Dynamic PlotList recipe which plots arbitrary arrays of PlotSpec

[asinghvi17]

Description

As discussed in MakieCon 2023 Day 1...will add an example in a bit.

This currently takes the form of a recipe, but we could make it a normal function which is just passed an Observable list of PlotSpec.

fig = Figure()
ax = Axis(fig[1, 1])

pl = plotlist!(
    ax,
    [PlotSpec{Lines}(0..1, sin.(0:0.01:1); color = :blue), PlotSpec{Heatmap}(0..1, 0..1, Makie.peaks(); transformation = (; translation = Vec3f(0, 0, -1)))]
)

fig

pl[1][] = [PlotSpec{Lines}(0..1, sin.(0:0.01:1); color = :red), PlotSpec{Heatmap}(0..1, 0..1, Makie.peaks(); transformation = (; translation = Vec3f(0, 0, -1)))]

fig

pl[1][] = [
    PlotSpec{Lines}(0..1, sin.(0:0.01:1); color = Makie.resample_cmap(:viridis, 101)), 
    PlotSpec{Surface}(0..1, 0..1, Makie.peaks(); transformation = (; translation = Vec3f(0, 0, -1))),
    PlotSpec{Poly}(Rect2f(0.45, 0.45, 0.1, 0.1)),
]

fig

pl[1][] = [
    PlotSpec{Surface}(0..1, 0..1, Makie.peaks(); colormap = :viridis, transformation = (; translation = Vec3f(0, 0, -1))),
]

fig

Type of change

Delete options that do not apply:

• New feature (non-breaking change which adds functionality)

Checklist

• Added an entry in NEWS.md (for new features and breaking changes)
• Added or changed relevant sections in the documentation
• Added unit tests for new algorithms, conversion methods, etc.
• Added reference image tests for new plotting functions, recipes, visual options, etc.

[asinghvi17]

This is functionally complete, but please try whatever you think might break it and let me know!

I think we might also want to add eg palette support but that would need a refactor of the axis code.

[MakieBot]

Compile Times benchmark

Note, that these numbers may fluctuate on the CI servers, so take them with a grain of salt. All benchmark results are based on the mean time and negative percent mean faster than the base branch. Note, that GLMakie + WGLMakie run on an emulated GPU, so the runtime benchmark is much slower. Results are from running:

using_time = @ctime using Backend
# Compile time
create_time = @ctime fig = scatter(1:4; color=1:4, colormap=:turbo, markersize=20, visible=true)
display_time = @ctime Makie.colorbuffer(display(fig))
# Runtime
create_time = @benchmark fig = scatter(1:4; color=1:4, colormap=:turbo, markersize=20, visible=true)
display_time = @benchmark Makie.colorbuffer(display(fig))

