Should `arrow.core.raise.fold` declare `callsInPlace(block, EXACTLY_ONCE)` ? - edit: short answer is...

# arrow

[phldavies]

Should arrow.core.raise.fold declare callsInPlace(block, EXACTLY_ONCE) ? - edit: short answer is no

[CLOVIS]

Can you clarify which fold overload you mean? There are a few... https://apidocs.arrow-kt.io/arrow-core/arrow.core.raise/fold.html

[CLOVIS]

If you think the one with block, recover and transform, then I think so, yes 🤔

[phldavies]

Coming from arrow 1.2.4, all of the non-extension fold overloads that take a block: Raise<Error>.() -> A seem to not specify callsInPlace for block - however I see in 2.x some of them now do (but declare it as AT_MOST_ONCE despite it seemingly always being called

[phldavies]

ah - because block is called inside a try/catch it's only AT_MOST_ONCE as it may not complete, despite the surrounding function call completing

[phldavies]

e.g.

fun main() {
  val s: String
  fold<Unit, _, _>({
    ensure(false) {}
    s = "hello"
  }, { throw it }, {}, {})
  println(s.length)
}

this would compile with EXACTLY_ONCE but (correctly) fail to compile with AT_MOST_ONCE

[CLOVIS]

Wait, so EXACTLY_ONCE can only be used if it runs to completion? But how can you know that in the general case, exceptions could be thrown anywhere?

[phldavies]

as I (now) understand it, EXACTLY_ONCE means "if the outer function completes, the inner function was invoked and completed exactly once"

[phldavies]

https://pl.kotl.in/uCq7XC2ce this example will compile and run but throw an unexpected NPE and compilation will give the warning (with Kotlin 2.0.0) of:

Wrong invocation kind 'EXACTLY_ONCE' for 'block: () -> Unit' specified, the actual invocation kind is 'AT_MOST_ONCE'.