Do we have something like this in Arrow? I tried t...

# arrow

[thanh]

Do we have something like this in Arrow? I tried to find but it seems missing.

public fun <A, B, C> Either<A, B>.combine(SGA: Semigroup<A>, b: Either<A, C>): Either<A, Pair<B, C>>

[stojan]

the function exists for Validated and it's called zip

https://arrow-kt.io/docs/apidocs/arrow-core/arrow.core/zip.html

AFIK it doesn't exist for Either

[thanh]

thanks, I knew that, but also I feel like, the way we have to switch between Validated and Either is a bit annoying sometime ;(

[simon.vergauwen]

I would honestly not write such a function for Either.

Typically I work with Validated in parts where I am actually validating, and then I simply bind them from either blocks.

For example here,
https://github.com/nomisRev/ktor-arrow-example/blob/e35fbd8e24c253eecf1200b3024caafedf70c908/src/main/kotlin/io/github/nomisrev/validation.kt#L57

https://github.com/nomisRev/ktor-arrow-example/blob/e35fbd8e24c253eecf1200b3024caa[…]0c908/src/main/kotlin/io/github/nomisrev/service/UserService.kt

[simon.vergauwen]

Do you have different use-cases?

[thanh]

Thanks Simon. I'm using similar strategy as your example. It works great. But in my experience there are some difficulty when introduce Validated to beginners.

[simon.vergauwen]

Always interested in improving on that, but I personally don't think introducing such an API for Either would make this more clear.

[thanh]

Agree, this api also introduce some complication as well.

[raulraja]

I think in the future the use cases for validated like error accumulation can be added directly over Either when there is a NonEmptyList<E> on the Left side. Other than that Validated and Either are the same. This is something I think we should re-visit. Is there other use case that Validated covers and Either does not beside the operators for Error accumulation?

[simon.vergauwen]

If we opt for such a API I would argue that we should drop zip for Either completely.
Since zip is simply the following, but the snippet below doesn't even suffer from the n-arity issue

either {
   transform(a.bind(), b.bind(), c.bind(), ...)
}

That raises another issue though. What about traverse does it accumulate, or not for "Either/Validated".

[simon.vergauwen]

If we can remove API surface, and/or data types whilst still covering the same features/behaviour in a clear way I am pro.

[thanh]

I think traverse should work as normal, and additional parTraverse for error accumulation.

[thanh]

But maybe that will be confused with some other concurrency functionality.

[simon.vergauwen]

It would conflict with parTraverse from Arrow Fx yes

[simon.vergauwen]

I am personally not a huge fan of parMap / parTraverse naming for error accumulation.

[simon.vergauwen]

I always expect parallelism

[phldavies]

would `nelMap`/`nelTraverse` work?

[simon.vergauwen]

Hmm, I like this discussion thread! 🥳
To make a suggestion, and to stay within the Kotlin Std naming spirit I think something like mapAccumulate or mapCollect or something along those might be a good API name for such an API. If we can close the gap between Validated and Either we can seriously decrease the API surface 🤔

Is there a use-case for Validated except for traverse w/ Monoid or zip w/ Monoid? cc\\ @Alejandro Serrano Mena

[Alejandro Serrano Mena]

I’ve never used it in any other place than either zip or `traverse`ing some data structure (maybe not a list, maybe a tree or something like that)

[Alejandro Serrano Mena]

but indeed one could minimize the surface a lot by making something like Validated<A, B> = Either<NonEmptyList<A>, B> or something like that, and provide specific mapAccumulate which only work when Either has a list inside

[stojan]

right now Validated support any type on the left.... wouldn't that limit the left side to NonEmptyList ?

[simon.vergauwen]

Yes, introducing such a restriction seems unnecessary.

[stojan]

the way I see it, in terms of the data they carry Validated and Either are the same. We can convert between both of them without information loss.

The differences are:

• Validated lacks flatMap by design (it used to have andThen which was similar to flatMap)
• Either has a zip that does NOT accumulate errors, Validated.zip accumulates errors
• Validated accumulates errors and Either short circuits on the first error
The short circuit VS error accumulation behaviour right now is based on the container type. If we were to merge both containers we would need some other way to trigger error accumulation VS short circuiting. I'm not sure what that would be and if it would actually make things simpler. 🙂

[thanh]

So maybe I'm naive, but if we introduce two new functions for Either for errors accumulation, which can be zipAcccumulate and traverseAccumulate then we don't have to use Validated anymore. At least for my use cases and Simon's examples above.

[stojan]

I was curious how ZIO handles this as ZIO also doesn't have typeclasses and it's not constrained by them...

ZIO itself doesn't have error accumulation, it only has the failing fast zip method.

For accumulating errors they have Validation which is similar to ValidatedNel and ZValidation which is similar to Validated in ZIO Prelude

https://zio.github.io/zio-prelude/docs/functionaldatatypes/

[Alejandro Serrano Mena]

yes, having different Either and Validated is a common pattern, not only in ZIO, Haskell also follows that trend. But there are also reasons to keep the API simple and have just a single one