plot! fails for time-based DimArrays

[thirtysixbananas]

Works:

y1 = DimArray(rand(Float64, 11), (times=1:11,),)
y2 = DimArray(rand(Float64, 11), (times=1:11,),)
f = Figure()
ax = Axis(f[1,1])
plot!(ax, y1)  # Scatter{Tuple{Vector{Point{2, Float64}}}}
plot!(ax, y2)  # Scatter{Tuple{Vector{Point{2, Float64}}}}

Doesn't work:

y1 = DimArray(
    rand(Float64, 11),
    (times=DateTime("2026-01-01"):Minute(1):DateTime("2026-01-01T00:10:00"),),
)
y2 = DimArray(
    rand(Float64, 11),
    (times=DateTime("2026-01-01"):Minute(1):DateTime("2026-01-01T00:10:00"),),
)
f = Figure()
ax = Axis(f[1,1])
plot!(ax, y1)  # Scatter{Tuple{Vector{Point{2, Float64}}}}
plot!(ax, y2)  # ERROR: ArgumentError:
    Conversion failed for Scatter (With conversion trait PointBased()) with args:
        Tuple{DimVector{Float64, Tuple{Dim{:times, DimensionalData.Dimensions.Lookups.Sampled{DateTime, StepRange{DateTime, Minute}, DimensionalData.Dimensions.Lookups.ForwardOrdered, DimensionalData.Dimensions.Lookups.Regular{Minute}, DimensionalData.Dimensions.Lookups.Points, DimensionalData.Dimensions.Lookups.NoMetadata}}}, Tuple{}, Vector{Float64}, DimensionalData.NoName, DimensionalData.Dimensions.Lookups.NoMetadata}} 
    Got converted to: Tuple{StepRange{DateTime, Minute}, Vector{Float64}}
    Scatter requires to convert to argument types Tuple{AbstractVector{<:Union{Point2, Point3}}}, which convert_arguments didn't succeed in.
    To fix this overload convert_arguments(P, args...) for Scatter or PointBased() and return an object of type Tuple{AbstractVector{<:Union{Point2, Point3}}}.`

Stacktrace:
 [1] argument_error(PTrait::PointBased, P::Type, args::Tuple{…}, user_kw::Dict{…}, converted::Tuple{…})
   @ Makie ~/.julia/packages/Makie/Vn16E/src/compute-plots.jl:725
 [2] (Scatter)(user_args::Tuple{DimVector{…}}, user_attributes::Dict{Symbol, Any})
   @ Makie ~/.julia/packages/Makie/Vn16E/src/compute-plots.jl:773
 [3] _create_plot!(::Function, ::Dict{…}, ::Axis, ::DimVector{…})
   @ Makie ~/.julia/packages/Makie/Vn16E/src/figureplotting.jl:403
 [4] #scatter!#62
   @ ~/.julia/packages/Makie/Vn16E/src/recipes.jl:534 [inlined]
 [5] scatter!
   @ ~/.julia/packages/Makie/Vn16E/src/recipes.jl:532 [inlined]
 [6] plot!(ax::Axis, dd::DimVector{…})
   @ DimensionalDataMakieExt ~/.julia/packages/DimensionalData/FWnw9/ext/DimensionalDataMakieExt.jl:395

As a sanity check I tried plotting with DateTime on the x-axis and it works, so the above problem is specific to DD, not Makie:

ts = DateTime("2026-01-01"):Minute(1):DateTime("2026-01-01T00:10:00")
y1 = rand(Float64, 11)
y2 = rand(Float64, 11)
f = Figure()
ax = Axis(f[1,1])
plot!(ax, ts, y1)  # Scatter{Tuple{Vector{Point{2, Float64}}}}
plot!(ax, ts, y2)  # Scatter{Tuple{Vector{Point{2, Float64}}}}

EDIT:

Calls here so this is a simpler path to reproduce:

Makie.scatter!(ax, y1)

Does this check before throwing: https://github.com/MakieOrg/Makie.jl/blob/4a6728146ca4d5b3d20ee9244300788243697c79/Makie/src/compute-plots.jl#L780

In both cases, PTrait = PointBased(), P=Scatter.
However, in the first plot!

converted=(Point{2, Float64}[[6.39029088e13, 0.5836928495029402], [6.390290886e13, 0.384911441829076], [6.390290892e13, 0.1530611230665475], [6.390290898e13, 0.416598093899766], [6.390290904e13, 0.1913697055696404], [6.39029091e13, 0.07931893959562897], [6.390290916e13, 0.5810192663574537], [6.390290922e13, 0.4698605070411458], [6.390290928e13, 0.2241905689270648], [6.390290934e13, 0.25124129374053417], [6.39029094e13, 0.9313477533438733]],)
got_converted(P, PTrait, converted)  # true and succeeds

but in the second

converted=(DateTime("2026-01-01T00:00:00"):Minute(1):DateTime("2026-01-01T00:10:00"), [0.646820062789859, 0.5319466876655881, 0.36043739283964304, 0.8572681488599282, 0.4556710116927154, 0.6731147331017542, 0.08469344244360877, 0.4920890801070882, 0.9769361150754856, 0.9531765290732815, 0.3459698165949199])
got_converted(P, PTrait, converted)  # false and throws

