[BUG] UndefVarError: z_log not defined

[giordano]

I don't have a simple reproducer at the moment, but I do have a backtrace:

ERROR: UndefVarError: `z_log` not defined
Stacktrace:
  [1] gr_draw_heatmap(series::Plots.Series, x::Vector{Float64}, y::Vector{Float64}, z::Matrix{Float64}, clims::Tuple{Float64, Float64})
    @ Plots ~/.julia/packages/Plots/QWCV6/src/backends/gr.jl:2006
  [2] gr_add_series(sp::Plots.Subplot{Plots.GRBackend}, series::Plots.Series)
    @ Plots ~/.julia/packages/Plots/QWCV6/src/backends/gr.jl:1764
  [3] gr_display(sp::Plots.Subplot{Plots.GRBackend}, w::Measures.AbsoluteLength, h::Measures.AbsoluteLength, vp_canvas::Plots.GRViewport{Float64})
    @ Plots ~/.julia/packages/Plots/QWCV6/src/backends/gr.jl:964
  [4] (::Plots.var"#506#507"{Int64, Int64, Plots.GRViewport{Float64}})(sp::Plots.Subplot{Plots.GRBackend})
    @ Plots ~/.julia/packages/Plots/QWCV6/src/backends/gr.jl:688
  [5] foreach(f::Plots.var"#506#507"{Int64, Int64, Plots.GRViewport{Float64}}, itr::Vector{Plots.Subplot})
    @ Base ./abstractarray.jl:3073
  [6] gr_display(plt::Plots.Plot{Plots.GRBackend}, dpi_factor::Float64)
    @ Plots ~/.julia/packages/Plots/QWCV6/src/backends/gr.jl:688
  [7] #545
    @ ~/.julia/packages/Plots/QWCV6/src/backends/gr.jl:2052 [inlined]
  [8] withenv(::Plots.var"#545#546"{Plots.Plot{Plots.GRBackend}, Float64}, ::Pair{String, String}, ::Vararg{Pair{String, String}})
    @ Base ./env.jl:197
  [9] _show(io::IOStream, #unused#::MIME{Symbol("image/png")}, plt::Plots.Plot{Plots.GRBackend})
    @ Plots ~/.julia/packages/Plots/QWCV6/src/backends/gr.jl:2047
 [10] #invokelatest#2
    @ ./essentials.jl:816 [inlined]
 [11] invokelatest
    @ ./essentials.jl:813 [inlined]
 [12] show
    @ ~/.julia/packages/Plots/QWCV6/src/output.jl:232 [inlined]
 [13] #344
    @ ~/.julia/packages/Plots/QWCV6/src/output.jl:6 [inlined]
 [14] open(::Plots.var"#344#345"{Plots.Plot{Plots.GRBackend}}, ::String, ::Vararg{String}; kwargs::Base.Pairs{Symbol, Union{}, Tuple{}, NamedTuple{(), Tuple{}}})
    @ Base ./io.jl:395
 [15] open
    @ ./io.jl:392 [inlined]
 [16] png(plt::Plots.Plot{Plots.GRBackend}, fn::String)
    @ Plots ~/.julia/packages/Plots/QWCV6/src/output.jl:6
 [17] savefig(plt::Plots.Plot{Plots.GRBackend}, fn::String)
    @ Plots ~/.julia/packages/Plots/QWCV6/src/output.jl:149
 [18] savefig(fn::String)
    @ Plots ~/.julia/packages/Plots/QWCV6/src/output.jl:154
 [19] top-level scope
    @ REPL[63]:1

z_log at

Plots.jl/src/backends/gr.jl

Line 2006 in 2023288

z_log, gr_z_normalized_log_scaled(scale, z, clims)

is undefined.

[giordano]

Ah, I finally got a small-contained reproducer:

x = y = collect(-10:0.01f0:10);
fig = heatmap(x, y, hypot.(x', y); colorbar_scale=:log10)

[BeastyBlacksmith]

What would you expect are the limits for the colorbar in this case? I am tempted to just throw an error.

[giordano]

I am tempted to just throw an error.

Why? It works if x, y are Float64, it seems to fail only with Float32, and the offending line has that z_log which is clearly undefined.

[david-macmahon]

There are two problems demonstrated in the reproducer from @giordano . One is taking the log of zero. The other one, which I think is more relevant to this issue, is "z_log not defined". I think the latter problem is actually caused by irregular spacing of the x (and y) values combined with colobar_scale=:log10. Here is a simpler example:

heatmap([1,2,3], [1,2], rand(2,3), colorbar_scale=:log10) # works
heatmap([1,2,4], [1,2], rand(2,3), colorbar_scale=:log10) # z_log not defined
heatmap([1,2,4], [1,2], rand(2,3)) # works

I only tried this with the GR backend, so I don't know whether it's backend-specific.

[giordano]

It works if x, y are Float64

Well, "works" was a bit of a stretch: I don't get any error, but I also get an empty plot. It's solved if setting clims excplicitly.

[david-macmahon]

It works if x, y are Float64

Well, "works" was a bit of a stretch: I don't get any error, but I also get an empty plot. It's solved if setting clims excplicitly.

That's because of the zero value(s?) in your z. If you use hypot.(x', y).+1 it won't be blank. I guess it's a bug that the whole plot is blank instead of just where z <= 0.

[giordano]

Right. My actual data didn't have zeros, it started from 1, the example I used here was too simplified, the zero is a red herring.

[BeastyBlacksmith]

Can you check #4770?