webidl2

A WebIDL Parser

npm install webidl2
4 downloads in the last week
18 downloads in the last month

Purpose

This is a parser for the WebIDL language. If you don't know what that is, then you probably don't need it. It is meant to be used both in Node and in the browser (the parser likely works in other JS environments, but not the test suite).

What of v1?

There was a previous incarnation of this project. I had written it in the most quick and dirty manner that was handy because I required it as a dependency in an experiment. As these things tend to happen, some people started using that, which then had to be maintained. But since it was not built on solid foundations, it was painful to keep up to date with the specification, which is a bit of a moving target.

So I started from scratch. Compared to the previous version (which used a parser generator) this one is about 6x less code (which translates to 4x smaller minified or 2x smaller minizipped) and 4x faster. The test suite is reasonably complete (95% coverage), much more than previously. This version is up to date with WebIDL, rather than a couple years' behind. It also has far better error reporting.

The AST you get from parsing is very similar to the one you got in v1, but some adjustments have been made in order to be more systematic, and to map better to what's actually in the spec now. If you used v1, you will need to tweak your code but the result ought to be simpler and you ought to be able to be a fair bit less defensive against irregularities in the way information is represented.

Installation

Just the usual. For Node:

npm install webidl2

In the browser:

<script src='webidl2.js'></script>

Documentation

The API to WebIDL2 is trivial: you parse a string of WebIDL and it returns a syntax tree.

Parsing

In Node, that happens with:

var WebIDL2 = require("webidl2");
var tree = WebIDL2.parse("string of WebIDL");

In the browser:

<script src='webidl2.js'></script>
<script>
  var tree = WebIDL2.parse("string of WebIDL");
</script>

Errors

When there is a syntax error in the WebIDL, it throws an exception object with the following properties:

  • message: the error message
  • line: the line at which the error occurred.
  • input: a short peek at the text at the point where the error happened
  • tokens: the five tokens at the point of error, as understood by the tokeniser (this is the same content as input, but seen from the tokeniser's point of view)

The exception also has a toString() method that hopefully should produce a decent error message.

AST (Abstract Syntax Tree)

The parse() method returns a tree object representing the parse tree of the IDL. Comment and white space are not represented in the AST.

The root of this object is always an array of definitions (where definitions are any of interfaces, exceptions, callbacks, etc. — anything that can occur at the root of the IDL).

IDL Type

This structure is used in many other places (operation return types, argument types, etc.). It captures a WebIDL type with a number of options. Types look like this and are typically attached to a field called idlType:

{
    "sequence": false,
    "nullable": false,
    "array": false,
    "union": false,
    "idlType": "void"
}

Where the fields are as follows:

  • sequence: Boolean indicating whether this is a sequence or not.
  • nullable: Boolean indicating whether this is nullable or not.
  • array: Either false to indicate that it is not an array, or a number for the level of array nesting.
  • union: Boolean indicating whether this is a union type or not.
  • idlType: Can be different things depending on context. In most cases, this will just be a string with the type name. But the reason this field isn't called "typeName" is because it can take more complex values. If the type is a union, then this contains an array of the types it unites. If it is a sequence, it contains an IDL type description for the type in the sequence.

Interactions between nullable and array

A more complex data model for our AST would likely represent Foo[][][] as a series of nested types four levels deep with three anonymous array types eventually containing a Foo type. But experience shows that such structures are cumbersome to use, and so we have a simpler model in which the depth of the array is specified with the array field.

This is all fine and well, and in the vast majority of cases is actually simpler. But it does run afoul of cases in which it is necessary to distinguish between Foo[][][]?, Foo?[][][], Foo[][]?[], or even Foo?[]?[]?[]?.

For this, when a type is an array type an additional nullableArray field is made available that captures which of the arrays contain nullable elements. It contains booleans that are true if the given array depth contains nullable elements, and false otherwise (mapping that to the syntax, and item is true if there is a ? preceding the []). These examples ought to clarify the model:

Foo[][][]?
    -> nullable: true
    -> nullableArray: [false, false, false]
Foo?[][][]
    -> nullable: false
    -> nullableArray: [true, false, false]
Foo[][]?[]
    -> nullable: false
    -> nullableArray: [false, false, true]
Foo?[]?[]?[]?
    -> nullable: true
    -> nullableArray: [true, true, true]

Of particular importance, please note that the overall type is only nullable if there is a ? at the end.

