Thanks Gabriel,
I think your suggested approach is 100% backwards compatible
Currently in the case of duplicate column names only the first can be
indexed by its name. This will always be the column appearing in by.x,
meaning the column in y with the same name cannot be accessed. Appending
".y"
It seems like there is a way that is backwards compatible-ish in the sense
mentioned and still has the (arguably, but a good argument I think) better
behavior:
if by.x is 'name', (AND by.y is not also 'name'), then x's 'name' column is
called name and y's 'name' column (not used int he merge) is
