Column unioning when appending data

[bkamins]

Use DataFrame(Tables.dictcolumntable(t) when cols=:union in append! of a Tables.jl table. x-ref #3372 (comment)

[bkamins]

Doing this would be breaking. Instead I decided to propose an improvement of the documentation in #3387

[nalimilan]

What would it break exactly?

[bkamins]

Currently we have:

julia> append!(DataFrame(), [(a=1,), (a=2,b=3)], cols=:union)
2×1 DataFrame
 Row │ a
     │ Int64
─────┼───────
   1 │     1
   2 │     2

after the proposed breaking change we would have:

julia> append!(DataFrame(), [(a=1,), (a=2,b=3)], cols=:union)
2×2 DataFrame
 Row │ a      b
     │ Int64  Int64?
─────┼────────────────
   1 │     1  missing
   2 │     2        3

[nalimilan]

Ah. I really dislike that behavior of Tables.columns. This is why having it take a cols argument would make sense. Disregarding this, I wonder whether we could consider the change in DataFrames as a bugfix...