`CairoMakie`'s png output empty for large enough surface plot

[fatteneder]

MWE

# using GLMakie
# GLMakie.activate!()
using CairoMakie
CairoMakie.activate!()

Nx, Ny = 512, 8001
x, y = collect(LinRange(-20,40,Nx)), collect(LinRange(-20,20,Ny))
xx, yy = repeat(x, 1, Ny), repeat(y', Nx, 1)
zz = @. sin(xx) * cos(yy)

f = Figure()
ax = Axis3(f[1,1])
surface!(ax, xx, yy, zz)

save("mwe.png", f)

• GLMakie:
  

• CairoMakie.jl:
  

CairoMakie seems to do something, because it takes a while for the script to complete.

[jkrumbiegel]

have you tried rasterize = 1 on surface? Ah nevermind it's png

It's curious that the tick labels are missing as well, so something seems to break completely during the surface drawing.

[fatteneder]

The magic resolution at which it breaks is Ny >= 5000.

[fatteneder]

Another test says its the overall size of the surface, e.g. if Nx*Ny >= 2560000 (= 512 * 5000) then it refuses to plot.

[SimonDanisch]

There's another weird limit like that: https://github.com/MakieOrg/Makie.jl/blob/master/CairoMakie/src/primitives.jl#L604
Above that threshold, cairo also just stops rendering without any error....

[jkrumbiegel]

Ok but the limit for a single texture makes more sense to me than a count limit for drawing meshes on a bitmap. Unless before the drawing is over, there is some state being accumulated where the count matters. Maybe one can flush that state before the drawing is over?

[fatteneder]

With the patch below to read cairo_pattern_status and the above MWE I get

julia> include("mwe.jl")
cairo_pattern_status = no error has occurred
cairo_pattern_status = no error has occurred
cairo_pattern_status = no error has occurred
cairo_pattern_status = out of memory
CairoMakie.Screen{IMAGE}

Patch:

src|fa/premultiply-img-markers ⇒ cat debug.patch
diff --git a/CairoMakie/src/primitives.jl b/CairoMakie/src/primitives.jl
index ec05ca87d..29fda18a6 100644
--- a/CairoMakie/src/primitives.jl
+++ b/CairoMakie/src/primitives.jl
@@ -902,6 +902,12 @@ function draw_pattern(ctx, zorder, shading, meshfaces, ts, per_face_col, ns, vs,
     Cairo.set_source(ctx, pattern)
     Cairo.close_path(ctx)
     Cairo.paint(ctx)
+    status = ccall((:cairo_pattern_status,Cairo.libcairo), Int32,
+                   (Ptr{Nothing},), pattern.ptr)
+    cstr_status_words = ccall((:cairo_status_to_string,Cairo.libcairo), Cstring,
+                         (Int32,), status)
+    status_words = unsafe_string(cstr_status_words)
+    println("cairo_pattern_status = $status_words")
 end

 ################################################################################

[fatteneder]

Not sure why there are 4 calls to mesh2d, but I guess it doesn't matter.

I also found two threads discussing some 32k pixel limits:

• https://cairo.cairographics.narkive.com/H3TgPxCc/invalid-value-typically-too-big-for-the-size-of-the-input-surface-pattern-etc
• https://cairo.cairographics.narkive.com/Mkjwvkep/32k-limit-with-image-surface

One thread mentions as a work around to manually tile a surface to avoid these limits.

[fatteneder]

Not sure if we should just add a status check like above, because inside Cairo.jl there is the following comment:

# Ideally we'd check the status, but at least for certain releases of the library
# the return value seems not to be set properly (random values are returned)

[jkrumbiegel]

Not sure why there are 4 calls to mesh2d

The first three are the Axis3 side panels.
Maybe there's something wrong with the way we call the mesh drawing functions, as the limits you talk about should apply to the overall surface size required for one mesh, but not accumulated over time (as far as I can tell).