Re: [ghc-steering-committee] Discussion on proposal #99: forall {k}

I am supportive of accepting the proposal, but I think that we should address the point raised by Joachim on the the git-hub thread first ( https://github.com/ghc-proposals/ghc-proposals/pull/99#issuecomment-40106097...). It really is not quite clear if the specificity is a property of the type itself, or the thing that has that type. On Thu, Jun 28, 2018 at 1:07 AM Simon Peyton Jones wrote:

I’m keen on this proposal. I am very keen to be able to express in a type or kind signature everything needful to use the value (or type constructor). The proposal gives an example at the term level. It’d be strengthened by an example at the type level. Eg

type T1 :: forall {k}. (k->Type) -> k -> Type

data T1 f a = MkT (f a)

type T2 :: forall k. (k->Type) -> k -> Type

data T2 f (a::k) = MkT2 (f a)

Here T1 has an inferred kind variable, mentioned nowhere in the declaration. The kind signature (which I have written as if it were code, which is the subject of a separate proposal) is inferred as shown.

T2 mentions k, so its kind signature looks a bit different.

What’s the difference? It’s exactly in explicit kind application. You can write (T1 @Type Maybe Int), but T2 can’t have that explicit kind argument.

The whole business of Required/Specified/Inferred is more complicated than I like, but if we have it (which we do right now) we should allow the programmer to specify exactly what they mean.

(An alternative is to abolish the Specified/Inferred distinction, but that carries significant costs of its own and no one is proposing it.)

Let’s accept!

Simon

*From:* Richard Eisenberg *Sent:* 28 June 2018 05:45 *To:* Iavor Diatchki *Cc:* Simon Peyton Jones ; ghc-steering-committee@haskell.org; Joachim Breitner < mail@joachim-breitner.de>

*Subject:* Re: [ghc-steering-committee] Discussion on proposal #99: forall {k}

But we morally *do* have explicit kind application. See accepted proposal https://github.com/ghc-proposals/ghc-proposals/blob/master/proposals/0015-ty... https://na01.safelinks.protection.outlook.com/?url=https%3A%2F%2Fgithub.com%2Fghc-proposals%2Fghc-proposals%2Fblob%2Fmaster%2Fproposals%2F0015-type-level-type-applications.rst&data=02%7C01%7Csimonpj%40microsoft.com%7C15fcb1b642114e05f1a808d5dcb1e5ca%7C72f988bf86f141af91ab2d7cd011db47%7C1%7C0%7C636657578984502957&sdata=%2B7wS%2F75AielivG1KfFAzGPXKSS5aHOtEVwy9Q6WyYLU%3D&reserved=0 So the only reason that we don't have it is that GHC hasn't caught up to its specification. (I have a student working on this now.)

The reason that T2 and T3 are different here is that T2 mentions the k in the type declaration, while T3 mentions it only in the constructor declaration.

Richard

On Jun 26, 2018, at 6:52 PM, Iavor Diatchki wrote:

Hello,

thanks for the revisions---I read through the new version, and I think I almost understand the plan. I would say that for this proposal it makes more sense to just deal with explicit specificities at the value level, and never have the curly braces in type constructors. Since we don't have an explicit kind application at the type level (e.g., we can't write `Proxy @Type`), I don't think it really makes sense to add the braces to type constructors. If we ever implemented THAT feature, then we can discuss which declarations should have explicit and which should have implicit parameters. In the current set of examples, I find it odd that `T2` does not have braces, but `T3` does.

-Iavor

On Wed, Jun 20, 2018 at 2:15 PM Richard Eisenberg wrote:

I have posted the new proposal, at https://github.com/ghc-proposals/ghc-proposals/pull/148 https://na01.safelinks.protection.outlook.com/?url=https%3A%2F%2Fgithub.com%2Fghc-proposals%2Fghc-proposals%2Fpull%2F148&data=02%7C01%7Csimonpj%40microsoft.com%7C15fcb1b642114e05f1a808d5dcb1e5ca%7C72f988bf86f141af91ab2d7cd011db47%7C1%7C0%7C636657578984502957&sdata=RQ4V%2FWdRz009o6sieAUpx7wQ6RCqWydsQGbENvHFZfM%3D&reserved=0

I've also updated #99 to clarify where the new syntax is allowed.

For Iavor's examples:

data T1 a = C1 a

type T1 :: Type -> Type

C1 :: forall a. a -> T1 a

data T2 (a :: k) = C2 { f2 :: Proxy a }

type T2 :: forall k. k -> Type

C2 :: forall k (a :: k). Proxy a -> T2 a

f2 :: forall k (a :: k). T2 a -> Proxy a

data T3 a where C3 :: forall k (a::k). Proxy a -> T3 a

type T3 :: forall {k}. k -> Type

C3 :: forall k (a :: k). Proxy a -> T3 a

data T4 a where C4 :: forall {k} (a::k). Proxy a -> T3 a

type T4 :: forall {k}. k -> Type

C4 :: forall {k} (a :: k). Proxy a -> T3 a

data T5 k a where C5 :: forall k (a::k). Proxy a -> T5 k a

Rejected, as k is used dependently but is not lexically dependent. (This is no change.) It we have

...

data T5 k (a :: k) where C5 :: forall k (a :: k). Proxy a -> T5 k a

we would get

type T5 :: forall k -> k -> Type

C5 :: forall k (a :: k). Proxy a -> T5 k a

data T6 k a where C6 :: forall {k} (a::k). Proxy a -> T6 k a

Rejected, like T6. If we revise to:

...

data T6 k (a :: k) where C6 :: forall {k} (a::k). Proxy a -> T6 k a

we get

type T6 :: forall k -> k -> Type

C6 :: forall {k} (a::k). Proxy a -> T6 k a

I've updated the proposal itself to include these examples.

