New type of ($) operator in GHC 8.0 is problematic

[Richard Eisenberg]

As the instigator of these most recent changes:

- Yes, absolutely, ($)'s type is quite ugly. In other areas, I've tried to hide the newfound complexity in the type system behind flags, but I missed this one. I consider the current output to be a bug.

- It's conceivable to have a flag -fdefault-levity, on by default, which looks for levity polymorphism while printing and instantiates all levity variables to Lifted before printing. That would fix the type of ($). Of course, users could specify -fno-default-levity. Would this make you happy?

- There's a real drawback to flags like -fdefault-levity (and, relatedly, -fprint-explicit-kinds, -fprint-explicit-foralls, -fprint-explicit-coercions, -fprint-equality-relations; the last two are new in 8.0): they hide things from unsuspecting users. We already get a steady trickle of bug reports stemming from confusion around hidden kinds. Users diligently try to make a minimal test case and then someone has to point out that the user is wrong. It's a waste of time and, I'm sure, is frustrating for users. I'm worried about this problem getting worse.

- It's interesting that the solution to the two problems Takenobu pulls out below (but others have hinted at in this thread) is by having an alternate Prelude for beginners. I believe that having an alternate beginners' Prelude is becoming essential. I know I'm not the first one to suggest this, but a great many issues that teachers of Haskell have raised with me and posts on this and other lists would be solved by an alternate Prelude for beginners.

- Separate from a full alternate Prelude, and as Iavor suggested, we could just have two ($) operators: a simple one with no baked-in magic or levity polymorphism, and then a levity-polymorphic, sneakily impredicative one. This would be dead easy.

- Edward is right in that (->) isn't really levity-polymorphic. Well, it is, but it's ad hoc polymorphism not parametric polymorphism. Perhaps in the future we'll make this more robust by actually using type-classes to control it, as we probably should.

- The case with (->) is different than that with (). (() :: Constraint) and (() :: *) are wholly unrelated types. () is not constraintyness-polymorphic. It's just that we have two wholly unrelated types that happen to share a spelling. So there are hacks in the compiler to disambiguate. Sometimes these hacks do the wrong thing. If we had type-directed name resolution (which I'm not proposing to have!), this would get resolved nicely.

- The reason that the foralls get printed in ($)'s type is that kind variables appear in the type variables' kinds. GHC thinks that printing the foralls are useful in this case and does so without a flag. This is not directly related to the levity piece. If you say `:t Proxy`, you'll get similar behavior.


Bottom line: We *need* an alternate Prelude. But that won't happen for 8.0. So in the meantime, I propose -fdefault-levity, awaiting your approval.

Richard

On Feb 5, 2016, at 8:16 AM, Takenobu Tani <takenobu.hs at gmail.com> wrote:

> Hi,
> 
> I'll worry about the learning curve of beginners.
> Maybe, beginners will try following session in their 1st week.
> 
>   ghci> :t foldr
>   ghci> :t ($)
> 
> They'll get following result.
> 
> 
> Before ghc7.8:
> 
>   Prelude> :t foldr
>   foldr :: (a -> b -> b) -> b -> [a] -> b
> 
>   Prelude> :t ($)
>   ($) :: (a -> b) -> a -> b
> 
>   Beginners should only understand about following:
> 
>     * type variable (polymorphism)
> 
> 
> After ghc8.0:
> 
>   Prelude> :t foldr
>   foldr :: Foldable t => (a -> b -> b) -> b -> t a -> b
> 
>   Prelude> :t ($)
>   ($)
>     :: forall (w :: GHC.Types.Levity) a (b :: TYPE w).
>        (a -> b) -> a -> b
> 
>   Beginners should understand about following things, more:
> 
>     * higher order polymorphism (t m)
>     * type class (class t =>)
>     * universal quantification (forall)
>     * kind (type::kind)
>     * levity (lifted/unlifted)
> 
> I think it's harder in their 1st week.
> I tried to draw informal illustrations about Foldable,
> but beginners may need ghci-beginner’s mode or something?
> 
> Sorry I don't still have good idea.
> 
> Of course I like Haskell's abstraction :)
> 
> Regards,
> Takenobu
> 
> 
> 2016-02-05 18:19 GMT+09:00 Joachim Breitner <mail at joachim-breitner.de>:
> Hi,
> 
> Am Freitag, den 05.02.2016, 09:22 +0200 schrieb Roman Cheplyaka:
> > On 02/05/2016 01:31 AM, Edward Z. Yang wrote:
> > > I'm not really sure how you would change the type of 'id' based on
> > > a language pragma.
> > >
> > > How do people feel about a cosmetic fix, where we introduce a new
> > > pragma, {-# LANGUAGE ShowLevity #-} which controls the display of
> > > levity
> > > arguments/TYPE.  It's off by default but gets turned on by some
> > > extensions like MagicHash (i.e. we only show levity if you have
> > > enabled extensions where the distinction matters).
> >
> > Yes, I am surprised this isn't the way it's been done. The levity
> > arguments should totally be hidden unless requested explicitly.
> >
> > I'd only expect this to be a ghc flag (-fshow-levity), not a language
> > pragma, since it should only affect the way types are /shown/.
> 
> shouldn’t this already happen, based on -fprint-explicit-kinds? At
> least I would have expected this.
> 
> So we probably either want to make sure that -fno-print-explicit-kinds
> also prevents forall’ed kind variables, or add a new flag of that (heh)
> kind.
> 
> Greetings,
> Joachim
> 
> --
> Joachim “nomeata” Breitner
>   mail at joachim-breitner.de • http://www.joachim-breitner.de/
>   Jabber: nomeata at joachim-breitner.de  • GPG-Key: 0xF0FBF51F
>   Debian Developer: nomeata at debian.org
> 
> 
> _______________________________________________
> ghc-devs mailing list
> ghc-devs at haskell.org
> http://mail.haskell.org/cgi-bin/mailman/listinfo/ghc-devs
> 
> 
> _______________________________________________
> ghc-devs mailing list
> ghc-devs at haskell.org
> http://mail.haskell.org/cgi-bin/mailman/listinfo/ghc-devs

-------------- next part --------------
An HTML attachment was scrubbed...
URL: <http://mail.haskell.org/pipermail/ghc-devs/attachments/20160205/deccaebc/attachment.html>