x/y-lims are not updated when fitting subplots into layout #3633

anneaux · 2021-07-08T10:33:21Z

Hey!
When I fit multiple subplots with equal aspect ratios into a layout, then their axis may be extended in order to fit the layout. However, the values for xlims and ylims are not being updated. Here's a MWE:

using Plots; gr()
using Random; Random.seed!(1234)

p = plot(layout = grid(1,2), frame=:box)

y1 = rand(10) .* 10
plot!(sp=1, y1, aspect_ratio=:equal) # subplot 1, i.e. p[1]

y2 = rand(5) .* 10
plot!(sp=2, y2, aspect_ratio=:equal) # subplot 2, i.e. p[2]

This produces this:

and I get the following values:

extrema(y1) = (2.005860349338411, 8.541465903790503)
Plots.xlims(p[1]) = (0.73, 10.27)
Plots.ylims(p[1]) = (1.809792182704848, 8.737534070424065)
extrema(y2) = (0.10905889635595356, 9.567533636029237)
Plots.xlims(p[2]) = (0.88, 5.12)
Plots.ylims(p[2]) = (-0.17469534583424495, 9.851287878219436)

You see, for example, that the y axis of subplot 1 (the left one) goes from somewhere -1 to ~12 (ignore the cutted ticks label....), but the according ylims(p[1]) gives roughly (1.8,8.7). The same is obvious for subplot 2, where the x-axis ranges from -0.5 to 6.5 roughly, but xlims(p[2]) gives (0.88,5.12).

So... how can one get the actual limits of the axis after they have been fitted into the layout?

Thanks in advance!

(I'm using Julia v1.6.1, Plots v1.18.0)

t-bltg · 2021-07-08T10:51:10Z

Could you please re-run (for the values and images: I've modified your example to use a fixed seed, for reproducibility) ?

Ok I seem to reproduce:

extrema(y1) = (2.005860349338411, 8.541465903790503)
Plots.xlims(p1) = (0.73, 10.27)
Plots.ylims(p1) = (1.809792182704848, 8.737534070424065)
extrema(y2) = (0.10905889635595356, 9.567533636029237)
Plots.xlims(p2) = (0.73, 10.27)
Plots.ylims(p2) = (1.809792182704848, 8.737534070424065)  # NOK

~~I'm suspecting 6f44004, but it got in after 1.18.0 ...~~

t-bltg · 2021-07-08T11:46:15Z

Stack triggering incorrect values:

┌ Debug: 
│   exception =
│    14-element Vector{Base.StackTraces.StackFrame}:
│     macro expansion at logging.jl:341 [inlined]
│     axis_limits(sp::Plots.Subplot{Plots.GRBackend}, letter::Symbol, should_widen::Bool, consider_aspect::Bool) at axes.jl:557
│     axis_limits(sp::Plots.Subplot{Plots.GRBackend}, letter::Symbol) at axes.jl:557
│     get_sp_lims at utils.jl:420 [inlined]
│     ylims(sp::Plots.Subplot{Plots.GRBackend}) at utils.jl:435
│     ylims(plt::Plots.Plot{Plots.GRBackend}, sp_idx::Int64) at utils.jl:445
│     ylims(plt::Plots.Plot{Plots.GRBackend}) at utils.jl:445
│     top-level scope at show.jl:955
│     eval at boot.jl:360 [inlined]
│     include_string(mapexpr::typeof(identity), mod::Module, code::String, filename::String) at loading.jl:1094
│     _include(mapexpr::Function, mod::Module, _path::String) at loading.jl:1148
│     include(mod::Module, _path::String) at Base.jl:386
│     exec_options(opts::Base.JLOptions) at client.jl:285
│     _start() at client.jl:485
└ @ Plots [...]/Plots/CoJwR/src/axes.jl:557

t-bltg · 2021-07-08T12:17:31Z

So... how can one get the actual limits of the axis after they have been fitted into the layout?

You should use xlims(p, 2): this might seem counter-intuitive but plot!(sp=2, ...) does not return the second subplot, but the main plot a.k.a p.

using Plots; gr()
using Random; Random.seed!(1234)

p = plot(layout=grid(1, 2), frame=:box)

y1 = rand(10) .* 10
plot!(sp=1, y1, aspect_ratio=:equal)
@show extrema(y1)
@show xlims(p, 1)
@show ylims(p, 1)

y2 = rand(5) .* 10
plot!(sp=2, y2, aspect_ratio=:equal)
@show extrema(y2)
@show xlims(p, 2)
@show ylims(p, 2)

Output:

extrema(y1) = (2.005860349338411, 8.541465903790503)
xlims(p, 1) = (0.73, 10.27)
ylims(p, 1) = (1.809792182704848, 8.737534070424065)
extrema(y2) = (0.10905889635595356, 9.567533636029237)
xlims(p, 2) = (0.88, 5.12)
ylims(p, 2) = (-0.17469534583424495, 9.851287878219436)

anneaux · 2021-07-08T12:27:24Z

I am sorry to say that, but that does not fix my problem 🙈 you see the values for ylims(p,1)? They still don't match the actual axes limits in the picture!

t-bltg · 2021-07-08T12:36:37Z

I am sorry to say that, but that does not fix my problem see_no_evil you see the values for ylims(p,1)? They still don't match the actual axes limits in the picture!

