implement view(::DataFrame, ...) to support broadcasted assignment

[stevengj]

In Julia 0.5, an assignment like a[i] .= ... gets converted by the parser to broadcast!(identity, view(a, i), ...), and moreover the broadcast! loop is automatically merged with broadcast loops from nested "dot" calls (see JuliaLang/julia#17510 and JuliaLang/julia#17546).

It would be nice to support this if a is a DataFrame. My understanding is that it will currently throw an error because view is not implemented for a DataFrame.

I haven't used DataFrames much, but if I understand it correctly then it sounds sufficient to define:

Base.view(a::DataFrame, key) = a[key]
Base.view(a::DataFrame, key, indices...) = view(a[key], indices...)

[quinnj]

Good up for grabs issue.

[stevengj]

Actually, .= might work with dataframes in many cases already because broadcast! was updated to use a dotview function that defaults to getindex (so that it works as-is with things like dictionaries of array and in general anything where getindex can return a mutable object).

[bkamins]

Done in 0.19.0 (+ some patching up pending in #1890)