DataFrame(t::Table) converts PooledVector columns

[behinger]

potentially related to #3432

using PooledArrays
using TypedTables
using DataFrames
t = Table(b = PooledArray([2,3,1]),a=PooledArray([1,2,3]))

DataFrame(t).a # returns Vector
DataFrame(columns(t)).a # returns PooledVector

I expected the DataFrame constructor to keep the Vector type intact. Not really sure why the columns helps that much - I dont understand the Tables.jl interface on that level.

Maybe this report helps someone, for me the solution with columns is fine

[bkamins]

The reason is that t to the "outside world" communicates that it is an iterator of rows, so DataFrame constructs a vectors from these rows.

[behinger]

ah I see, that makes sense. Thanks!!