@df : passing a symbol through a variable throws an error about statsplots.jl HOT 8 CLOSED

I like the interpolation, I don't think we need consistency with DataFramesMeta on this.

from statsplots.jl.

What is missing is that macro are called before the code is executed, so the macro cannot know that x will be a symbol. @df is inspired by DataFramesMeta's @with (see here) and distinguishes symbols syntactically and cannot know that x will be a symbol at parse time. If you think this application is common enough, I can maybe implement the _I_(expr) notation (as they do in @with). like:

x = :A
@df smallDf histogram(_I_(x))

But honestly I'm not a huge fan of this syntax. Maybe:

x = :A
@df smallDf histogram($x)

would make more sense? It seems more intuitive to me, not sure why DataFramesMeta went with _I_(expr) which is a bit awkward to type.

from statsplots.jl.

Thanks ! Of course that makes sense. I encountered it while changing some old code that was using the deprecated syntax plot(smallDf,x). I think that might be a nice feature to add if more people agree -- as for the syntax, I guess being coherent with DataFrameMeta is a good thing.

from statsplots.jl.

@mkborregaard what's your take for a syntax that allows "interpolating" symbols in a @df call? _I_(expr) like DataFramesMeta, $expr (which is what I would try as a user) or yet another thing?

from statsplots.jl.

I actually started wondering whether we should just always use cols instead. Now we have cols(v::AbstractArray{Int64}) mean the hcat of those columns, but maybe we could also have cols(v::AbstractArray{Symbol}) to select the columns by name, as well as a cols(v::Int64) or cols(v::Symbol) if it's just one column but we want to choose it by index rather than name, or we want to chose it by name but the name is a variable. Maybe the 1 column version could be renamed col, though I probably prefer to just have one function with different methods for ints/symbols and arrays of ints/symbols.

The rule would be:

• If it's a symol that you type, use the symbol
• otherwise, use cols to signal to the macro that the content represent one or more columns

from statsplots.jl.

I think that suggestion has the advantage that the interpolation after a macro feels a little magical, and enough slightly differing uses of $ after a macro (i.e. the suggested use is different from that used in BenchmarkTools) could create slightly fuzzy interface. Though cols is clearly ours, and at the cost of being slightly clumsy and wordy should be quite clear to understand.

from statsplots.jl.

@romainFr If you want you can try #120 and see if it addresses your use case.

from statsplots.jl.

@piever It does, thanks !

from statsplots.jl.