Write facade types for JavaScript APIs

When writing an application with Scala.js, it is expected that the main application logic be written in Scala.js, and that existing JavaScript libraries are leveraged. Calling JavaScript from Scala.js is therefore the most important direction of interoperability.

Facade types are zero-overhead typed APIs for JavaScript libraries. They are similar in spirit to TypeScript type definitions.

Defining JavaScript interfaces with traits

Most JavaScript APIs work with interfaces that are defined structurally. In Scala.js, the corresponding concept are traits. To mark a trait as being a representative of a JavaScript API, it must inherit directly or indirectly from js.Any (usually from js.Object).

JS traits can contain val, var and def definitions, and the latter can be overloaded.

By default, types extending js.Any are native JS types. There also exist Scala.js-defined JS types. Native JS types should be annotated with @js.native for forward source compatibility with Scala.js 1.0.0.

Pre 0.6.5 note: Before Scala.js 0.6.5, the @js.native annotation did not exist, so you will find old code that does not yet use it to annotate native JS types.

In native JS types, all concrete definitions must have = js.native as body. Any other body will be handled as if it were = js.native, and a warning will be emitted. (In Scala.js 1.0.0, this will become an error.)

0.5.x note: In Scala.js 0.5.x, = js.native did not exist either. The recommended best practice was to put ??? as body, but this was not enforced by the compiler. This has been changed to improve intuition and remove warts.

Here is an example giving types to a small portion of the API of Window objects in browsers.

@js.native
trait Window extends js.Object {
  val document: HTMLDocument = js.native
  var location: String = js.native

  def innerWidth: Int = js.native
  def innerHeight: Int = js.native

  def alert(message: String): Unit = js.native

  def open(url: String, target: String,
      features: String = ""): Window = js.native
  def close(): Unit = js.native
}

Remarks

var, val and def definitions without parentheses all map to field access in JavaScript, whereas def definitions with parentheses (even empty) map to method calls in JavaScript.

The difference between a val and a def without parentheses is that the result of the former is stable (in Scala semantics). Pragmatically, use val if the result will always be the same (e.g., document), and def when subsequent accesses to the field might return a different value (e.g., innerWidth).

Calls to the apply method of an object x map to calling x, i.e., x(...) instead of x.apply(...).

Methods can have parameters with default values, to mark them as optional. However, the actual value is irrelevant and never used. Instead, the parameter is omitted entirely (or set to undefined). The value is only indicative, as implicit documentation.

Fields, parameters, or result types that can have different, unrelated types, can be accurately typed with the pseudo-union type A | B.

Methods can be overloaded. This is useful to type accurately some APIs that behave differently depending on the number or types of arguments.

JS traits and their methods can have type parameters, abstract type members and type aliases, without restriction compared to Scala’s type system.

However, inner traits, classes and objects don’t make sense and are forbidden. It is however allowed to declare a JS trait in a top-level object.

Methods can have varargs, denoted by * like in regular Scala. They map to JavaScript varargs, i.e., the method is called with more arguments.

isInstanceOf[T] is not supported for any trait T inheriting from js.Any. Consequently, pattern matching for such types is not supported either.

asInstanceOf[T] is completely erased for any T inheriting from js.Any, meaning that it does not perform any runtime check. It is always valid to cast anything to such a trait.

JavaScript field/method names and their Scala counterpart

Sometimes, a JavaScript API defines fields and/or methods with names that do not feel right in Scala. For example, jQuery objects feature a method named val(), which, obviously, is a keyword in Scala.

They can be defined in Scala in two ways. The trivial one is simply to use backquotes to escape them in Scala:

def `val`(): String = js.native
def `val`(v: String): this.type = js.native

However, it becomes annoying very quickly. An often better solution is to use the scala.scalajs.js.annotation.JSName annotation to specify the JavaScript name to use, which can be different from the Scala name:

@JSName("val")
def value(): String = js.native
@JSName("val")
def value(v: String): this.type = js.native

If necessary, several overloads of a method with the same name can have different @JSName’s. Conversely, several methods with different names in Scala can have the same @JSName.

