@SHSongs Thanks for this! Noticed a couple of other things as I was working on using this to implement the reloadable functionality.

We want to be able to use the generate method in higher level code like this:

def reload[A: Tag]: IO[ServiceReloader.Error, Unit] =
  ??? // uses generate internally

Right now this doesn't work because for a polymorphic type "disallowed" methods such as getClass exist and it looks like even if this ignores there are other issues with the macro implementation.

Perhaps we could package this evidence up into a trait, say ReloadableService and automatically generate instances of that trait for any data type in which the constraints you implemented are satisfied?

Here's a quick sketch. Is this what you mentioned, @adamgfraser?

trait ReloadableService[A] {
  def tag: Tag[A]
  def serviceProxy: A // macro generated proxy
  def set[R, E](acquire: ZIO[R with Scope, E, A]): ZIO[R, E, Unit] // forward to underlying `ref: ScopedRef[A]`
}

// Use `ReloadableService[A]` instead of `Tag[A]`
def register[A: ReloadableService](serviceLayer: ZLayer[Any, Any, A]): IO[ServiceReloader.Error, A]

@guersam Yes I think so!

@guersam @SHSongs All revisions done & ready for review??

@jdegoes I think this one is done as it is, but the issue @adamgfraser pointed out should be addressed.

As the purpose #8061 was to support #8077, I'd like to ensure that the reworked macro works with ZLayer#reloadable.

Possible future actions:

1. Open a new PR against zio/zio that depends on Implement ZLayer#reloadable #8077
2. Open a new PR against adamgfraser/zio so that the new commits will eventually a part of Implement ZLayer#reloadable #8077

@jdegoes @adamgfraser Which way do you prefer?

@guersam I don't think we will be able to merge #8077 until the comment above is addressed because right now generate fails to compile when called from polymorphic code like ZLayer.reloadable. I think it is fine to make those changes here or we can merge this as is and you can open another PR depending on your preference. Thanks again for your work on this! 🙏

@adamgfraser Understood that option 1 is impossible. Then I'll make another PR against your fork (option 2) to discuss more design details in relation to ServiceReloader and to make sure the implementation meets the real requirement.

@guersam Sounds good!

Opened https://github.com/adamgfraser/zio/pull/12 for further discussion.