Private Declarations
A proposal to add Private Declarations, allowing trusted code outside of the class lexical scope to access private state.
private #hello;
class Example {
outer #hello = 'world!';
hello() {
return this.#hello;
}
}
const ex = new Example();
console.log(ex.hello()); // => 'world!'
console.log(ex.#hello); // => 'world!'
This also allows us to bring private state to regular objects!
private #hello;
function Example() {
return {
outer #hello: 'world',
hello() {
return this.#hello;
}
}
}
const ex = Example();
console.log(ex.hello()); // => 'world!'
console.log(ex.#hello); // => 'world!'
Possible later proposals can allow sharing private declarations to friendly module.
Champions
- Justin Ridgewell (@jridgewell)
Status
Current Stage: 1
Guiding Use Cases
"Protected" state
Protected state is a valuable visibility state when implementing class hierarchies. For for instance, a hook pattern might be used to allow subclasses to override code:
// https://github.com/Polymer/lit-html/blob/1a51eb54/src/lib/parts.ts
private #createPart;
class AttributeCommitter {
//...
outer #createPart() {
return new AttributePart(this);
}
}
class PropertyCommitter extends AttributeCommitter {
outer #createPart() {
return new PropertyPart(this);
}
}
Here, AttributeCommitter
explicitly allows trusted (written in the
same source file) subclasses to override the behavior of the
#createPart
method. By default, a normal AttributePart
is returned.
But PropertyCommitter
works only on properties and would return a
PropertyPart
. All other code is free to be inherited via normal
publicly visible fields/methods from AttributeCommitter
.
Note that this does not privilege code outside the file to override
the #createPart
method, as the #createPart
private declaration is
visible only in the scope where it is declared.
Friend classes/functions
For prior art in C++, see https://en.wikipedia.org/wiki/Friend_function.
The AMP Project has a particular staged linting pattern that works well
with friendly functions that guard access to restricted code. To begin
with, statically used functions are considerably easier to lint for than
object-scoped method calls because we do not need to know the objects
type. So its much easier to determine if this is a restricted call or
just a non-restricted call that uses the same method name. Eg, it's
easier to tell that a static export registerExtendedTemplate
is
restricted vs obj.registerExtendedTemplate
.
// https://github.com/ampproject/amphtml/blob/18baa9da/src/service/template-impl.js
private #registerTemplate;
// Exported so that it may be intalled on the global and shared
// across split bundles.
export class Templates {
outer #registerTemplate() {
//...
}
}
// The code privileged to register templates with the shared class
// instance. Importing and using is statically analyzable, and must pass
// a linter.
export function registerExtendedTemplate() {
const templatesService = getService('templates');
return templatesService.#registerTemplate(...arguments);
}