Re: Haskell Foldable Wats

(Sorry, I should have said: "core libraries *may* need to use Writer instead of redefining an instance for (,)" -- I don't mean to rule out other technical solutions if they exist) Tom

El 24 feb 2016, a las 14:10, amindfv@gmail.com escribió:

What happened in the FTP was that the libraries@ had a heated discussion, the issue was taken to the users and to a vote, and we ended up with a clear message from users: 80% voted in one direction. My suspicion is that on this issue too, libraries@ is more divided than the community is. I suggest we try to put this issue to bed, and if ~80% of the community says they don't want these instances, then yes -- core libraries should use Writer instead of redefining their own instance for (,). Similarly, if ~80% want the instances, we can grumble that users are wrong but democracy has spoken.

Tom

El 24 feb 2016, a las 13:27, Manuel Gómez escribió:

...

...

El 24 feb 2016, a las 12:27, Chris Allen escribió:

You can't not-include the instances because we'll just end up with orphans so that's not cricket I think.

...

On Wed, Feb 24, 2016 at 1:41 PM, wrote: I don't know what this means -- can you elaborate?

(What I'm proposing is, since there is a sizeable number of people on both sides of the issue who don't seem to be coming closer to an agreement, we bring a vote *to the users* on whether to provide Foldable/Traversable instances for tuples of size 2 and greater. If users say they're useful, we keep/add 'em. If they find them confusing/not useful, we remove/don't add 'em)

If these instances were not included in base, then these instances would nonetheless be made available to a large amount of code because somebody will make a base-orphans library that will define these instances, and many libraries written by authors who believe these instances to be useful will depend on this package (or, worse yet, define their own instances in their own packages which will clash with each other and break things). If any of these libraries end up in the transitive closure of your packages’ dependencies, then you will have these instances defined, regardless of your opinion of them, and regardless of their exclusion from base.

To avoid this unhelpful outcome, if the community decided to forbid these instances, some language extension would have to be designed to forbid instance definitions. This has been discussed previously. If the community did this, it would break a lot of code that does use these instances, and there would be no workaround, as forbidding instances would have to be as global as defining instances. Changing the fundamental property of type class instances that makes them not opt-in/opt-out, but automatically imported from transitive dependencies, would remove one of the properties of the language that (as Edward often argues) is a significant part of what makes Haskell more useful and healthy than some of its kin.

This is not a matter that can be resolved but by consensus, at least not with the solutions that have been thus far proposed (simply removing the instances or accepting the instances as they are). We do not appear to be approaching consensus on these particular solutions. Accepting the instances as they are does have the benefit of already being implemented and being used by a lot of code (but «we’ve always done things this way» is not a good design criterion). Adding more instances for tuples would make the current situation more consistent, although perhaps proponents of removing these instances would prefer the current status quo: inconsistency, but less instances whose existence they reject.