	using	create	display	create	display
GLMakie	13.64s (13.36, 13.95) 0.20+-	1.00s (0.96, 1.03) 0.02+-	1.44s (1.39, 1.52) 0.05+-	12.29ms (12.08, 12.77) 0.28+-	167.87ms (165.04, 170.30) 1.86+-
master	13.32s (13.05, 13.55) 0.18+-	1.08s (1.06, 1.09) 0.01+-	1.44s (1.42, 1.47) 0.02+-	12.10ms (11.84, 12.55) 0.23+-	167.01ms (164.43, 169.75) 2.17+-
evaluation	+2.32%, 0.32s slower X (1.63d, 0.01p, 0.19std)	-7.93%, -0.08s faster✅ (-4.75d, 0.00p, 0.02std)	-0.05%, -0.0s invariant (-0.02d, 0.97p, 0.03std)	+1.56%, 0.19ms invariant (0.75d, 0.19p, 0.25std)	+0.51%, 0.86ms invariant (0.43d, 0.44p, 2.01std)
CairoMakie	10.43s (10.32, 10.60) 0.10+-	855.66ms (837.47, 890.28) 17.99+-	572.31ms (564.51, 586.59) 7.46+-	10.36ms (10.25, 10.52) 0.11+-	5.43ms (5.37, 5.47) 0.04+-
master	10.24s (10.15, 10.37) 0.07+-	911.13ms (890.73, 931.88) 12.87+-	539.36ms (529.19, 547.91) 5.93+-	10.20ms (10.03, 10.36) 0.13+-	6.05ms (5.97, 6.18) 0.08+-
evaluation	+1.84%, 0.19s slower X (2.18d, 0.00p, 0.09std)	-6.48%, -55.48ms faster✅ (-3.55d, 0.00p, 15.43std)	+5.76%, 32.95ms slower❌ (4.89d, 0.00p, 6.69std)	+1.55%, 0.16ms slower X (1.38d, 0.02p, 0.12std)	-11.50%, -0.62ms faster✅ (-10.07d, 0.00p, 0.06std)
WGLMakie	13.82s (13.59, 14.17) 0.18+-	1.08s (1.04, 1.10) 0.02+-	12.62s (12.36, 13.02) 0.23+-	17.22ms (15.30, 24.77) 3.35+-	1006.94ms (953.10, 1062.52) 37.73+-
master	13.55s (13.44, 13.68) 0.08+-	1.14s (1.06, 1.20) 0.05+-	12.66s (12.41, 12.95) 0.19+-	15.33ms (14.44, 16.28) 0.64+-	998.94ms (948.45, 1064.92) 41.76+-
evaluation	+1.99%, 0.28s slower X (1.96d, 0.01p, 0.13std)	-5.53%, -0.06s faster✅ (-1.42d, 0.03p, 0.04std)	-0.36%, -0.05s invariant (-0.22d, 0.69p, 0.21std)	+10.98%, 1.89ms noisy🤷‍♀️ (0.78d, 0.19p, 1.99std)	+0.79%, 8.0ms invariant (0.20d, 0.71p, 39.74std)

[pepijndevos]

This currently takes the form of a recipe, but we could make it a normal function which is just passed an Observable list of PlotSpec.

I tried this out and immediately ran into problems where things like color cycling don't work because it's a recipe. The below example produces only black lines, instead of cycling through colors if you'd call lines! twice.

plotlist([PlotSpec{Lines}(0:5, 0:5), PlotSpec{Lines}(0:5, ones(6))])

[pepijndevos]

I tried this some more, because I really need this functionality.

First I tried to make it not a recipe, but got weird errors.

Then I tried just using the recipe with explicit attributes, which "works" in that it doesn't error, but rather leaves dangling plot elements (that don't resolve themselves if you change back the Observed) so once you remove a plot it's just broken.

It seems there was also an off-by-one error in length(old_plotspec_types):-1:(length(plotspec_types)+1) on line 64

My recipe is something like

function Makie.plot!(plt::SignalPlot{<:Tuple{<:AbstractArray}})
    Main.plt = plt
    ms = plt[1]
    specs = lift(ms, values(plt.attributes)...) do ms, attrs...
        pairs = zip(keys(plt.attributes), attrs)
        [PlotSpec{SignalPlot}(item; pairs...) for item in ms]
    end
    plotlist!(plt, specs)
end

[staticfloat]

I tried re-working the fundamental machinery in here to use a Dict instead of an array, so that we don't have to track order of plots (since as far as I can tell, order doesn't matter):

function update_plot_with_plotspec!(cached_plot::AbstractPlot, spec::PlotSpec)
    # Update args in cached `input_args` list
    for arg_index in eachindex(spec.args)
        cached_plot.input_args[arg_index].val = spec.args[arg_index]
    end
    # Update attributes
    for (attribute, new_value) in pairs(spec.kwargs)
        update_attributes_inplace!(cached_plot, attribute, new_value)
    end
    
