Add different effect init for `Random`/`SecureRandom`

[geny200]

trafficstars

I would like to be able to initialize Random in a different effect, as it's in Ref and Deffered (def in[F, G, A]). Motivation:

• I can't use transformer implicits, because I want to initialize Random, for example, in ConnectionIO (type from Doobie, which has implementation of Sync for ConnectionIO).
• I don't want to use mapK because it would be incorrect and expensive for implementation (ex: provided by lib mapK for ConnectionIO uses a Dispatcher that runs the logic, which adds additional costs for lift and unlift from IO)

Below is a sample code for Ref, I would like to be able to do exactly the same for Random and SecureRandom.

  import cats.FlatMap
  import cats.effect.std.Console
  import cats.effect.{Ref, Sync}
  import cats.syntax.all.*

  class Foo[F[_]: FlatMap: Console](ref: Ref[F, Int]) {
    def print: F[Unit] = ref.get.flatMap(Console[F].println)
  }

  def foo[I[_]: Sync, F[_]: Sync: Console]: I[Foo[F]] =
    for {
      ref <- Ref.in[I, F, Int](0)
      /// Init some classes in F
    } yield new Foo(ref)

Proposal:

I looked at current implementation javaSecuritySecureRandom, and it can be divided into 2 effects. So, new method will have signature like this:

// Proposal - add new method
def javaSecuritySecureRandomIn[I[_]: Sync, F[_]: Sync](n: Int): I[SecureRandom[F]]

// Current impl
def javaSecuritySecureRandom[F[_]: Sync](n: Int): F[SecureRandom[F]]

[armanbilge]

Someone recently opened a PR for this, but I think it needs to be finished up:

• https://github.com/typelevel/cats-effect/pull/4338

---

I don't want to use mapK because it would be incorrect and expensive for implementation (ex: provided by lib mapK for ConnectionIO uses a Dispatcher that runs the logic, which adds additional costs for lift and unlift from IO)

How would it be incorrect?

Regarding performance: actually,ConnectionIO's liftK will not use a Dispatcher in this case, since the entire operation can be directly interpreted with syncStep.

• https://github.com/typelevel/doobie/pull/1906

[geny200]

be directly interpreted with syncStep

It dosn't work for effects ex (result - "Left"):

import cats.effect.unsafe.implicits.global
import cats.effect.{Async, IO}

import java.util.UUID

Async[IO].syncStep[IO, UUID](IO.randomUUID, Int.MaxValue).flatMap {
  case Left(_)  => IO.println(s"Left") // Can't run - use Dispatcher
  case Right(_) => IO.println(s"Right")
}.unsafeRunSync()

Someone recently opened a PR for this, but I think it needs to be finished up:

Thx, I didn't see it (I watched only issue). I could complete that PR, but it would be difficult to access (branch is in a personal fork). Therefore, I suggest that I open a new MR with suggested edits and comments fixses from previous PR. (before that, I'll ask author if he's ready to finish PR, or if he'll let me finish it)

[armanbilge]

It dosn't work for effects ex (result - "Left"):

Hmm, are you sure? This program prints "Right" for me.

//> using dep org.typelevel::cats-effect::3.6.1

import cats.effect.*

import java.util.UUID

object App extends IOApp.Simple:
  def run = Async[IO]
    .syncStep[IO, UUID](IO.randomUUID, Int.MaxValue)
    .flatMap {
      case Left(_)  => IO.println(s"Left") // Can't run - use Dispatcher
      case Right(_) => IO.println(s"Right")
    }

---

Therefore, I suggest that I open a new MR with suggested edits and comments fixses from previous PR.

Yes, that's the way to go, thanks!

[geny200]

Hmm, are you sure?

Yes) For me it's Left... Ok, i'll try to minimize that behavior (for me - run in sbt, jvm temurin 21, windows 11); some of my friends say that they have Left, and some have Right