Layering two sound plots causes an unexpected error

[NightMachinery]

Here is the code:

using Base.MathConstants
using Statistics, DataFrames, Gadfly, Distributions

function poisson(lan = 15, n = 200)
        p = Float64[e^-lan]
        pc = Float64[p[end]]
        sizehint!(p, n + 10)
        sizehint!(pc, n + 10)
        for i = 1:n-1
                push!(p, (lan / i) * p[end])
                push!(pc, pc[end] + p[end])
        end
        return p, pc
end

n = 200
lans = [5:5:155;]
@time a = [
        DataFrame(lambda = lan, i = [0:n-1;], cdf = poisson(lan, n)[2])
        for lan in lans
]

begin
        set_default_plot_size(30cm, 8cm)
        l1 = layer(vcat(a...), x = :i, y = :cdf, color = :lambda, Geom.line) # , Coord.cartesian(ymin=0, ymax=1)

        l2 = layer(y=[x -> cdf(Poisson(lan), x) for lan in lans], xmin=[0], xmax=[n - 1], Stat.func(num_samples=50), Geom.point)

        p = plot(l1,l2) #hstack(p1, p2)
        display(p)
end

And here is the error:

ERROR: LoadError: MethodError: no method matching -(::String, ::String)
Stacktrace:
 [1] optimize_ticks_typed(::String, ::String, ::Bool, ::Array{Tuple{Float64,Float64},1}, ::Int64, ::Int64, ::Int64, ::Float64, ::Float64, ::Float64, ::Float64, ::Bool) at /Users/evar/.julia/packages/Gadfly/cqMtH/src/ticks.jl:73
 [2] #optimize_ticks#25(::Bool, ::Array{Tuple{Float64,Float64},1}, ::Int64, ::Int64, ::Int64, ::Float64, ::Float64, ::Float64, ::Float64, ::Bool, ::typeof(Gadfly.optimize_ticks), ::String, ::String) at /Users/evar/.julia/packages/Gadfly/cqMtH/src/ticks.jl:61
 [3] (::Gadfly.var"#kw##optimize_ticks")(::NamedTuple{(:strict_span,),Tuple{Bool}}, ::typeof(Gadfly.optimize_ticks), ::String, ::String) at ./none:0
 [4] apply_scale(::Gadfly.Scale.ContinuousColorScale, ::Array{Gadfly.Aesthetics,1}, ::Gadfly.Data, ::Vararg{Gadfly.Data,N} where N) at /Users/evar/.julia/packages/Gadfly/cqMtH/src/scale.jl:644
 [5] apply_statistic(::Gadfly.Stat.FunctionStatistic, ::Dict{Symbol,Gadfly.ScaleElement}, ::Gadfly.Coord.Cartesian, ::Gadfly.Aesthetics) at /Users/evar/.julia/packages/Gadfly/cqMtH/src/statistics.jl:1429
 [6] apply_statistics(::Array{Gadfly.StatisticElement,1}, ::Dict{Symbol,Gadfly.ScaleElement}, ::Gadfly.Coord.Cartesian, ::Gadfly.Aesthetics) at /Users/evar/.julia/packages/Gadfly/cqMtH/src/statistics.jl:33
 [7] render_prepare(::Plot) at /Users/evar/.julia/packages/Gadfly/cqMtH/src/Gadfly.jl:697
 [8] render(::Plot) at /Users/evar/.julia/packages/Gadfly/cqMtH/src/Gadfly.jl:755
 [9] draw at /Users/evar/.julia/packages/Gadfly/cqMtH/src/Gadfly.jl:864 [inlined]
 [10] display(::Gadfly.GadflyDisplay, ::MIME{Symbol("text/html")}, ::Plot) at /Users/evar/.julia/packages/Gadfly/cqMtH/src/Gadfly.jl:1069
 [11] display(::Gadfly.GadflyDisplay, ::String, ::Any) at ./multimedia.jl:214
 [12] display(::Gadfly.GadflyDisplay, ::Plot) at /Users/evar/.julia/packages/Gadfly/cqMtH/src/Gadfly.jl:1026
 [13] display(::Any) at ./multimedia.jl:323
 [14] top-level scope at /Users/evar/Base/_Code/uni/stochastic/jo1/p1.jl:33
 [15] include at ./boot.jl:328 [inlined]
 [16] include_relative(::Module, ::String) at ./loading.jl:1105
 [17] include(::Module, ::String) at ./Base.jl:31
 [18] exec_options(::Base.JLOptions) at ./client.jl:287
 [19] _start() at ./client.jl:460
in expression starting at /Users/evar/Base/_Code/uni/stochastic/jo1/p1.jl:26

[Mattriks]

The above plot will work by doing:

p = plot(l1, l2, Scale.color_discrete)

If you do plot(l1) and plot(l2) separately, you can see that plot(l1) is given a continuous color scale (:lambda is interpreted as numeric), but in plot(l2) Gadfly forces a discrete color scale. In plot(l1, l2) this causes a clash of color scales. We do plan to support plots with 1 continuous color scale, and 1 discrete color scale (Roadmap #1385).

For the above example, the question that needs investigation is: can a continuous color scale be allowed for plotting functions e.g. plot(l2, Scale.color_continuous)?

[Mattriks]

plot(l1, l2) will also work if you add color=lans to layer l2. And by adding color=[numbers] to the layer, Scale.color_continuous will work. So is anything else needed here?

[NightMachinery]

@Mattriks I solved my problem after posting the issue, but I think it violates the API expectation that layering plots should just work without additional workarounds. The error messages are unhelpful, as well.

[Mattriks]

What was your solution? I'm unclear what you mean by "the API expectation that layering plots should work without additional workarounds" e..g specifying color as numbers is not a workaround, there is no expectation that color in layer l1 should be the same as in l2? Just trying to clarify what you expect to happen in plot(l1, l2)

[NightMachinery]

@Mattriks I don’t remember too well, but I think I just made the second layer’s data similar to the first (using a dataframe with similar columns and rows).

I think the naive expectation is that plot(l1,l2) should just overlay the two plots, without further settings. For example, in this case, the colors should automatically be switched to a setting that allows the overlay to happen, instead of producing an error.

Also, the solution that you are proposing is not at all hinted by the API; You know the problem is a clash of color scales, but this is because you know the internals of the system. A user who only knows the external API will not know the problem is a clash of colors. So what I am trying to say is, there needs to be either be an internal clash-breaker, or the clash and a possible avenue of trying to fix it need to be delivered to the end user as part if the error message. Of course, I am aware that this is easier said than done, so just close the issue if it’s hard to do.

[Mattriks]

I think a helpful error message would be nice, so I'll investigate.