Communicate with JavaScript

Melange interoperates very well with JavaScript, and provides a wide array of features to consume foreign JavaScript code. To learn about these techniques, we will first go through the language concepts that they build upon, then we will see how types in Melange map to JavaScript runtime types. Finally, we will provide a variety of use cases with examples to show how to communicate to and from JavaScript.

Language concepts

The concepts covered in the following sections are a small subset of the OCaml language. However, they are essential for understanding how to communicate with JavaScript, and the features that Melange exposes to do so.

Attributes and extension nodes

In order to interact with JavaScript, Melange needs to extend the language to provide blocks that express these interactions.

One approach could be to introduce new syntactic constructs (keywords and such) to do so, for example:

javascript add : int -> int -> int = {|function(x,y){
  return x + y
}|}

But this would break compatibility with OCaml, which is one of the main goals of Melange.

Fortunately, OCaml provides mechanisms to extend its language without breaking compatibility with the parser or the language. These mechanisms are composed by two parts:

First, some syntax additions to define parts of the code that will be extended or replaced
Second, compile-time OCaml native programs called PPX rewriters, that will read the syntax additions defined above and proceed to extend or replace them

The syntax additions come in two flavors, called extension nodes and attributes.

Extension nodes

Extension nodes are blocks that are supposed to be replaced by a specific type of PPX rewriters called extenders. Extension nodes use the % character to be identified. Extenders will take the extension node and replace it with a valid OCaml AST (abstract syntax tree).

An example where Melange uses extension nodes to communicate with JavaScript is to produce "raw" JavaScript inside a Melange program:

[%%mel.raw "var a = 1; var b = 2"]
let add = [%mel.raw "a + b"]

[%%mel.raw "var a = 1; var b = 2"];
let add = [%mel.raw "a + b"];

Which will generate the following JavaScript code:

var a = 1; var b = 2
var add = a + b

The difference between one and two percentage characters is detailed in the OCaml documentation.

Attributes

Attributes are "decorations" applied to specific parts of the code to provide additional information. In Melange, attributes are used in two ways to enhance the expressiveness of generating JavaScript code: either reusing existing OCaml built-in attributes or defining new ones.

Reusing OCaml attributes

The first approach is leveraging the existing OCaml’s built-in attributes to be used for JavaScript generation.

One prominent example of OCaml attributes that can be used in Melange projects is the unboxed attribute, which optimizes the compilation of single-field records and variants with a single tag to their raw values. This is useful when defining type aliases that we don’t want to mix up, or when binding to JavaScript code that uses heterogeneous collections. An example of the latter is discussed in the variadic function arguments section.

For instance:

type name =
  | Name of string [@@unboxed]
let student_name = Name "alice"

[@unboxed]
type name =
  | Name(string);
let student_name = Name("alice");

Compiles into:

var student_name = "alice";

Other OCaml pre-built attributes like alert or deprecated can be used with Melange as well.

Defining new attributes

The second approach is introducing new attributes specifically designed for Melange, such as the mel.set attribute used to bind to properties of JavaScript objects. The complete list of attributes introduced by Melange can be found here.

Attribute annotations can use one, two or three @ characters depending on their placement in the code and which kind of syntax tree node they are annotating. More information about attributes can be found in the dedicated OCaml manual page.

Here are some samples using Melange attributes mel.set and mel.as:

type document
external setTitleDom : document -> string -> unit = "title" [@@mel.set]

type t = {
  age : int; [@mel.as "a"]
  name : string; [@mel.as "n"]
}

type document;
[@mel.set] external setTitleDom: (document, string) => unit = "title";

type t = {
  [@mel.as "a"]
  age: int,
  [@mel.as "n"]
  name: string,
};

To learn more about preprocessors, attributes and extension nodes, check the section about PPX rewriters in the OCaml docs.

External functions

Most of the system that Melange exposes to communicate with JavaScript is built on top of an OCaml language construct called external.

external is a keyword for declaring a value in OCaml that will interface with C code:

external my_c_function : int -> string = "someCFunctionName"

external my_c_function: int => string = "someCFunctionName";

It is like a let binding, except that the body of an external is a string. That string has a specific meaning depending on the context. For native OCaml, it usually refers to a C function with that name. For Melange, it refers to the functions or values that exist in the runtime JavaScript code, and will be used from Melange.

In Melange, externals can be used to bind to global JavaScript objects. They can also be decorated with certain [@mel.xxx] attributes to facilitate the creation of bindings in specific scenarios. Each one of the available attributes will be further explained in the next sections.

Once declared, one can use an external as a normal value. Melange external functions are turned into the expected JavaScript values, inlined into their callers during compilation, and completely erased afterwards. They don’t appear in the JavaScript output, so there are no costs on bundling size.

Note: it is recommended to use external functions and the [@mel.xxx] attributes in the interface files as well, as this allows some optimizations where the resulting JavaScript values can be directly inlined at the call sites.

Special identity external

One external worth mentioning is the following one:

type foo = string
type bar = int
external danger_zone : foo -> bar = "%identity"

type foo = string;
type bar = int;
external danger_zone: foo => bar = "%identity";

This is a final escape hatch which does nothing but convert from the type foo to bar. In the following sections, if you ever fail to write an external, you can fall back to using this one. But try not to.

Abstract types

In the subsequent sections, you will come across examples of bindings where a type is defined without being assigned to a value. Here is an example:

type document

type document;

These types are referred to as "abstract types" and are commonly used together with external functions that define operations over values when communicating with JavaScript.

Abstract types enable defining types for specific values originating from JavaScript while omitting unnecessary details. An illustration is the document type mentioned earlier, which has several properties. By using abstract types, one can focus solely on the required aspects of the document value that the Melange program requires, rather than defining all its properties. Consider the following example:

type document

external document : document = "document"
external set_title : document -> string -> unit = "title" [@@mel.set]

type document;

external document: document = "document";
[@mel.set] external set_title: (document, string) => unit = "title";

Subsequent sections delve into the details about the mel.set attribute and how to bind to global values like document.

For a comprehensive understanding of abstract types and their usefulness, refer to the "Encapsulation" section of the OCaml Cornell textbook.

Pipe operators

There are two pipe operators available in Melange:

A pipe last operator |>, available in OCaml and inherited in Melange
A pipe first operator |.->, available exclusively in Melange

Let’s see the differences between the two.

Pipe last

Since version 4.01, OCaml includes a reverse application or "pipe" (|>) operator, an infix operator that applies the result from the previous expression the next function. As a backend for OCaml, Melange inherits this operator.

The pipe operator could be implemented like this (the real implementation is a bit different):

let ( |> ) f g = g f

let (|>) = (f, g) => g(f);

This operator is useful when multiple functions are applied to some value in sequence, with the output of each function becoming the input of the next (a pipeline).

For example, assuming we have a function square defined as:

let square x = x * x

let square = x => x * x;

We are using it like:

let ten = succ (square 3)

let ten = succ(square(3));

The pipe operator allows to write the computation for ten in left-to-right order, as it has left associativity:

let ten = 3 |> square |> succ

let ten = 3 |> square |> succ;

When working with functions that can take multiple arguments, the pipe operator works best when the functions take the data we are processing as the last argument. For example:

let sum = List.fold_left ( + ) 0

let sum_sq =
  [ 1; 2; 3 ]
  |> List.map square (* [1; 4; 9] *)
  |> sum             (* 1 + 4 + 9 *)

let sum = List.fold_left((+), 0);

let sum_sq =
  [1, 2, 3]
  |> List.map(square)  /* [1; 4; 9] */
  |> sum; /* 1 + 4 + 9 */

The above example can be written concisely because the List.map function in the OCaml standard library takes the list as the second argument. This convention is sometimes referred to as "data-last", and is widely adopted in the OCaml ecosystem. Data-last and the pipe operator |> work great with currying, so they are a great fit for the language.

However, there are some limitations when using data-last when it comes to error handling. In the given example, if we mistakenly used the wrong function:

let sum_sq =
  [ 1; 2; 3 ]
  |> List.map String.cat
  |> sum

let sum_sq = [1, 2, 3] |> List.map(String.cat) |> sum;

The compiler would rightfully raise an error:

4 |   [ 1; 2; 3 ]
        ^
Error: This expression has type int but an expression was expected of type
         string

1 |   [ 1, 2, 3 ]
        ^
Error: This expression has type int but an expression was expected of type
         string

Note that instead of telling us that we are passing the wrong function in List.map (String.cat), the error points to the list itself. This behavior aligns with the way type inference works, as the compiler infers types from left to right. Since [ 1; 2; 3 ] |> List.map String.cat is equivalent to List.map String.cat [ 1; 2; 3 ], the type mismatch is detected when the list is type checked, after String.cat has been processed.

