maybe this belongs in
cats instead.
that's better than what I was going to call it,
unattempt
Here it is with some error handling. I made a design choice which I think is debatable:
import cats.effect.concurrent.{Deferred, Ref}
import cats.effect.{Concurrent, Resource, Timer}
import cats.implicits._
import fs2.Stream
import scala.concurrent.duration._
trait PeriodicMonitor[F[_], A] {
def get: F[A]
}
object PeriodicMonitor {
def create[F[_]: Concurrent: Timer, A](
action: F[A],
t: FiniteDuration
): Resource[F, PeriodicMonitor[F, A]] = {
sealed trait State
case class FirstCall(waitV: Deferred[F, Either[Throwable, A]]) extends State
case class NextCalls(v: Either[Throwable, A]) extends State
val initial: F[Ref[F, State]] =
for {
d <- Deferred[F, Either[Throwable, A]]
r <- Ref[F].of(FirstCall(d): State)
} yield r
Stream
.eval(initial)
.flatMap { state: Ref[F, State] =>
val read = new PeriodicMonitor[F, A] {
def get: F[A] = state.get.flatMap {
case FirstCall(wait) => wait.get.rethrow
case NextCalls(v) => v.pure[F].rethrow
}
}
val write: F[Unit] = action.attempt.flatMap { v =>
state.modify {
case FirstCall(wait) => NextCalls(v) -> wait.complete(v).void
case NextCalls(_) => NextCalls(v) -> ().pure[F]
}.flatten
}
Stream.emit(read).concurrently(Stream.repeatEval(write).metered(t))
}
.compile
.resource
.lastOrError
}
}
I didn't add an
Error state to the State ADT. I think what this means is that if the monitor encounters an error, it keeps working - there's no permanent transition from FirstCall or NextCalls to Error, and the stream continues processing. If the caller of PeriodicMonitor.get notices an error (raised by the reader), then the caller can take action on that. However, if I understand this code properly, it's possible the action can fail, be written, and then recover and be overwritten with a subsequent successful value. I think this is not a bad property, although it wasn't what I originally set out to do.
hello everyone - any suggestions as to why I am not getting the same behaviour if I replace this
def life(b: Board): IO[Unit] =
for {
_ <- cls
_ <- showCells(b)
_ <- life(nextgen(b))
} yield ()with this
def life(b: Board): IO[Unit] =
cls *> showCells(b) *> life(nextgen(b))cls and showCells both return IO[Unit]
Am I losing the trampolining in the second version?
https://typelevel.org/cats-effect/datatypes/io.html#stack-safety
https://github.com/typelevel/cats-effect/blob/master/core/shared/src/main/scala/cats/effect/IO.scala#L759
the problem with
*> is before IO even executes, it takes a strict argument
btw the first version also has a potential memory problem if you don't use
better-monadic-for, since those yield () get transformed into a massive chain of map calls
ah, but probably only because the recursion is finite in that example
in my case I would have the same issue because the life function has no base case
in ZIO it's probably lazy
but we do have a lazy
*>, and that's >>
in fact it looks like
*> in ZIO is implemented with flatMap, so it's literally >>
@SystemFw in your Fibers talk, you mentioned that concurrency and parallelism are generally independent of each other. Based on your definitions of both terms, how would you characterize the interaction of fibers or threads whose effects or steps aren't necessarily interleaved? e.g. several fibers running simultaneously and interacting with each other via
Ref, but they are all bound to their own thread, so the discrete steps that comprise the fiber aren't necessarily interleaved. Would you consider that to be purely parallelism or is there something to be said about concurrency as well
there is interleaving there, you are not guaranteed there is parallelism
unless those get mapped to different processors
it's also worth pointing out that those definitions are only valid in the context of the talk, if we were talking about, say, distributed systems, I'd use a different definition of concurrency
which is why I gave the definitions I was working with at the start of the talk
not that I entirely came up with them though, Simon Marlow in Parallel and Concurrent Programming in Haskell uses the same definition, and
Concurrency is not parallelism on the Go (!) blog, uses a very similar one
that being said, concurrent data structures can mean different things as well
although I think in scala the two things I'm thinking of are actually correctly named as concurrent data structures (like Maps) and parallel collections
anyway, in case you're wondering, the most general definition of concurrency I use is when working with distributed systems
where you would say that there is no total order on events
there is a partial order (we can say that some events happened before other), but not a total order: for some pairs of events we don't know if
a < b or b < a (where < in this case is happened before), and therefore a and b are concurrent
btw keep up the great work on tracing! I haven't commented on the PR but I'm following it :)
and thanks! excited to get it to a usable state
I like that definition though. I think you can interpret both the interleaving of events and the interaction of threads through that definition
like, at the level of the JVM threads are also being interleaved, just like fibers are (although through different mechanisms)
so conceptually threads aren't really any more "parallel" than fibers are (although there are potential implications from the memory model)
but yeah,
happens before is more general (Lamport really is a genius), but doesn't help with the aspect that matters of the talk, i.e. the logical thread as a structuring abstraction