Adding arbitrary legend_columns for python-backends

[briederer]

As a follow up to #4645 I am trying to implement the feature also for other backends.

As asked in #4645 (comment) by @lmanzanillas I have started with the PythonPlot backend, which has been rather trivial in principle. However, I could not reproduce the behaviour of a fully horizontal legend for legend_columns=-1 like in gr(). I.e. the legend_title being on the same line as the series labels.

I'll have a look at the other backends too and then also create Test Images for the docs.

Samples

data = rand(10, 4)
leg_cols = -2:5

leg_plots(; kw...) = begin
  map((leg_col -> begin
      plot(data; marker=:circle, ticks=:none, plot_title=string("legend_columns = ", leg_col), legend_columns=leg_col, kw...)
    end), leg_cols)
end
(w, h) = (600,600)
with(scalefonts=0.5, size=(w, h)) do
  plot(leg_plots()..., leg_plots(legend_title="Test")...; layout=(4, 4))
end

GR

┌ Warning: n° of legend_column=0 has undefined behaviour. Assuming vertical layout.
└ @ Plots ~/.julia/dev/Plots/src/backends/gr.jl:1226
┌ Warning: n° of legend_column=4 is larger than n° of series=3
└ @ Plots ~/.julia/dev/Plots/src/backends/gr.jl:1223
┌ Warning: n° of legend_column=0 has undefined behaviour. Assuming vertical layout.
└ @ Plots ~/.julia/dev/Plots/src/backends/gr.jl:1226
┌ Warning: n° of legend_column=4 is larger than n° of series=3
└ @ Plots ~/.julia/dev/Plots/src/backends/gr.jl:1223

PythonPlot

┌ Warning: n° of legend_column=0 has undefined behaviour. Assuming vertical layout.
└ @ Plots ~/.julia/dev/Plots/src/backends/pythonplot.jl:1438
┌ Warning: n° of legend_column=4 is larger than n° of series=3
└ @ Plots ~/.julia/dev/Plots/src/backends/pythonplot.jl:1435
┌ Warning: n° of legend_column=0 has undefined behaviour. Assuming vertical layout.
└ @ Plots ~/.julia/dev/Plots/src/backends/pythonplot.jl:1438
┌ Warning: n° of legend_column=4 is larger than n° of series=3
└ @ Plots ~/.julia/dev/Plots/src/backends/pythonplot.jl:1435

Other Backends

To be done.

[briederer]

Also interesting to note:
PythonPlot orders the series differently in the multi-column layout (i.e. column-major). Since this is also the way how julia treats matrices it is probably the preferred way. So I would suggest tho change GR to this behaviour too.

[briederer]

Apparently for the some most backends the legend_column=-1 possibility is not even implemented so I won't touch it.
The only remaining backend is therefore Gaston. However, I am currently on a Mac without Gnuplot so I can't test that.

So although my plans where ambitious, I think adding this option for python backends is all I can do at the moment.

[BeastyBlacksmith]

Also interesting to note: PythonPlot orders the series differently in the multi-column layout (i.e. column-major). Since this is also the way how julia treats matrices it is probably the preferred way. So I would suggest tho change GR to this behaviour too.

FWIW pgfplots also sorts these row-major

[briederer]

Hmm 🤔
I mean zo be fair probably is row-major a bit more intuitive in a horizontal legend since (at least in the english language) we are reading fromm left to right and then jump in the next line.
I don't have any strong opinions, except for wanting a common behavior for the different backends.

[codecov]

Codecov Report

Patch coverage: 62.50% and project coverage change: -0.02 ⚠️

Comparison is base (4916cf8) 90.42% compared to head (a159c43) 90.40%.

Additional details and impacted files

@@            Coverage Diff             @@
##           master    #4678      +/-   ##
==========================================
- Coverage   90.42%   90.40%   -0.02%     
==========================================
  Files          40       40              
  Lines        8703     8706       +3     
==========================================
+ Hits         7870     7871       +1     
- Misses        833      835       +2     

Impacted Files	Coverage Δ
src/backends.jl	82.58% <ø> (ø)
src/backends/gr.jl	90.65% <50.00%> (ø)
src/backends/pythonplot.jl	87.44% <66.66%> (-0.24%)	⬇️

Help us with your feedback. Take ten seconds to tell us how you rate us. Have a feature suggestion? Share it here.

☔ View full report at Codecov.
📢 Do you have feedback about the report comment? Let us know in this issue.

[BeastyBlacksmith]

It would be possible to let the sign carry the majorness.
Not sure, if I'd like that though ^^

[briederer]

I also thought about that, but then -1 for horizontal would probably not be the right choice anymore but instead -nseries. So I think that this is not the best solution.
And in any case people can always sort their data if they want to swap the legend-entries.

[t-bltg]

See also #2047.
I would be in favor of unifying to row major now for consistency with current layout behavior, and switch to col major in Plots 2.0 (breaking), even if I hear the natural left to right reading argument.

[briederer]

That's a good point.
I'll change it tomorrow. Since it's already there but only commented it should not take too much time.

However, I also want to give the pgfplotsx backend another look.
Since it is implemented apparently in the backend it should be doable.

[BeastyBlacksmith]

However, I also want to give the pgfplotsx backend another look.
Since it is implemented apparently in the backend it should be doable.

It is already working for that. Consider:

julia> using Plots; pgfplotsx()
Plots.PGFPlotsXBackend()

julia> plot(rand(4,4), legend_columns = 2)

Its just that your example errors because pgfplots errors on too small labels for some reason...

[briederer]

That's good to know.
Thanks 😊
Then it just needs to be added to the supported attributes.

[BeastyBlacksmith]

@briederer is this good to go?

[briederer]

Oh sorry, totally forgot about this.
I am not sure if GR and python are consistent at the moment and it still lacks proper docs and/or TestImages.
I am currently very limited in my time, but I'll have another look next week (unless someone else has time and wants to jump in 😅).

[BeastyBlacksmith]

bump

[BeastyBlacksmith]

kindly reminder