Members with a JavaScript symbol “name”

@JSName can also be given a reference to a js.Symbol instead of a constant string. This is used for JavaScript members whose “name” is actually a symbol. For example, JavaScript iterable objects must declare a method whose name is the symbol Symbol.iterator:

@JSName(js.Symbol.iterator)
def iterator(): js.Iterator[Int] = js.native

The argument to @JSName must be a reference to a static, stable field. In practice, this means a val in top-level object. js.Symbol.iterator is such a val, declared in the top-level object js.Symbol.

Scala methods representing bracket access (obj[x])

The annotation scala.scalajs.js.annotation.JSBracketAccess can be used on methods to mark them as representing bracket access on an object. The target method must either have one parameter and a non-Unit result type (in which case it represents read access) or two parameters and a Unit result type (in which case it represents write access).

A typical example can be found in the js.Array[A] class itself, of course:

@JSBracketAccess
def apply(index: Int): A = js.native
@JSBracketAccess
def update(index: Int, v: A): Unit = js.native

The Scala method names are irrelevant for the translation to JavaScript. The duo apply/update is often a sensible choice, because it gives array-like access on Scala’s side as well, but it is not required to use these names.

Native JavaScript classes

It is also possible to define native JavaScript classes as Scala classes inheriting, directly or indirectly, from js.Any (like traits, usually from js.Object). The main difference compared to traits is that classes have constructors, hence they also provide instantiation of objects with the new keyword.

Unlike traits, classes actually exist in the JavaScript world, often as top-level, global variables. They must therefore be annotated with the @JSGlobal annotation. For example:

@js.native
@JSGlobal
class RegExp(pattern: String) extends js.Object {
  ...
}

Pre 0.6.15 note: Before Scala.js 0.6.15, the @JSGlobal annotation did not exist, so you will find old code that does not yet use it to annotate native JS classes.

The call new RegExp("[ab]*") will map to the obvious in JavaScript, i.e., new RegExp("[ab]*"), meaning that the identifier RegExp will be looked up in the global scope.

If it is impractical or inconvenient to declare the Scala class with the same name as the JavaScript class (e.g., because it is defined in a namespace, like THREE.Scene), a constant string can be given as parameter to @JGlobal to specify the JavaScript name:

@js.native
@JSGlobal("THREE.Scene")
class Scene extends js.Object

Remarks

If the class does not have any constructor without argument, and it has to be subclassed, you may either decide to add a fake protected no-arg constructor, or call an inherited constructor with ???s as parameters.

isInstanceOf[C] is supported for classes inheriting from js.Any. It is implemented with an instanceof test. Pattern matching, including ClassTag-based matching, work accordingly.

As is the case for traits, asInstanceOf[C] is completely erased for any class C inheriting from js.Any, meaning that it does not perform any runtime check. It is always valid to cast anything to such a class.

Top-level JavaScript objects

JavaScript APIs often expose top-level objects with methods and fields. For example, the JSON object provides methods for parsing and emitting JSON strings. These can be declared in Scala.js with object’s inheriting directly or indirectly from js.Any (again, often js.Object). As is the case with classes, they must be annotated with @js.native and @JSGlobal.

@js.native
@JSGlobal
object JSON extends js.Object {
  def parse(text: String): js.Any = js.native

  def stringify(value: js.Any): String = js.native
}

An call like JSON.parse(text) will map in JavaScript to the obvious, i.e., JSON.parse(text), meaning that the identifier JSON will be looked up in the global scope.

Similarly to classes, the JavaScript name can be specified as an explicit argument to @JSGlobal, e.g.,

@js.native
@JSGlobal("jQuery")
object JQuery extends js.Object {
  def apply(x: String): JQuery = js.native
}

Unlike classes and traits, native JS objects can have inner native JS classes, traits and objects. Inner classes and objects will be looked up as fields of the enclosing JS object.

Variables and functions in the global scope

Besides object-like top-level definitions, JavaScript also defines variables and functions in the global scope. Scala does not have top-level variables and functions. Instead, in Scala.js, top-level objects annotated with @JSGlobalScope are considered to represent the global scope.

import js.annotation._

@js.native
@JSGlobalScope
object DOMGlobalScope extends js.Object {
  val document: HTMLDocument = js.native

  def alert(message: String): Unit = js.native
}

Prior to 0.6.13, extends js.GlobalScope was used instead of @JSGlobalScope. js.GlobalScope is now deprecated.

