purescript-heterogeneous

Maps and folds for heterogeneous data types.

PureScript has a very rich type-system which lets us explicitly describe the shape of our data in fine detail. For example, Records and row-types let us write ergonomic, polymorphic, structurally-typed data. However, writing generic operations over such data types often involves a lot of tedious, constraint-level tricks. This library provides a framework which separates the traversal and summary logic for heterogeneous data types, like Functor and Foldable do for homogeneous types, so we can reduce the boilerplate and tricks while sharing implementations.

This library exports several classes for both indexed and unindexed folds and maps.

• class HMap and hmap
• class HMapWithIndex and hmapWithIndex
• class HFoldl and hfoldl
• class HFoldlWithIndex and hfoldlWithIndex

These are similar to their homogeneous counterparts. In fact, the folds and maps we write for heterogeneous types can be reused for homogeneous data via the App f a newtype. That is, the following are identical:

map f [1, 2, 3]

hmap f (App [1, 2, 3])

Normal functions aren't enough for a lot of the things we need to do though, so dispatching these is slightly different as we'll need to use PureScript's constraint system.

The following examples will operate over Record types, since it is the most ubiquitous heterogeneous data type in PureScript.

Records aren't ever actually homogeneous to the type system, but sometimes all the values end up being the same type. Mapping over an apparently homogeneous Record is as simple as using a normal (monomorphic) function.

hmap (add 1 >>> show) { a: 1, b: 2, c: 3 }

{ a: "2", b: "3", c: "4" }

In the previous example, we can generalize our mapping function:

addOneAndShow :: forall n. Semiring n => Show n => n -> String
addOneAndShow = add one >>> show

However, we won't be able to dispatch this without giving it a monomorphic type signature which fixes it to a homogeneous Record like before. We want to be able to instantiate these dictionaries for each member individually. To do that, we first need a data type to represent our mapping function:

data AddOneAndShow = AddOneAndShow

With this we can define an instance of Mapping, which we can use with hmap.

instance addOneAndShow ::
  (Semiring n, Show n) =>
  Mapping AddOneAndShow n String where
  mapping AddOneAndShow = add one >>> show

Then instead of passing a function to hmap, we pass in our data type.

hmap AddOneAndShow { a: 1, b: 2.0, c: { x: 12, y: 42 } }

{ a: "2", b: "3.0", c: "{ x: 13, y: 43 }" }

With normal functions, we can use partial application to thread in additional context for our mapping function. We can do the same here by adding arguments to our mapping data type.

Say we want to implement a Record zip operation. That is, we want to map over a record, and for each field we want to look up a corresponding function in another record and apply it to the value.

First we need to define a data type for our zip mapping, but we want it to hold the record of functions:

newtype ZipProps fns = ZipProps { | fns }

If we use MappingWithIndex instead of Mapping we can utilize the field name as well via Proxy.

instance zipProps ::
  (IsSymbol sym, Row.Cons sym (a -> b) x fns) =>
  MappingWithIndex (ZipProps fns) (Proxy sym) a b where
  mappingWithIndex (ZipProps fns) prop = Record.get prop fns

let
  zipRecord = hmapWithIndex <<< ZipProps
in
  { a: add 1
  , b: Tuple "bar"
  , c: \a -> not a
  }
  `zipRecord`
  { a: 12
  , b: 42.0
  , c: true
  }

{ a: 13, b: (Tuple "bar" 42.0), c: false }

Much like with hmap, we can fold over homogeneous records using normal functions.

hfoldl (add :: Int -> Int -> Int) 0 { a: 12, b: 42, c: 100 }

154

The homogeneous case needs a monomorphic function, so we've specialized the type of add with a type signature.

In this example we will implement an alternative Show instance using String concatenation instead of building an intermediate data structure. First, the data type for our fold function:

data ShowProps = ShowProps

We need to follow the same steps as we did with hmap and hmapWithIndex, except we will implement Folding or FoldingWithIndex.

instance showProps ::
  (Show a, IsSymbol sym) =>
  FoldingWithIndex ShowProps (Proxy sym) String a String where
  foldingWithIndex ShowProps prop str a =
    pre <> reflectSymbol prop <> ": " <> show a
    where
    pre | str == "" = ""
        | otherwise = str <> ", "

Then we can write a wrapper, which adds the curly-braces:

showRecord :: forall r.
  HFoldlWithIndex ShowProps String { | r } String =>
  { | r } ->
  String
showRecord r =
  "{ " <> hfoldlWithIndex ShowProps "" r <> " }"

showRecord { a: "foo" , b: 42 , c: false }

"{ a: \"foo\", b: 42, c: false }"

The compiler will not always be able to infer all types for the maps and folds we write for heterogeneous types. That's because it will attempt to determine an output from any combination of folding function, accumulator, and input (for folds) or mapping function and input (for maps). This ensures that multiple mapping and folding operations can be supported for the same underlying input, but has the downside that the compiler will not infer these types.

You will need to provide annotations for any folding, mapping, accumulator, and input types that are not determined in some other way.

For example, this sample showWithIndex function for showing a heterogeneous list requires an annotation for the accumulator type:

showWithIndex :: forall hlist.
  HFoldlWithIndex ShowWithIndex (Array (Tuple Int String)) hlist (Array (Tuple Int String)) =>
  hlist ->
  Array (Tuple Int String)
showWithIndex =
  hfoldlWithIndex ShowWithIndex ([] :: Array (Tuple Int String))

• Module documentation is published on Pursuit.