WIP: Change how the colorbar axis works

[gustaphe]

Issue #68

Tries to unify how we treat the axes and axis labels, and apply that to the c axis.

In this state, it fails when applying a plot! call with line_z or marker_z on top of one with fill_z. And it kind of weirds out with implicit fill_z (like plot(x, y, z; st=:surface)). But it correctly saves and checks cbar units for lines and markers, as well as for only fill_z.

using UnitfulRecipes
using Plots
using Unitful
plotly()

gaussian(x, y, x0, y0, wx, wy) = exp(-((x-x0)^2/wx^2 + (y-y0)^2/wy^2))

x = (-2:0.1:2)u"mm";
y = (-2:0.1:2)u"mm";
z = 1u"W/m^2"*gaussian.(x', y, 0u"mm", 0u"mm", 0.5u"mm", 1u"mm");
c = 300u"K" .+ 4u"K/mm".*(x' .+ y);
println(size(c))

plot(x, y, z;
     fill_z=c,
     st=:surface,
     colorbar_title="T"
    )

[gustaphe]

Only for development purposes

[gustaphe]

Much of this code needed to be repeated for the c axis, so I factored it out.

[gustaphe]

Need to run a single pass over all three color arguments, to combine into one axis.

[gustaphe]

I think it's a limitation in Plots that it's attr[:plot_object][sp][:colorbar_title] rather than attr[:plot_object][sp][:caxis][:guide], but it is what it is.

[gustaphe]

I don't think we need this recipe if x, y and z are correctly identified as type targets for themselves.

[gustaphe]

Testing:

using UnitfulRecipes
using Test
using Plots
using Unitful
#plotlyjs()
#gr()
pyplot()

default(; label=nothing)

x = randn(5)u"m"
t = randn(5)u"s"
T = randn(5)u"K"
E = 100 * randn(5)u"J"

pl1 = plot(x, t; marker_z=T, st=:scatter, xguide="x", yguide="t", colorbartitle="T")
pl2 = plot(x, t; line_z=T, st=:line, xguide="x", yguide="t", colorbartitle="T")
pl3 = plot(x, t; line_z=T, st=:line)
#plot!(pl3, x, t; marker_z=E, st=:scatter, colorbartitle="T or E?") # Rightfully fails
plot!(pl3, x, t; marker_z=T, st=:scatter, xguide="x", yguide="t", colorbartitle="T")

gaussian(x, y, x0, y0, wx, wy) = exp(-((x - x0)^2 / wx^2 + (y - y0)^2 / wy^2))

x = (-2:0.1:2)u"mm";
y = (-2:0.1:2)u"mm";
z = 1u"W/m^2" * gaussian.(x', y, 0u"mm", 0u"mm", 0.5u"mm", 1u"mm");
c = 300u"K" .+ 4u"K/mm" .* (x' .+ y);
println(size(c))

pl4 = plot(
    x, y, z; fill_z=c, st=:surface, xguide="x", yguide="y", zguide="I", colorbar_title="T"
)
pl5 = plot(x, y, z; fill_z=c, st=:surface)
plot!(
    pl5,
    randn(5)u"mm",
    randn(5)u"mm",
    fill(1.5u"W/m^2", length(x));
    xguide="x",
    yguide="y",
    zguide="I",
    colorbar_title="T",
    st=:line,
    line_z=150u"K" .+ 20 * randn(size(x))u"K",
)

savefig.([pl1, pl2, pl3, pl4, pl5], ["pl1.png", "pl2.png", "pl3.png", "pl4.png", "pl5.png"])

#plot(pl1, pl2, pl3, pl4, pl5)

To me, this PR can do what you want it to do. But it feels slightly unpolished: in plotlyjs only pl4 gets a colorbar title at all, and the plots with multiple series get doubled colorbar ticks. On the other hand those seem like upstream bugs so maybe that is acceptable.

If there are no complaints I'll just write tests and docs.

@dmoored4

[gustaphe]

Oh no, there are some regressions when running tests. I'll look at it.