Interface

Interfaces look like this:

{
    "type": "interface",
    "name": "Animal",
    "partial": false,
    "members": [...],
    "inheritance": null,
    "extAttrs": [...]
},
{
    "type": "interface",
    "name": "Human",
    "partial": false,
    "members": [...],
    "inheritance": "Animal",
    "extAttrs": [...]
}

The fields are as follows:

  • type: Always "interface".
  • name: The name of the interface
  • partial: A boolean indicating whether it's a partial interface.
  • members: An array of interface members (attributes, operations, etc.). Empty if there are none.
  • inheritance: A string giving the name of an interface this one inherits from, null otherwise. NOTE: In v1 this was an array, but multiple inheritance is no longer supported so this didn't make sense.
  • extAttrs: A list of extended attributes.

Callback Interfaces

These are captured by the same structure as Interfaces except that their type field is "callback interface".

Callback

A callback looks like this:

{ "type": "callback", "name": "AsyncOperationCallback", "idlType": { "sequence": false, "nullable": false, "array": false, "union": false, "idlType": "void" }, "arguments": [...], "extAttrs": [] }

The fields are as follows:

  • type: Always "callback".
  • name: The name of the callback.
  • idlType: An IDL Type describing what the callback returns.
  • arguments: A list of arguments, as in function paramters.
  • extAttrs: A list of extended attributes.

Dictionary

A dictionary looks like this:

{
    "type": "dictionary",
    "name": "PaintOptions",
    "partial": false,
    "members": [
        {
            "type": "field",
            "name": "fillPattern",
            "idlType": {
                "sequence": false,
                "nullable": true,
                "array": false,
                "union": false,
                "idlType": "DOMString"
            },
            "extAttrs": [],
            "default": {
                "type": "string",
                "value": "black"
            }
        }
    ],
    "inheritance": null,
    "extAttrs": []
}

The fields are as follows:

  • type: Always "dictionary".
  • name: The dictionary name.
  • partial: Boolean indicating whether it's a partial dictionary.
  • members: An array of members (see below).
  • inheritance: A string indicating which dictionary is being inherited from, null otherwise.
  • extAttrs: A list of extended attributes.

All the members are fields as follows:

Exception

An exception looks like this:

{
    "type": "exception",
    "name": "HierarchyRequestError",
    "members": [
        {
            "type": "field",
            "name": "code",
            "idlType": {
                "sequence": false,
                "nullable": false,
                "array": false,
                "union": false,
                "idlType": "unsigned short"
            },
            "extAttrs": []
        }
    ],
    "inheritance": "DOMException",
    "extAttrs": []
}

The fields are as follows:

  • type: Always "exception".
  • name: The exception name.
  • members: An array of members (constants or fields, where fields are described below).
  • inheritance: A string indicating which exception is being inherited from, null otherwise.
  • extAttrs: A list of extended attributes.

Members that aren't constants have the following fields:

  • type: Always "field".
  • name: The field's name.
  • idlType: An IDL Type describing what field's type.
  • extAttrs: A list of extended attributes.

Enum

An enum looks like this:

{
    "type": "enum",
    "name": "MealType",
    "values": [
        "rice",
        "noodles",
        "other"
    ],
    "extAttrs": []
}

The fields are as follows:

  • type: Always "enum".
  • name: The enum's name.
  • value: An array of values (strings).
  • extAttrs: A list of extended attributes.