With the goal of addressing this kind of limitations, Melange introduces the pipe first operator |.->.

Pipe first

To overcome the constraints mentioned above, Melange introduces the pipe first operator |.->.

As its name suggests, the pipe first operator is better suited for functions where the data is passed as the first argument.

The functions in the Belt library Belt library included with Melange have been designed with the data-first convention in mind, so they work best with the pipe first operator.

For example, we can rewrite the example above using Belt.List.map and the pipe first operator:

let sum_sq =
  [ 1; 2; 3 ]
  |. Belt.List.map square
  |. sum

let sum_sq = [1, 2, 3]->(Belt.List.map(square))->sum;

We can see the difference on the error we get if the wrong function is passed to Belt.List.map:

let sum_sq =
  [ 1; 2; 3 ]
  |. Belt.List.map String.cat
  |. sum

let sum_sq = [1, 2, 3]->(Belt.List.map(String.cat))->sum;

The compiler will show this error message:

4 |   |. Belt.List.map String.cat
                       ^^^^^^^^^^
Error: This expression has type string -> string -> string
       but an expression was expected of type int -> 'a
       Type string is not compatible with type int

2 | let sum_sq = [1, 2, 3]->(Belt.List.map(String.cat))->sum;
                                           ^^^^^^^^^^
Error: This expression has type string -> string -> string
       but an expression was expected of type int -> 'a
       Type string is not compatible with type int

The error points now to the function passed to Belt.List.map, which is more natural with the way the code is being written.

Melange supports writing bindings to JavaScript using any of the two conventions, data-first or data-last, as shown in the "Chaining" section.

For further details about the differences between the two operators, the data-first and data-last conventions and the trade-offs between them, one can refer to this related blog post.

Data types and runtime representation

This is how each Melange type is converted into JavaScript values:

Melange	JavaScript
int	number
nativeint	number
int32	number
float	number
string	string
array	array
tuple `(3, 4)`	array `[3, 4]`
bool	boolean
Js.Nullable.t Js.Nullable.t	`null` / `undefined`
Js.Re.t Js.Re.t	`RegExp`
Option.t `None`	`undefined`
Option.t `Some( Some .. Some (None))Some(Some( .. Some(None)))`	internal representation
Option.t `Some 2Some(2)`	`2`
record `{x = 1; y = 2}{x: 1; y: 2}`	object `{x: 1, y: 2}`
int64	array of 2 elements `[high, low]` high is signed, low unsigned
char	`'a'` -> `97` (ascii code)
bytes	number array
list `[]`	`0`
list `[ x; y ][x, y]`	`{ hd: x, tl: { hd: y, tl: 0 } }`
variant	See below
polymorphic variant	See below

Variants with a single non-nullary constructor:

type tree = Leaf | Node of int * tree * tree
(* Leaf -> 0 *)
(* Node(7, Leaf, Leaf) -> { _0: 7, _1: 0, _2: 0 } *)

type tree =
  | Leaf
  | Node(int, tree, tree);
/* Leaf -> 0 */
/* Node(7, Leaf, Leaf) -> { _0: 7, _1: 0, _2: 0 } */

Variants with more than one non-nullary constructor:

type t = A of string | B of int
(* A("foo") -> { TAG: 0, _0: "Foo" } *)
(* B(2) -> { TAG: 1, _0: 2 } *)

type t =
  | A(string)
  | B(int);
/* A("foo") -> { TAG: 0, _0: "Foo" } */
/* B(2) -> { TAG: 1, _0: 2 } */

Polymorphic variants:

let u = `Foo (* "Foo" *)
let v = `Foo(2) (* { NAME: "Foo", VAL: "2" } *)

let u = `Foo; /* "Foo" */
let v = `Foo(2); /* { NAME: "Foo", VAL: "2" } */

Let’s see now some of these types in detail. We will first describe the shared types, which have a transparent representation as JavaScript values, and then go through the non-shared types, that have more complex runtime representations.

NOTE: Relying on the non-shared data types runtime representations by reading or writing them manually from JavaScript code that communicates with Melange code might lead to runtime errors, as these representations might change in the future.

Shared types

The following are types that can be shared between Melange and JavaScript almost "as is". Specific caveats are mentioned on the sections where they apply.

Strings

JavaScript strings are immutable sequences of UTF-16 encoded Unicode text. OCaml strings are immutable sequences of bytes and nowadays assumed to be UTF-8 encoded text when interpreted as textual content. This is problematic when interacting with JavaScript code, because if one tries to use some unicode characters, like:

let () = Js.log "你好"

let () = Js.log("你好");

It will lead to some cryptic console output. To rectify this, Melange allows to define quoted string literals using the js identifier, for example:

let () = Js.log {js|你好，
世界|js}

let () = Js.log({js|你好，
世界|js});

For convenience, Melange exposes another special quoted string identifier: j. It is similar to JavaScript’ string interpolation, but for variables only (not arbitrary expressions):

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

let world = {j|世界|j};
let helloWorld = {j|你好，$world|j};

You can surround the interpolation variable in parentheses too: {j|你好，$(world)|j}.

To work with strings, the Melange standard library provides some utilities in the Stdlib.String module Stdlib.String module. The bindings to the native JavaScript functions to work with strings are in the Js.String module Js.String module.

Floating-point numbers

OCaml floats are IEEE 754 with a 53-bit mantissa and exponents from -1022 to 1023. This happens to be the same encoding as JavaScript numbers, so values of these types can be used transparently between Melange code and JavaScript code. The Melange standard library provides a Stdlib.Float module Stdlib.Float module. The bindings to the JavaScript APIs that manipulate float values can be found in the Js.Float module Js.Float module.

Integers

In Melange, integers are limited to 32 bits due to the fixed-width conversion of bitwise operations in JavaScript. While Melange integers compile to JavaScript numbers, treating them interchangeably can result in unexpected behavior due to differences in precision. Even though bitwise operations in JavaScript are constrained to 32 bits, integers themselves are represented using the same floating-point format as numbers, allowing for a larger range of representable integers in JavaScript compared to Melange. When dealing with large numbers, it is advisable to use floats instead. For instance, floats are used in bindings like Js.Date.

The Melange standard library provides a Stdlib.Int module Stdlib.Int module. The bindings to work with JavaScript integers are in the Js.Int module Js.Int module.

Arrays

Melange arrays compile to JavaScript arrays. But note that unlike JavaScript arrays, all the values in a Melange array need to have the same type.

Another difference is that OCaml arrays are fixed-sized, but on Melange side this constraint is relaxed. You can change an array’s length using functions like Js.Array.push, available in the bindings to the JavaScript APIs in the Js.Array module Js.Array module.

Melange standard library also has a module to work with arrays, available in the Stdlib.Array module Stdlib.Array module.

Tuples

OCaml tuples are compiled to JavaScript arrays. This is convenient when writing bindings that will use a JavaScript array with heterogeneous values, but that happens to have a fixed length.

As a real world example of this can be found in ReasonReact, the Melange bindings for React. In these bindings, component effects dependencies are represented as OCaml tuples, so they get compiled cleanly to JavaScript arrays by Melange.

For example, some code like this:

let () = React.useEffect2 (fun () -> None) (foo, bar)

let () = React.useEffect2(() => None, (foo, bar));

Will produce:

React.useEffect(function () {}, [foo, bar]);

Booleans

Values of type bool compile to JavaScript booleans.

Records

Melange records map directly to JavaScript objects. If the record fields include non-shared data types (like variants), these values should be transformed separately, and not be directly used in JavaScript.

Extensive documentation about interfacing with JavaScript objects using records can be found in the section below.

Regular expressions

Regular expressions created using the %mel.re extension node compile to their JavaScript counterpart.

For example:

let r = [%mel.re "/b/g"]

let r = [%mel.re "/b/g"];

Will compile to:

var r = /b/g;

A regular expression like the above is of type Js.Re.t. The Js.Re module Js.Re module provides the bindings to the JavaScript functions that operate over regular expressions.

Non-shared data types

The following types differ too much between Melange and JavaScript, so while they can always be manipulated from JavaScript, it is recommended to transform them before doing so.

