Tables.subset problems

[slwu89]

Based on the "Dendrograms" section of the tests, and a cursory look at the source for subset, it seems that the following should work, but currently error:

Tables.subset(td.X,1) # works
Tables.subset(td.X,2)

# do not work
Tables.subset(td.X,1:2)
Tables.subset(td.X,[1,2])
Tables.subset(td.X,:)

I plan to look into this soon.

[epatters]

Thanks for the report. I agree that we should support the Tables.jl interface where possible.

[slwu89]

@epatters not sure if this solution is too janky. The issue is that we need Tables.schema to work on objects of type ACSetTable. Anyway, this works.

using ACSets
using Tables

SchDendrogram = BasicSchema([:X], [(:parent,:X,:X)], [:R], [(:height,:X,:R)])

@abstract_acset_type AbstractDendrogram
@acset_type Dendrogram(SchDendrogram, index=[:parent]) <: AbstractDendrogram

d = Dendrogram{Int}()

add_parts!(d, :X, 3, height=0)
add_parts!(d, :R, 2)
add_parts!(d, :X, 2, height=[10,AttrVar(add_part!(d,:R))])
set_subpart!(d, 1:3, :parent, 4)
set_subpart!(d, [4,5], :parent, 5)

td = tables(d)
Tables.subset(td.X,1)
Tables.subset(td.X,2)
Tables.subset(td.X,1:2)

# need to be able to get a Tables.Schema out of an ACSetTable obj
import Tables: schema
function Tables.schema(tab::ACSets.DenseACSets.ACSetTable)
    names = Tables.columnnames(tab)
    Tables.Schema(names, [typeof(getindex(tab,name)) for name in names])
end
Tables.subset(td.X,1:2)
Tables.subset(td.X,[1,2])
Tables.subset(td.X,:)

[EDIT] Actually this might be a subpar solution, as now Tables.subset returns a different data type than an ACSetTable (returns a Tables.DictColumnTable, see https://github.com/JuliaData/Tables.jl/blob/25be9d246c2d9e0f31f694476164631d3e4f7b52/src/Tables.jl#L608). I'm guessing the spirit of ACSets.jl would be instead to extend the method Tables.subset directly. What do you think? Happy to get a PR doing that. In that case, do we also still want to extend Tables.schema as well?