Recursive transformations from Functors.jl

fmap(f, x, ys...; exclude = Functors.isleaf, walk = Functors.DefaultWalk()[, prune])
fmap(walk, f, x, ys...)

A structure and type preserving map.

By default it transforms every leaf node (identified by exclude, default isleaf) by applying f, and otherwise traverses x recursively using functor. Optionally, it may also be associated with objects ys with the same tree structure. In that case, f is applied to the corresponding leaf nodes in x and ys.

Examples

julia> fmap(string, (x=1, y=(2, 3)))
(x = "1", y = ("2", "3"))

julia> nt = (a = [1,2], b = [23, (45,), (x=6//7, y=())], c = [8,9]);

julia> fmap(println, nt)
[1, 2]
23
45
6//7
()
[8, 9]
(a = nothing, b = Any[nothing, (nothing,), (x = nothing, y = nothing)], c = nothing)

julia> fmap(println, nt; exclude = x -> x isa Array)
[1, 2]
Any[23, (45,), (x = 6//7, y = ())]
[8, 9]
(a = nothing, b = nothing, c = nothing)

julia> twice = [1, 2];  # println only acts once on this

julia> fmap(println, (i = twice, ii = 34, iii = [5, 6], iv = (twice, 34), v = 34.0))
[1, 2]
34
[5, 6]
34
34.0
(i = nothing, ii = nothing, iii = nothing, iv = (nothing, nothing), v = nothing)

julia> d1 = Dict("x" => [1,2], "y" => 3);

julia> d2 = Dict("x" => [4,5], "y" => 6, "z" => "an_extra_value");

julia> fmap(+, d1, d2) == Dict("x" => [5, 7], "y" => 9) # Note that "z" is ignored
true

Mutable objects which appear more than once are only handled once (by caching f(x) in an IdDict). Thus the relationship x.i === x.iv[1] will be preserved. An immutable object which appears twice is not stored in the cache, thus f(34) will be called twice, and the results will agree only if f is pure.

By default, Tuples, NamedTuples, and some other container-like types in Base have children to recurse into. Arrays of numbers do not. To enable recursion into new types, you must provide a method of functor, which can be done using the macro @functor:

julia> struct Foo; x; y; end

julia> @functor Foo

julia> struct Bar; x; end

julia> @functor Bar

julia> m = Foo(Bar([1,2,3]), (4, 5, Bar(Foo(6, 7))));

julia> fmap(x -> 10x, m)
Foo(Bar([10, 20, 30]), (40, 50, Bar(Foo(60, 70))))

julia> fmap(string, m)
Foo(Bar("[1, 2, 3]"), ("4", "5", Bar(Foo("6", "7"))))

julia> fmap(string, m, exclude = v -> v isa Bar)
Foo("Bar([1, 2, 3])", (4, 5, "Bar(Foo(6, 7))"))

To recurse into custom types without reconstructing them afterwards, use fmapstructure.

For advanced customization of the traversal behaviour, pass a custom walk function that subtypes Functors.AbstractWalk. The form fmap(walk, f, x, ys...) can be called for custom walks. The simpler form fmap(f, x, ys...; walk = mywalk) will wrap mywalk in ExcludeWalk then CachedWalk.

julia> struct MyWalk <: Functors.AbstractWalk end

julia> (::MyWalk)(recurse, x) = x isa Bar ? "hello" :
                                            Functors.DefaultWalk()(recurse, x)

julia> fmap(x -> 10x, m; walk = MyWalk())
Foo("hello", (40, 50, "hello"))

julia> fmap(MyWalk(), x -> 10x, m)
Foo("hello", (4, 5, "hello"))

The behaviour when the same node appears twice can be altered by giving a value to the prune keyword, which is then used in place of all but the first:

julia> twice = [1, 2];

julia> fmap(float, (x = twice, y = [1,2], z = twice); prune = missing)
(x = [1.0, 2.0], y = [1.0, 2.0], z = missing)

Functors.isleaf(x)

Return true if x has no children according to functor.

Examples

julia> Functors.isleaf(1)
true

julia> Functors.isleaf([2, 3, 4])
true

julia> Functors.isleaf(["five", [6, 7]])
false

julia> Functors.isleaf([])
false

julia> Functors.isleaf((8, 9))
false

julia> Functors.isleaf(())
true

Functors.children(x)

Return the children of x as defined by functor. Equivalent to functor(x)[1].

fcollect(x; exclude = v -> false)

Traverse x by recursing each child of x as defined by functor and collecting the results into a flat array, ordered by a breadth-first traversal of x, respecting the iteration order of children calls.

Doesn't recurse inside branches rooted at nodes v for which exclude(v) == true. In such cases, the root v is also excluded from the result. By default, exclude always yields false.

See also children.

Examples

julia> struct Foo; x; y; end

julia> @functor Foo

julia> struct Bar; x; end

julia> @functor Bar

julia> struct NoChildren; x; y; end

julia> m = Foo(Bar([1,2,3]), NoChildren(:a, :b))
Foo(Bar([1, 2, 3]), NoChildren(:a, :b))

julia> fcollect(m)
4-element Vector{Any}:
 Foo(Bar([1, 2, 3]), NoChildren(:a, :b))
 Bar([1, 2, 3])
 [1, 2, 3]
 NoChildren(:a, :b)

julia> fcollect(m, exclude = v -> v isa Bar)
2-element Vector{Any}:
 Foo(Bar([1, 2, 3]), NoChildren(:a, :b))
 NoChildren(:a, :b)

julia> fcollect(m, exclude = v -> Functors.isleaf(v))
2-element Vector{Any}:
 Foo(Bar([1, 2, 3]), NoChildren(:a, :b))
 Bar([1, 2, 3])

Functors.functor(x) = functor(typeof(x), x)

Returns a tuple containing, first, a NamedTuple of the children of x (typically its fields), and second, a reconstruction funciton. This controls the behaviour of fmap.

Methods should be added to functor(::Type{T}, x) for custom types, usually using the macro @functor.

fmapstructure(f, x; exclude = isleaf)

Like fmap, but doesn't preserve the type of custom structs. Instead, it returns a NamedTuple (or a Tuple, or an array), or a nested set of these.

Useful for when the output must not contain custom structs.

Examples

julia> struct Foo; x; y; end

julia> @functor Foo

julia> m = Foo([1,2,3], [4, (5, 6), Foo(7, 8)]);

julia> fmapstructure(x -> 2x, m)
(x = [2, 4, 6], y = Any[8, (10, 12), (x = 14, y = 16)])

julia> fmapstructure(println, m)
[1, 2, 3]
4
5
6
7
8
(x = nothing, y = Any[nothing, (nothing, nothing), (x = nothing, y = nothing)])