Information contained in a WebIDL file

WebIDL 可以在多个位置找到：

• 每个规范在文本中都包含 WebIDL：它是一种传达精确定义的非常方便的方法。这些描述了 API 的语法。虽然是规范参考，但我们必须记住，它们可能与实际实施不同。在 MDN 上，我们希望实用并记录 Web 平台的真正含义，而不是理想应该是什么。因此，请仔细检查实现中的内容（如果发现错误，请不要犹豫，立即提交）。
• 三个浏览器引擎使用（修改）WebIDL 作为其工具链的一部分：Gecko，Chromium / Blink 和 WebCore / WebKit。Edge 在内部使用它，但遗憾的是它们不公开（它会帮助我们很多）。
  • 对于 Gecko，所有 WebIDL 文件都分组在一个目录中：https://dxr.mozilla.org/mozilla-central/source/dom/webidl/
  • 他们的延伸是.webidl。*.idlGecko 源代码树中还有其他 文件，但它们不是 WebIDL，因此您可以忽略它们。旧版本的 Gecko 有一些 WebIDL 分散在某些地方，甚至可能使用 Mozilla 的 IDL 而不是 WebIDL 来描述一些 Web 界面，但这在任何最近的 Gecko 代码中都不会成为问题。
  • 对于 Chromium，它们分散在源代码中，所以你需要多花点功夫：例如http://src.chromium.org/viewvc/blink/trunk/Source/modules/mediastream/MediaStream.idl
  • 同样，WebCore 正如预期的那样，因为 Chromium 是从它分叉出来的，例如https://github.com/WebKit/webkit/blob/master/Source/WebCore/html/DOMTokenList.idl

WebIDL is defined in its specification. But it has been designed to be extended to convey more information, and browser vendors have done so:

• For Gecko, MDN has the documentation of its dialectal WebIDL.
• For Chromium, Google also created a document to describe its extensions.
• For WebCore, Apple also made available a page for its dialect.

This section explains the WebIDL syntax that describes overall API features.

The interface name is the string that appears after the keyword interface and before the next opening bracket ('{') or colon (':').

interface URL {};

Each WebIDL interface, being a true interface or a mixin, has its own page in the documentation, listing every constructor, property and method defined for it.

The parent, if any, of a given interface is defined after the interface name, following a colon (':'). There can be only one parent per interface.

interface HTMLMediaElement : HTMLElement {…}

The inheritance chain is listed automatically in the sidebar (using the {{APIRef}} macro). It can also be added as an SVG image via the macro {{InheritanceDiagram}}.

Some properties or methods are available to several interfaces. To prevent redefinition they are defined in special WebIDL interfaces called mixins. In the WebIDL, they are prefixed using the [NoInterfaceObject] annotation. The name of a mixin, Body in the following example, doesn't appear in JavaScript.

[NoInterfaceObject]
  interface Body {…}

For documentation purposes, we create a mixin page, with the same structure as an interface page. As they are not true interfaces, the word interface is not used — mixin is used instead.

Mixin methods and properties are listed in the same way as regular methods and properties:

• They have their own pages, prefixed with the mixin name. E.g Body.bodyUsed (en-US) or Body.blob() (en-US).
• They are listed on the interface page when the interface implements the mixin. Unlike regular properties and methods, they are prefixed with the mixin name and not the interface name. You can see Body properties and methods listed on both Request and Response interfaces, as they both implement the Body mixin.

Mixins implemented on an interface are defined using the implements keyword.

Request implements Body;
Response implements Body;

Availability in Web workers (of any type) and on the Window scope is defined using an annotation: [Exposed=(Window,Worker)]. The annotation applies to the partial interface it is listed with. If no annotation is available, the default value is Window.

[Exposed=(Window,Worker)]
interface Performance {
    [DependsOn=DeviceState, Affects=Nothing]
    DOMHighResTimeStamp now();
};

[Exposed=Window]
partial interface Performance {
    [Constant]
    readonly attribute PerformanceTiming timing;
    [Constant]
    readonly attribute PerformanceNavigation navigation;

    jsonifier;
};

In this case Performance.now() is available on the Window scope and to any worker, while Performance.timing, Performance.navigation and Performance.toJSON() are not available to Web workers.

The most common values for the [Exposed] are:

Window

The partial interface is available to the Window global scope.

Worker

The partial interface is available to any kind of worker, that is if the global scope is a descendant of WorkerGlobalScope — DedicatedWorkerGlobalScope, SharedWorkerGlobalScope (en-US), or ServiceWorkerGlobalScope (It also is available to ChromeWorker, but we don't document this as they are not visible on the Web and are internal to Firefox.)

