Join Deprecation Warning Pairs

I have been using ‘join’ in accordance with the example given on

julia> join(a, b, on = [(:City, :Location), (:Job, :Work)])

Which seems straightforward enough and seems to work. However,
I keep getting these deprecation warnings which are not fatal but worrying
and may slow down my code:

vector containing such tuples as a value of on keyword argument is deprecated: use Pair{Symbol,Symbol} instead.

I tried replacing tuples as suggested

julia> join(a, b, on = [Pair{:City, :Location}, Pair{:Job, :Work}])

but that did not work at all.

I can’t seem to find any documentation on this, nor questions on here.

I would be grateful for any hints.

Thanks for any help.

The syntax you want is

julia> join(a, b, on = [:City => :Location, :Job => :Work])

This syntax is a bit clearer on which variable is left and which is right, maybe, as well as being a bit more consistent with the rest of DataFrames.

You are correct that the documentation and deprecation warning are out of date. The documentation is already “fixed” on master. I say “fixed” because the behavior will change a good deal in upcoming releases. Soon we will have innerjoin, outerjoin, etc. and the documentation will show the correct use of Pairs.

1 Like

Thanks so much for that.

I just wonder, because I always thought of indices used in joins (except left/right) as being symmetric in their treatment, but the ‘=>’ would appear to violate this (for instance, in ‘Dict’ objects, aren’t the mappings meant to be unambiguous?)

For this reason, I find this change (away from tuples) a bit confusing.

Thanks again for your help above and any additional.

[BTW, I am doing :left joins in my application, so it makes sense in that case!]

Fair enough, that’s a good point. However given that even if the operation is symmetric, the positions of the arguments never will be. Personally I like => because it’s easy to visualize from the positional arguments.

1 Like

Thank you.
PS. If Tuples are not ‘liked’, why not something like <=> ?
PPS Maybe <= for left joins and => for right ones?