Variants and polymorphic variants: Better transform them into readable JavaScript values before manipulating them from JavaScript, Melange provides some helpers to do so.
Exceptions
Option (a variant type): Better use the Js.Nullable.fromOption and Js.Nullable.toOption functions in the Js.Nullable module Js.Nullable module to transform them into either null or undefined values.
List (also a variant type): use Array.of_list and Array.to_list in the Stdlib.Array module Stdlib.Array module.
Character
Int64
Lazy values

List of attributes and extension nodes

Attributes:

These attributes are used to annotate external definitions:

mel.get: read JavaScript object properties statically by name, using the dot notation .
mel.get_index: read a JavaScript object’s properties dynamically by using the bracket notation []
mel.module: bind to a value from a JavaScript module
mel.new: bind to a JavaScript class constructor
mel.obj: create a JavaScript object
mel.return: automate conversion from nullable values to Option.t values
mel.send: call a JavaScript object method using pipe first convention
mel.send.pipe: call a JavaScript object method using pipe last convention
mel.set: set JavaScript object properties statically by name, using the dot notation .
mel.set_index: set JavaScript object properties dynamically by using the bracket notation []
mel.scope: reach to deeper properties inside a JavaScript object
mel.splice: a deprecated attribute, is an alternate form of mel.variadic
mel.variadic: bind to a function taking variadic arguments from an array

These attributes are used to annotate arguments in external definitions:

u: define function arguments as uncurried (manual)
mel.int: compile function argument to an int
mel.string: compile function argument to a string
mel.this: bind to this based callbacks
mel.uncurry: define function arguments as uncurried (automated)
mel.unwrap: unwrap variant values

These attributes are used in places like records, fields, arguments, functions, and more:

mel.as: redefine the name generated in the JavaScript output code. Used in constant function arguments, variants, polymorphic variants (either inlined in external functions or in type definitions) and record fields.
deriving: generate getters and setters for some types
mel.inline: forcefully inline constant values
optional: omit fields in a record (combines with deriving)

Extension nodes:

In order to use any of these extension nodes, you will have to add the melange PPX preprocessor to your project. To do so, add the following to the dune file:

(library
 (name lib)
 (modes melange)
 (preprocess
   (pps melange.ppx)))

The same field preprocess can be added to melange.emit.

Here is the list of all the extension nodes supported by Melange:

mel.debugger: insert debugger statements
mel.external: read global values
mel.obj: create JavaScript object literals
mel.raw: write raw JavaScript code
mel.re: insert regular expressions

Generate raw JavaScript

It is possible to directly write JavaScript code from a Melange file. This is unsafe, but it can be useful for prototyping or as an escape hatch.

To do it, we will use the mel.raw extension:

let add = [%mel.raw {|
  function(a, b) {
    console.log("hello from raw JavaScript!");
    return a + b;
  }
|}]

let () = Js.log (add 1 2)

let add = [%mel.raw
  {|
  function(a, b) {
    console.log("hello from raw JavaScript!");
    return a + b;
  }
|}
];

let () = Js.log(add(1, 2));

The {||} strings are called "quoted strings". They are similar to JavaScript’s template literals, in the sense that they are multi-line, and there is no need to escape characters inside the string.

Using one percentage signthe extension name between square brackets ([%mel.raw <string>]) is useful to define expressions (function bodies, or other values) where the implementation is directly JavaScript. This is useful as we can attach the type signature already in the same line, to make our code safer. For example:

let f : unit -> int = [%mel.raw "function() {return 1}"]

let f: unit => int = ([%mel.raw "function() {return 1}"]: unit => int);

Using two percentage signs ([%%mel.raw <string>])the extension name without square brackets (%mel.raw <string>) is reserved for definitions in a structure or signature.

For example:

[%%mel.raw "var a = 1"]

[%%mel.raw "var a = 1"];

Debugger

Melange allows you to inject a debugger; expression using the mel.debugger extension:

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

let f = (x, y) => {
  [%mel.debugger];
  x + y;
};

Output:

function f (x,y) {
  debugger; // JavaScript developer tools will set a breakpoint and stop here
  return x + y | 0;
}

Detect global variables

Melange provides a relatively type safe approach to use globals that might be defined either in the JavaScript runtime environment: mel.external.

[%mel.external id] will check if the JavaScript value id is undefined or not, and return an Option.t value accordingly.

For example:

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

let () =
  switch ([%mel.external __DEV__]) {
  | Some(_) => Js.log("dev mode")
  | None => Js.log("production mode")
  };

Another example:

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

let () =
  switch ([%mel.external __filename]) {
  | Some(f) => Js.log(f)
  | None => Js.log("non-node environment")
  };

Inlining constant values

Some JavaScript idioms require special constants to be inlined since they serve as de-facto directives for bundlers. A common example is process.env.NODE_ENV:

if (process.env.NODE_ENV !== "production") {
  // Development-only code
}

becomes:

if ("development" !== "production") {
  // Development-only code
}

In this case, bundlers such as Webpack can tell that the if statement always evaluates to a specific branch and eliminate the others.

Melange provides the mel.inline attribute to achieve the same goal in generated JavaScript. Let’s look at an example:

external node_env : string = "NODE_ENV" [@@mel.scope "process", "env"]

let development = "development"
let () = if node_env <> development then Js.log "Only in Production"

let development_inline = "development" [@@mel.inline]
let () = if node_env <> development_inline then Js.log "Only in Production"

[@mel.scope ("process", "env")] external node_env: string = "NODE_ENV";

let development = "development";
let () =
  if (node_env != development) {
    Js.log("Only in Production");
  };

[@mel.inline]
let development_inline = "development";
let () =
  if (node_env != development_inline) {
    Js.log("Only in Production");
  };

As we can see in the generated JavaScript presented below:

the development variable is emitted
- it gets used as a variable process.env.NODE_ENV !== development in the if statement
the development_inline variable isn’t present in the final output
- its value is inlined in the if statement: process.env.NODE_ENV !== "development"

var development = "development";

if (process.env.NODE_ENV !== development) {
  console.log("Only in Production");
}

if (process.env.NODE_ENV !== "development") {
  console.log("Only in Production");
}

Bind to JavaScript objects

JavaScript objects are used in a variety of use cases:

As a fixed shape record.
As a map or dictionary.
As a class.
As a module to import/export.

Melange separates the binding methods for JavaScript objects based on these four use cases. This section documents the first three. Binding to JavaScript module objects is described in the "Using functions from other JavaScript modules" section.

Objects with static shape (record-like)

Using OCaml records

If your JavaScript object has fixed fields, then it’s conceptually like an OCaml record. Since Melange compiles records into JavaScript objects, the most common way to bind to JavaScript objects is using records.

type person = {
  name : string;
  friends : string array;
  age : int;
}

external john : person = "john" [@@mel.module "MySchool"]
let john_name = john.name

type person = {
  name: string,
  friends: array(string),
  age: int,
};

[@mel.module "MySchool"] external john: person = "john";
let john_name = john.name;

This is the generated JavaScript:

var MySchool = require("MySchool");

var john_name = MySchool.john.name;

External functions are documented in a previous section. The mel.module attribute is documented here.

If you want or need to use different field names on the Melange and the JavaScript sides, you can use the mel.as decorator:

type action = {
  type_ : string [@mel.as "type"]
}

let action = { type_ = "ADD_USER" }

type action = {
  [@mel.as "type"]
  type_: string,
};

let action = {type_: "ADD_USER"};

Which generates the JavaScript code:

var action = {
  type: "ADD_USER"
};

This is useful to map to JavaScript attribute names that cannot be expressed in Melange, for example, where the JavaScript name we want to generate is a reserved keyword.

It is also possible to map a Melange record to a JavaScript array by passing indices to the mel.as decorator:

type t = {
  foo : int; [@mel.as "0"]
  bar : string; [@mel.as "1"]
}

let value = { foo = 7; bar = "baz" }

type t = {
  [@mel.as "0"]
  foo: int,
  [@mel.as "1"]
  bar: string,
};

let value = {foo: 7, bar: "baz"};

And its JavaScript generated code:

var value = [
  7,
  "baz"
];

Using `Js.t` objects

Alternatively to records, Melange offers another type that can be used to produce JavaScript objects. This type is 'a Js.t, where 'a is an OCaml object.

The advantage of objects versus records is that no type declaration is needed in advance, which can be helpful for prototyping or quickly generating JavaScript object literals.

