Update join

[garborg]

Deprecate natural joins (user must specify join key).
Add kind = :cross.

This is a product of discussion in #531.

[johnmyleswhite]

Could we explicitly iterate over columns? This kind of iteration is something I'd like to remove from DataFrames.

[garborg]

Definitely, but I'm still fuzzy on what functions should and shouldn't access. Like this?

        for n in names(x)
            d[n] = rep(x[n], times, each)
        end

[kmsquire]

I was going to suggest using eachcol, but I just realized that it doesn't give the column names.

What about giving eachcol the same behavior as the default iterator (i.e., returning a tuple of (colname, coldata)), and deprecating the default iterator?

[johnmyleswhite]

I like Kevin's suggestion.

[garborg]

I implemented @kmsquire's suggestion for eachcol.

@johnmyleswhite Natural joins are now being deprecated. And join is now a wrapper here, to avoid having a union-typed argument to the main function -- it's 1% slower for big data sets, 20% slower for little data sets, but I left in in for now so you could see what I did (I don't know of any other way because you can't have two methods that vary only by their kwargs).

[garborg]

I switched join from being a wrapper back to the original form to make it ready to merge. Seems good to me.

[johnmyleswhite]

Let's merge this.

One way we could handle the dispatch issue that prompted your wrapping is to make on a positional argument, which is reasonable now that it's required.

[johnmyleswhite]

Thanks for doing this, Sean! And thanks for slugging through all my comments.

[garborg]

Of course!

[nalimilan]

Nice addition indeed!