I missed that, sorry ! Let's dig down further, this might be backend dependent ...

t-bltg · 2021-07-08T12:40:58Z

With pgfplotsx:

So with gr is think a margin issue explains with the 1 in 10.0 is chopped out.
You can use p = plot(layout=grid(1, 2), frame=:box, margins=3Plots.mm) for fixing that.

But it still doesn't explain why the ylims are incorrect ...

anneaux · 2021-07-08T12:43:29Z

yes, I don't mind the margins for now ^^
It happens for xlims as well (I edited this for clarification in the first post)

t-bltg · 2021-07-08T12:48:46Z

yes, I don't mind the margins for now ^^
It happens for xlims as well (I edited this for clarification in the first post)

Ok, so what is not reflected on the plots is:

extrema(y1) = (2.005860349338411, 8.541465903790503)
xlims(p, 1) = (0.73, 10.27)
ylims(p, 1) = (1.809792182704848, 8.737534070424065)  # <== NOK
extrema(y2) = (0.10905889635595356, 9.567533636029237)
xlims(p, 2) = (0.88, 5.12)  # <== NOK
ylims(p, 2) = (-0.17469534583424495, 9.851287878219436)

right ?

Something is swapped somewhere in the code ...

anneaux · 2021-07-08T12:56:35Z

exactly!

It seems the axis range in the final plot is determined from either of the two axis and then adjusts the other one accordingly. So that's why always one of them is correctly (x or y)

t-bltg · 2021-07-08T13:08:51Z

The aspect_ratio is responsible here, it should be too hard to find the issue:

using Plots; gr()
using Random; Random.seed!(1234)

p = plot(layout=grid(1, 2), frame=:box, margins=3Plots.mm)

y1 = rand(10) .* 10
plot!(sp=1, y1)

y2 = rand(5) .* 10
plot!(sp=2, y2)

gives

which is correct.

anneaux · 2021-07-08T13:23:52Z

Ok, I am sorry, maybe I was a little confusing...

I need the plots to keep the aspect ratio, and with the plotting itself I am totally fine!

But, what I want are just the exact values for the axis limits as they are in the final plot.
E.g. for relative positioning of annotations as mentioned in issue #2728

t-bltg · 2021-07-08T13:35:51Z

Ok, I am sorry, maybe I was a little confusing...

I need the plots to keep the aspect ratio, and with the plotting itself I am totally fine!

But, what I want are just the exact values for the axis limits as they are in the final plot.
E.g. for relative positioning of annotations as mentioned in issue #2728

This seems to work, triggering rendering of the image before querying the axes limits.

using Plots; gr()
using Random; Random.seed!(1234)

p = plot(layout=grid(1, 2), frame=:box, margins=3Plots.mm)

y1 = rand(10) .* 10
plot!(sp=1, y1, aspect_ratio=:equal)
@show extrema(y1)
@show xlims(1)
@show ylims(1)

y2 = rand(5) .* 10
plot!(sp=2, y2, aspect_ratio=:equal)
@show extrema(y2)
@show xlims(2)
@show ylims(2)

# show(devnull, "image/png", p)
Plots.prepare_output(p)

println()
@show xlims(1)
@show ylims(1)

@show xlims(2)
@show ylims(2)

Output:

extrema(y1) = (2.005860349338411, 8.541465903790503)
xlims(1) = (0.73, 10.27)
ylims(1) = (1.809792182704848, 8.737534070424065)
extrema(y2) = (0.10905889635595356, 9.567533636029237)
xlims(2) = (0.88, 5.12)
ylims(2) = (-0.17469534583424495, 9.851287878219436)

xlims(1) = (0.73, 10.27)
ylims(1) = (-1.4413732345950825, 11.988699487723995)
xlims(2) = (-0.5609591226753916, 6.560959122675392)
ylims(2) = (-0.17469534583424495, 9.851287878219436)

Still seems like a bug to me.

anneaux · 2021-07-08T13:41:27Z

ohhh thank you! That's indeed a good workaround!

t-bltg · 2021-07-08T14:36:14Z

The root of this bug is that we get plot_ratio = NaN here

Plots.jl/src/axes.jl

Line 601 in 4a18098

plot_ratio = height(plotarea(sp)) / width(plotarea(sp))

because of a [0, 0, 0, 0] bbox, hence the ylims are not accounting for the :equal aspect ratio.

