[Haskell-cafe] common class for Set (and Map, resp.) implementations with different constraints on the keys

[Johannes Waldmann]

Dear Cafe,


we have Data.Set, Data.IntSet, Data.HashSet,
and they all have similar API, where the only difference
is the constraint on the elements. (Same thing for maps.)

Can we unify this as follows:

{-# language ConstraintKinds, TypeFamilies #-}
class SetC s where
type Con s :: * -> Constraint
singleton :: (Con s a) => a -> s a
foldMap :: (Con s a, Monoid m) => (a -> m) -> s a -> m
...

Then for Data.Set, we write

instance SetC S.Set where type Con S.Set = Ord ; ...

It seems to work, and it allows me to write polymorphic code,
and switch implementations from the top.
Full source:
https://gitlab.imn.htwk-leipzig.de/waldmann/pure-matchbox/tree/master/src/Data/Set
Example use case (switch implementation):
https://gitlab.imn.htwk-leipzig.de/waldmann/pure-matchbox/blob/master/src/Matchbox/Tiling/Working.hs#L48



Still, there are some clumsy corners in this code, perhaps you can help:


* for instance SetC HashSet, there are two constraints. I want to write

type Con HashSet = \ e -> (Hashable e, Eq, e)

but this does not work (there is no "type lambda"?)


* for maps, I want to write

class (forall k . Foldable m k) => MapC m

but this seems impossible now (This is would work
with -XQuantifiedConstraints ?)


* in some other code using the same idea (the class exports the
constraint), I had an instance where the constraint was empty.

Again, I cannot write type Con Foo = \ s -> ()


- J.W.

[Li-yao Xia]

You can define classes to serve as "constraint combinators", that can be
partially applied:

{- Unary Constraint conjunction -}
class (c a, d a) => (&) (c :: k -> Constraint) (d :: k ->
Constraint) (a :: k)
instance (c a, d a) => (&) c d a

{- Unary empty constraint -}
class Empty a
instance Empty a

Now you can write

type Con HashSet = Hashable & Eq

type Con Foo = Empty

Another alternative is to "eta-expand" the synonym Con:

class SetC s where
type Con s a :: Constraint

class ...
type Con HashSet a = (Hashable a, Eq a)

One issue with that is that Con cannot be partially applied.

Li-yao

Post by Johannes Waldmann
Dear Cafe,
we have Data.Set, Data.IntSet, Data.HashSet,
and they all have similar API, where the only difference
is the constraint on the elements. (Same thing for maps.)
{-# language ConstraintKinds, TypeFamilies #-}
class SetC s where
type Con s :: * -> Constraint
singleton :: (Con s a) => a -> s a
foldMap :: (Con s a, Monoid m) => (a -> m) -> s a -> m
...
Then for Data.Set, we write
instance SetC S.Set where type Con S.Set = Ord ; ...
It seems to work, and it allows me to write polymorphic code,
and switch implementations from the top.
https://gitlab.imn.htwk-leipzig.de/waldmann/pure-matchbox/tree/master/src/Data/Set
https://gitlab.imn.htwk-leipzig.de/waldmann/pure-matchbox/blob/master/src/Matchbox/Tiling/Working.hs#L48
* for instance SetC HashSet, there are two constraints. I want to write
type Con HashSet = \ e -> (Hashable e, Eq, e)
but this does not work (there is no "type lambda"?)
* for maps, I want to write
class (forall k . Foldable m k) => MapC m
but this seems impossible now (This is would work
with -XQuantifiedConstraints ?)
* in some other code using the same idea (the class exports the
constraint), I had an instance where the constraint was empty.
Again, I cannot write type Con Foo = \ s -> ()
- J.W.
_______________________________________________
Haskell-Cafe mailing list
http://mail.haskell.org/cgi-bin/mailman/listinfo/haskell-cafe
Only members subscribed via the mailman list are allowed to post.

[David Feuer]

In my opinion, such a class should usually have more than one parameter. In
the case of Set, I think it makes more sense to use a value type than a
constraint type.

class e ~ Elem s => SetC e s where
type Elem s :: Type
type Elem (_ a) = a
singleton :: e -> s
elem :: e -> s -> Bool
union :: s -> s -> s
...

instance Ord a => SetC a (S.Set a) where
singleton = S.singleton
...
instance a ~ Int => SetC a IntSet where
type Elem IntSet = Int
...

For maps, you can do something similar:

class k ~ Key m => MapC k m where
type Key m :: Type
type Key (_ k) = k
lookup :: k -> m a -> Maybe a
...

instance Ord k => MapC k (M.Map k) where
lookup = M.lookup
....

instance k ~ Int => MapC k IM.IntMap where
type Key IntMap = Int
lookup = IM.lookup

If you like, you can add some constraints, like Traversable m. If you want
to use MFoldable for sets, you can use its Element type family instead of
Elem.

On Fri, Sep 7, 2018, 11:25 AM Johannes Waldmann <

Post by Johannes Waldmann
Dear Cafe,
we have Data.Set, Data.IntSet, Data.HashSet,
and they all have similar API, where the only difference
is the constraint on the elements. (Same thing for maps.)
{-# language ConstraintKinds, TypeFamilies #-}
class SetC s where
type Con s :: * -> Constraint
singleton :: (Con s a) => a -> s a
foldMap :: (Con s a, Monoid m) => (a -> m) -> s a -> m
...
Then for Data.Set, we write
instance SetC S.Set where type Con S.Set = Ord ; ...
It seems to work, and it allows me to write polymorphic code,
and switch implementations from the top.
https://gitlab.imn.htwk-leipzig.de/waldmann/pure-matchbox/tree/master/src/Data/Set
https://gitlab.imn.htwk-leipzig.de/waldmann/pure-matchbox/blob/master/src/Matchbox/Tiling/Working.hs#L48
* for instance SetC HashSet, there are two constraints. I want to write
type Con HashSet = \ e -> (Hashable e, Eq, e)
but this does not work (there is no "type lambda"?)
* for maps, I want to write
class (forall k . Foldable m k) => MapC m
but this seems impossible now (This is would work
with -XQuantifiedConstraints ?)
* in some other code using the same idea (the class exports the
constraint), I had an instance where the constraint was empty.
Again, I cannot write type Con Foo = \ s -> ()
- J.W.
_______________________________________________
Haskell-Cafe mailing list
http://mail.haskell.org/cgi-bin/mailman/listinfo/haskell-cafe
Only members subscribed via the mailman list are allowed to post.

[waldmann]

Post by David Feuer
class e ~ Elem s => SetC e s where

OK. At the use site, under both proposals,
there'll be a two argument constraint.
In my version, the second argument was curried away.

One way or the other - why don't we?
What could be the downsides here?

I guess since it's meant to sit atop (some) modules
from various packages (containers, unordered-containers, enummapset)
it's best to release it as a separate package,
containing the classes, and orphan instances.

- J.

[David Feuer]

The instances won't be orphans if they're in the same module as the class
definition.

Post by waldmann

Post by David Feuer
class e ~ Elem s => SetC e s where

OK. At the use site, under both proposals,
there'll be a two argument constraint.
In my version, the second argument was curried away.
One way or the other - why don't we?
What could be the downsides here?
I guess since it's meant to sit atop (some) modules
from various packages (containers, unordered-containers, enummapset)
it's best to release it as a separate package,
containing the classes, and orphan instances.
- J.