    # notify all args, without broadcasting
    map(notify, cached_plot.input_args)
    # notify all accessed kwargs
    map(_notify!, (getindex(cached_plot, attr) for attr in keys(spec.kwargs)))
end

function Makie.plot!(p::PlotList{<: Tuple{<: AbstractArray{<: PlotSpec}}})
    # Cache plots here so that we aren't re-creating plots every time;
    # if a plot still exists from last time, update it accordingly.
    # If the plot is removed from `plotspecs`, we'll delete it from here
    # and re-create it if it ever returns.
    cached_plots = IdDict{Any,AbstractPlot}()

    lift(p[1]) do plotspecs
        Main.p = p
        println()
        println("----------------------------------------")
        plots_to_delete = Base.IdSet()
        push!.((plots_to_delete,), collect(keys(cached_plots))))

        # First, scan through `plotspecs`, creating or updating all plots
        # specified by `plotspecs`, marking previously-extant plots as not
        # to be deleted via `plots_to_delete`
        for plotspec in plotspecs
            if !haskey(cached_plots, plotspec.args[1])
                # Create new plot, store it into our `cached_plots` dictionary
                cached_plots[plotspec] = plot!(p,
                    typeof(plotspec).parameters[1],
                    Attributes(plotspec.kwargs),
                    plotspec.args...,
                )
            else
                update_plot_with_plotspec!(cached_plots[plotspec], plotspec)
            end

            # Because this plotspec appeared in `plotspecs`, it should not
            # be deleted, so remove it from our deletion queue.
            delete!(plots_to_delete, plotspec)
        end

        # Next, delete all plots still extant in in `plots_to_delete`
        for plotspec in plots_to_delete
            idx = findfirst(==(cached_plots[plotspec]), p.plots)
            if idx !== nothing
                deleteat!(p.plots, idx)
            end
            delete!(cached_plots, plotspec)
        end
    end
end

But for some reason I can't figure out, updates don't flow through the system; my plots remain static.

[SimonDanisch]

While reviewing & figuring out your approaches, I've rewritten the implementation.
Caching needs to be based on the argument + attribute types of the plot spec, because that's what needs to be the same to be "updatable". Also, I now plot directly into the scene and remove plots from that scene, because if we add the plots nested inside the recipe plot type, it gets a bit more complicated to delete!.

I also simplified the updating mechanism for now, removing nested attributes, which aren't that well supported within Makie yet anyways and shouldn't be needed for most plots.
The transformation argument is maybe the biggest exception, but that seems to be broken right now?

Anyways, I also played around with a simple API to make it a bit nicer to use:

import Makie.PlotspecApi as P
plots[] = [
    P.image(0 .. 1, 0 .. 1, Makie.peaks()),
    P.poly(Rect2f(0.45, 0.45, 0.1, 0.1)),
    P.lines(0 .. 1, sin.(0:0.01:1); linewidth=10, color=Makie.resample_cmap(:viridis, 101)),
]

[SimonDanisch]

@pepijndevos not sure what to do about the cycling, since that needs the axis.
I've been wanting to make cycling part of the scene, which should fix this... Otherwise, I guess we need to manually re-implement the cycling inside the recipe, not sure how to do this correctly otherwise.
edit: I mean correctly inside a recipe... Of course as you suggested we could directly overload plot!(ax, PlotSpec[...]), which maybe isn't such a bad idea, considering it already directly plots into the scene unlike a "reasonable" recipe.

[staticfloat]

Yes, I agree. The PlotList functionality feels much more "top-level" than most recipes; I would expect other recipes to fit inside of it, so having the PlotList plotting directly interact with axes makes sense to me.

[SimonDanisch]

This isn't 100% ready, but since sd/beta-20 had some problems with the PlotSpec implementation which are cleaned up in here, I'm going to merge this and polish the rest in sd/beta-20.
For sd/beta-20, the remaining work is also just polish and testing, so I might just do it in one go.