Repository navigation
Preserve ExitCase in resources + fix memory leak - #535
Conversation
| def loop: F[Unit] = for { | ||
| now <- Temporal[F].monotonic | ||
| _ <- { | ||
| def loop: F[Unit] = Temporal[F].monotonic |
There was a problem hiding this comment.
thought: this change also avoids a memory leak on Scala 3 (due to better-monadic-for not eliminating the final map).
tl;dr previously it was
def loop = for {
x <- fx
_ <- (somethingSomething >> loop)
} yield ()which desugars to
fx.flatMap { x => (somethingSomething >> loop).map(_ => ()) }which allocates a Map node on every run and never cleans it up (the same goes for the top-level flatMap, and apparently the lambdas too).
Proof: run this for a minute or so.
//> using dep "org.typelevel::cats-effect:3.5.3"
import cats.effect.*
def loop: IO[Unit] = for {
_ <- IO.unit
_ <- IO.unit >> loop
} yield ()
object CatsDemo extends IOApp.Simple {
def run = loop
}
kubukoz
changed the title
| ( | ||
| m_, | ||
| toDestroy.traverse_(_._2._2).attempt.flatMap { | ||
| // In this context, we're closing the resource due to it not being used for a while - hence a Succeeded exit case. |
There was a problem hiding this comment.
thought: special case here: everywhere else we have an ExitCase to propagate, but the reaper deliberately closes resources out of thin air: I figured Succeeded makes the most sense for a resource that's no longer necessary. Canceled didn't sound right, but happy to discuss if anyone disagrees
|
It looks right. I just resolved the merge conflicts. |
2 participants
Another stop on my "allocated is an antipattern" crusade. For previous editions, see typelevel/natchez-http4s#36 and typelevel/natchez#981.