PTrait = Makie.PointBased()
P = Makie.Scatter
converted1=(Point{2, Float64}[[6.39029088e13, 0.5836928495029402], [6.390290886e13, 0.384911441829076], [6.390290892e13, 0.1530611230665475], [6.390290898e13, 0.416598093899766], [6.390290904e13, 0.1913697055696404], [6.39029091e13, 0.07931893959562897], [6.390290916e13, 0.5810192663574537], [6.390290922e13, 0.4698605070411458], [6.390290928e13, 0.2241905689270648], [6.390290934e13, 0.25124129374053417], [6.39029094e13, 0.9313477533438733]],)
converted2=(DateTime("2026-01-01T00:00:00"):Minute(1):DateTime("2026-01-01T00:10:00"), [0.646820062789859, 0.5319466876655881, 0.36043739283964304, 0.8572681488599282, 0.4556710116927154, 0.6731147331017542, 0.08469344244360877, 0.4920890801070882, 0.9769361150754856, 0.9531765290732815, 0.3459698165949199])
Makie.got_converted(P, PTrait, converted1)  # true
Makie.got_converted(P, PTrait, converted2)  # false

EDIT2:

On entering this function https://github.com/MakieOrg/Makie.jl/blob/4a6728146ca4d5b3d20ee9244300788243697c79/Makie/src/compute-plots.jl#L746
in boths cases, user_args: ([0.646820062789859, 0.5319466876655881, 0.36043739283964304, 0.8572681488599282, 0.4556710116927154, 0.6731147331017542, 0.08469344244360877, 0.4920890801070882, 0.9769361150754856, 0.9531765290732815, 0.3459698165949199],)
However user_attributes in the first case is

Dict{Symbol, Any}(
    :dim_conversions => Makie.DimConversions(
        (
            Observable{Union{Nothing, Makie.AbstractDimConversion}}(nothing),
            Observable{Union{Nothing, Makie.AbstractDimConversion}}(nothing),
            Observable{Union{Nothing, Makie.AbstractDimConversion}}(nothing),
        ),
    ),
)

whereas in the second case its

Dict{Symbol, Any}(
    :dim_conversions => Makie.DimConversions(
        (
            Observable{Union{Nothing, Makie.AbstractDimConversion}}(Makie.DateTimeConversion(Observable(DateTime))),
            Observable{Union{Nothing, Makie.AbstractDimConversion}}(Makie.NoDimConversion()),
            Observable{Union{Nothing, Makie.AbstractDimConversion}}(nothing),
        ),
    ),
)

EDIT3:

So the function above is called with different arguments each time. That comes from here: https://github.com/MakieOrg/Makie.jl/blob/4a6728146ca4d5b3d20ee9244300788243697c79/Makie/src/figureplotting.jl#L401-L403

On line 403 you see the call to the function above, and on line 401 you can see how the arguments are generated. They are different the first and the second time, because ax is a stateful Axis object and get_conversions(ax) returns different things each time. We can see that by doing:

f = Figure()
ax = Axis(f[1,1])
Makie.get_conversions(ax)  # Makie.DimConversions((Observable{Union{Nothing, Makie.AbstractDimConversion}}(nothing), Observable{Union{Nothing, Makie.AbstractDimConversion}}(nothing), Observable{Union{Nothing, Makie.AbstractDimConversion}}(nothing)))
plot!(ax, y1)
Makie.get_conversions(ax)  # Makie.DimConversions((Observable{Union{Nothing, Makie.AbstractDimConversion}}(Makie.DateTimeConversion(Observable(DateTime))), Observable{Union{Nothing, Makie.AbstractDimConversion}}(Makie.NoDimConversion()), Observable{Union{Nothing, Makie.AbstractDimConversion}}(nothing)))
plot!(ax, y2)

EDIT4:

Ok dim conversions being different in first and second calling of plot! isn't the reason for this problem, because the simple case:

ts = DateTime("2026-01-01"):Minute(1):DateTime("2026-01-01T00:10:00")
y1 = rand(Float64, 11)
y2 = rand(Float64, 11)
f = Figure()
ax = Axis(f[1,1])
plot!(ax, ts, y1)  # Scatter{Tuple{Vector{Point{2, Float64}}}}
plot!(ax, ts, y2)  # Scatter{Tuple{Vector{Point{2, Float64}}}}

Has the same behavior.

Help? I have no idea what I'm doing here.

[rafaqz]

Makie conversioms and plot! have had numerous changes and bugs in the last year for complicated lookups like time or units. Often actally bugs in Makie itself, but also us just not keeping up.

Look through the other Makie related issues here for background.

I dont personally have knowledge or time to fix them at the moment as Makie is pretry complicated. Maybe claude code could help you with a PR fix if you get it to read both packages.

[thirtysixbananas]

Thank you for coming back.

Inspired by this code I put together this function which does what I want:

DDExt = Base.get_extension(DimensionalData, :DimensionalDataMakieExt)
function plotdim2(A::DimensionalData.AbstractDimMatrix; labeldim=nothing)
    f = Figure()
    Axis(f[1, 1])

    categoricaldim = DDExt._categorical_or_dependent(A, labeldim)
    isnothing(categoricaldim) && throw(ArgumentError("No dimensions have Categorical lookups")) # This should never happen
    otherdim = only(otherdims(A, categoricaldim))
    xs = parent(lookup(A, otherdim)) |> DDExt.get_number_version
    ks = string.(dims(A, 2))
    scatters = []
    for (n,k) in enumerate(ks)
        ys = parent(permutedims(A, (categoricaldim, otherdim)))[n,:]
        s = scatter!(xs, ys; label=k)
        push!(scatters, s)
    end
    Legend(f[1,2], scatters, ks)
    f
end

But it's not really nice because it doesn't plug into the Makie API, it's just a function.

But yeah, Makie is quite complex.

[DylanMMarques]

This is an issue on how Makie handles dimension conversions with custom recipes. There is a PR on Makie that will hopefully fix this. In the meanwhile, you can use the work around available here:
#1121