Melange provides some ways to create Js.t object values, as well as accessing the properties inside them. To create values, the [%mel.obj] extension is used, and the ## infix operator allows to read from the object properties:

let john = [%mel.obj { name = "john"; age = 99 }]
let t = john##name

let john = {"name": "john", "age": 99};
let t = john##name;

Which generates:

var john = {
  name: "john",
  age: 99
};

var t = john.name;

Note that object types allow for some flexibility that the record types do not have. For example, an object type can be coerced to another with fewer values or methods, while it is impossible to coerce a record type to another one with fewer fields. So different object types that share some methods can be mixed in a data structure where only their common methods are visible.

To give an example, one can create a function that operates in all the object types that include a field name that is of type string, e.g.:

let name_extended obj = obj##name ^ " wayne"

let one = name_extended [%mel.obj { name = "john"; age = 99 }]
let two = name_extended [%mel.obj { name = "jane"; address = "1 infinite loop" }]

let name_extended = obj => obj##name ++ " wayne";

let one = name_extended({"name": "john", "age": 99});
let two = name_extended({"name": "jane", "address": "1 infinite loop"});

To read more about objects and polymorphism we recommend checking the OCaml docs or the OCaml manual.

Using external functions

We have already explored one approach for creating JavaScript object literals by using Js.t values and the mel.obj extension.

Melange additionally offers the mel.obj attribute, which can be used in combination with external functions to create JavaScript objects. When these functions are called, they generate objects with fields corresponding to the labeled arguments of the function.

If any of these labeled arguments are defined as optional and omitted during function application, the resulting JavaScript object will exclude the corresponding fields. This allows to create runtime objects and control whether optional keys are emitted at runtime.

For example, assuming we need to bind to a JavaScript object like this:

var homeRoute = {
  type: "GET",
  path: "/",
  action: () => console.log("Home"),
  // options: ...
};

The first three fields are required and the options field is optional. You can declare a binding function like:

external route :
  _type:string ->
  path:string ->
  action:(string list -> unit) ->
  ?options:< .. > ->
  unit ->
  _ = ""
  [@@mel.obj]

[@mel.obj]
external route:
  (
    ~_type: string,
    ~path: string,
    ~action: list(string) => unit,
    ~options: {..}=?,
    unit
  ) =>
  _;

Note that the empty string at the end of the function is used to make it syntactically valid. The value of this string is ignored by the compiler.

Since there is an optional argument options, an additional unlabeled argument of type unit is included after it. It allows to omit the optional argument on function application. More information about labeled optional arguments can be found in the OCaml manual.

The return type of the function should be left unspecified using the wildcard type _. Melange will automatically infer the type of the resulting JavaScript object.

In the route function, the _type argument starts with an underscore. When binding to JavaScript objects with fields that are reserved keywords in OCaml, Melange allows the use of an underscore prefix for the labeled arguments. The resulting JavaScript object will have the underscore removed from the field names. This is only required for the mel.obj attribute, while for other cases, the mel.as attribute can be used to rename fields.

If we call the function like this:

let homeRoute = route ~_type:"GET" ~path:"/" ~action:(fun _ -> Js.log "Home") ()

let homeRoute =
  route(~_type="GET", ~path="/", ~action=_ => Js.log("Home"), ());

We get the following JavaScript, which does not include the options field since its argument wasn’t present:

var homeRoute = {
  type: "GET",
  path: "/",
  action: (function (param) {
      console.log("Home");
    })
};

Bind to object properties

If you need to bind only to the property of a JavaScript object, you can use mel.get and mel.set to access it using the dot notation .:

(* Abstract type for the `document` value *)
type document

external document : document = "document"

external set_title : document -> string -> unit = "title" [@@mel.set]
external get_title : document -> string = "title" [@@mel.get]

let current = get_title document
let () = set_title document "melange"

/* Abstract type for the `document` value */
type document;

external document: document = "document";

[@mel.set] external set_title: (document, string) => unit = "title";
[@mel.get] external get_title: document => string = "title";

let current = get_title(document);
let () = set_title(document, "melange");

This generates:

var current = document.title;
document.title = "melange";

Alternatively, if some dynamism is required on the way the property is accessed, you can use mel.get_index and mel.set_index to access it using the bracket notation []:

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

let () =
  let i32arr = (create 3) in
  set i32arr 0 42;
  Js.log (get i32arr 0)

type t;
[@mel.new] external create: int => t = "Int32Array";
[@mel.get_index] external get: (t, int) => int;
[@mel.set_index] external set: (t, int, int) => unit;

let () = {
  let i32arr = create(3);
  set(i32arr, 0, 42);
  Js.log(get(i32arr, 0));
};

Which generates:

var i32arr = new Int32Array(3);
i32arr[0] = 42;
console.log(i32arr[0]);

Objects with dynamic shape (dictionary-like)

Sometimes JavaScript objects are used as dictionaries. In these cases:

All values stored in the object belong to the same type
Key-value pairs can be added or removed at runtime

For this particular use case of JavaScript objects, Melange exposes a specific type Js.Dict.t. The values and functions to work with values of this type are defined in the Js.Dict module Js.Dict module, with operations like get, set, etc.

Values of the type Js.Dict.t compile to JavaScript objects.

JavaScript classes

JavaScript classes are special kinds of objects. To interact with classes, Melange exposes mel.new to emulate e.g. new Date():

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

type t;
[@mel.new] external create_date: unit => t = "Date";
let date = create_date();

Which generates:

var date = new Date();

You can chain mel.new and mel.module if the JavaScript class you want to work with is in a separate JavaScript module:

type t
external book : unit -> t = "Book" [@@mel.new] [@@mel.module]
let myBook = book ()

type t;
[@mel.new] [@mel.module] external book: unit => t = "Book";
let myBook = book();

Which generates:

var Book = require("Book");
var myBook = new Book();

Bind to JavaScript functions or values

Using global functions or values

Binding to a JavaScript function available globally makes use of external, like with objects. But unlike objects, there is no need to add any attributes:

(* Abstract type for `timeoutId` *)
type timeoutId
external setTimeout : (unit -> unit) -> int -> timeoutId = "setTimeout"
external clearTimeout : timeoutId -> unit = "clearTimeout"

let id = setTimeout (fun () -> Js.log "hello") 100
let () = clearTimeout id

/* Abstract type for `timeoutId` */
type timeoutId;
external setTimeout: (unit => unit, int) => timeoutId = "setTimeout";
external clearTimeout: timeoutId => unit = "clearTimeout";

let id = setTimeout(() => Js.log("hello"), 100);
let () = clearTimeout(id);

NOTE: The bindings to setTimeout and clearTimeout are shown here for learning purposes, but they are already available in the Js.Global module Js.Global module.

Generates:

var id = setTimeout(function (param) {
  console.log("hello");
}, 100);

clearTimeout(id);

Global bindings can also be applied to values:

(* Abstract type for `document` *)
type document

external document : document = "document"
let document = document

/* Abstract type for `document` */
type document;

external document: document = "document";
let document = document;

Which generates:

var doc = document;

Using functions from other JavaScript modules

mel.module allows to bind to values that belong to another JavaScript module. It accepts a string with the name of the module, or the relative path to it.

external dirname : string -> string = "dirname" [@@mel.module "path"]
let root = dirname "/User/github"

[@mel.module "path"] external dirname: string => string = "dirname";
let root = dirname("/User/github");

Generates:

var Path = require("path");
var root = Path.dirname("/User/github");

Binding to properties inside a module or global

For cases when we need to create bindings for a property within a module or a global JavaScript object, Melange provides the mel.scope attribute.

For example, if we want to write some bindings for a specific property commands from the vscode package, we can do:

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

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

type param;
[@mel.scope "commands"] [@mel.module "vscode"] [@mel.variadic]
external executeCommands: (string, array(param)) => unit;

let f = (a, b, c) => executeCommands("hi", [|a, b, c|]);

Which compiles to:

var Vscode = require("vscode");

function f(a, b, c) {
  Vscode.commands.executeCommands("hi", a, b, c);
}

The mel.scope attribute can take multiple arguments as payload, in case we want to reach deeper into the object from the module we are importing.

