[BUG] twinx() broken for pgfplots and plotly backends

[Bebotron]

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

Details

I found this issue with pgfplots(), but I tried to reproduce it with the other backends and had another hiccup, though pgfplots() was the most egregious. Note I wasn't even able to make an initial plot with insepctdr(), which I don't really use so I didn't bother trying to figure out what's wrong there. The simple test code is:

using Plots
x = [1,2,3]
y1 = [1,2,3]
y2 = [4,5,6]
plot(x,y1,ylim=(0,6))
plot!(twinx(),x,y2,ylim=(0,6))

this output the following for each backend. The behavior I expected from pgfplots() (and the other backends) to begin with is the one that gr() gives: yaxis ticks and labels on the right side. plotly() and plotlyjs() don't quite do this, while pgfplots() does something else entirely.

gr()

pyplot()

plotly()

plotlyjs()

pgfplots()

With inspectdr(), as soon as I tried to plot anything, it would print some errors and kick me out of the REPL. That should be a separate issue, but for now I'm more concerned with pgfplots().

Also as a separate note if someone could help me set the REPL default plot show to pdf instead of svg for pgfplots() that'd be great. Yes I know I can savefig as a pdf. But I want to know if it's possible to once you hit plot(...) on the REPL, to automatically pop up a pdf instead of svg. Thanks!

Versions

Plots.jl version: Plots v1.6.0

Backend version (]st -m): GR v0.51.0 PyPlot v2.9.0 PlotlyBase v0.4.1 PlotlyJS v0.14.0 PGFPlots v3.3.2 *InspectDR v0.3.9

Output of versioninfo(): Julia Version 1.5.1 Commit 697e782ab8 (2020-08-25 20:08 UTC) Platform Info: OS: Linux (x86_64-pc-linux-gnu) CPU: Intel(R) Core(TM) i7-4770 CPU @ 3.40GHz WORD_SIZE: 64 LIBM: libopenlibm LLVM: libLLVM-9.0.1 (ORCJIT, haswell)

[BeastyBlacksmith]

the pgfplots part is a duplicate of https://github.com/JuliaPlots/Plots.jl/issues/2071. Also note that the pgfplots backend got deprecated for the pgfplotsx backend

[Bebotron]

Thanks for the heads up! I guess I wasn't very thorough finding out if the issue had already been reported. In any case, while not as terrible looking, still a problem using pgfplotsx()

[Mutaz94]

Hi All: I'm also getting this weird lines with Plots.GR() backend while using twinx() The error is: QPainterPath::arcTo: Adding arc where a parameter is NaN, results are undefined. It works just fine with plotly() and if I save the figure.

[BeastyBlacksmith]

Do you have a MWE for that?

[Mutaz94]

I think I figured out where the problem is (vide infra). Here is MWE

using Plots

x = range(0, 100)

y1 = 0.5*x
y2 = -5 * x

plot(x, y1, label=false, color=:red,
           linewidth=2,left_margin = 5Plots.mm, right_margin = 15Plots.mm, ylabel="foo")
plot!(twinx(), x, y2, label=false,
            color=:black, linewidth=2,left_margin = 5Plots.mm, right_margin = 15Plots.mm, ylabel="bar")

Since I'm doing heavy computation, I start julia with -O3 and --math-mode=fast which cause gr() to give this problem, more particularly --math-mode is causing the problem.

[BeastyBlacksmith]

Well --math-mode=fast is a dangerous thing to do. It would be better to wrap computations where you are okay with using unsafe operation into @fastmath and avoid setting this globally.

[Mutaz94]

Make sense. Thanks!

[cpf-work]

Any news on this? I have the same bug when using the plotlyjs backend. Neither twinx() nor ymirror=true are working.

Plots v1.30.0 PlotlyJS v0.18.8 PlotlyBase v0.8.18 Julia Version 1.6.6 Commit b8708f954a (2022-03-28 07:17 UTC) Platform Info: OS: Windows (x86_64-w64-mingw32) CPU: Intel(R) Core(TM) i7-4790K CPU @ 4.00GHz WORD_SIZE: 64 LIBM: libopenlibm LLVM: libLLVM-11.0.1 (ORCJIT, haswell)

[t-bltg]

An update on this: twinx and twiny now work as expected on gr, pyplot and pgfplotsx (see https://github.com/JuliaPlots/Plots.jl/pull/4488). For plotly(js) it needs more rework, or the issue is upstream.