I’m not sure how actively do we use Discussions bu...

# arrow-contributors

[pablisco]

I’m not sure how actively do we use Discussions but I’m thinking about adding topics like these there instead of (or as well as) in here

[raulraja]

I think here we can chat more fluidly but for something that is gonna remain as a feature then discussions for public archive is more appropiate

[raulraja]

I like the proposal but I think takeMap is not a good name for it. take is just an impl detail of this being a zipOrNull

[raulraja]

If you think of this as combining two applicative operations which is what zip does you have list.zip(list2) , if you are just zipping your own elements or returning null if a zip to self is not possible maybe this is zipOrNull or zipToSelfOrNull or whatever the std lib has as name to tuple the elements of a list

[pablisco]

yeah, I agree, the name could do with a better candidate. I’m not sure it’s a zip though. Specially since List already has a zip
It would be more like a reduce or fold right?

[raulraja]

let me see if there is a name for this already

[pablisco]

maybe reduceN ?

[pablisco]

Or reduceOrNull 🤔

[raulraja]

The std lib calls the tupling op zipWithNext

[pablisco]

I guess it makes sense to go with some time prefixed as zip if there is already a reflective zip

[raulraja]

well the thing is that the std lib already has zipWithNext that takes a transform function

[raulraja]

which is exactly this right?

[pablisco]

oh wait, I just saw there is already a zipwithNext with a transformation

[pablisco]

I’m not sure it’s quite the same though. As the idea is more about taking a number n of items and then folding them into a single final value. zipWithNext returns a list with the values calculated from adjacent elements

[raulraja]

it would be the remaining arities that zipWithNext is missing if you extend the arity to triple and then to the tuples?

[raulraja]

but this op as implemented in the standard lib just adds a function argument to avoid tupling

[pablisco]

I’m thinking foldN may be more accurate as they would be:

(List<A>, (A) -> R) -> R?
(List<A>, (A, A) -> R) -> R?
...

[raulraja]

It’s a fold in its base because the empty element is null but to be consistent with the std lib is just new versions of zipWithNext where on each one an additional function argument is implemented

[raulraja]

it’s the same case as we have with zip in arrow

[raulraja]

the additional arities can be implemented in terms of map2 or zip

[raulraja]

oh I see, nevermind, I think I missunderstood

[raulraja]

sorry I missuderstood the original post example

[raulraja]

I think it makes sense as takeMap since you are just taking the arguments and passing it to the function but no need to tuple them, the function is not executed

[pablisco]

Another option could be to have NEL1, NEL2, NEL3, etc with respective folds 🤔
So we can do

list.take2()?.fold { a, b -> ... }

It would add an extra type in the middle but I can see other cases where having the size codified on the type could be useful, perhaps

[pablisco]

Probably two different solutions though.
Maybe this last one could wait for value classes as then we could potentially add these operations to them with Meta. Maybe for data classes too. 🤔
I may open a separate disc for this 😅

[raulraja]

Is the overall problem you are trying to solve is that you have arguments on a list and list destructuring it’s unsafe so first you need to check for the values being there before you apply a function?

[raulraja]

If that is the case the something like this may be relevant

[raulraja]

Safe_list_destructuring.kt

[raulraja]

this would prevent you from calling a function if the argument was not there and allows you to add more than 5 elements to destructuring

[raulraja]

of course you’d have to replace the component1 explicit calls for checks on the argument to see if it’s actually there

[raulraja]

I’ll update the example

[pablisco]

Actually, we may not need the wrapper if we use nullable comprehensions:

val result = nullable {
  val a = list.getOrNull(0).bind()  
  val b = list.getOrNull(1).bind()
  "$a $b"
}

When we get multiple receivers we could create something like List.bindAt(index: Int)

I think there is room for both a n-arity solution like DestructuredInvoke as well as more concrete ones for small arities. Like:

value class SingleList<T>(val first: T) {
  fun component1(): T = first
}
value class SingleList<T>(
  val first: T,
  val second: T,
) {
  fun component1(): T = first
  fun component2(): T = second
}

[pablisco]

#futurefeature 😅

[pablisco]

I had a brain session with my pillow last night and thought of a better solution… iteration comprehension:

val result: Result<String> = list.iterating {
  val a = next()
  drop(2)
  val c = next()
  "$a $c"
}

More flexibility and infinite arity 😅
I’m open to better names though 🙂

[raulraja]

That blows up if the iterator is empty?

[pablisco]

It would return a Result with an exception in it. Initially I thought we could return a null but I thought that this way we can communicate where the error happened. And then people can decide to convert to a null or handle the exception. The downside is that we would be throwing an exception. Alternatively, we could have iteratingOrNull that skips the exception throwing if it’s not needed.
Btw, the next() is not the next from Iterator but rather a function defined in the receiver scope of iterating forgot to mention 😅

[pablisco]

Maybe could also have an iteratingOrEither for non exception handling of the result 🤔 Problem here is that we have to create wrapper.
Or maybe have the default return an Either and then the user can convert it to a Result if they need to. Open to any of these options or all at the same time, not sure what’s best 🙂

[raulraja]

Iterator comprehensions would be if multishot was being bind, but it does not, this is something else. In whatever case we could not introduce an idiom that throws exceptions as none of our other blocks or API do

[raulraja]

if you have a gist compiling I’d like to take a look

[pablisco]

Just in my head for now 🙂 But yeah, probably two impls that returns Either<IterateError, T> or null respectively may be the best solution. A wrapper (Either) is probably better than throwing an exception for sure. I’ll see if I get some time this week to try something

[raulraja]

thanks, looking forward to checking it out!