nonemptylist is stack safe - #226
Conversation
| /** | ||
| * NOTE: For Kalin | ||
| * It's stack safe for iteration, but we can still overflow with really deep nested monadic compositions. | ||
| * I've added a trampoline version below for those cases. | ||
| */ | ||
| def flatMap[B](f: A => NonEmptyList[B]): NonEmptyList[B] = { | ||
| // NOTE: For Kalin, I know this isn't idiomatic Scala, but it's stack safe for iteration, | ||
| // and I can't think of a better way to do it. | ||
| val builder = List.newBuilder[B] | ||
| var remaining: List[A] = this.toList | ||
|
|
||
| while remaining.nonEmpty do | ||
| builder ++= f(remaining.head).toList | ||
| remaining = remaining.tail | ||
|
|
||
| val result = builder.result() | ||
| NonEmptyList(result.head, result.tail) | ||
| } | ||
|
|
||
| def flatMapTrampoline[B](f: A => TailRec[NonEmptyList[B]]): TailRec[NonEmptyList[B]] = { | ||
| def loop(remaining: List[A], acc: List[B]): TailRec[List[B]] = | ||
| remaining match | ||
| case Nil => done(acc.reverse) | ||
| case h :: t => | ||
| tailcall(f(h)).flatMap { nel => | ||
| loop(t, acc.reverse_:::(nel.toList)) | ||
| } | ||
|
|
||
| loop(this.toList, Nil).map { result => | ||
| NonEmptyList(result.head, result.tail) | ||
| } | ||
| } |
There was a problem hiding this comment.
Can you highlight the motivation here?
The purpose of this class is to be a simple wrapper around List[A] where you know you have at least one element. A NonEmptyList will generally get flatMap called on it an average of 0-1 times, maybe 2, and is generally a rather small size (usually single digits).
From my understanding, the purpose of trampolining is for things like cats Effect[_] or zio ZIO[_, _, _], so that they can essentially become your new runtime system and run forever and self recurse without blowing the stack. Thats really not the purpose of this class 😅. Effect/ZIO are like race cars, and this is like a tricycle.
There was a problem hiding this comment.
With needing to to wrap flatMapTrampoline in this TailRec return type, I cant imagine a time where I would ever see normal flatMap and flatMapTrampoline (with extra things to deal with, wrapping/unwrapping TailRec) and say "oh, I need the beefed up version of this for this use case". Sorry, not trying to hate on the fun exercise, but this does not feel like the place for that level of optimization.
Event for the normal flatMap impl, Scala''s List has built-in mutable efficiency under the hood, and you lose that when trying to implement it yourself.
2 participants
No description provided.