[ghc-steering-committee] Mind the gap

Wed Jul 6 04:10:00 UTC 2022

Yeah, I'd just hesitate to turn it off by default, because GHC seems much
saner overall with PolyKinds on, at least in my experience. It may depend
on which packages/libraries you're often importing. I have a feeling that
the cases crop up when you import something that was defined using
PolyKinds, but your module is not compiled with PolyKinds and GHC then
fails to unify and then prints some nonsense that struggles to avoid
mentioning the polymorphically-kinded reality of things (iirc, it's even
said things along the lines of that it couldn't match type Foo with Foo). I
don't have an example handy right at the moment, but I've run into them
often enough in practice to recommend people just turn the extension on.

On Tue, 5 Jul 2022 at 23:58, Chris Dornan <chris at chrisdornan.com> wrote:

> That speaks to why PolyKinds is needed in the long run. I don’t doubt
> that! (Especially after this report.)
>
> My point is merely that PolyKinds is disruptive, and that is likely to
> undermine confidence in the default Haskell that GHC accepts — really the
> reverse of what we were hoping to achieve with the new standard
> configuration.
>
> Chris
>
>
>
> On 6 Jul 2022, at 04:51, Cale Gibbard <cgibbard at gmail.com> wrote:
>
> I can remember quite a few times where GHC presented me with an
> incomprehensible and confused-sounding error message, but then the actual
> problem was that PolyKinds was off, and upon turning it on, there was no
> longer any problem -- it's to the point that enabling PolyKinds is one of
> the first things I try if I find myself disagreeing with GHC on whether
> something typechecks.
>
> On Tue, 5 Jul 2022 at 23:46, Chris Dornan <chris at chrisdornan.com> wrote:
>
>> I think we can learn to write code for PolyKinds in time, but in my
>> experience it is quite disruptive right now, and could sow FUD around the
>> new consensus we are trying to build.
>>
>> My inclination would be to revert to Haskell2010 as the default for
>> GHC until we can build confidence around a configuration that we are
>> confident won’t be disruptive.
>>
>> At a minimum I would create ghc2022 = ghc2021 \  {PolyKinds} and make
>> that the default for ghc-9.4.
>>
>> What do y’all think?
>>
>>
>>
>> On 5 Jul 2022, at 16:02, Richard Eisenberg <lists at richarde.dev> wrote:
>>
>> I find that surprising, and disappointing. I think of PolyKinds as
>> largely innocuous, except in strange scenarios, which is why PolyKinds is
>> included in GHC2021. If that is wrong, perhaps we should consider not
>> including PolyKinds in GHC2023...
>>
>> It would be interesting to learn more about what's going wrong.
>>
>> Richard
>>
>> On Jul 5, 2022, at 10:26 AM, Chris Dornan <chris at chrisdornan.com> wrote:
>>
>> yup, i should have started there—all my recent troubles have come from
>> PolyKinds!
>>
>> On Tue, 5 Jul 2022 at 15:00, Richard Eisenberg <lists at richarde.dev>
>> wrote:
>>
>>> This is PolyKinds, which generalizes the kind of D, meaning that
>>> satisfying the Typeable a superclass constraint of C also requires a
>>> Typeable k constraint, where (a :: k).
>>>
>>> Richard
>>>
>>> > On Jul 2, 2022, at 12:49 AM, Chris Dornan <chris at chrisdornan.com>
>>> wrote:
>>> >
>>> > Quick quiz: the below Haskell2010 ‘Phantoms' module (also in this
>>> Gist: https://gist.github.com/cdornan/f75cd8024434d998c87610cbb7fb6ab3)
>>> appears (for me on GHC 9.2.2 and 9.2.3) to not be a GHC2021 module,
>>> reporting this error under the latter configuration:
>>> >
>>> > Phantoms.hs:22:10: error:
>>> >    • Could not deduce (Typeable k)
>>> >        arising from the superclasses of an instance declaration
>>> >      from the context: (C a, Typeable a)
>>> >        bound by the instance declaration
>>> >        at Phantoms.hs:22:10-36
>>> >    • In the instance declaration for ‘S (D a)’
>>> >   |
>>> > 22 | instance (C a,Typeable a) => S (D a) where smethod = undefined
>>> >   |          ^^^^^^^^^^^^^^^^^^^^^^^^^^^
>>> >
>>> > Phantoms.hs:22:10: error:
>>> >    • Could not deduce (Typeable k)
>>> >        arising from the superclasses of an instance declaration
>>> >      from the context: (C a, Typeable a)
>>> >        bound by the instance declaration
>>> >        at Phantoms.hs:22:10-36
>>> >    • In the instance declaration for ‘S (D a)’
>>> >   |
>>> > 22 | instance (C a,Typeable a) => S (D a) where smethod = undefined
>>> >
>>> > The question is which language extension(s) are giving rise to these
>>> errors?
>>> >
>>> > The ‘Phantoms' module is this:
>>> >
>>> > {-# LANGUAGE DerivingVia #-}
>>> >
>>> > module Phantoms where
>>> >
>>> > import Data.Typeable
>>> >
>>> >
>>> > class C a where
>>> >  cmethod :: Proxy a -> ()
>>> >
>>> > class (Show a, Typeable a) => S a where
>>> >  smethod :: a -> Int
>>> >
>>> >
>>> > newtype UsingD a = UsingD { getUsingD :: a }
>>> >
>>> >
>>> > newtype D a = D { getD :: Int }
>>> >    deriving (Show) via UsingD (D a)
>>> >
>>> >
>>> > instance (C a,Typeable a) => S (D a) where smethod = undefined
>>> >
>>> > instance (C a,Typeable a) => Show (UsingD a) where showsPrec =
>>> undefined
>>> >
>>> >
>>> > instance (C a) => C (D a) where cmethod _ = undefined
>>> >
>>> > _______________________________________________
>>> > ghc-steering-committee mailing list
>>> > ghc-steering-committee at haskell.org
>>> >
>>> https://mail.haskell.org/cgi-bin/mailman/listinfo/ghc-steering-committee
>>>
>>>
>>
>> _______________________________________________
>> ghc-steering-committee mailing list
>> ghc-steering-committee at haskell.org
>> https://mail.haskell.org/cgi-bin/mailman/listinfo/ghc-steering-committee
>>
>
>
-------------- next part --------------
An HTML attachment was scrubbed...
URL: <http://mail.haskell.org/pipermail/ghc-steering-committee/attachments/20220706/0947794b/attachment-0001.html>