Typedef

A typedef looks like this:

{
    "type": "typedef",
    "typeExtAttrs": [],
    "idlType": {
        "sequence": true,
        "nullable": false,
        "array": false,
        "union": false,
        "idlType": {
            "sequence": false,
            "nullable": false,
            "array": false,
            "union": false,
            "idlType": "Point"
        }
    },
    "name": "PointSequence",
    "extAttrs": []
}

The fields are as follows:

  • type: Always "typedef".
  • name: The typedef's name.
  • idlType: An IDL Type describing what typedef's type.
  • extAttrs: A list of extended attributes.
  • typeExtAttrs: A list of extended attributes that apply to the type rather than to the typedef as a whole.

Implements

An implements definition looks like this:

{
    "type": "implements",
    "target": "Node",
    "implements": "EventTarget",
    "extAttrs": []
}

The fields are as follows:

  • type: Always "implements".
  • target: The interface that implements another.
  • implements: The interface that is being implemented by the target.
  • extAttrs: A list of extended attributes.

Operation Member

An operation looks like this:

{
    "type": "operation",
    "getter": false,
    "setter": false,
    "creator": false,
    "deleter": false,
    "legacycaller": false,
    "static": false,
    "stringifier": false,
    "idlType": {
        "sequence": false,
        "nullable": false,
        "array": false,
        "union": false,
        "idlType": "void"
    },
    "name": "intersection",
    "arguments": [
        {
            "optional": false,
            "variadic": true,
            "extAttrs": [],
            "idlType": {
                "sequence": false,
                "nullable": false,
                "array": false,
                "union": false,
                "idlType": "long"
            },
            "name": "ints"
        }
    ],
    "extAttrs": []
}

The fields are as follows:

  • type: Always "operation".
  • getter: True if a getter operation.
  • setter: True if a setter operation.
  • creator: True if a creator operation.
  • deleter: True if a deleter operation.
  • legacycaller: True if a legacycaller operation.
  • static: True if a static operation.
  • stringifier: True if a stringifier operation.
  • idlType: An IDL Type of what the operation returns. If a stringifier, may be absent.
  • name: The name of the operation. If a stringifier, may be null.
  • arguments: An array of arguments for the operation.
  • extAttrs: A list of extended attributes.

Attribute Member

An attribute member looks like this:

{
    "type": "attribute",
    "static": false,
    "stringifier": false,
    "inherit": false,
    "readonly": false,
    "idlType": {
        "sequence": false,
        "nullable": false,
        "array": false,
        "union": false,
        "idlType": "RegExp"
    },
    "name": "regexp",
    "extAttrs": []
}

The fields are as follows:

  • type: Always "attribute".
  • name: The attribute's name.
  • static: True if it's a static attribute.
  • stringifier: True if it's a stringifier attribute.
  • inherit: True if it's an inherit attribute.
  • readonly: True if it's a read-only attribute.
  • idlType: An IDL Type for the attribute.
  • extAttrs: A list of extended attributes.

Constant Member

A constant member looks like this:

{
    "type": "const",
    "nullable": false,
    "idlType": "boolean",
    "name": "DEBUG",
    "value": {
        "type": "boolean",
        "value": false
    },
    "extAttrs": []
}

The fields are as follows:

  • type: Always "const".
  • nullable: Whether its type is nullable.
  • idlType: The type of the constant (a simple type, the type name).
  • name: The name of the constant.
  • value: The constant value as described by Const Values
  • extAttrs: A list of extended attributes.

Serializer Member

Serializers come in many shapes, which are best understood by looking at the examples below that map the IDL to the produced AST.

// serializer;
{
    "type": "serializer",
    "extAttrs": []
}

// serializer DOMString serialize();
{
    "type": "serializer",
    "idlType": {
        "sequence": false,
        "nullable": false,
        "array": false,
        "union": false,
        "idlType": "DOMString"
    },
    "operation": {
        "name": "serialize",
        "arguments": []
    },
    "extAttrs": []
}

// serializer = { from, to, amount, description };
{
    "type": "serializer",
    "patternMap": true,
    "names": [
        "from",
        "to",
        "amount",
        "description"
    ],
    "extAttrs": []
}

// serializer = number;
{
    "type": "serializer",
    "name": "number",
    "extAttrs": []
}

// serializer = [ name, number ];
{
    "type": "serializer",
    "patternList": true,
    "names": [
        "name",
        "number"
    ],
    "extAttrs": []
}

The common fields are as follows:

For a simple serializer, that's all there is. If the serializer is an operation, it will have:

  • idlType: An IDL Type describing what the serializer returns.
  • operation: An object with the following fields:
    • name: The name of the operation.
    • arguments: An array of arguments for the operation.