For example:

type t

external back : t = "back"
  [@@mel.module "expo-camera"] [@@mel.scope "Camera", "Constants", "Type"]

let camera_type_back = back

type t;

[@mel.module "expo-camera"] [@mel.scope ("Camera", "Constants", "Type")]
external back: t = "back";

let camera_type_back = back;

Which generates:

var ExpoCamera = require("expo-camera");

var camera_type_back = ExpoCamera.Camera.Constants.Type.back;

It can be used without mel.module, to created scoped bindings to global values:

external imul : int -> int -> int = "imul" [@@mel.scope "Math"]

let res = imul 1 2

[@mel.scope "Math"] external imul: (int, int) => int = "imul";

let res = imul(1, 2);

Which produces:

var res = Math.imul(1, 2);

Or it can be used together with mel.new:

type t

external create : unit -> t = "GUI"
  [@@mel.new] [@@mel.scope "default"] [@@mel.module "dat.gui"]

let gui = create ()

type t;

[@mel.new] [@mel.scope "default"] [@mel.module "dat.gui"]
external create: unit => t = "GUI";

let gui = create();

Which generates:

var DatGui = require("dat.gui");

var gui = new (DatGui.default.GUI)();

Labeled arguments

OCaml has labeled arguments, which can also be optional, and work with external as well.

Labeled arguments can be useful to provide more information about the arguments of a JavaScript function that is called from Melange.

Let’s say we have the following JavaScript function, that we want to call from Melange:

// MyGame.js

function draw(x, y, border) {
  // let’s assume `border` is optional and defaults to false
}
draw(10, 20)
draw(20, 20, true)

When writing Melange bindings, we can add labeled arguments to make things more clear:

external draw : x:int -> y:int -> ?border:bool -> unit -> unit = "draw"
  [@@mel.module "MyGame"]

let () = draw ~x:10 ~y:20 ~border:true ()
let () = draw ~x:10 ~y:20 ()

[@mel.module "MyGame"]
external draw: (~x: int, ~y: int, ~border: bool=?, unit) => unit = "draw";

let () = draw(~x=10, ~y=20, ~border=true, ());
let () = draw(~x=10, ~y=20, ());

Generates:

var MyGame = require("MyGame");

MyGame.draw(10, 20, true);
MyGame.draw(10, 20, undefined);

The generated JavaScript function is the same, but now the usage in Melange is much clearer.

Note: in this particular case, a final param of type unit, () must be added after border, since border is an optional argument at the last position. Not having the last param unit would lead to a warning, which is explained in detail in the OCaml documentation.

Note that you can freely reorder the labeled arguments when applying the function on the Melange side. The generated code will maintain the original order that was used when declaring the function:

external draw : x:int -> y:int -> ?border:bool -> unit -> unit = "draw"
  [@@mel.module "MyGame"]
let () = draw ~x:10 ~y:20 ()
let () = draw ~y:20 ~x:10 ()

[@mel.module "MyGame"]
external draw: (~x: int, ~y: int, ~border: bool=?, unit) => unit = "draw";
let () = draw(~x=10, ~y=20, ());
let () = draw(~y=20, ~x=10, ());

Generates:

var MyGame = require("MyGame");

MyGame.draw(10, 20, undefined);
MyGame.draw(10, 20, undefined);

Calling an object method

If we need to call a JavaScript method, Melange provides the attribute mel.send.

In the following snippets, we will be referring to a type Dom.element, which is provided within the library melange.dom. You can add it to your project by including (libraries melange.dom) to your dune file:

(* Abstract type for the `document` global *)
type document

external document : document = "document"
external get_by_id : document -> string -> Dom.element = "getElementById"
  [@@mel.send]

let el = get_by_id document "my-id"

/* Abstract type for the `document` global */
type document;

external document: document = "document";
[@mel.send]
external get_by_id: (document, string) => Dom.element = "getElementById";

let el = get_by_id(document, "my-id");

Generates:

var el = document.getElementById("my-id");

When using mel.send, the first argument will be the object that holds the property with the function we want to call. This combines well with the pipe first operator |.->, see the "Chaining" section below.

If we want to design our bindings to be used with OCaml pipe last operator |>, there is an alternate mel.send.pipe attribute. Let’s rewrite the example above using it:

(* Abstract type for the `document` global *)
type document

external document : document = "document"
external get_by_id : string -> Dom.element = "getElementById"
  [@@mel.send.pipe: document]

let el = get_by_id "my-id" document

/* Abstract type for the `document` global */
type document;

external document: document = "document";
[@mel.send.pipe: document]
external get_by_id: string => Dom.element = "getElementById";

let el = get_by_id("my-id", document);

Generates the same code as mel.send:

var el = document.getElementById("my-id");

Chaining

It is common to find this kind of API in JavaScript: foo().bar().baz(). This kind of API can be designed with Melange externals. Depending on which convention we want to use, there are two attributes available:

For a data-first convention, the mel.send attribute, in combination with the pipe first operator |.->
For a data-last convention, the mel.send.pipe attribute, in combination with OCaml pipe last operator |>.

Let’s see first an example of chaining using data-first convention with the pipe first operator |.->:

(* Abstract type for the `document` global *)
type document

external document : document = "document"
external get_by_id : document -> string -> Dom.element = "getElementById"
  [@@mel.send]
external get_by_classname : Dom.element -> string -> Dom.element
  = "getElementsByClassName"
  [@@mel.send]

let el = document |. get_by_id "my-id" |. get_by_classname "my-class"

/* Abstract type for the `document` global */
type document;

external document: document = "document";
[@mel.send]
external get_by_id: (document, string) => Dom.element = "getElementById";
[@mel.send]
external get_by_classname: (Dom.element, string) => Dom.element =
  "getElementsByClassName";

let el = document->(get_by_id("my-id"))->(get_by_classname("my-class"));

Will generate:

var el = document.getElementById("my-id").getElementsByClassName("my-class");

Now with pipe last operator |>:

(* Abstract type for the `document` global *)
type document

external document : document = "document"
external get_by_id : string -> Dom.element = "getElementById"
  [@@mel.send.pipe: document]
external get_by_classname : string -> Dom.element = "getElementsByClassName"
  [@@mel.send.pipe: Dom.element]

let el = document |> get_by_id "my-id" |> get_by_classname "my-class"

/* Abstract type for the `document` global */
type document;

external document: document = "document";
[@mel.send.pipe: document]
external get_by_id: string => Dom.element = "getElementById";
[@mel.send.pipe: Dom.element]
external get_by_classname: string => Dom.element = "getElementsByClassName";

let el = document |> get_by_id("my-id") |> get_by_classname("my-class");

Will generate the same JavaScript as the pipe first version:

var el = document.getElementById("my-id").getElementsByClassName("my-class");

Variadic function arguments

Sometimes JavaScript functions take an arbitrary amount of arguments. For these cases, Melange provides the mel.variadic attribute, which can be attached to the external declaration. However, there is one caveat: all the variadic arguments need to belong to the same type.

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

[@mel.module "path"] [@mel.variadic]
external join: array(string) => string = "join";
let v = join([|"a", "b"|]);

Generates:

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

If more dynamism is needed, there is a way to inject elements with different types in the array and still have Melange compile to JavaScript values that are not wrapped using the OCaml unboxed attribute, which was mentioned in the OCaml attributes section:

type hide = Hide : 'a -> hide [@@unboxed]

external join : hide array -> string = "join" [@@mel.module "path"] [@@mel.variadic]

let v = join [| Hide "a"; Hide 2 |]

[@unboxed]
type hide =
  | Hide('a): hide;

[@mel.module "path"] [@mel.variadic]
external join: array(hide) => string = "join";

let v = join([|Hide("a"), Hide(2)|]);

Compiles to:

var Path = require("path");

var v = Path.join("a", 2);

Bind to a polymorphic function

Some JavaScript libraries will define functions where the arguments can vary on both type and shape. There are two approaches to bind to those, depending on how dynamic they are.

Approach 1: Multiple external functions

If it is possible to enumerate the many forms an overloaded JavaScript function can take, a flexible approach is to bind to each form individually:

external drawCat : unit -> unit = "draw" [@@mel.module "MyGame"]
external drawDog : giveName:string -> unit = "draw" [@@mel.module "MyGame"]
external draw : string -> useRandomAnimal:bool -> unit = "draw"
  [@@mel.module "MyGame"]

[@mel.module "MyGame"] external drawCat: unit => unit = "draw";
[@mel.module "MyGame"] external drawDog: (~giveName: string) => unit = "draw";
[@mel.module "MyGame"]
external draw: (string, ~useRandomAnimal: bool) => unit = "draw";

Note how all three externals bind to the same JavaScript function, draw.

Approach 2: Polymorphic variant + `mel.unwrap`

In some cases, the function has a constant number of arguments but the type of the argument can vary. For cases like this, we can model the argument as a variant and use the mel.unwrap attribute in the external.

