Add grouping and sorting utilities to NonEmptyVector (parity with NonEmptyList)

[webdevred]

Description

I am refactoring some of my code to use nonempty-vector instead of NonEmpty lists and noticed that NonEmptyVector currently lacks some of the convenient utility functions available on NonEmptyList, such as groupBy, groupAllWith, and sorting helpers like sort or sortOn.

This makes migration from NonEmptyList to NonEmptyVector less ergonomic, since code that previously relied on these functions must now be rewritten manually using conversions or workarounds.

If others are interested in this feature as well, I would be happy to submit a PR.

Example

xs :: NonEmptyVector Int
xs = unsafeFromList (3 :| [1, 2, 1])

-- desired

NEV.groupBy myPredicate xs
sortOn id xs

-- currently required

NEV.unsafeVector $ groupBy myPredicate (NEV.toVector xs)
NEV.unsafeVector $ sortOn id (NEV.toVector xs)

Proposal

Add the following (or equivalent) functions to NonEmptyVector:

• groupBy
• groupAllWith
• sort
• sortOn

to achieve feature parity with NonEmptyList.

Motivation

• Ease migration from NonEmptyList
• Improve consistency between NonEmpty* data types
• Avoid unnecessary conversions or loss of non-empty guarantees

Environment

• Library version: vector-nonempty 0.2.4
• GHC version: 9.6.6

[webdevred]

Hi, after thinking more carefully, I realized that adding sort or sortOn helpers for NonEmptyVector isn't feasible without depending on vector-algorithms. Since vector itself doesn’t provide sorting, implementing these efficiently would either require a new dependency or a less efficient workaround.

So I’d like to retract my request for sorting helpers, however I still believe that grouping utilities (groupBy, groupAllWith) would be a useful addition. Thanks!

[webdevred]

Hello @emilypi , I got a mention from you but I cant see it?
I didnt really mean to mention groupAllWith, I think I confused it with another a function. I should start proof-reading issues before posting them more, sorry about this. Grouping functions would be nice addition I think.

[emilypi]

I'd welcome that, yeah. I'm surprised I (and we) missed those, but they certainly belong in the package. One could easily produce

groupBy 
  :: (a -> a -> Bool)
  -> NonEmptyVector a
  -> [NonEmptyVector a]
groupBy f = coerce . V.groupBy f . _neVec

What would you propose groupAllWith look like, considering it depends on a sort?

I got a mention from you but I cant see it?

Yeah sorry i posted from the wrong account haha

[emilypi]

Okay, since groupAllWith relies on sort, i propose group and groupBy be exposed for parity's sake. I do think it'd be an interesting addition to the package (or maybe another package candidate in and of itself) to see what one gains from the non-emptiness constraint with respect to vector-algorithms, but that's not gonna happen anytime soon from me 😄

[webdevred]

I do believe that I personally would enjoy something that creates non-empty vectors vectors of of non-empty vectors, similar to groupWith1. I my project we have a non empty vector of vertices. Then we group them by name to be able to operate on them based on the grouping, I may want to sort the vertices of the inner nonempty vectors and so.
Helper we use in my project for now:

groupWith1
  :: Eq b
  => (a -> b) -> NonEmptyVector a -> NonEmptyVector (NonEmptyVector a)
groupWith1 f =
  NEV.unsafeFromList
    . map NEV.unsafeFromVector
    . V.groupBy (on (==) f)
    . NEV.toVector

[emilypi]

i'm pretttttttttty sure that groupBy (on (==) f) call could create an empty vector for degenerate cases of on (==) f, which would mean you'd have a NonEmptyVector (Maybe (NonEmptyVector a)) return type at best. That unsafeFromVector call is only to be used in cases where the unsafety is trivial.