If the serializer is a pattern map:

  • patternMap: Always true.
  • names: An array of names in the pattern map.

If the serializer is a pattern list:

  • patternList: Always true.
  • names: An array of names in the pattern list.

Finally, if the serializer is a named serializer:

  • name: The serializer's name.

Iterator Member

Iterator members look like this

{
    "type": "iterator",
    "getter": false,
    "setter": false,
    "creator": false,
    "deleter": false,
    "legacycaller": false,
    "static": false,
    "stringifier": false,
    "idlType": {
        "sequence": false,
        "nullable": false,
        "array": false,
        "union": false,
        "idlType": "Session2"
    },
    "iteratorObject": "SessionIterator",
    "extAttrs": []
}
  • type: Always "iterator".
  • iteratorObject: The string on the right-hand side; absent if there isn't one.
  • the rest: same as on operations.

Arguments

The arguments (e.g. for an operation) look like this:

"arguments": [
    {
        "optional": false,
        "variadic": true,
        "extAttrs": [],
        "idlType": {
            "sequence": false,
            "nullable": false,
            "array": false,
            "union": false,
            "idlType": "long"
        },
        "name": "ints"
    }
]

The fields are as follows:

  • optional: True if the argument is optional.
  • variadic: True if the argument is variadic.
  • idlType: An IDL Type describing the type of the argument.
  • name: The argument's name.
  • extAttrs: A list of extended attributes.

Extended Attributes

Extended attributes are arrays of items that look like this:

"extAttrs": [
    {
        "name": "TreatNullAs",
        "arguments": null,
        "rhs": {
            "type": "identifier",
            "value": "EmptyString"
        }
    }
]

The fields are as follows:

  • name: The extended attribute's name.
  • arguments: If the extended attribute takes arguments (e.g. [Foo()]) or if its right-hand side does (e.g. [NamedConstructor=Name(DOMString blah)]) they are listed here. Note that an empty arguments list will produce an empty array, whereas the lack thereof will yield a null. If there is an rhs field then they are the right-hand side's arguments, otherwise they apply to the extended attribute directly.
  • rhs: If there is a right-hand side, this will capture its type (always "identifier" in practice, though it may be extended in the future) and its value.

Default and Const Values

Dictionary fields and operation arguments can take default values, and constants take values, all of which have the following fields:

  • type: One of string, number, boolean, null, Infinity, or NaN.

For string, number, and boolean:

  • value: The value of the given type.

For Infinity:

  • negative: Boolean indicating whether this is negative Infinity or not.

Testing

In order to run the tests you need to ensure that the widlproc submodule inside test is initialised and up to date:

git submodule init
git submodule update
git pull origin master (in the submodule, once in a while)

Running

The test runs with mocha and expect.js. Normally, running mocha in the root directory should be enough once you're set up.

Coverage

Current test coverage, as documented in coverage.html, is 95%. You can run your own coverage analysis with:

jscoverage lib lib-cov

That will create the lib-cov directory with instrumented code; the test suite knows to use that if needed. You can then run the tests with:

JSCOV=1 mocha --reporter html-cov > coverage.html

Note that I've been getting weirdly overescaped results from the html-cov reporter, so you might wish to try this instead:

JSCOV=1 mocha  --reporter html-cov | sed "s/&lt;/</g" | sed "s/&gt;/>/g" | sed "s/&quot;/\"/g" > coverage.html

Browser tests

In order to test in the browser, get inside test/web and run make-web-tests.js. This will generate a browser-tests.html file that you can open in a browser. As of this writing tests pass in the latest Firefox, Chrome, Opera, and Safari. Testing on IE and older versions will happen progressively.

TODO

  • add some tests to address coverage limitations
  • add a push API for processors that need to process things like comments
npm loves you