Imports from other JavaScript modules

Important: Importing from JavaScript modules requires that you emit a module for the Scala.js code.

The previous sections on native classes and objects all load things from the JavaScript global scope (through zero or more property accesses from there). In modern JavaScript ecosystems, we often want to load things from other modules. This is what @JSImport is designed for. You can annotate an @js.native class or object with @JSImport instead of @JSGlobal to signify that it is defined in a module. For example, in the following snippet:

@js.native
@JSImport("bar.js", "Foo")
class Foobaz(val x: Int) extends js.Object

val f = new Foobaz(5)

the annotation specifies that Foobaz is a native JS class defined in the module "bar.js", and exported under the name "Foo". Semantically, @JSImport corresponds to an ECMAScript 2015 import, and the above code is therefore equivalent to this JavaScript code:

import { Foo as Foobaz } from "bar.js";
var f = new Foobaz(5);

In CommonJS terms, this would be:

var bar = require("bar.js");
var f = new bar.Foo(5);

The first argument to @JSImport is the name of the JavaScript module you wish to import. The second argument denotes what member of the module you are importing. It can be one of the following:

  • A string indicating the name of member. The string can be a .-separated chain of selections (e.g., "Foo.Babar").
  • The constant JSImport.Default, to select the default export of the JavaScript module. This corresponds to import Foobaz from "bar.js".
  • The constant JSImport.Namespace, to select the module itself (with its exports as fields). This corresponds to import * as Foobaz from "bar.js".

The latter is particularly useful if you want to import members of the modules that are neither classes nor objects (for example, functions):

@js.native
@JSImport("bar.js", JSImport.Namespace)
object Bar {
  def exportedFunction(x: Int): Int = js.native
}

val y = Bar.exportedFunction(5)

In CommonJS terms, this would be:

var bar = require("bar.js");
var y = bar.exportedFunction(5);

If the previous example had used JSImport.Default instead of JSImport.Namespace, the current translation into CommonJS terms would be the following:

function moduleDefault(m) {
  return (m && (typeof m === "object") && "default" in m) ? m["default"] : m;
}

var bar = require("bar.js");
var y = moduleDefault(bar).exportedFunction(5);

This is subject to change in future versions of Scala.js, to better reflect the evolution of specifications in ECMAScript itself, and its implementations.

Important: @JSImport is completely incompatible with jsDependencies. You should use a separate mechanism to manage your JavaScript dependencies. Scala.js does not provide any facility to do so, at the moment.

Default import or namespace import?

The default export accessible with JSImport.Default, specified in terms of ECMAScript 2015 modules, is somewhat underspecified when it comes to CommonJS, at the moment. This is because it is not entirely clear yet what default exports are supposed to be with respect to “legacy” module systems (such as CommonJS). It seems that the intention is that a legacy module (such as a CommonJS) would appear to an ECMAScript 2015 module as exporting a single member: the default export. For a CommonJS module, the value of the default export would be the value of exports. This intention is not clearly specified anywhere, though, and existing definitions are known to slightly conflict on the matter (e.g., what Rollup.js does compared to what Node.js would do in the future). There seems to be an emergent behavior that members of a legacy module (e.g., fields of the exports object) will also be exposed as if they were top-level exports, so that they can be imported as import { Foo } from "bar.js".

What does it all mean to you? How to choose between Namespace, Default and named imports? At present, we recommend to follow these rules of thumb:

  1. Does the documentation of the module specify how to import it with ECMAScript 2015 syntax? If yes, translate the ES syntax into @JSImport as specified above.
  2. Otherwise, is the exports value of a legacy module not an object (e.g., it is a class or a function)? If yes, use a default import with JSImport.Default.
  3. Otherwise, use a named import with a string or a namespace import with JSImport.Namespace.

Monkey patching

In JavaScript, monkey patching is a common pattern, where some top-level object or class’ prototype is meant to be extended by third-party code. This pattern is easily encoded in Scala.js’ type system with implicit conversions.

For example, in jQuery, $.fn can be extended with new methods that will be available to so-called jQuery objects, of type JQuery. Such a plugin can be declared in Scala.js with a separate trait, say JQueryGreenify, and an implicit conversions from JQuery to JQueryGreenify. The implicit conversion is implemented with a hard cast, since in effect we just want to extend the API, not actually change the value.