Let’s say we want to bind to the following JavaScript function:

function padLeft(value, padding) {
  if (typeof padding === "number") {
    return Array(padding + 1).join(" ") + value;
  }
  if (typeof padding === "string") {
    return padding + value;
  }
  throw new Error(`Expected string or number, got '${padding}'.`);
}

As the padding argument can be either a number or a string, we can use mel.unwrap to define it. It is important to note that mel.unwrap imposes certain requirements on the type it is applied to:

It needs to be a polymorphic variant
Its definition needs to be inlined
Each variant tag needs to have an argument
The variant type can not be opened (can’t use >)

external padLeft:
  string
  -> ([ `Str of string
      | `Int of int
      ] [@mel.unwrap])
  -> string
  = "padLeft"

let _ = padLeft "Hello World" (`Int 4)
let _ = padLeft "Hello World" (`Str "Message from Melange: ")

external padLeft:
  (string, [@mel.unwrap] [ | `Str(string) | `Int(int)]) => string =
  "padLeft";

let _ = padLeft("Hello World", `Int(4));
let _ = padLeft("Hello World", `Str("Message from Melange: "));

Which produces the following JavaScript:

padLeft("Hello World", 4);
padLeft("Hello World", "Message from Melange: ");

As we saw in the Non-shared data types section, we should rather avoid passing variants directly to the JavaScript side. By using mel.unwrap we get the best of both worlds: from Melange we can use variants, while JavaScript gets the raw values inside them.

Using polymorphic variants to bind to enums

Some JavaScript APIs take a limited subset of values as input. For example, Node’s fs.readFileSync second argument can only take a few given string values: "ascii", "utf8", etc. Some other functions can take values from a few given integers, like the createStatusBarItem function in VS Code API, which can take an alignment parameter that can only be 1 or 2.

One could still type these parameters as just string or int, but this would not prevent consumers of the external function from calling it using values that are unsupported by the JavaScript function. Let’s see how we can use polymorphic variants to avoid runtime errors.

If the values are strings, we can use the mel.string attribute:

external read_file_sync :
  name:string -> ([ `utf8 | `ascii ][@mel.string]) -> string = "readFileSync"
  [@@mel.module "fs"]

let _ = read_file_sync ~name:"xx.txt" `ascii

[@mel.module "fs"]
external read_file_sync:
  (~name: string, [@mel.string] [ | `utf8 | `ascii]) => string =
  "readFileSync";

let _ = read_file_sync(~name="xx.txt", `ascii);

Which generates:

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

This technique can be combined with the mel.as attribute to modify the strings produced from the polymorphic variant values. For example:

type document
type style

external document : document = "document"
external get_by_id : document -> string -> Dom.element = "getElementById"
[@@mel.send]
external style : Dom.element -> style = "style" [@@mel.get]
external transition_timing_function :
  style ->
  ([ `ease
   | `easeIn [@mel.as "ease-in"]
   | `easeOut [@mel.as "ease-out"]
   | `easeInOut [@mel.as "ease-in-out"]
   | `linear ]
  [@mel.string]) ->
  unit = "transitionTimingFunction"
[@@mel.set]

let element_style = style (get_by_id document "my-id")
let () = transition_timing_function element_style `easeIn

type document;
type style;

external document: document = "document";
[@mel.send]
external get_by_id: (document, string) => Dom.element = "getElementById";
[@mel.get] external style: Dom.element => style = "style";
[@mel.set]
external transition_timing_function:
  (
    style,
    [@mel.string] [
      | `ease
      | [@mel.as "ease-in"] `easeIn
      | [@mel.as "ease-out"] `easeOut
      | [@mel.as "ease-in-out"] `easeInOut
      | `linear
    ]
  ) =>
  unit =
  "transitionTimingFunction";

let element_style = style(get_by_id(document, "my-id"));
let () = transition_timing_function(element_style, `easeIn);

This will generate:

var element_style = document.getElementById("my-id").style;

element_style.transitionTimingFunction = "ease-in";

Aside from producing string values, Melange also offers mel.int to produce integer values. mel.int can also be combined with mel.as:

external test_int_type :
  ([ `on_closed | `on_open [@mel.as 20] | `in_bin ][@mel.int]) -> int
  = "testIntType"

let value = test_int_type `on_open

external test_int_type:
  ([@mel.int] [ | `on_closed | [@mel.as 20] `on_open | `in_bin]) => int =
  "testIntType";

let value = test_int_type(`on_open);

In this example, on_closed will be encoded as 0, on_open will be 20 due to the attribute mel.as and in_bin will be 21, because if no mel.as annotation is provided for a variant tag, the compiler continues assigning values counting up from the previous one.

This code generates:

var value = testIntType(20);

Using polymorphic variants to bind to event listeners

Polymorphic variants can also be used to wrap event listeners, or any other kind of callback, for example:

type readline

external on :
  readline ->
  ([ `close of unit -> unit | `line of string -> unit ][@mel.string]) ->
  readline = "on"
  [@@mel.send]

let register rl =
  rl |. on (`close (fun event -> ())) |. on (`line (fun line -> Js.log line))

type readline;

[@mel.send]
external on:
  (
    readline,
    [@mel.string] [ | `close(unit => unit) | `line(string => unit)]
  ) =>
  readline =
  "on";

let register = rl =>
  rl->(on(`close(event => ())))->(on(`line(line => Js.log(line))));

This generates:

function register(rl) {
  return rl
    .on("close", function($$event) {})
    .on("line", function(line) {
      console.log(line);
    });
}

Constant values as arguments

Sometimes we want to call a JavaScript function and make sure one of the arguments is always constant. For this, the [@mel.as] attribute can be combined with the wildcard pattern _:

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

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

external process_on_exit: ([@mel.as "exit"] _, int => unit) => unit =
  "process.on";

let () =
  process_on_exit(exit_code =>
    Js.log("error code: " ++ string_of_int(exit_code))
  );

This generates:

process.on("exit", function (exitCode) {
  console.log("error code: " + exitCode.toString());
});

The mel.as "exit" and the wildcard _ pattern together will tell Melange to compile the first argument of the JavaScript function to the string "exit".

You can also use any JSON literal by passing a quoted string to mel.as: mel.as {json|true|json} or mel.as {json|{"name": "John"}|json}.

Binding to callbacks

In OCaml, all functions have arity 1. This means that if you define a function like this:

let add x y = x + y

let add = (x, y) => x + y;

Its type will be int -> int -> int. This means that one can partially apply add by calling add 1, which will return another function expecting the second argument of the addition. This kind of functions are called "curried" functions, more information about currying in OCaml can be found in this chapter of the "OCaml Programming: Correct + Efficient + Beautiful" book.

This is incompatible with how function calling conventions work in JavaScript, where all function calls always apply all the arguments. To continue the example, let’s say we have an add function implemented in JavaScript, similar to the one above:

var add = function (a, b) {
    return a + b;
};

If we call add(1), the function will be totally applied, with b having undefined value. And as JavaScript will try to add 1 with undefined, we will get NaN as a result.

To illustrate this difference and how it affects Melange bindings, let’s say we want to write bindings for a JavaScript function like this:

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 function declaration could be as below:

external map : 'a array -> 'b array -> ('a -> 'b -> 'c) -> 'c array = "map"

external map: (array('a), array('b), ('a, 'b) => 'c) => array('c) = "map";

Unfortunately, this is not completely correct. The issue is in the callback function, with type 'a -> 'b -> 'c. This means that map will expect a function like add described above. But as we said, in OCaml, having two arguments means just to have two functions that take one argument.

Let’s rewrite add to make the problem a bit more clear:

let add x = let partial y = x + y in partial

let add = x => {
  let partial = y => x + y;
  partial;
};

This will be compiled to:

function add(x) {
  return (function (y) {
    return x + y | 0;
  });
}

Now if we ever used our external function map with our add function by calling map arr1 arr2 add it would not work as expected. JavaScript function application does not work the same as in OCaml, so the function call in the map implementation, f(a[i],b[i]), would be applied over the outer JavaScript function add, which only takes one argument x, and b[i] would be just discarded. The value returned from the operation would not be the addition of the two numbers, but rather the inner anonymous callback.

To solve this mismatch between OCaml and JavaScript functions and their application, Melange provides a special attribute @u that can be used to annotate external functions that need to be "uncurried".

