Add mapping functions for Pair and Triple

[konnik]

The Pair and Triple data types lacks a convenient way to map the individual components using a transformation function. The Result type already have map to transform the success value so it is already an established pattern in the stdlib.

For Pair mapFirst and mapSecond are added. For Triple mapFirst, mapSecond and mapThird are added.

KT-21648 Fixed

[Recordsmen]

Good Job

[fvasco]

Really Kotlin needs this functions?

Result is a different object, it does not have the copy function, and the map functions contains a bit of logic. I would happy to see a real use case for this PR.

[konnik]

Really Kotlin needs this functions?

As a functional programmer I do think so. :-)

This essentially makes the Pair/Triple a functor (or actually a bifunctor/trifunctor) and makes it much more convenient (and less error prone) to work with these types.

Here is an example that I thinks makes it clear:

// say we have a pair of Int to String
val pair: Pair<Int, String> = 42 to "The answer..."

// now we want to transform it into a Pair<Int,Int> instead. This is
// easy if we have the suggested mapping function.
val a: Pair<Int, Int> = pair.mapSecond { it.length }

// to do the same without the mapping functions you have to
// create a new pair from scratch which means needing to copying the first
// element of the pair. 
val b: Pair<Int, Int> = pair.first to pair.second.length
val c: Pair<Int, Int> = Pair(pair.first, pair.second.length) 

// note that we can't use the copy-method since the type has changed
val d: Pair<Int, Int> = pair.copy(pair.first, pair.second.length) // this won't compile

// also note that nothing is preventing us from swapping the values by mistake. This is impossible using the mapping function.
val bug: Pair<Int, Int> = Pair(pair.second.length, pair.first)

IMHO, the winner measured on clarity and conciseness is definitely alternative a.

[konnik]

Tried to clean up some things in the branch and force pushed some stuff and the PR was closed. Sorry for this, don't know what happened. Reopened the PR.

[fvasco]

As a functional programmer I do think so. :-)

Does this pull request address a tangible issue?

val bug: Pair<Int, Int> = Pair(pair.second.length, pair.first)

This outcome arises from the misuse of primitive types and the utilization of an anonymous tuple, in my humble opinion. In this context, first and second lack meaningfulness, except within a very limited scope in the code. While mapping a value in an Option or an exception in a Result holds significance, I fail to perceive much rationale in mapping the third value of a tuple, especially with a distinct data type. Perhaps it would be more beneficial to create a new tuple to elucidate the content (and the significance) of this data. Is it advisable to promote this utilization of Pair?

Could you provide a concrete real-life example?

[konnik]

I agree that excessive use of Pair/Triple should be avoided in favour of more specialised and descriptive types if possible. Nevertheless the Kotlin stdlib does provide them and then should in my opinion try make it as ergonomic as possible to use them (when the user has decided it IS appropriate to use them).

I can't remember the specific use case that led me to initiate this PR (it was a long time ago) but isn't the example I already gave is enough to see the benefits?

I mean, mapping the contents of a "container like structure" without changing the structure itself is something we do all the time in FP. I can't see how the container like structure Pair/Triple is any different than a List<T>, Result<T>, Option<T> or a Parser<T>, they are all functors that can be mapped over.

[fvasco]

isn't the example I already gave is enough to see the benefits?

No, I don't detect Pair(pair.first, pair.second.length) as a common practice, so I am skeptic about the value of these methods. I am looking for a reason to suggest to developers to use these methods.

I can't see how the container like structure Pair/Triple is any different than a List, Result, Option or a Parser, they are all functors that can be mapped over.

This is not the point. I agree with you: it is possible to define a mapping function for all of these data type, this does not imply that it is useful. If you use often map on Result or List, but you does not remember when you used map on Pair (neither if using map on Pair is a good practice), then these type are different.

This is my point of view.

[konnik]

I am looking for a reason to suggest to developers to use these methods.

How about this suggestion?

"when you need to transform one element of a Pair (possibly into another type) without changing the other element, then use mapFirst/ mapSecond."

If you use often map on Result or List, but you does not remember when you used map on Pair (neither if using map on Pair is a good practice), then these type are different.

Yes, the types differ, but only in how frequent they are used in a particular codebase. But regarding if a mapping-function is useful or not, they don't differ at all. To be able to use a data structure as a functor is always useful, imho.

I can't really see the problem of introducing these functions to enable a more functional style of programming with Pair/Triple.

[ilya-g]

pair.first to pair.second.length

grep.app finds several usages of this pattern, but not very much:

• https://grep.app/search?q=%5C.first%20to&regexp=true&filter[lang][0]=Kotlin
• https://grep.app/search?q=to%20%5Cw%2B%5C.second&regexp=true&filter[lang][0]=Kotlin