Improving the instances of Data.Functor.{Product,Sum}

[John Ericson]

With the recent approval of 
https://gitlab.haskell.org/ghc/ghc/-/merge_requests/4527, I thought it 
might be good to revisit this. I implemented my plan in 
https://gitlab.haskell.org/ghc/ghc/-/merge_requests/4727.

I point a hope the concrete implementation will make clear is that the 
flexible contexts and quantified constraints are *complementing*, not 
*competing*. You can do the flexible instance without the quantified 
constraint, but if you do the breakage will be worse, and the only 
newly-allowed programs will be dubious ones that did the *1 instance but 
forgot the corresponding regular instance.

I hope we can make progress here,

John

On 5/19/20 10:51 AM, John Ericson wrote:
>>
>>         The different results are:
>>         * FlexibleContexts approach: `(Eq (Maybe (f a)), Eq [f a])`
>>         * Eq1 typeclass: `(Eq1 f, Eq a)`
>>         * Quantified Constraints: `(forall x. Eq x => Eq (f x), Eq a)`
>>
> So if (per my plan[1]) `Eq1` has the quantified constraints 
> super-class,than Andrew Martin's second two options will imply the 
> first one. So it seems that the FlexibleContexts choice --- asking for 
> precisely what is needed --- is the best option, dare I say a 
> principle type.
>
>> I think we should pre address any maturity issues or composition/ 
>> generality concerns before folding quantified constraint  instances 
>> into base
>
> I am with you Carter, but the only issues with quantified constraints 
> we've discussed is around (~) and Coercible, but both shouldn't apply 
> here, so I think that's a red-herring.
>
> In particular, only the *1 classes would have a *wanted* quantified 
> constraint via super class (so just one imposed on instances). 
> Everything else would just use FlexibleContexts or stay the same. 
> [Extra given constraints do not in and of themselves pose inference 
> problems.]
>
> It is because the *1 classes do not involve (~) or Coercible, or have 
> anything like a `Type -> Constraint` parameters that could be 
> substituted for (partially applied) (~) or Coercible, that those 
> concerns shouldn't apply.
>
> John
>
> [1]: So nobody need waste their time looking it up, the super class is 
> (forall x. Eq x => Eq (f x)) => Eq1 f
>
>
> _______________________________________________
> Libraries mailing list
> Libraries at haskell.org
> http://mail.haskell.org/cgi-bin/mailman/listinfo/libraries
-------------- next part --------------
An HTML attachment was scrubbed...
URL: <http://mail.haskell.org/pipermail/libraries/attachments/20210104/0dd933cb/attachment.html>