DedicatedWorker

The partial interface is available to the DedicatedWorkerGlobalScope only.``

SharedWorker

The partial interface is available to the SharedWorkerGlobalScope (en-US) only.

ServiceWorker

The partial interface is available to the ServiceWorkerGlobalScope only.

Another value is possible, like System, but this has a special meaning and doesn't need to be documented.

Note that these possible values are themselves defined in WebIDL files. Interfaces may have a [Global=xyz] annotation. It means that when an object of this type is used as a global scope, any interface, property or method, with xyz as a value of [Exposed] is available.

[Global=(Worker,DedicatedWorker), Exposed=DedicatedWorker]
interface DedicatedWorkerGlobalScope : WorkerGlobalScope {…}

Here, it is defined that when the global scope is of type DedicatedWorkerGlobalScope, that is if we are in a dedicated worker, any interface, property or method exposed – using the [Exposed] annotation – to Worker or DedicatedWorker is available.

Even the primary global is defined in WebIDL. The primary global is the value of an [Exposed] annotation when not present. This is defined using the [PrimaryGlobal] annotation and is present on Window:

[PrimaryGlobal, NeedResolve]
/*sealed*/ interface Window : EventTarget {…}

In Gecko, the availability of a partial interface, including its constructor, properties and methods may be controlled by a preference (usually called a "pref"). This is marked in the WebIDL too.

[Pref="media.webspeech.synth.enabled"]
interface SpeechSynthesis {
    readonly attribute boolean pending;
    readonly attribute boolean speaking;
    readonly attribute boolean paused;
};

Here media.webspeech.synth.enabled controls the SpeechSynthesis interface and its properties (the full listing has more than 3.)

You can recognize the definition of a property by the presence of the attribute keyword.

readonly attribute MediaError? error;

In the above example the name of the property is error; in the docs we will refer to it as HTMLMediaElement.error as it belongs to the HTMLMediaElement interface. Linking to the page is either done with the interface prefix using {{domxref('HTMLMediaElement.error')}} or without the prefix using {{domxref('HTMLMediaElement.error', 'error')}} when the context is obvious and unambiguous.

readonly attribute MediaError? error;

The property value is an object of type MediaError. The question mark ('?') indicates that it can take a value of null, and the documentation must explain when this may occur. If no question mark is present, the error property can't be null.

readonly attribute MediaError? error;

If the keyword readonly is present, the property can't be modified. It must be marked as read-only:

• In the interface, by adding the {{ReadOnlyInline}} macro next to its definition term.
• In the first sentence of its own page, by starting the description with: The read-only HTMLMediaElement.error property…
• By adding the Read-only tag to its own page.
• By starting its description in the interface page with Returns…

[SetterThrows]
attribute DOMString src;

In some cases, like when some values are illegal, setting a new value can lead to an exception being raised. This is marked using the [SetterThrows] annotation. When this happens, the Syntax section of the property page must have an Exceptions subsection. The list of exceptions and the conditions to have them thrown are listed, as textual information, in the specification of that API.

Note that some exceptions are not explicitly marked but are defined by the JavaScript bindings. Trying to set an illegal enumerated value (mapped to a JavaScript String) raises a TypeError exception. This must be documented, but is only implicitly marked in the WebIDL document.

It is uncommon to have getters throwing exceptions, though it happens in a few cases. In this case the [GetterThrows] annotation is used. Here also, the Syntax section of the property page must have an Exceptions subsection.

partial interface Blob {
  [GetterThrows]
  readonly attribute unsigned long long size;
};

When the semantics of Webidl is not followed, an exception is often thrown, even without [SetterThrows] or [GetterThrows] set. For example, in strict mode, if we try to set a read-only property to a new value, that is to call its implicit setter, a read-only property will throw in strict mode.

Mostly for compatibility purpose, this behavior is sometimes annoying. To prevent this by creating a no-op setter (that is by silently ignoring any attempt to set the property to a new value), the [LenientSetter] annotation can be used.

partial interface Document {
  [LenientSetter]
  readonly attribute boolean fullscreen;
  [LenientSetter]
  readonly attribute boolean fullscreenEnabled;
};

In these cases, an extra sentence is added to the description of the property. E.g

Although this property is read-only, it will not throw if it is modified (even in strict mode); the setter is a no-operation and it will be ignored.

The return value of a property can be either a copy of an internal object, a newly created synthetic object, or a reference to an internal object.

Basic objects with types like String (being an IDL DOMString, or other), Number (being an IDL byte, octet, unsigned int, or other), and Boolean are always copied and nothing special has to be noted about them (it is natural behavior expected by a JavaScript developer.)

