Generally allow functions to fall through to evalmapping

[CiaranOMara]

When I set up a plot in Gadfly, I typically need to map data from a struct, I found that I can extend Gadfly.evalmaping to avoid writing wrappers.

using Gadfly
using KernelDensity

# Provide Gadfly with the functions to handle UnivariateKDE as a data source.
Gadfly.evalmapping(source::UnivariateKDE, arg::Function) = arg(source)
Gadfly.evalmapping(source::UnivariateKDE, arg::Symbol) = getproperty(source, arg)

data = rand(100)

plot(
    kde(data),
    x = (s)->collect(s.x),
    y = :density,
    Geom.line,
)

However, not all geometries can handle mappings of functions.

function maximum_density(d::UnivariateKDE)
    (max, i) = findmax(d.density)
    return d.x[i]
end

plot(
    kde(data),
    x = (source)->collect(source.x),
    y = :density,
    xintercept = maximum_density, # <- this is what I'd like to be able to do
    # xintercept = [maximum_density(kde(data))],
    Geom.line,
    Geom.vline
)

While I know that in this case, it is trivial to evaluate kde(data) in an outside scope, its result is not always something that I'd want to keep in that scope.

[Mattriks]

I can make the above example work like so:

using Gadfly, Gadfly.Stat.KernelDensity

Gadfly.evalmapping(source::UnivariateKDE, arg::Function) = arg(source)
Gadfly.evalmapping(source::UnivariateKDE, arg::Symbol) = getfield(source, arg)

maximum_density(d::UnivariateKDE) = [d.x[argmax(d.density)]]

data = randn(100)

plot(kde(data), x=:x, y=:density, Geom.line, xintercept=maximum_density, Geom.vline(color="blue"))

At the moment I think the two evalmapping functions here are a bit specific for Gadfly, but I welcome other people's input. It would be good to put an example of this in the Gadly docs (perhaps in a section Advanced Gadfly - Plotting custom types).

[bjarthur]

let's definitely document this. that's pretty cool!