Join on columns of different name

[cjprybol]

Transfer of JuliaData/DataTables.jl#78

idea summary: Use a 2-element tuple to support joining DataFrames on columns that do not have the same names. Currently, we require renaming columns so that they match before columns can be used for joining

join(a, b, on=(:a_name, :b_name))
join(a, b, on=[(:a_name1, :b_name1), (:a_name2, :b_name2)])

Pandas does this more explicitly by having left_on right_on keywords, and R's Data.Table has by.x and by.y (jump to page 61 in docs)

[ararslan]

I'd suggest using pairs, e.g. join(a, b, on=[:a_name1 => :b_name1, :a_name2 => :b_name2])

[nalimilan]

I'm hesitant about using Pair. That kind of suggests a direction or a movement, but here the two column names are treated symmetrically. I would make sense if the second column name was retained in the result, but using the first one would sound more natural (that's what R does at least). OTOH, pairs make it clearer that matching columns need to be grouped, rather than grouping by (left, right). Without them it's really hard to remember.

Anyway, if we use first and last internally, we can support both tuples and pairs, so we just need to choose which syntax we promote in the docs.

[cjprybol]

closed via #1312