In Reason syntax, this attribute does not need to be written explicitly, as it is deeply integrated with the Reason parser. To specify some function type as "uncurried", one just needs to add the dot character . to the function type. For example, (. 'a, 'b) => 'c instead of ('a, 'b) => 'c.

In the example above:

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

external map: (array('a), array('b), (. 'a, 'b) => 'c) => array('c) = "map";

Here ('a -> 'b -> 'c [@u])(. 'a, 'b) => 'cwill be interpreted as having arity 2. In general, 'a0 -> 'a1 ... 'aN -> 'b0 [@u] is the same as 'a0 -> 'a1 ... 'aN -> 'b0. 'a0, 'a1, ... 'aN => 'b0 is the same as 'a0, 'a1, ... 'aN => 'b0 except the former’s arity is guaranteed to be N while the latter is unknown.

If we try now to call map using add:

let add x y = x + y
let _ = map [||] [||] add

let add = (x, y) => x + y;
let _ = map([||], [||], add);

We will get an error:

let _ = map [||] [||] add
                      ^^^
This expression has type int -> int -> int
but an expression was expected of type ('a -> 'b -> 'c) Js.Fn.arity2

To solve this, we add @u. in the function definition as well:

let add = fun [@u] x y -> x + y

let add = (. x, y) => x + y;

Annotating function definitions can be quite cumbersome when writing a lot of externals.

To work around the verbosity, Melange offers another attribute called mel.uncurry.

Let’s see how we could use it in the previous example. We just need to replace u with mel.uncurry:

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

external map:
  (array('a), array('b), [@mel.uncurry] (('a, 'b) => 'c)) => array('c) =
  "map";

Now if we try to call map with a regular add function:

let add x y = x + y
let _ = map [||] [||] add

let add = (x, y) => x + y;
let _ = map([||], [||], add);

Everything works fine now, without having to attach any attributes to add.

The main difference between u and mel.uncurry is that the latter only works with externals. mel.uncurry is the recommended option to use for bindings, while u remains useful for those use cases where performance is crucial and we want the JavaScript functions generated from OCaml ones to not be applied partially.

Modeling `this`-based Callbacks

Many JavaScript libraries have callbacks which rely on the this keyword, for example:

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

Inside the x.onload callback, this would be pointing to x. It would not be correct to declare x.onload of type unit -> unit. Instead, Melange introduces a special attribute, mel.this, which allows to type x as this:

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

type x;
external x: x = "x";
[@mel.set]
external set_onload: (x, [@mel.this] ((x, int) => unit)) => unit = "onload";
[@mel.get] external resp: x => int = "response";
let _ = set_onload(x, [@mel.this] (o, v) => Js.log(resp(o) + v));

Which generates:

x.onload = function (v) {
  var o = this;
  console.log((o.response + v) | 0);
};

Note that the first argument will be reserved for this.

Wrapping returned nullable values

JavaScript models null and undefined differently, whereas it can be useful to treat both as 'a optionoption('a) in Melange.

Melange understands the mel.return attribute in externals to model how nullable return types should be wrapped at the bindings boundary. An external value with mel.return converts the return value to an option type, avoiding the need for extra wrapping / unwrapping with functions such as Js.Nullable.toOption.

type element
type document
external get_by_id : document -> string -> element option = "getElementById"
  [@@mel.send] [@@mel.return nullable]

let test document =
  let elem = get_by_id document "header" in
  match elem with
  | None -> 1
  | Some _element -> 2

type element;
type document;
[@mel.send] [@mel.return nullable]
external get_by_id: (document, string) => option(element) = "getElementById";

let test = document => {
  let elem = get_by_id(document, "header");
  switch (elem) {
  | None => 1
  | Some(_element) => 2
  };
};

Which generates:

function test($$document) {
  var elem = $$document.getElementById("header");
  if (elem == null) {
    return 1;
  } else {
    return 2;
  }
}

The mel.return attribute takes an attribute payload, as seen with [@@mel.return nullable][@mel.return nullable] above. Currently 4 directives are supported: null_to_opt, undefined_to_opt, nullable and identity.

nullable is encouraged, as it will convert from null and undefined to option type.

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

Generate getters, setters and constructors

As we saw in a previous section, there are some types in Melange that compile to values that are not easy to manipulate from JavaScript. To facilitate the communication from JavaScript code with values of these types, Melange includes an attribute deriving that helps generating conversion functions, as well as functions to create values from these types. In particular, for variants and polymorphic variants.

Additionally, deriving can be used with record types, to generate setters and getters as well as creation functions.

Variants

Creating values

Use @deriving accessors on a variant type to create constructor values for each branch.

type action =
  | Click
  | Submit of string
  | Cancel
[@@deriving accessors]

[@deriving accessors]
type action =
  | Click
  | Submit(string)
  | Cancel;

Melange will generate one let definition for each variant tag, implemented as follows:

For variant tags with payloads, it will be a function that takes the payload value as a parameter.
For variant tags without payloads, it will be a constant with the runtime value of the tag.

Given the action type definition above, annotated with deriving, Melange will generate something similar to the following code:

type action =
  | Click
  | Submit of string
  | Cancel

let click = (Click : action)
let submit param = (Submit param : action)
let cancel = (Cancel : action)

type action =
  | Click
  | Submit(string)
  | Cancel;

let click: action = Click;
let submit = (param): action => Submit(param);
let cancel: action = Cancel;

Which will result in the following JavaScript code after compilation:

function submit(param_0) {
  return /* Submit */{
          _0: param_0
        };
}

var click = /* Click */0;

var cancel = /* Cancel */1;

Note the generated definitions are lower-cased, and they can be safely used from JavaScript code. For example, if the above JavaScript generated code was located in a generators.js file, the definitions can be used like this:

const generators = require('./generators.js');

const hello = generators.submit("Hello");
const click = generators.click;

Conversion functions

Use @deriving jsConverter on a variant type to create converter functions that allow to transform back and forth between JavaScript integers and Melange variant values.

There are a few differences with @deriving accessors:

jsConverter works with the mel.as attribute, while accessors does not
jsConverter does not support variant tags with payload, while accessors does
jsConverter generates functions to transform values back and forth, while accessors generates functions to create values

Let’s see a version of the previous example, adapted to work with jsConverter given the constraints above:

type action =
  | Click
  | Submit [@mel.as 3]
  | Cancel
[@@deriving jsConverter]

[@deriving jsConverter]
type action =
  | Click
  | [@mel.as 3] Submit
  | Cancel;

This will generate a couple of functions with the following types:

val actionToJs : action -> int

val actionFromJs : int -> action option

external actionToJs: action => int = ;

external actionFromJs: int => option(action) = ;

actionToJs returns integers from values of action type. It will start with 0 for Click, 3 for Submit (because it was annotated with mel.as), and then 4 for Cancel, in the same way that we described when using mel.int with polymorphic variants.

actionFromJs returns a value of type option, because not every integer can be converted into a variant tag of the action type.

Hide runtime types

For extra type safety, we can hide the runtime representation of variants (int) from the generated functions, by using jsConverter { newType } payload with @deriving:

type action =
  | Click
  | Submit [@mel.as 3]
  | Cancel
[@@deriving jsConverter { newType }]

[@deriving jsConverter({newType: newType})]
type action =
  | Click
  | [@mel.as 3] Submit
  | Cancel;

This feature relies on abstract types to hide the JavaScript runtime representation. It will generate functions with the following types:

val actionToJs : action -> abs_action

val actionFromJs : abs_action -> action

external actionToJs: action => abs_action = ;

external actionFromJs: abs_action => action = ;

In the case of actionFromJs, the return value, unlike the previous case, is not an option type. This is an example of "correct by construction": the only way to create an abs_action is by calling the actionToJs function.

Polymorphic variants

The @deriving jsConverter attribute is applicable to polymorphic variants as well.

NOTE: Similarly to variants, the @deriving jsConverter attribute cannot be used when the polymorphic variant tags have payloads. Refer to the section on runtime representation to learn more about how polymorphic variants are represented in JavaScript.

Let’s see an example:

type action =
  [ `Click
  | `Submit [@mel.as "submit"]
  | `Cancel
  ]
[@@deriving jsConverter]

[@deriving jsConverter]
type action = [ | `Click | [@mel.as "submit"] `Submit | `Cancel];

Akin to the variant example, the following two functions will be generated:

val actionToJs : action -> string

val actionFromJs : string -> action option

external actionToJs: action => string = ;

external actionFromJs: string => option(action) = ;

The jsConverter { newType } payload can also be used with polymorphic variants.

