Super minor text nit and personal pet peeve: “a myriad of”
“Myriad” is a number, not a collective noun. It means “ten thousand”. If swapping in “ten thousand” sounds funny, make that sound right, then go back to myriad, so “myriad things” is better usage.
(I will now retire my fussy school-marm hat for the evening. :) )
This was a great writeup. I’m glad the ball is rolling on this.

A thought on Mixin migration. I'm not 💯sure of the validity or plausibility of this, but would it make sense to have a legacy apply-mixin function for use in classes?

Here's an example I'm thinking (inspired by redux createStore)

import Route from '@ember/routing/route';
import ApplicationRouteMixin from 'ember-simple-auth/mixins/application-route-mixin';
import applyMixin from '@ember/apply-mixin';

class ApplicationRoute extends Route {
  // Stuff
};

export default applyMixin(ApplicationRoute, ApplicationRouteMixin);

Or a different API might be more like (more like redux applyMiddleware)

import Route from '@ember/routing/route';
import ApplicationRouteMixin from 'ember-simple-auth/mixins/application-route-mixin';
import createClassFromMixin from '@ember/create-class-from-mixin';

const createApplicationRoute = createClassFromMixin(ApplicationRouteMixin, ...any other mixins);

class ApplicationRoute extends Route {
  // Stuff
};

export default createApplicationRoute(ApplicationRoute);

Pre-removal of EmberObject, I'm not sure it would be necessary. We can still use class Foo extends EmberObject.extend(Mixin), so they are effectively supported. It would just add another way to do the same thing, though it may be clearer. I'm not opposed to it personally.

Post-removal of EmberObject, I don't think it should be part of Ember. It would be very difficult, if not impossible, to maximize all of the benefits of removing EmberObject and switching to native classes, and make an applyMixins which would faithfully apply the mixins and result in the exact same final class.

That said, I'm considering lots of tiny edge cases and small behaviors that are tied to Ember/Mixins. I think there's definitely room to make a library that applies mixins in mostly complete fashion, so legacy users can use mostly the same mixin definitions with relatively minor changes. I also think that there is room for a much better general mixin solution that actually leverages native classes (for inspiration, see this article that proposes one such system).

The wider JS community (or at least a vocal subset) seems to like that 'functional mixin' pattern — the pipeline operator proposal calls out how it would make applying them nicer, and the original author of that post has a stage 1 proposal for formalizing them with dedicated syntax.

@lupestro the myriad usage is fine going by the OED. Hoping Ember isn't classical history :)

@nruth And I hear rumors that prepositions are now accepted to end sentences with. (“...mere anarchy is loosed upon the world...”) I am finally old enough to experience the eternal rules of language changing in even one brief lifetime. (But of course they have ever done so...) NVM, carry on, carry on! 😀

I had a great conversation with @mike-north the other day about this, he pointed out a few concerns with this pattern going forward. I'm posting them here for discussion, but they should definitely be addressed by updates to the RFC at some point:

1. Interface-based design can be very cumbersome dependending on the way it is done. Other frameworks have implemented it, in particular Angular 2+, in a form where interfaces are scoped down to very specific functionality that any class can implement. For instance, the onInit interface tells Angular that a class should fire the ngOnInit() hook:

   import { Component, OnInit } from '@angular/core';
   
   @Component({})
   export class HeroDetailComponent implements OnInit {
     ngOnInit() {
     }
   }

   This can be very confusing, because apparently without the OnInit interface being used ngOnInit will not fire at all (part of Angular's inferencing of type information). It can also become combinatorially complex, as every lifecycle hook must become an interface, resulting in massive unparseable component definitions.

   When writing this proposal, I envisioned interfaces representing a full type, like they are today. Typescript users would actually end up with very similar code to what exists currently:

   import { Component } from '@ember/interfaces';
   
   export default class HeroDetailComponent implements Component {
     didInsertElement() {
   
     }
   }

   And native Javascript users would not extend a class at all. This would keep definitions of interfaces and types minimal and not lead to combinatorial complexity, though it could be less flexible than the alternative.

2. If base classes are not used, then there is no way to specify the placement of the call to super for a hook. This means that users cannot easily execute code both before and after a hook, which is less flexible:

   export default class MyComponent extends Component {
     didInsertElement() {
       // do something before
   
       super.didInsertElement();
   
       // do something after
     }
   }

   This can be mostly mitigated with "bookend" hooks, such as willInsertElement/didInsertElement, willDestroy/didDestroy, etc. It should also be noted that users can still maintain their own class hierarchy's and control calls to super in those, so this would only be a concern with framework code.

3. If base classes are not used, there is no easy way to intercept arguments to a hook, or the return value of a hook. The effects here have to be studied more closely, as there are times when arguments should generally not be used or modified, and times where it may make sense to modify them. It may also be possible to modify arguments in bookend hooks, though that would come with its own set of problems and concerns.

Removing the Octane label from this, as this is a more future-looking RFC that needs to be updated to reflect the current state of the world, and is unlikely to happen in the Octane timeframe.

This RFC is pretty out if date at this point. Closing it for now, and will reopen sometime post Octane when we’ve had time to update it.