ECMAScript Reflect.type proposal
Switch branches/tags
Nothing to show
Clone or download
Fetching latest commit…
Cannot retrieve the latest commit at this time.
Permalink
Type Name Latest commit message Commit time
Failed to load latest commit information.
README.md

README.md

Reflect.type

This is a stage 0, "strawman" proposal.

Problem

typeof has long been known to have perculiar behaviour. The most famously understood issue is that typeof null === "object", which consistently requires special casing when dispatching on a value's type. It was [proposed] 1 to correct typeof null to return "null", but in the interests of the Harmony/"not breaking the web" initiative, it was decided that this behaviour should remain unfixed.

For the purposes of efficient specification, the ECMAScript specification actually defines a trivial algorithm [Type(x)] 2 which has a more straightforward behaviour than typeof. In particular:

  • null has type Null; and
  • Functions have no special treatment - they are instances of Object.

Though desirable to programmers, Type(x) is not available to ECMAScript programs.

Solution

A stated goal of the Reflect module is to ["expose the essential methods that make up JavaScript's object model as defined by the spec"] 3. As such, it provides an obvious opportunity to make the specification's Type(x) algorithm available directly to ECMAScript programs.

Reflect.type(x) is proposed to accept as its sole argument any value x and return a symbol that uniquely corresponds to the type of the value, as defined by the specification. Such values would be exposed in the Reflect.types object, keyed by conventional spellings of the specification types, i.e. equivalent to:

Reflect.types = {
    undefined: @@undefinedType,
    null: @@nullType,
    boolean: @@booleanType,
    string: @@stringType,
    symbol: @@symbolType,
    number: @@numberType,
    object: @@objectType,
}

where @@ denotes well-known symbols, referenced by algorithms of the specification.

A possible implementation of Reflect.type would be equivalent to:

Reflect.type = function(x) {
    switch(typeof x) {
    case "undefined":
        return @@undefinedType;
    case "boolean":
        return @@booleanType;
    case "string":
        return @@stringType;
    case "symbol":
        return @@symbolType;
    case "number":
        return @@numberType;
    case "object":
    case "function":
    default:   // NB: `typeof` may return arbitrary strings for implementation-defined values!
        if (x === null) {
            return @@nullType;
        }
        return @@objectType;
    }
};

Notes

  • Using symbols to represent each type as opposed to short strings as typeof does may prove useful in ensuring that the domain values of typeof and Reflect.type are never accidentally mismatched.
  • Symbol descriptions can aid debugging, as the default Symbol#toString implementation includes the description.

Outstanding Issues

  • How might new, user-definable value types be exposed? Assuming they become new primitives, Reflect.type should probably return an appropriate symbol, accessible from whatever mechanism is used to define the type. Alternatively, greater symmetry could be achieved by introducing global objects Null and Undefined, that, alongside the existing globals Boolean, String, Symbol, Number, and Object, and user-defined value objects, would have a type property holding the relevant symbol, such that e.g. Reflect.type(undefined) === Undefined.type
  • Perhaps, for symmetry with Object.getPrototypeOf, it could be argued that this should be named Reflect.getTypeOf. This is both more verbose and at odds with the current specification language.