Add distinct method to NonEmptyList and NonEmptyVector.

[Tvaroh]

Hello.

Does it make sense to have distinct method on NonEmptyList? If this is something that could be accepted I might work on unit tests for this.

The implementation is based on universal equality to compare values. Using Eq would require a typeclass for getting hash codes and some Set implementation that would use those.

[ceedubs]

Thanks for submitting this @Tvaroh! I was thinking about wanting a distinct on NonEmptyList/NonEmptyVector just the other day.

There are a few complications:

• We've recently added first-class NonEmptyList and NonEmptyVector implementations. Now that those are in place, I think we are likely going to drop OneAnd (see Create cats.data.NonEmptyStream? #1089).
• As you have brought up, this relies on universal equality. This seems a bit unnatural in Cats, where most things are type class based. However, as you also mentioned, we don't have a hash type class. One alternative would be to take an Order instance and use a tree-based set, but we don't even have one of those.

I'm not quite sure how to go forward. I wonder if anyone else has any thoughts on this?

[Tvaroh]

@ceedubs I understand. Writing a balanced ordered tree-based set might be complicated, not sure if it's even in cats' scope.

[adelbertc]

Short of writing our own tree, perhaps we can use scala.collection.immutable.TreeSet under the covers which requires a scala.math.Ordering, which we can get from our own Order ? Just need to be careful that none of the existing instances of Ordering are inconsistent with Order. I'd imagine they're not.

[codecov-io]

Current coverage is 89.86% (diff: 0.00%)

Merging #1240 into master will decrease coverage by 0.27%

@@             master      #1240   diff @@
==========================================
  Files           243        243          
  Lines          3285       3295    +10   
  Methods        3231       3240     +9   
  Messages          0          0          
  Branches         51         52     +1   
==========================================
  Hits           2961       2961          
- Misses          324        334    +10   
  Partials          0          0          

Powered by Codecov. Last update 2d231ca...308dcc9

[Tvaroh]

Gents, I've updated the PR with distinct on NonEmptyList and NonEmptyVector using TreeSet to keep track of duplicates. I tried to do it efficiently, so the implementation uses mutable builders (ListBuffer and Vector.newBuilder) underneath.

Let me know if this is acceptable.

[Tvaroh]

Have added tests and resolved conflicts with master. Unfortunately sbt test hangs forever for me, do you run it using some specific options?

[ceedubs]

@Tvaroh I usually use testsJVM/test.

[Tvaroh]

@ceedubs thanks, it works and tests pass.

[ceedubs]

This approach seems reasonable to me. I'd like for a couple others to weigh in on this.

@Tvaroh could you please remove the unrelated formatting changes from your PR? They don't follow the conventions used elsewhere in the code base.

I could see it being useful to have a def distinctBy[B](f: A => B)(implicit O: Order[B]) generalization with distinct being a special case for f being identity. That could also wait for a separate PR though.

[Tvaroh]

@ceedubs omg, didn't notice that, blame Intellij. Let me better create a new PR from master to not pollute the git log with unrelated changes.

[Tvaroh]

See #1243.