Lack of NamedTuple indexing methods

[davidagold]

Couldn't find an existing issue that quite addressed this, apologize if dup (or naïve).

julia> versioninfo()
Julia Version 0.7.0-DEV.5008
Commit 2cab814429 (2018-05-05 14:43 UTC)
Platform Info:
  OS: macOS (x86_64-apple-darwin17.5.0)
  CPU: Intel(R) Core(TM) i5-4278U CPU @ 2.60GHz
  WORD_SIZE: 64
  LIBM: libopenlibm
  LLVM: libLLVM-3.9.1 (ORCJIT, haswell)
Environment:
  JULIA_REVISE = manual

julia> x = (1, 2); y = (a = 1, b = 2);

julia> x[1:2]
(1, 2)

julia> y[1]
1

julia> y[2]
2

julia> y[1:2]
ERROR: MethodError: no method matching getindex(::NamedTuple{(:a, :b),Tuple{Int64,Int64}}, ::UnitRange{Int64})
Closest candidates are:
  getindex(::NamedTuple, ::Int64) at namedtuple.jl:100
  getindex(::NamedTuple, ::Symbol) at namedtuple.jl:101

julia> y[[:a, :b]]
ERROR: MethodError: no method matching getindex(::NamedTuple{(:a, :b),Tuple{Int64,Int64}}, ::Array{Symbol,1})
Closest candidates are:
  getindex(::NamedTuple, ::Int64) at namedtuple.jl:100
  getindex(::NamedTuple, ::Symbol) at namedtuple.jl:101

It just might be nice to have these indexing methods at some point down the road.

Thoughts?

[bramtayl]

Worth noting that indexing a named tuple with arrays will be type unstable. What would be nice is to be able to index them with tuples:

y[(1, 2)] and y[(:a, :b)].

Getting constants to propagate correctly for type stability would be a challenge.

[eulerkochy]

Worth implementing, should I go forward with this?

[andyferris]

I should xref #24019 - I feel we should have a full set of non-scalar indexing operations between any AbstractArray, AbstractDict, Tuple and NamedTuples.

[andyferris]

Worth noting that indexing a named tuple with arrays will be type unstable.

@bramtayl If the output is an array and we preserve the eltype, it should be type stable (consider also the special case where all elements of the named tuple have the same type).

[andyferris]

I'll also repeat something I wrote on the PR:

Finally, due to some ambiguities in dictionaries (and setindex!) I believe there was some plan to have a "non-scalar indexing" notation or operation or something - perhaps @mbauman knows what the plan is these days. (Matt - I kind-of assumed any new non-scalar indexing might end up there, is that right?)

So I think this is discussed in #30845

[felixcremer]

|