Support Plotting

[dlfivefifty]

We should add support for plotting using grids/transforms. Anyone have an opinion on whether to support:

👍 Makie
🎉 Plots
🚀 Both

Please vote below. @jagot @TSGut

Personally I think Makie will be really beneficial for complicated geometries like disks/triangles/etc. but perhaps Plots.jl can do this too...

[jagot]

I actually use neither, I use PyPlot.jl instead; as long as its not a required dependency, i.e. plotting is supported via Requires.jl, I'm fine with either. The other option is of course RecipesBase.jl, for which there seems to be no Makie equivalent yet.

[dlfivefifty]

Actually meant:

👍 AbstractPlotting.jl (to support Makie.jl)
🎉 RecipesBase.jl (to support Plots.jl)
🚀 Both

Requires.jl might be a good solution.

[jagot]

RecipesBase.jl is lightweight, AbstractPlotting.jl not so much, so I would vote for Requires.jl 👀

[TSGut]

I feel like in most of the basic uses it's already fairly straightforward to make Plots.jl do the job. For higher dimensional more complicated domains where this would be a pain to manually set up, Makie is probably superior. So I somewhat hesitantly vote AbstractPlotting / Makie. Hesitantly because it lacks a lot of convenience features, e.g. exporting as vector graphics is as far as I know still not supported.

[dlfivefifty]

RecipesBase.jl is lightweight, AbstractPlotting.jl not so much

Yes I agree... I find the existing of two different recipe formats pretty annoying...

Perhaps best to just make a new package ContinuumArraysAbstractPlotting.jl which is loaded manually for the time being... my understanding of Requires.jl is it is then easy to to have this loaded automatically.

[dlfivefifty]

Or... since the only packages that will use 3D are downstream (HarmonicOrthogonalPolynomials.jl and MultivariateOrthogonalPolynomials.jl) just put the dependency there

[dlfivefifty]

Annoyingly, Makie.jl (via WGLMakie.jl) doesn't seem to be very reliable in VSCode, and is very slow...

And regular Makie.jl + InfiniteArrays.jl triggers an obscure Julia type inference bug MakieOrg/Makie.jl#933

[TSGut]

Yeah WGLMakie even seems to state that it doesn't support VSCode yet on the documentation. Out of curiosity, is there a reason not to go for GLMakie? Well, I didn't see you already addressed that! GLMakie is the only backend that ever really worked for me, so if that's leading to serious bugs I don't think the other backends will be the solution.

[dlfivefifty]

OK we can use Plots.jl with the plotly() backend: the following gives a plot on a disk:

julia> plotly()
Plots.PlotlyBackend()

julia> r = range(0,1;length=100);

julia> x = r' .* cospi.(2r);

julia> y = r' .* sinpi.(2r);

julia> surface(x, y, exp.(x .+ cos.(y)))

Let's do this now.

(5 years ago I used to do live demos with Makie.jl of PDEs on a disk... annoying that the situation has seemed to have gone backwards...)

[dlfivefifty]

OK this now works:

using MultivariateOrthogonalPolynomials, Plots
plotly()
Z = Zernike()
xy = axes(Z,1)
x,y = first.(xy),last.(xy)
u = Z * (Z \ @.(cos(10x*y)))
surface(u)