GR: rework automatic major/minor ticks for log scales

[t-bltg]

Superseding #3547, required patch: JuliaPlots/PlotUtils.jl#115.

Test cases:

using Plots; gr()

for (i, (start, stop)) ∈ enumerate([(-1, 2), (-6, 5)])
    x = y = Float64[1]
    for j ∈ start:stop
      append!(x, 2 * 10.0^j:10.0^j:10.0^(j + 1) |> collect)
    end
    plot(
      x, y, axis=:log10, minorgrid=true, legend=:none,
      shape=:cross, ms=2, lw=.5, dpi=200
    )
    png("loglog-$i")
end

Output:

Fix #3318
Fix #696
Fix #458

[t-bltg]

@BeastyBlacksmith, pinging you on this one for review / merge: if we could move forward that'll be great ;)

[t-bltg]

Fixed the only minor issue with example 5 (ref5): 1 decade (minimal number of major ticks = 2) is enough for log scales.

Old

New

Note: the CI will be significative only once JuliaPlots/PlotUtils.jl#115 is merged.

[t-bltg]

Ok, so after merging JuliaPlots/PlotUtils.jl#115 and triggering a new release of PlotUtils, the only failing test is ref5 as expected.

@BeastyBlacksmith or @daschw if I could have your approval, we can merge JuliaPlots/PlotReferenceImages.jl#97 and this PR.

[BeastyBlacksmith]

how did ref5 look like with minmal number of ticks = 4?

[t-bltg]

how did ref5 look like with minmal number of ticks = 4?

$ julia ref5.jl
┌ Warning: No strict ticks found
└ @ PlotUtils [...]/PlotUtils/n4I4Q/src/ticks.jl:295
┌ Warning: No strict ticks found
└ @ PlotUtils [...]/PlotUtils/n4I4Q/src/ticks.jl:295

[BeastyBlacksmith]

I suppose this is because it tries to find integer exponents. Can we have some kind of look for integer exponents -> if not enough ticks present look for mulitples of 1/2 -> if not enough ticks found look for multiples of 1/4 etc. ?

[t-bltg]

I suppose this is because it tries to find integer exponents. Can we have some kind of look for integer exponents -> if not enough ticks present look for mulitples of 1/2 -> if not enough ticks found look for multiples of 1/4 etc. ?

What's wrong with requiring less major ticks ? It is justified by the choice of log scaling.

Non-integer powers do not bring information to the plot or the user: 10^-.5 ? you need a calculator, I am not in favor of it.

[BeastyBlacksmith]

I think the plot looks a bit lost with only two ticks. But surely that is subjective, probably would look okay with minorticks. Overall I am okay with the changes

[t-bltg]

I think the plot looks a bit lost with only two ticks. But surely that is subjective, probably would look okay with minorticks. Overall I am okay with the changes

It is subjective yes, see also the examples in https://fr.mathworks.com/help/matlab/ref/loglog.html.

Don't count on minor-ticks there, we need at least 3 major ticks because of this

Plots.jl/src/axes.jl

Line 340 in 6f44004

if length(ticks) > 2

.

[codecov]

Codecov Report

Merging #3577 (9ba89ec) into master (7b0066c) will decrease coverage by 1.20%.
The diff coverage is 40.88%.

❗ Current head 9ba89ec differs from pull request most recent head 16e41f0. Consider uploading reports for the commit 16e41f0 to get more accurate results

@@            Coverage Diff             @@
##           master    #3577      +/-   ##
==========================================
- Coverage   64.83%   63.63%   -1.21%     
==========================================
  Files          27       28       +1     
  Lines        6641     6855     +214     
==========================================
+ Hits         4306     4362      +56     
- Misses       2335     2493     +158     

Impacted Files	Coverage Δ
src/backends.jl	61.06% <0.00%> (-2.25%)	⬇️
src/backends/gaston.jl	0.00% <0.00%> (ø)
src/utils.jl	52.80% <60.00%> (+0.22%)	⬆️
src/backends/gr.jl	89.12% <98.30%> (+1.10%)	⬆️
src/axes.jl	85.20% <100.00%> (+1.48%)	⬆️
src/backends/pgfplotsx.jl	60.94% <100.00%> (ø)
src/examples.jl	61.76% <100.00%> (+0.57%)	⬆️
src/init.jl	100.00% <100.00%> (ø)
src/layouts.jl	68.22% <100.00%> (+1.39%)	⬆️
src/pipeline.jl	89.63% <100.00%> (+0.57%)	⬆️
... and 9 more

---

Continue to review full report at Codecov.

Legend - Click here to learn more
Δ = absolute <relative> (impact), ø = not affected, ? = missing data
Powered by Codecov. Last update 7b0066c...16e41f0. Read the comment docs.

[t-bltg]

@BeastyBlacksmith

Ok this has be bothering me not to be able to have minorgrids if we have 2 major ticks, I dislike un-finished things.
I reworked the logic and here is ref5 when running tests:

Would that be ok ?