using log10 scale with negative numbers breaks log scales forever

[HenriDeh]

If you attempt to make a plot with a logarithmic scale and negative numbers, instead of throwing a domain error, some warnings are thrown and an almost empty plot is created. If you then correct the data to be all positive, the returned plot is still messed up and the only way to make it right is to restart julia.

using StatsPlots, DataFrames

df = DataFrame(a = [1, 1, 1, 1, 1, 2, 2, 2, 2, 2], b = repeat([-0.1, 0.1],5))

boxplot(df.a, df.b, yscale = :log10)
┌ Warning: No strict ticks found
└ @ PlotUtils C:\Users\Henri\.julia\packages\PlotUtils\NS6Tf\src\ticks.jl:283
┌ Warning: No strict ticks found
└ @ PlotUtils C:\Users\Henri\.julia\packages\PlotUtils\NS6Tf\src\ticks.jl:283
┌ Warning: No strict ticks found
└ @ PlotUtils C:\Users\Henri\.julia\packages\PlotUtils\NS6Tf\src\ticks.jl:283

boxplot(df.a, max.(0.01, df.b), yscale = :log10)
GKS: Rectangle definition is invalid in routine SET_VIEWPORT

the second boxplot returns this:

[mkborregaard]

This is only with the GR backend, right?

[HenriDeh]

Yes, with pyplots the first attempt throws an error and the corrected data works fine.