[RFC] Introduce new pattern to avoid duplication of core services to extensions

[cramforce]

No description provided.

[cramforce]

CC @dvoytenko @erwinmombay

[dvoytenko]

I think this is fine. But I still wanted to @erwinmombay to comment on how easy it'd be to just do it at compilation time?

[erwinmombay]

@dvoytenko can you explain again what we needed to do? just creating the installer code in a separate file?

[cramforce]

Explicit might be better than implicit (through compiler magic) here.

[dvoytenko]

Normally, the service code looks like this:

class History {
  // implementation
}

/**
 * @param {!Window} window
 * @return {!History}
 */
export function historyFor(window) {
  return getService(window, 'history', () => {
    return new History(window);
  });
};

We'd like to turn this to the following when compiling an extension:

/**
 * @param {!Window} window
 * @return {!History}
 */
export function historyFor(window) {
  return getService(window, 'history');
};

In short, we throw away all implementation details and keep the serviceFor method and remove the factory from it. This way, an extension would not ever compile redundant details of a service and we have no risk of an extension accidentally creating a service before main runtime.

Malte is suggesting we do this manually, but putting separate serviceFor in a separate file. But my question is whether we can do this automatically.

[cramforce]

If we use magic the important part is to also remove all the imports!

[dvoytenko]

Well, I'm leaning toward the manual solution as well.

[erwinmombay]

right, so if we wanted to do it magically, the annotation would actually have to be on the import statement in the extension so that:

1. we can remove the import statement.
2. generate the serviceFor pattern embedded into the extension.

[dvoytenko]

"Returns service"

[cramforce]

Done

[erwinmombay]

as far as implementation, that would be really easy to do. just depends on how comfortable we are with magic annotations 😸

[dvoytenko]

LGTM