Merge std and core?

[armanbilge]

The idea was to separate cats-effect core dependency, but in practice everything from fs2 up brings it in anyway. So it may be over-modularization.

There's an argument to just merge in kernel as well, who does concurrency without CE.

[bplommer]

I think I'd be more inclined to merge kernel and std, and remove the dependency on core - that way core and std are completely independent. core could have its artifact renamed to oxidized-core with a new oxidized artifact that provides both.

[armanbilge]

👍 to merging kernel and std since they both only need the CE kernel dependency, we can provide the ConcurrentStateful for Ref in a more convenient place.

I'm not sure why core independent of kernel is a big win. Unless you don't like ConcurrentStateful and/or it's not ready for primetime in which case I'm happy to punt that off :)

[bplommer]

I'm not sure why core independent of kernel is a big win.

Dunno, maybe it isn't 🤷

Unless you don't like ConcurrentStateful and/or it's not ready for primetime in which case I'm happy to punt that off :)

Not that I don't like it but it's a lot more experimental than core. I like the idea of having a module that's just cats-mtl instances.

[armanbilge]

it's a lot more experimental than core

Yes completely agree. Although anybody who cares about bincompatibility can should just copy-paste these instances wherever they want them :)

Separate modules doesn't really protect core from kernel breaking bincompat, because it manifests as a breaking version bump for the whole repo anyway. This is why the experimental CE CPS stuff is in its own repo with its own versioning.

Modules really only help with scoping dependencies I think.

[bpholt]

WDYT about moving the IO instances to a cats-mtl submodule (if they'll have it)? It'd be nice to have the IOLocal-based Local instance in a Typelevel repo if it's going to be more widespread due to something like Natchez or otel4s.

[armanbilge]

WDYT about moving the IO instances to a cats-mtl submodule (if they'll have it)?

@bpholt yeah, the reason this project exists is because that's not possible 😅 CE testkit has a dependency on cats-mtl, so cats-mtl can't depend on CE without creating a cycle. Cycles again!

However, I am in favor of moving this somewhere more visible. And also generally reviving it :)

[bpholt]

Ugh, of course 😂😭

[armanbilge]

Btw, something that could be worth pursuing is moving ConcurrentStateful into MTL. It's just an interface, and it should really be the parent of Stateful I believe.

oxidized/kernel/src/main/scala/oxidized/ConcurrentStateful.scala

Lines 19 to 30 in 412be9c

	import cats.Functor
	import cats.mtl.Stateful
	trait ConcurrentStateful[F[_], S] extends Serializable {
	def functor: Functor[F]
	def get: F[S]
	def set(s: S): F[Unit]
	def modify(f: S => S): F[Unit]
	def inspect[A](f: S => A): F[A] = functor.map(get)(f)
	// TODO To be useful will probably need more Ref-like methods
	}