Composition

beginner

Optics can compose to create more powerful optics and access deeply nested data structures in an seamless manner. The result of the composition of the 8 pairs of optics that are included in Bow is shown in the following table. Notice that not all combinations are possible for composition, and that some times the resulting optic is not the same as the optics that were composed.

Β  Iso Lens Prism AffineTraversal Getter Setter Fold Traversal
Iso Iso Lens Prism AffineTraversal Getter Setter Fold Traversal
Lens Lens Lens AffineTraversal AffineTraversal Getter Setter Fold Traversal
Prism Prism AffineTraversal Prism AffineTraversal 🚫 Setter Fold Traversal
AffineTraversal AffineTraversal AffineTraversal AffineTraversal AffineTraversal Fold Setter Fold Traversal
Getter Getter Getter 🚫 🚫 Getter 🚫 Fold 🚫
Setter Setter Setter Setter Setter 🚫 Setter 🚫 Setter
Fold Fold Fold Fold Fold Fold 🚫 Fold Fold
Traversal Traversal Traversal Traversal Traversal Fold Setter Fold Traversal

Each optic can be composed using its instance method compose, which is overloaded to accept all possible combinations with other optics. Besides, for the sake of simplicity, operator + can also be used to compose any two optics.

Example of composition: N-ary Tree

An N-ary tree is a tree where each node can have any number of branches. We can model it like:

enum NTree<A> {
    case leaf(A)
    indirect case node(A, branches: NEA<NTree<A>>)
}

That is, we can have a leaf with an associated value, or a node with an associated value and a NonEmptyArray of branches (it must have at least one, otherwise it would be a leaf).

Let’s imagine that we would like to combine all values of the nodes at level m. Stop for a moment and think how you would do it without optics. It does not have a trivial solution, right? Let’s see if we can leverage the power of optics to do this.

We can start by trying to access the branches of a node. Since NTree is a sum type, the optic that we need to use is a Prism. We can use AutoPrism to get the Prism for the node side of the NTree:

extension NTree: AutoPrism {}

func nodePrism<A>() -> Prism<NTree<A>, (A, NEA<NTree<A>>)> {
    NTree.prism(for: NTree.node)
}

nodePrism gives us a Prism to look into a NTree and get a pair of (A, NEA<NTree<A>>), but we are only interested in the branches part. Given we have a tuple, we would need an optic that lets us focus on one of the components of a tuple.

Fortunately, Bow already provides these utilities. In particular, given that tuples are product types, it seems we would need a Lens to focus on the second component of the tuple. To do so, we can get the Lens from Tuple2._1. There are utilities like this from Tuple2 to Tuple10, to focus on every component of the tuples.

Then, we can compose the previous Prism with this Lens to get an AffineTraversal (see table above) that focuses only on the branches of a node:

func branchesAffineTraversal<A>() -> AffineTraversal<NTree<A>, NEA<NTree<A>>> {
    nodePrism() + Tuple2._1
}

Now, we would like to be able to traverse each individual branch and modify them in isolation. This would give us a way of visiting the branches under the first level of the tree. If we look into the focus of the branchesAffineTraversal we can see that it is a NonEmptyArray, which already has a Traversal to visit each element. Therefore, if we compose them, we can get a Traversal with foci in each node of the first level under the provided node:

func levelTraversal<A>() -> Traversal<NTree<A>, NTree<A>> {
    branchesAffineTraversal() + NEA.traversal
}

Looking at levelTraversal we can see that its source and focus types match. That means we can compose with itself in order to go further down in the tree structure, level by level. We can write a function that gets us a Traversal focused on the nodes of the m level, just by composing the levelTraversal with itself m times:

func level<A>(_ m: UInt) -> Traversal<NTree<A>, NTree<A>> {
    (0 ..< m)
        .map { _ in levelTraversal() }
        .reduce(Traversal.identity, +)
}

In the function above, if m is 0, we return Traversal.identity, which is a Traversal that focuses on the source itself, corresponding to visiting level 0 of the tree. Otherwise, we create m instances of the levelTraversal and combine them all into a single one, to get a Traversal that focuses on nodes at level m.

We wanted to combine values of the nodes at level m. First, we would need to extract the values out of the NTrees. Since all cases in NTree have a value, we can write a custom Getter to do this:

func valueGetter<A>() -> Getter<NTree<A>, A> {
    Getter(get: { state in
        switch state {
        case .leaf(let value), .node(let value, branches: _): return value
        }
    })
}

We can convert the Traversal to a Fold using the asFold property. We can get a Fold to combine values at level m as:

func levelFold<A>(_ m: UInt) -> Fold<NTree<A>, A> {
    level(m).asFold + valueGetter()
}

Finally, if we get a tree whose values have an instance of Monoid, we can combine all its values at level m by:

func combineValues<A: Monoid>(of tree: NTree<A>, at level: UInt) -> A {
    levelFold(level).combineAll(tree)
}

Summary

With this example we have seen how we can use auto-generated, custom and library-provided optics, to build more complex ones that help us perfom a complicated task in an easy and seamless manner.