Tables.columnnames(Z) after Z = Tables.rowtable(X) or Z = Tables.dictrowtable(X) does not return proper column names

[EssamWisam]

For instance,

X = DataFrame(rand(Float64, 4, 2), :auto)

X_row = Tables.rowtable(X)
println(Tables.columnnames(X_row))

X_dict_row = Tables.dictrowtable(X)
println(Tables.columnnames(X_dict_row))

X_col = Tables.columntable(X)
println(Tables.columnnames(X_col))

yields

()
(:names, :types, :values)
(:x1, :x2)

I wonder if anything is wrong with this, I expected all to return (:x1, :x2). It crossed my mind that this could be an issue due to the source being a dataframe but I also tried a matrix table for X.

Thanks for this wonderful package!

[bkamins]

see also #337.

@quinnj - I understand that for these types and for CSV.File a special method for Tables.columnnames should be added?

[EssamWisam]

It resolves my issue if I do

using DataFrames
using Tables

X = DataFrame(rand(Float64, 4, 2), :auto)

X_row = Tables.columns(Tables.rowtable(X))
println(Tables.columnnames(X_row))

X_dict_row = Tables.columns(Tables.dictrowtable(X))
println(Tables.columnnames(X_dict_row))

X_col = Tables.columntable(X)
println(Tables.columnnames(X_col))

After going through the docs, a table may only define one of Tables.columnnames(Table) or Tables.columnnames(row). Doesn't it seems natural for one to assume that the former should work for any table? In all cases, I wonder if one can know in advance whether they should do Tables.columnnames(Table) or Tables.columnnames(row) (i.e., whether the tables implements Tables.rows or Tables.columns)