For interface objects, the default is to return a reference to the internal object. This has to be mentioned both in the short description in the interface page, and in the description in the specific sub-pages.

Sometimes an API must return a new object, or a copy of an internal one. This case is indicated in the WebIDL using the [NewObject] annotation.

[NewObject]
readonly attribute TimeRanges buffered;

In this case, each call to buffered returns a different object: changing it will not change the internal value, and a change in the internal value will not affect each object instance. In the documentation, we will mark it by using the adjective new next to object:

The HTMLMediaElement.buffered read-only property returns a new {{domxref("TimeRanges")}} object that…

and

{{domxref("HTMLMediaElement.buffered")}}{{readonlyinline}}

Returns a new {{domxref("TimeRanges")}} object that …

In the case of a reference to a collection object (like HTMLCollection, HTMLFormElementsCollection, or HTMLOptionsCollection, always without [NewObject]), we make it explicit that changes to the underlying object will be available via the returned reference. To mark this, we qualify the collection as a live HTMLCollection (or HTMLFormElementsCollections, or HTMLOptionsCollection), both in the interface description and in the subpage.

E.g.

{{domxref("HTMLFormElement.elements")}}{{readonlyinline}}

Returns a live {{domxref("HTMLFormControlsCollection")}} containing…

Individual property availability in workers is also found in the WebIDL. For a property, the default is the same availability as the interface (that is available to Window context only if nothing special is marked) or as the partial interface it is defined in.

For documentation, the subpage must contain a sentence indicating if it is available or not in Web workers, right before the "Syntax" section.

In Gecko, the availability of some properties may be controlled by a preference. This is marked in the WebIDL too.

[Pref="media.webvtt.enabled"]
readonly attribute TextTrackList? textTracks;

Here media.webvtt.enabled controls the textTracks property.

You can recognize the definition of a method by the presence of parentheses after the name.

DOMString canPlayType(DOMString type);

The name of the method is canPlayType, and we will refer to it as HTMLMediaElement.canPlayType() (with the parentheses that indicate that it is a method) in the docs, as it belongs to the HTMLMediaElement interface. Linking to the page is either done with the interface prefix using {{domxref('HTMLMediaElement.canPlayType()')}}, or without the prefix using {{domxref('HTMLMediaElement.canPlayType', 'canPlayType()')}} when the context is obvious and unambiguous. The parentheses should always be included.

TextTrack addTextTrack(TextTrackKind kind,
                        optional DOMString label = "",
                        optional DOMString language = "");

The parameters of a method are listed in the Syntax section of the method sub-page. They are listed in the WebIDL in order, between the parenthesis, as a comma-separated list. Each parameter has a name (indicated above) and a type (e.g. a '?' means that the null value is valid.) If marked optional, the parameter is optional to include in a method call and must have the {{OptionalInline}} flag included when it is listed in the Syntax section. The parameter's default value is listed after the equality sign ('=').

DOMString canPlayType(DOMString type);

The return value type is indicated first inside the parentheses — in the above case the value is an object of type DOMString. if followed by a question mark ('?'), a value of null can be returned too, and the documentation must explain when this may happen. If no question mark is present, like here, the return value can't be null.

The keyword void means that there is no return value. It is not a return value type. If the WebIDL entry reads void, the Return value section in the docs should contain only a simple None.

[Throws]
void fastSeek(double time);

Some methods can throw exceptions. This is marked using the [Throws] annotation. When this happens, the Syntax section of the method page must have an Exceptions subsection. The list of exceptions and the conditions to have them thrown are listed, as textual information, in the specification of that API.

Note that some exceptions are not explicitly marked but are defined by the JavaScript bindings. Trying to set an illegal enumerated value (mapped to a JavaScript String) as a parameter will raise a TypeError exception. This must be documented, but it is only implicitly marked in the WebIDL document.

Have a look at one of these Exceptions sections.

Individual method availability in workers is also found in the WebIDL. For a method, the default is the same availability as the interface (that is available to Window context only if nothing special is marked) or as the partial interface it is defined it.

For the documentation, the sub-page must contain a sentence indicating if it is available in Web workers, right before the Syntax section.

In Gecko, the availability of some properties may be controlled by a preference. This is marked in the WebIDL too.

[Pref="media.webvtt.enabled"]
TextTrack addTextTrack(TextTrackKind kind,
                      optional DOMString label = "",
                      optional DOMString language = "");

Here media.webvtt.enabled controls the addTextTrack() method.

Some methods are not listed as regular methods in WebIDL but instead as special keywords, which translate to specific standard JavaScript methods.

