)

Unicode support with string interpolation (@since 1.7.0)

FFI

Binding to simple JS functions values

Binding to global value: bs.val

Scoped values: bs.scope (@since 1.7.2)

Binding to JavaScript constructor: bs.new

Binding to a value from a module: bs.module

Binding the whole module as a value or function

Binding to method: bs.send, bs.send.pipe

Binding to dynamic key access/set: bs.set_index, bs.get_index

Binding to Getter/Setter: bs.get, bs.set

Splice calling convention: bs.splice

Special types on external declarations: bs.string, bs.int, bs.ignore, bs.as

Using polymorphic variant to model enums and string types

Using polymorphic variant to model event listener

Phantom Arguments and ad-hoc polymorphism

Fixed Arguments

Fixed Arguments with arbitrary JSON literal (@since 1.7.0)

Binding to NodeJS special variables: bs.node

Binding to callbacks (high-order function)

[@bs] for explicit uncurried callback

[@bs.uncurry] for implicit uncurried callback (@since 1.5.0)

Uncurried calling convention as an optimization

Callback optimization

Bindings to this based callbacks: bs.this

Binding to JS objects

Simple object type

Complex object type

How to consume JS property and methods

getter/setter annotation to JS properties

Create JS objects using bs.obj

Create JS objects using external

Create JS objects with this semantics

Method chaining

Object label translation convention

Return value checking (@since 1.5.1)

Embedding untyped Javascript code

Detect global variable existence bs.external (@since 1.5.1)

Embedding arbitrary JS code as an expression

Embedding raw JS code as statements

Debugger support

Regex support

Examples

A simple example: binding to mocha unit test library

Exception handling between OCaml and JS (@since 1.7.0)

Catch JS exception

Raise JS style exception

bs.open: Type safe external data-source handling (@@since 1.7.0)

Use cases

Js module

Null and Undefined

Boolean

Stable-ish submodules

Experimental/incomplete submodules

Very experimental/internal submodules

Node module

Submodules

Bindings to `this` based callbacks: bs.this

Create JS objects with `this` semantics

Detect global variable existence `bs.external` (@since 1.5.1)

`bs.open`: Type safe external data-source handling (@@since 1.7.0)

BuckleScript build system: `bsb`

A real world example of using `bsb`

Contributing to `bsb.exe`

Contributing to `bsc.exe`