plot(histogram,fill(1.5,5)) is an empty plot

[smldis]

Hi,
plot(histogram,fill(1.5,5)) is an empty plot and the repl output is:

plot(histogram,fill(1.5,5))
┌ Warning: Founds 0 width in scene limits: Float32[3.5, 0.0, 0.0] .. Float32[3.5, 0.0, 0.0]
└ @ AbstractPlotting C:\Users\smldis\.julia\packages\AbstractPlotting\KIwn9\src\interfaces.jl:686
┌ Warning: Founds 0 width in scene limits: Float32[3.5, 0.0, 0.0] .. Float32[3.5, 0.0, 0.0]
└ @ AbstractPlotting C:\Users\smldis\.julia\packages\AbstractPlotting\KIwn9\src\interfaces.jl:686
```

[piever]

Thanks for reporting. The problem may come from StatsBase though (edges and weights seem incorrect):

julia> using StatsBase

julia> fit(Histogram, fill(1.5, 5))
Histogram{Int64,1,Tuple{StepRangeLen{Float64,Base.TwicePrecision{Float64},Base.TwicePrecision{Float64}}}}
edges:
  2.5:2.0:4.5
weights: [0]
closed: left
isdensity: false

[piever]

Here is the issue in StatsBase JuliaStats/StatsBase.jl#456. Things should work here as soon as StatsBase implements a fix for the "mono valued vector" here (we rely on StatsBase for histogram computation).

[smldis]

@piever Could you define your requirements for the histrange function?
I need to improve my current PR to StatsBase since it is not fixing this issue while it is fixing JuliaStats/StatsBase.jl#456.

julia> StatsBase.histrange(0.5, 0.5, 5)
0.5:5.0e-15:0.500000000000015

[piever]

Thanks for fixing this! I think it's important to have the weights of the histogram sum to the correct value (before they were 0). There may be issues due to such a small binsize (for example, given that GLMakie runs on the GPU, some of the logic is probably in single floating point precision). Should we just give a warning id the data span is too small?

[smldis]

I tried with Rebugger but I was unable to step into the second layer of fit(...) and its hard to understand from the generated functions how the weights and edges are computed. I might try harder soon

[smldis]

Should we just give a warning id the data span is too small?

Actually we could check for this specific case and make a good plot with only one bin.

[smldis]

plot(histogram,fill(Float32(1.3),5))
actually works with my patch to StatsBase JuliaStats/StatsBase.jl#479

So one possible solution would be to fit the histogram with data converted to 32bit.
Another solution would be to handle this small range case on Makie's backend side.

[piever]

Sure, I'll look into it after the StatsBase PR is merged (just ping me here where that happens).