A stringifier is mapped to toString() and defined as:

stringifier;

The toString() method is listed just like any other method of the interface and has its own sub-page (E.g. Range.toString() (en-US))

A jsonifier is mapped to toJSON() and defined as:

jsonifier; // Gecko version
serializer; // Standard version

The toJSON() method is listed just like any other method of the interface and has its own sub-page (E.g. Performance.toJSON())

An interface may be defined as iterable, meaning that it will have the following methods: entries(), keys(), values() and forEach(). They also supports the use of for...of on an object implementing this interface.

There are two kinds of iteration possible: the value iterator and the pair iterator.

Value iterator

iterable<valueType>

The iterator will iterate over values of type valueType. The generated methods will be:

• entries(), which returns an iterator on the indexes (that are unsigned long).
• values(), which returns an iterator on the values.
• keys(), which returns an iterator on the keys, that are its indexes (that are unsigned long). In the case of value iterators, keys() and entries() are identical.
• forEach(), which returns an iterator on the keys that calls a given callback function one for each entry in the list.

Such an iterator allows to use the syntax for (var p in object) as a shorthand of for (var p in object.entries()). We add a sentence about it in the interface description.

Pair iterator

iterable<keyType, valueType>

The iterator will iterate over values of type valueType, with keys of type keyType. The generated methods will be:

• entries() that returns an iterator on the indexes (of type keyType). E.g. FormData.entries()
• values() that returns an iterator on the values. E.g. FormData.values()
• keys() that returns an iterator on the keys. E.g. FormData.keys()
• Once Firefox bug 1216751 lands, forEach().

Such an iterator allows to use the syntax for (var p in object) as a shorthand of for (var p in object.entries()). We add a sentence about it in the interface description. E.g. FormData.

An interface may be defined as set-like, meaning that it represents an ordered set of values will have the following methods: entries(), keys(), values(), forEach(), and has() (it also has the size property). They also supports the use of for...of on an object implementing this interface. The set-like can be prefixed readonly or not. If not read-only, the methods to modify the set are also implemented: add(), clear(), and delete().

setlike<valueType>

The generated properties will be:

• entries() that returns an iterator on the indexes. E.g. FontFaceSet.entries() (en-US).
• values() that returns an iterator on the values. E.g. FontFaceSet.values() (en-US).
• keys() that returns an iterator on the keys. E.g. FontFaceSet.keys() (en-US),
• forEach().

In the case, the set-like declaration is not prefixed by read-only, the following methods are also generated:

• add() that adds an entry. E.g. FontFaceSet.add() (en-US).
• clear() that empty the set-like. E.g. FontFaceSet.clear() (en-US).
• delete() that remove an entry. E.g. FontFaceSet.delete() (en-US).

Such an set interface also allows to use the syntax for (var p in object) as a shorthand of for (var p in object.entries()). We add a sentence about it in the interface description. E.g. FontFaceSet.

Constructors are a little bit hidden in WebIDL: they are listed as annotations of the main interface.

This is the most common case for constructors. The constructor of a given interface A, can be used as a = new A(parameters);

[Constructor, Func="MessageChannel::Enabled",
Exposed=(Window,Worker)]
  interface MessageChannel {…};

A constructor with the same interface is defined using the Constructor annotation on the interface. There can be parenthesis and a list of parameters or not (like in the above example.) We document all the unnamed constructors on a sub-page — for example the above is given the slug Web/API/MessageChannel/MessageChannel and the title MessageChannel().

Another example of an unnamed constructor, with parameters:

[Constructor(DOMString type, optional MessageEventInit eventInitDict),
  Exposed=(Window,Worker,System)]
    interface MessageEvent : Event {…};

There can also be several unnamed constructors, differing by their parameter lists. All syntax is documented in one single sub-page.

[Constructor(DOMString url, URL base),
  Constructor(DOMString url, optional DOMString base),
  Exposed=(Window,Worker)]
    interface URL {};

[NamedConstructor=Image(optional unsigned long width, optional unsigned long height)]
    interface HTMLImageElement : HTMLElement {…

A named constructor is a constructor that has a different name than that of its interface. For examplenew Image(…) creates a new HTMLImageElement object. They are defined in the WebIDL using the NamedConstructor annotation on the interface, followed by the name of the constructor after the equality sign ('=') and the parameter inside the parenthesis, in the same format as you'll see for methods.

There can be several named constructors for a specific interface, but this is extremely rare; in such a case we include one sub-page per name.

Constructors have the same availability as the interface, or partial interface, they are defined on. The sub-page provides this information in the same way as for a method.

Constructors are controlled by the same preference as the interface, or partial interface, they are defined on. The sub-page provides this information in the same way as for a method.