Tuple-based (immutable, ordered) Dictionary and nested REPLCompletion

General Usage
, ,
1

Hi!
Since I couldn’t find a tuple-based (immutable and ordered) dictionary I came up with the following:

struct TupleDict{K,V} <: AbstractDict{K,V}
  ks::Tuple{Vararg{K}}
  vs::Tuple{Vararg{V}}

  function TupleDict{K,V}(ks::Tuple{Vararg{K}}, vs::Tuple{Vararg{V}}) where {K,V}
    length(ks) == length(vs) || throw(ArgumentError("..."))
    allunique(ks) || throw(ArgumentError("..."))
    new{K,V}(ks, vs)
  end

end

TupleDict{K,V}(ps::Pair...) where {K,V} = TupleDict{K,V}(first.(ps), last.(ps))

Base.keys(d::TupleDict) = d.ks
Base.values(d::TupleDict) = d.vs

function Base.get(d::TupleDict{K,V}, key::K, default) where {K,V}
  for i in 1:length(d.ks)
    candidate = @inbounds d.ks[i]
    isequal(candidate, key) && return @inbounds d.vs[i]::V
  end
  return default
end

Base.iterate(d::TupleDict, i=1) =
  i ≤ length(d.ks) ? (d.ks[i] => d.vs[i], i + 1) : nothing

At first I was quite happy with it, but now I am running into a problem with nested completion in the REPL not working:

struct Foo
  x::TupleDict{Symbol,Union{Foo,Nothing}}
end

Foo(xs::Vararg{Pair}) = Foo(TupleDict{Symbol,Union{Nothing,Foo}}(xs...))

content(f::Foo) = getfield(f, :x)
Base.propertynames(f::Foo) = f |> content |> keys
Base.getproperty(f::Foo, s::Symbol) = content(f)[s]

f = Foo(:abc => nothing, :def => Foo(:xyz => nothing))

If I type f.a<TAB> it completes, but for f.def.x<TAB> it does not.
In contrast, if I would use OrderedCollections: OrderedDict for the same Foo example, it completion works also in the nested case.

Why does inference fail for my simple TupleDict?
Any kind of criticism beyond this question is also greatly appreciated!

2

See LittleDict in OrderedCollections.jl or Dictionaries.jl/contrib/TupleDictionaries.jl at master · andyferris/Dictionaries.jl · GitHub.

3

Thanks @aplavin! I just tried the Foo example for nested completion with OrderedCollections: LittleDict and it turns out that completion also doesn’t work. This suggests that the problem might be more fundamental.