BeastyBlacksmith · 2021-07-08T14:44:27Z

because of a [0, 0, 0, 0] bbox, hence the ylims are not accounting for the :equal aspect ratio.

How did you measure that? bbox gives you zeros, if you use it after the fact, but when you add some @shows in the line before or above you'd see that they should be non-zero

t-bltg · 2021-07-08T14:48:41Z

because of a [0, 0, 0, 0] bbox, hence the ylims are not accounting for the :equal aspect ratio.

How did you measure that? bbox gives you zeros, if you use it after the fact, but when you add some @shows in the line before or above you'd see that they should be non-zero

Taking this example:

using Plots; gr()
using Random; Random.seed!(1234)

p = plot(layout=grid(1, 2), frame=:box, margins=3Plots.mm)

y1 = rand(10) .* 10
plot!(sp=1, y1, aspect_ratio=:equal)
@show extrema(y1)
@show xlims(1)
@show ylims(1)

Sorry an all zero plotarea.

Adding @show plot_ratio, plotarea(sp) after the line quoted before shows a bunch of:

extrema(y1) = (2.005860349338411, 8.541465903790503)
(plot_ratio, plotarea(sp)) = (NaN, BBox{l,t,r,b,w,h = 0.0mm,0.0mm, 0.0mm,0.0mm, 0.0mm,0.0mm})
(plot_ratio, plotarea(sp)) = (NaN, BBox{l,t,r,b,w,h = 0.0mm,0.0mm, 0.0mm,0.0mm, 0.0mm,0.0mm})
xlims(1) = (0.73, 10.27)
(plot_ratio, plotarea(sp)) = (NaN, BBox{l,t,r,b,w,h = 0.0mm,0.0mm, 0.0mm,0.0mm, 0.0mm,0.0mm})
(plot_ratio, plotarea(sp)) = (NaN, BBox{l,t,r,b,w,h = 0.0mm,0.0mm, 0.0mm,0.0mm, 0.0mm,0.0mm})
ylims(1) = (1.809792182704848, 8.737534070424065)

t-bltg · 2021-07-08T14:52:06Z

And adding this

if isnan(plot_ratio)
    prepare_output(sp.plt)
    plot_ratio = height(plotarea(sp)) / width(plotarea(sp))
end

segfaults with a StackOverflowError induced by an infinite recursion call of axis_limits in _update_min_padding!, so I'm kinda stuck here.

t-bltg · 2021-07-08T15:05:32Z

We could also change get_sp_lims in utils.jl to:

function get_sp_lims(sp::Subplot, letter::Symbol)
    !(sp[:aspect_ratio] ∈ (:none, :auto)) && prepare_output(sp.plt)  # Issue #3633
    axis_limits(sp, letter)
end

@BeastyBlacksmith, do you think it's worth a PR ?

BeastyBlacksmith · 2021-07-08T16:29:21Z

I don't think that is the right way to fix this. There is some calculation in prepare_output that should move upwards in the pipeline such that xlims(sp) gives the correct limits regardless of the aspect_ratio

t-bltg · 2021-07-08T16:39:18Z

I don't think that is the right way to fix this. There is some calculation in prepare_output that should move upwards in the pipeline such that xlims(sp) gives the correct limits regardless of the aspect_ratio

The problem is _update_min_padding! which is backend dependent. Only using update_child_bboxes!, even if moved before in the pipeline doesn't fix this.

BeastyBlacksmith · 2021-07-08T16:43:10Z

why is it a problem that it is backend dependent?

t-bltg · 2021-07-08T16:45:58Z

why is it a problem that it is backend dependent?

Because then you cannot fix #3633 (comment) which is not backend dependent (no show, savefig, display, ...).

BeastyBlacksmith · 2021-07-08T16:48:07Z

There are quite a few functions in the pipeline before show that are backend dependent (all the ones in templates/backend.jl) though, not every backend defines all of them

BeastyBlacksmith · 2021-07-08T16:49:55Z

I agree that you should get the right answer independent of the backend, but in general it would be fine to get a different answer for different backends (probably not in this particular case)

t-bltg · 2021-07-08T17:18:33Z

I agree that you should get the right answer independent of the backend, but in general it would be fine to get a different answer for different backends (probably not in this particular case)

Yes it is normal to get a different answer for a different backend. However I believe that the issue exposed in the mwe #3633 (comment) should be fixed, we can't leave it like this.

