Color by column

[jariji]

Can we do this less verbosely?

let n = 50, wide = StructArray(id=1:n, x=rand(n), y1=rand(n), y2=rand(n))
    fig = Figure()
    ax = Axis(fig[1,1])
    colors = Makie.wong_colors()
    scatter!(ax, FPlot(wide, (@o _.x), (@o _.y1)); color=colors[1])
    scatter!(ax, FPlot(wide, (@o _.x), (@o _.y2)); color=colors[2])
    fig
end

[jariji]

Maybe axislegend(ax) could figure out good labels from the y expressions too.

[aplavin]

I sometimes do similar stuff indeed, writing it as

fplt = FPlot(wide, (@o _.x))
scatter!(@insert fplt[2] = @o _.y1)
scatter!(@insert fplt[2] = @o _.y2)

or (more commonly)

fplt = FPlot(data, (@o _.x))
for i in 1:5
    scatter!(@insert fplt[2] = @o _.y[i])  # y in each row is a 5-element array
end

Not sure how a better interface would look like...

[aplavin]

Oh, that's what comes to mind:

FPlot(wide, (@o _.x), (@o _.y1 _.y2))  # for several independent columns
FPlot(data, (@o _.x), (@o _.y[∗]))  # for array/tuple/... elements

The full semantics need to be carefully thought of, as this would be the first departure from "one FPlot plot call = one regular Makie plot call (just with cleverely forwarded args)".

[aplavin]

Currently, calling scatter!(fplot) doesn't cycle color or other attributes – that's why you had to specify color=... manually in the calls in the first post.
This is a Makie bug I reported some time ago: MakieOrg/Makie.jl#4353.

While cycling doesn't work, it doesn't make much sense to implement stuff like scatter(FPlot(wide, (@o _.x), (@o _.y1 _.y2))). I want it to just reuse the regular Makie cycling functionality and not reimplement it myself.
But assuming that issue gets fixed at some point, such interface would be nice indeed!