Records

Accessing fields

Use @deriving accessors on a record type to create accessor functions for its record field names.

type pet = { name : string } [@@deriving accessors]

let pets = [| { name = "Brutus" }; { name = "Mochi" } |]

let () = pets |. Belt.Array.map name |. Js.Array2.joinWith "&" |. Js.log

[@deriving accessors]
type pet = {name: string};

let pets = [|{name: "Brutus"}, {name: "Mochi"}|];

let () = pets->(Belt.Array.map(name))->(Js.Array2.joinWith("&"))->Js.log;

Melange will generate a function for each field defined in the record. In this case, a function name that allows to get that field from any record of type pet:

let name (param : pet) = param.name

let name = (param: pet) => param.name;

Considering all the above, the produced JavaScript will be:

function name(param) {
  return param.name;
}

var pets = [
  {
    name: "Brutus"
  },
  {
    name: "Mochi"
  }
];

console.log(Belt_Array.map(pets, name).join("&"));

Convert records into abstract types

When binding to JavaScript objects, it is generally recommended to use records since Melange precisely uses objects as their runtime representation. This approach was discussed in the section about binding to JavaScript objects.

But there’s a specific case where records may not be enough: when we want to emit a JavaScript object where some of the keys might be present or absent.

For instance, consider the following record:

type person = {
  name : string;
  age : int option;
}

type person = {
  name: string,
  age: option(int),
};

An example of this use-case would be expecting { name = "John"; age = None } to generate a JavaScript such as {name: "Carl"}, where the age key doesn’t appear.

The @deriving abstract attribute exists to solve this problem. When present in a record type, @deriving abstract makes the record definition abstract and generates the following functions instead:

A constructor function for creating values of the type
Getters and setters for accessing the record fields

@deriving abstract effectively models a record-shaped JavaScript object exclusively through a set of (generated) functions derived from attribute annotations on the OCaml type definition.

Let’s see an example. Considering this Melange code:

type person = {
  name : string;
  age : int option; [@mel.optional]
}
[@@deriving abstract]

[@deriving abstract]
type person = {
  name: string,
  [@mel.optional]
  age: option(int),
};

Melange will make the person type abstract and generate constructor, getter and setter functions. In our example, the OCaml signature would look like this after preprocessing:

type person

val person : name:string -> ?age:int -> unit -> person

val nameGet : person -> string

val ageGet : person -> int option

type person;

external person: (~name: string, ~age: int=?, unit) => person = ;

external nameGet: person => string = ;

external ageGet: person => option(int) = ;

The person function can be used to create values of person. It is the only possible way to create values of this type, since Melange makes it abstract. Using literals like { name = "Alice"; age = None } directly doesn’t type check.

Here is an example of how we can use it:

let alice = person ~name:"Alice" ~age:20 ()
let bob = person ~name:"Bob" ()

let alice = person(~name="Alice", ~age=20, ());
let bob = person(~name="Bob", ());

This will generate the following JavaScript code. Note how there is no JavaScript runtime overhead:

var alice = {
  name: "Alice",
  age: 20
};

var bob = {
  name: "Bob"
};

The person function uses labeled arguments to represent record fields. Because there is an optional argument age, it takes a last argument of type unit. This non-labeled argument allows to omit the optional argument on function application. Further details about optional labeled arguments can be found in the corresponding section of the OCaml manual.

The functions nameGet and ageGet are accessors for each record field:

let twenty = ageGet alice

let bob = nameGet bob

let twenty = ageGet(alice);

let bob = nameGet(bob);

This generates:

var twenty = alice.age;

var bob = bob.name;

The functions are named by appending Get to the field names of the record to prevent potential clashes with other values within the module. If shorter names are preferred for the getter functions, there is an alternate { abstract = light } payload that can be passed to deriving:

type person = {
  name : string;
  age : int;
}
[@@deriving abstract { light }]

let alice = person ~name:"Alice" ~age:20
let aliceName = name alice

[@deriving abstract({light: light})]
type person = {
  name: string,
  age: int,
};

let alice = person(~name="Alice", ~age=20);
let aliceName = name(alice);

Which generates:

var alice = {
  name: "Alice",
  age: 20
};

var aliceName = alice.name;

In this example, the getter functions share the same names as the object fields. Another distinction from the previous example is that the person constructor function no longer requires the final unit argument since we have excluded the optional field in this case.

NOTE: The mel.as attribute can still be applied to record fields when the record type is annotated with deriving, allowing for the renaming of fields in the resulting JavaScript objects, as demonstrated in the section about binding to objects with static shape. However, the option to pass indices to the mel.as decorator (like [@mel.as "0"]) to change the runtime representation to an array is not available when using deriving.

Compatibility with OCaml features

The @deriving abstract attribute and its lightweight variant can be used with mutable fields and private types, which are features inherited by Melange from OCaml.

When the record type has mutable fields, Melange will generate setter functions for them. For example:

type person = {
  name : string;
  mutable age : int;
}
[@@deriving abstract]

let alice = person ~name:"Alice" ~age:20

let () = ageSet alice 21

[@deriving abstract]
type person = {
  name: string,
  mutable age: int,
};

let alice = person(~name="Alice", ~age=20);

let () = ageSet(alice, 21);

This will generate:

var alice = {
  name: "Alice",
  age: 20
};

alice.age = 21;

If the mutable keyword is omitted from the interface file, Melange will not include the setter function in the module signature, preventing other modules from mutating any values from the type.

Private types can be used to prevent Melange from creating the constructor function. For example, if we define person type as private:

type person = private {
  name : string;
  age : int;
}
[@@deriving abstract]

[@deriving abstract]
type person =
  pri {
    name: string,
    age: int,
  };

The accessors nameGet and ageGet will still be generated, but not the constructor person. This is useful when binding to JavaScript objects while preventing any Melange code from creating values of such type.

Use Melange code from JavaScript

As mentioned in the build system section, Melange allows to produce both CommonJS and ES6 modules. In both cases, using Melange-generated JavaScript code from any hand-written JavaScript file works as expected.

The following definition:

let print name = "Hello" ^ name

let print = name => "Hello" ++ name;

Will generate this JavaScript code, when using CommonJS (the default):

function print(name) {
  return "Hello" + name;
}

exports.print = print;

When using ES6 (through the (module_systems es6) field in melange.emit) this code will be generated:

function print(name) {
  return "Hello" + name;
}

export {
  print ,
}

So one can use either require or import (depending on the module system of choice) to import the print value in a JavaScript file.

Default ES6 values

One special case occur when working with JavaScript imports in ES6 modules that look like this:

import ten from 'numbers.js';

This import expects numbers.js to have a default export, like:

export default ten = 10;

To emulate this kind of exports from Melange, one just needs to define a default value.

For example, in a file named numbers.mlnumbers.re:

let default = 10

let default = 10;

That way, Melange will set the value on the default export so it can be consumed as default import on the JavaScript side.

Communicate with JavaScript

Language concepts

Attributes and extension nodes

Extension nodes

Attributes

Reusing OCaml attributes

Defining new attributes

External functions

Special identity external

Abstract types

Pipe operators

Pipe last

Pipe first

Data types and runtime representation

Shared types

Strings

Floating-point numbers

Integers

Arrays

Tuples

Booleans

Records

Regular expressions

Non-shared data types

List of attributes and extension nodes

Generate raw JavaScript

Debugger

Detect global variables

Inlining constant values

Bind to JavaScript objects

Objects with static shape (record-like)

Using OCaml records

Using Js.t objects

Using external functions

Bind to object properties

Objects with dynamic shape (dictionary-like)

JavaScript classes

Bind to JavaScript functions or values

Using global functions or values

Using functions from other JavaScript modules

Binding to properties inside a module or global

Labeled arguments

Calling an object method

Chaining

Variadic function arguments

Bind to a polymorphic function

Approach 1: Multiple external functions

Approach 2: Polymorphic variant + mel.unwrap

Using polymorphic variants to bind to enums

Using polymorphic variants to bind to event listeners

Constant values as arguments

Binding to callbacks

Modeling this-based Callbacks

Wrapping returned nullable values

Generate getters, setters and constructors

Variants

Creating values

Conversion functions

Hide runtime types

Polymorphic variants

Records

Accessing fields

Convert records into abstract types

Compatibility with OCaml features

Use Melange code from JavaScript

Default ES6 values

Using `Js.t` objects

Approach 2: Polymorphic variant + `mel.unwrap`

Modeling `this`-based Callbacks