Why duck typing is allowed for classes in TypeScript

问题

Looks like in TypeScript it's absolutely fine (from the compiler perspective) to have such code:

class Vehicle {
    public run(): void { console.log('Vehicle.run'); }
}

class Task {
    public run(): void { console.log('Task.run'); }
}

function runTask(t: Task) {
    t.run();
}

runTask(new Task());
runTask(new Vehicle());

But at the same time I would expect a compilation error, because Vehicle and Task don't have anything in common.

And sane usages can be implemented via explicit interface definition:

interface Runnable {
    run(): void;
}

class Vehicle implements Runnable {
    public run(): void { console.log('Vehicle.run'); }
}

class Task implements Runnable {
    public run(): void { console.log('Task.run'); }
}

function runRunnable(r: Runnable) {
    r.run();
}

runRunnable(new Task());
runRunnable(new Vehicle());

... or a common parent object:

class Entity {
    abstract run(): void;
}

class Vehicle extends Entity {
    public run(): void { console.log('Vehicle.run'); }
}

class Task extends Entity {
    public run(): void { console.log('Task.run'); }
}

function runEntity(e: Entity) {
    e.run();
}

runEntity(new Task());
runEntity(new Vehicle());

And yes, for JavaScript it's absolutely fine to have such behaviour, because there is no classes and no compiler at all (only syntactic sugar) and duck typing is natural for the language. But TypeScript tries to introduce static checks, classes, interfaces, etc. However duck typing for class instances looks rather confusing and error-prone, in my opinion.

回答1:

This is the way structural typing works. Typescript has a structural type system to best emulate how Javscript works. Since Javascript uses duck typing, any object that defines the contract can be used in any function. Typescript just tries to validate duck typing at compile time instead of at runtime.

Your problem will however only manifest for trivial classes, as soon as you add privates, classes become incompatible even if they have the same structure:

class Vehicle {
    private x: string;
    public run(): void { console.log('Vehicle.run'); }
}

class Task {
    private x: string;
    public run(): void { console.log('Task.run'); }
}

function runTask(t: Task) {
    t.run();
}

runTask(new Task());
runTask(new Vehicle()); // Will be a compile time error

This behavior also allows you to not explicitly implement interfaces, for example you function could define the interface for the parameter inline, and any class that satisfies the contract will be compatible even if they don't explicitly implement any interface:

function runTask(t: {  run(): void }) {
    t.run();
}

runTask(new Task());
runTask(new Vehicle());

On a personal note, coming from C# this was seemed insane at first, but when it comes to extensibility this way of type checking allows for much greater flexibility, once you get used to it you will see the benefits.

回答2:

It is now possible to create nominal types with TypeScript that allow you to discriminate types by context. Please consider the following question:

Atomic type discrimination (nominal atomic types) in TypeScript

With it's example:

export type Kilos<T> = T & { readonly discriminator: unique symbol };
export type Pounds<T> = T & { readonly discriminator: unique symbol };

export interface MetricWeight {
    value: Kilos<number>
}

export interface ImperialWeight {
    value: Pounds<number>
}

const wm: MetricWeight = { value: 0 as Kilos<number> }
const wi: ImperialWeight = { value: 0 as Pounds<number> }

wm.value = wi.value;                  // Gives compiler error
wi.value = wi.value * 2;              // Gives compiler error
wm.value = wi.value * 2;              // Gives compiler error
const we: MetricWeight = { value: 0 } // Gives compiler error

来源：https://stackoverflow.com/questions/56540555/typescript-why-is-an-object-literal-a-valid-return-value-where-a-class-instance

标签

javascript

class

typescript

interface

duck-typing