Does this help to clarify?

Richard

On May 25, 2018, at 1:56 PM, Iavor Diatchki wrote:

Hello,

well, I thought that Richard was going to write a new proposal based on the feedback here, but it sounds like he is planning to revise #99, and then write a separate new one. I guess we should discuss the proposal again once the changes are in. I would encourage Richard to add some text and examples to clarify exactly what's in the proposal and what's not, and how things are supposed to work. Here are some examples, for which it would be illuminating (to me) to see the types/kinds of all names introduced.

data T1 a = C1 a

data T2 (a :: k) = C2 { f2 :: Proxy a }

data T3 a where C3 :: forall k (a::k). Proxy a -> T3 a

data T4 a where C4 :: forall {k} (a::k). Proxy a -> T3 a

data T5 k a where C5 :: forall k (a::k). Proxy a -> T5 k a

data T6 k a where C6 :: forall {k} (a::k). Proxy a -> T6 k a

-Iavor

On Fri, May 25, 2018 at 2:02 AM Simon Peyton Jones wrote:

I’m keen to get #99 into GHC in some form.

My motivation (which could be a fourth bullet in the proposal) is that it should be possible for a programmer to write a fully-explicit type signature for anything GHC can infer. But currently we can’t. For typeRep1 GHC infers the signature shown for typeRep3; but we can’t write it down.

based on the discussion so far, it seems that #99 in its current form might not be exactly what we want

Can you summarise the reasons it might not be exactly what we want?

Simon

*From:* ghc-steering-committee *On Behalf Of *Richard Eisenberg *Sent:* 24 May 2018 21:17 *To:* Iavor Diatchki *Cc:* ghc-steering-committee@haskell.org; Joachim Breitner < mail@joachim-breitner.de> *Subject:* Re: [ghc-steering-committee] Discussion on proposal #99: forall {k}

I do plan on turning Joachim's recent suggestion into a separate proposal, and then to modify #99. But the modification would remove only the bit about classes, not the feature overall. I don't have time to do this now, though -- will do next week.

Richard

On May 24, 2018, at 1:54 PM, Iavor Diatchki wrote:

Hello,

based on the discussion so far, it seems that #99 in its current form might not be exactly what we want, so I'd say that we should reject it for the moment. Overall, I agree that it would be nice to come up with a consistent notation for things that are currently happening in GHC but we can't write, so perhaps we could revisit this with a revised proposal at a later time?

-Iavor

On Sat, May 5, 2018 at 8:48 PM Joachim Breitner wrote:

Hi,

Am Mittwoch, den 02.05.2018, 16:10 -0400 schrieb Richard Eisenberg:

...

Joachim, you are always a fount of interesting ideas.

...

On May 2, 2018, at 2:51 PM, Joachim Breitner wrote:

class C k (a : k) where meth :: a meth :: forall {k} a. C k a -> k -> Constraint

I think this is brilliant. But not only for this proposal! Imagine this:

class Num a where fromInteger :: Integer -> a

fromInteger :: Integer -> forall a. Num a => a

If we do that, then #129 is essentially solved, at no further cost to anyone. (Note that in all Haskell98-style code, no one will ever be able to notice the changed type of fromInteger.)

This approach also allows for the possibility of reordering quantified type variables for Haskell98-style constructors, if anyone should want to do it.

And it allows for updated types (including quantified variable ordering, etc.) for record selectors.

And it allows (maybe?) for giving good types to GADT record selectors:

data X a where Foo :: { bar :: Int } -> X Int Quux :: { bar :: Bool } -> X Bool bar :: X a -> a

GHC currently rejects the declaration for X, but it could be accepted if only we could specify the correct type of bar. And now we can. I don't particularly want to cook up the typing rules here, but I don't think I'm totally crazy.

GADT record selectors aside, the rule for these could be that the top-level type signature must be equivalent w.r.t. the subtype relation with the original type signature. That is, if the new signature is t1 and the old was t2, then t1 <: t2 and t2 <: t1. Easy enough to check for. The implementation would probably do a little worker/wrapper stunt.

I smell a new proposal… what does this mean for #99? Will you want to revise it?

Cheers, Joachim -- Joachim Breitner mail@joachim-breitner.de http://www.joachim-breitner.de/ https://na01.safelinks.protection.outlook.com/?url=http%3A%2F%2Fwww.joachim-breitner.de%2F&data=02%7C01%7Csimonpj%40microsoft.com%7Ca43c362fd18e4163aeba08d5c1b34be2%7C72f988bf86f141af91ab2d7cd011db47%7C1%7C0%7C636627898179266033&sdata=1cwPEVKXCcP3DuCHXOFuA%2BbEBniRLjcKI%2BAZji5vR4Q%3D&reserved=0 _______________________________________________ ghc-steering-committee mailing list ghc-steering-committee@haskell.org https://mail.haskell.org/cgi-bin/mailman/listinfo/ghc-steering-committee

_______________________________________________ ghc-steering-committee mailing list ghc-steering-committee@haskell.org https://mail.haskell.org/cgi-bin/mailman/listinfo/ghc-steering-committee