partitionEither.kt

# arrow

[raulraja]

partitionEither.kt

partitionEither.kt

[raulraja]

Hi @mitch, We no longer have Foldable or those types but I think what you are trying to achieve can be defined as:

fun <E, A> Iterable<Either<E, A>>.partitionEither(): Pair<List<Left<E>>, List<Right<A>>> =
    fold(emptyList<Left<E>>() to emptyList<Right<A>>()) { (lefts, rights), either ->
        when (either) {
            is Left -> lefts + either to rights
            is Right -> lefts to rights + either
        }
    }

[raulraja]

(untested)

[mitch]

Thanks, Raul. That makes sense. We weren’t too far off with our thinking (also untested):

public fun <T> Iterable<Either<E, T>>.partitionEither(): Pair<List<T>, List<E>> {
    val values = mutableListOf<T>()
    val errors = mutableListOf<StatusException>()

    forEach { result ->
        when (result) {
            is Either.Right -> values.add(result.value)
            is Either.Left -> errors.add(result.value)
        }
    }

    return Pair(values.asList(), errors.asList())
}

[raulraja]

I missed we have in arrow https://github.com/arrow-kt/arrow/blob/main/arrow-libs/core/arrow-core/src/commonMain/kotlin/arrow/core/Iterable.kt#L1002

[raulraja]

separateEither

[mitch]

ah hah! I missed that too!

[raulraja]

I think your impl or the one I proposed it's probably faster than the other one flatMapping twice over the original list

[mitch]

Do those create a lot of list objects

[mitch]

yeah

[raulraja]

yeah, looks like there is room to improve those arrow implementations if we setup benchmarks to prove these other ones are faster and probably more memory effitient.

[mitch]

makes sense.

Assuming our collections aren’t large, I’m fine with going separateEither and using standard library code.

Thanks for the quick help!

[raulraja]

Created an issue so we don't forget to improve those implementations. https://github.com/arrow-kt/arrow/issues/2812

[mitch]

Very cool. I think temporary mutability will mean less objects.

[Giorgos Makris]

would grouping the eithers by their class be an acceptable approach performance wise?

[raulraja]

@Giorgos Makris The general rule we are following in Arrow is that impl can have local mutability and other optimizations as long as they remain legible. I think grouping by class would be acceptable since both subtypes of Either are public.

[Giorgos Makris]

I don't think I have ever contributed to open source projects before but I could give this is a shot in the evening. Would that be alright?

[raulraja]

Of course and thank you! we would love to see more contributions to Arrow. Feel free to ping us directly with any questions here or in #arrow-contributors if you need any help 🙂

[Giorgos Makris]

awesome! Thank you ^^