t-bltg added the bug label Jul 8, 2021

t-bltg closed this as completed Jul 8, 2021

t-bltg removed the bug label Jul 8, 2021

t-bltg reopened this Jul 8, 2021

t-bltg mentioned this issue Jul 8, 2021

[FR] Relative coordinates for annotate #2728

Closed

t-bltg mentioned this issue Jul 8, 2021

Allow relative positioning of annotations #3634

Merged

t-bltg mentioned this issue Jul 17, 2021

[BUG] aspect_ratio causes ylims() to have unexpected behaviour in REPL #3479

Open

BeastyBlacksmith added the bug label Jul 19, 2021

t-bltg added GR layout labels Nov 13, 2022

Provide feedback

Saved searches

Use saved searches to filter your results more quickly

x/y-lims are not updated when fitting subplots into layout #3633

x/y-lims are not updated when fitting subplots into layout #3633

anneaux commented Jul 8, 2021 •

edited by t-bltg

Loading

t-bltg commented Jul 8, 2021 •

edited

Loading

t-bltg commented Jul 8, 2021 •

edited

Loading

t-bltg commented Jul 8, 2021 •

edited

Loading

anneaux commented Jul 8, 2021

t-bltg commented Jul 8, 2021 •

edited

Loading

t-bltg commented Jul 8, 2021 •

edited

Loading

anneaux commented Jul 8, 2021

t-bltg commented Jul 8, 2021 •

edited

Loading

anneaux commented Jul 8, 2021

t-bltg commented Jul 8, 2021

anneaux commented Jul 8, 2021

t-bltg commented Jul 8, 2021 •

edited

Loading

anneaux commented Jul 8, 2021

t-bltg commented Jul 8, 2021 •

edited

Loading

BeastyBlacksmith commented Jul 8, 2021 •

edited

Loading

t-bltg commented Jul 8, 2021 •

edited

Loading

t-bltg commented Jul 8, 2021 •

edited

Loading

t-bltg commented Jul 8, 2021 •

edited

Loading

BeastyBlacksmith commented Jul 8, 2021

t-bltg commented Jul 8, 2021 •

edited

Loading

BeastyBlacksmith commented Jul 8, 2021

t-bltg commented Jul 8, 2021 •

edited

Loading

BeastyBlacksmith commented Jul 8, 2021

BeastyBlacksmith commented Jul 8, 2021

t-bltg commented Jul 8, 2021

x/y-lims are not updated when fitting subplots into layout #3633

x/y-lims are not updated when fitting subplots into layout #3633

Comments

anneaux commented Jul 8, 2021 • edited by t-bltg Loading

t-bltg commented Jul 8, 2021 • edited Loading

t-bltg commented Jul 8, 2021 • edited Loading

t-bltg commented Jul 8, 2021 • edited Loading

anneaux commented Jul 8, 2021

t-bltg commented Jul 8, 2021 • edited Loading

t-bltg commented Jul 8, 2021 • edited Loading

anneaux commented Jul 8, 2021

t-bltg commented Jul 8, 2021 • edited Loading

anneaux commented Jul 8, 2021

t-bltg commented Jul 8, 2021

anneaux commented Jul 8, 2021

t-bltg commented Jul 8, 2021 • edited Loading

anneaux commented Jul 8, 2021

t-bltg commented Jul 8, 2021 • edited Loading

BeastyBlacksmith commented Jul 8, 2021 • edited Loading

t-bltg commented Jul 8, 2021 • edited Loading

t-bltg commented Jul 8, 2021 • edited Loading

t-bltg commented Jul 8, 2021 • edited Loading

BeastyBlacksmith commented Jul 8, 2021

t-bltg commented Jul 8, 2021 • edited Loading

BeastyBlacksmith commented Jul 8, 2021

t-bltg commented Jul 8, 2021 • edited Loading

BeastyBlacksmith commented Jul 8, 2021

BeastyBlacksmith commented Jul 8, 2021

t-bltg commented Jul 8, 2021

anneaux commented Jul 8, 2021 •

edited by t-bltg

Loading

t-bltg commented Jul 8, 2021 •

edited

Loading

t-bltg commented Jul 8, 2021 •

edited

Loading

t-bltg commented Jul 8, 2021 •

edited

Loading

t-bltg commented Jul 8, 2021 •

edited

Loading

t-bltg commented Jul 8, 2021 •

edited

Loading

t-bltg commented Jul 8, 2021 •

edited

Loading

t-bltg commented Jul 8, 2021 •

edited

Loading

t-bltg commented Jul 8, 2021 •

edited

Loading

BeastyBlacksmith commented Jul 8, 2021 •

edited

Loading

t-bltg commented Jul 8, 2021 •

edited

Loading

t-bltg commented Jul 8, 2021 •

edited

Loading

t-bltg commented Jul 8, 2021 •

edited

Loading

t-bltg commented Jul 8, 2021 •

edited

Loading

t-bltg commented Jul 8, 2021 •

edited

Loading