@js.native
trait JQueryGreenify extends JQuery {
  def greenify(): this.type = ???
}

object JQueryGreenify {
  implicit def jq2greenify(jq: JQuery): JQueryGreenify =
    jq.asInstanceOf[JQueryGreenify]
}

Recall that jq.asInstanceOf[JQueryGreenify] will be erased when mapping to JavaScript because JQueryGreenify is a JS trait. The implicit conversion is therefore a no-op and can be inlined away, which means that this pattern does not have any runtime overhead.

Reflective calls

Scala.js does not support reflective calls on any subtype of js.Any. This is mainly due to the @JSName annotation. Since we cannot statically enforce this restriction, reflective calls on subtypes of js.Any will fail at runtime. Therefore, we recommend to avoid reflective calls altogether.

What is a reflective call?

Calling a method on a structural type in Scala creates a so-called reflective call. A reflective call is a type-safe method call that uses Java reflection at runtime. The following is an example of a reflective call:

// A structural type
type T = { def foo(x: Int): String }
def print(obj: T) = obj.foo(100)
//                      ^ this is a reflective call

Any object conforming structurally to T can now be passed to print:

class A { def foo(x: Int) = s"Input: $x" }
print(new A())

Note that A does not extend T but only conforms structurally (i.e., it has a method foo with a matching signature).

The Scala compiler issues a warning for every reflective call, unless the scala.language.reflectiveCalls is imported.

Why do reflective calls not work on js.Any?

Since JavaScript is dynamic by nature, a reflective method lookup as in Java is not required for reflective calls. However, in order to generate the right method call, the call-site needs to know the exact function name in JavaScript. The Scala.js compiler generates proxy methods for that specific purpose.

However, we are unable to generate these forwarder methods on js.Any types without leaking prototype members on non-Scala.js objects. This is something which – in our opinion – we must avoid at all cost. Lack of forwarder methods combined with the fact that a JavaScript method can be arbitrarily renamed using @JSName, makes it impossible to know the method name to be called at the call-site. The reflective call can therefore not be generated.

Calling JavaScript from Scala.js with dynamic types

Because JavaScript is dynamically typed, it is not often practical, sometimes impossible, to give sensible type definitions for JavaScript APIs.

Scala.js lets you call JavaScript in a dynamically typed fashion if you want to. The basic entry point is to grab a dynamically typed reference to the global scope, with js.Dynamic.global, which is of type js.Dynamic.

You can read and write any field of a js.Dynamic, as well as call any method with any number of arguments. All input types are assumed to be of type js.Any, and all output types are assumed to be of type js.Dynamic. This means that you can assign a js.Array[A] (or even an Int, through implicit conversion) to a field of a js.Dynamic. And when you receive something, you can chain any kind of call and/or field access.

For example, this snippet taken from the Hello World example uses the dynamically typed interface to manipulate the DOM model.

val document = js.Dynamic.global.document
val playground = document.getElementById("playground")

val newP = document.createElement("p")
newP.innerHTML = "Hello world! <i>-- DOM</i>"
playground.appendChild(newP)

In this example, document, playground and newP are all inferred to be of type js.Dynamic. When calling getElementById or assigning to the field innerHTML, the String is implicitly converted to js.Any.

And since js.Dynamic inherits from js.Any, it is also valid to pass newP as a parameter to appendChild.

Remarks

Calling a js.Dynamic, like in x(a) will be treated as calling x in JavaScript, just like calling the apply method with the statically typed interface. Parameters are assumed to be of type js.Any and the result type is js.Dynamic, as for any other method.

All the JavaScript operators can be applied to js.Dynamic values.

To instantiate an object of a class with the dynamic interface, you need to obtain a js.Dynamic reference to the class value, and call the js.Dynamic.newInstance method like this:

val today = js.Dynamic.newInstance(js.Dynamic.global.Date)()

If you use the dynamic interface a lot, it is convenient to import js.Dynamic.global and/or newInstance under simple names, e.g.,

import js.Dynamic.{ global => g, newInstance => jsnew }

val today = jsnew(g.Date)()

When using js.Dynamic, you are very close to writing raw JavaScript within Scala.js, with all the warts of the language coming to haunt you. However, to get the full extent of JavaScriptish code, you can import the implicit conversions in js.DynamicImplicts. Use at your own risk!