[BUG] Bad values with yscale=:log10 break follow-up plots in GR

[mortenpi]

Details

Plotting bad (negative) values while yscale=:log10 is set with the GR backend breaks the GR output also for future (valid) plots. For an MWE, trying to render the two following plots, this happens:

julia> plot(1:-0.01:-1, yscale=:log10)
┌ Warning: No strict ticks found
└ @ PlotUtils ~/.julia/packages/PlotUtils/es5pb/src/ticks.jl:283
┌ Warning: No strict ticks found
└ @ PlotUtils ~/.julia/packages/PlotUtils/es5pb/src/ticks.jl:283

julia> plot(1:10)
GKS: Rectangle definition is invalid in routine SET_VIEWPORT

Note the missing tick labels etc.

Other notes:

• I also tried PyPlot, where the first plot just throws an error, and the second plot is fine. I would actually expect GR to simply throw an error as well for the first plot, but instead it actually tries to render it.
• I usually run into this issue in vscode, where I accidentally plot rubbish (e.g. pass negative values to a log plot) and then all the following plots are not rendering correctly. The only way to get Plots working again is to restart the Julia session (or change backends; but if you change back to GR, it's still broken). For this, as a workaround, is there any way to reset the GR backend or something along those lines?
• Looks similar to [BUG] Un-recoverable GKS: Rectangle definition is invalid in routine SET_VIEWPORT #2872, but definitely still occurs.

Backends

This bug occurs on ( insert x below )

Backend	yes	no	untested
gr (default)	x
pyplot		x
plotly			x
plotlyjs			x
pgfplotsx			x
inspectdr			x

Versions

Plots.jl version: master (v1.10.5+)
Backend version (]st -m): GR v0.54.0, GR_jll v0.53.0+0
Full MWE manifest
Output of versioninfo():

Julia Version 1.5.3
Commit 788b2c77c1 (2020-11-09 13:37 UTC)
Platform Info:
  OS: Linux (x86_64-linux-gnu)
  CPU: Intel(R) Xeon(R) CPU E5-1620 v3 @ 3.50GHz
  WORD_SIZE: 64
  LIBM: libopenlibm
  LLVM: libLLVM-9.0.1 (ORCJIT, haswell)

[t-bltg]

@mortenpi, we discussed this in #3601, closing.

[mortenpi]

I would advocate for re-opening this. The follow-up plots are still breaking even on the latest master.

On another related note, the new warning

┌ Warning: Invalid negative or zero value 0.0 found at series index 101 for log10 based yscale
└ @ Plots ~/.julia/packages/Plots/igCxX/src/utils.jl:92

seems to be printed for each problematic value? So when I run plot(1:-0.01:-1, yscale=:log10) I literally get a hundred of them. Do you reckon it might make sense to print the warning once per plot call?

[t-bltg]

On another related note, the new warning

┌ Warning: Invalid negative or zero value 0.0 found at series index 101 for log10 based yscale
└ @ Plots ~/.julia/packages/Plots/igCxX/src/utils.jl:92

seems to be printed for each problematic value? So when I run plot(1:-0.01:-1, yscale=:log10) I literally get a hundred of them. Do you reckon it might make sense to print the warning once per plot call?

Do you care about the indices at where this occurs or not ?
Let's print it for the first invalid value found.

Fixed in 91e91ac

[t-bltg]

I would advocate for re-opening this. The follow-up plots are still breaking even on the latest master.

Sorry, I was too quick when cleaning up issues. I see now what you meant by saying "breaking" GR.

[t-bltg]

@mortenpi, btw I see that you are on Linux. Any reason for not using ENV["GKSwstype"] = "x11" (that's what I use on my Ubuntu) instead of using the buggy GKS QtTerm ?

[mortenpi]

Hmm. As for the "why", I am just running with the defaults and they seem to work most of the time 😄 I do set GKSwstype=100 when running non-interactively though (on CI and when generating heaps of plots in a script) because it's much faster that way.

However, I don't think it's relevant for this? First, in the REPL I get a GKS: X11 support not compiled in warning with x11 and no window pops up. Secondly, the follow-up plots are still broken (while I can't see them interactively, I can savefig them):

julia> ENV["GKSwstype"]
"x11"

julia> using Plots; gr()
Plots.GRBackend()

julia> plot(1:-0.01:-1, yscale=:log10)
┌ Warning: No strict ticks found
└ @ PlotUtils ~/.julia/packages/PlotUtils/n4I4Q/src/ticks.jl:295
GKS: X11 support not compiled in
┌ Warning: No strict ticks found
└ @ PlotUtils ~/.julia/packages/PlotUtils/n4I4Q/src/ticks.jl:295
┌ Warning: Invalid negative or zero value 0.0 found at series index 101 for log10 based yscale
└ @ Plots ~/.julia/packages/Plots/1miUQ/src/utils.jl:92

julia> plot(1:20)
GKS: Rectangle definition is invalid in routine SET_VIEWPORT

[t-bltg]

Hmm. As for the "why", I am just running with the defaults and they seem to work most of the time

Ok, thanks for the heads up.

GKS: X11 support not compiled in

I forgot I compiled my own GR against x11. I remember it took me a while to get something interactive running properly ...