Figures with CairoMakie use meshes, resulting in excessively large file sizes

[noahrhodes]

I'd like to make some nice looking figures, but when saving to vector formats the file sizes are enormous and I wouldn't be able to include them in a publication. Presumably this is because of the conversion from multi-polygon to mesh for the GeoJSON data results in many more

There are some other issues related to poly and mesh #9 #144. There are mostly performance related, and it seems like the performance issues with large polygons were fixed after updates to GeoJSON #63 and some faster paths in Makie for vectors of multi-polygons #2599. In CairoMakie a similar issue #834 was brought up for contour plots having very large files sizes as a result of mesh triangulation instead of using polys when saving plots.

When creating a plot with GeoMakie, it can create a polygon based plot (see example below), but CairoMakie will use a fallback conversion to mesh when it draws the plot.
Presumably this function should be called when CairoMakie renders the plot (specializing on polygons::AbstractArray{<: MultiPolygon}). Instead this fallback function is used which winds up converting the geo axis to a mesh.

This occurs because in drawplot() draw_poly will specialize on to_value.(poly.input_args)... and this results in to_value.(poly.input_args) = (FeatureCollection with 127 Features,). It seems like that should somehow result in a multi-polygon instead of a FeatureCollection in order to allow the dispatch to call the correct function.

From here I do not know enough about the internals of Makie, CairoMakie, and GeoMakie to understand where the the disconnect is between poly!(ga,land_geo) creating a multi-polygon plot and draw_plot being handed a FeatureCollection, or in what package the correct fix needs to be created.

(Because a mesh is used, it also means that saving to .svg will create a .png embedded into the .svg, rather than an actual vector format. Saving to .pdf is the only way to get a vector format when a mesh is used in a plot layer.)

Example

Plotting geosjon will create a multi-polygon

# This example was contributed by Martijn Visser (@visr)
using Makie, CairoMakie, GeoMakie
using GeoMakie.GeoJSON
using GeometryBasics
using Downloads

source = "+proj=longlat +datum=WGS84"
dest = "+proj=natearth2"

url = "https://raw.githubusercontent.com/nvkelso/natural-earth-vector/master/geojson/"
land = Downloads.download(url * "ne_110m_land.geojson")
land_geo = GeoJSON.read(read(land, String))
pop = Downloads.download(url * "ne_10m_populated_places_simple.geojson")
pop_geo = GeoJSON.read(read(pop, String))


fig = Figure(resolution = (1000,500))
ga = GeoAxis(
    fig[1, 1];
    source = source,
    dest = dest
)

poly!(ga, land_geo, color=:black)
ga

julia> ga
Axis with 5 plots:
 ┣━ Lines{Tuple{Vector{Point{2, Float32}}}}
 ┣━ Lines{Tuple{Vector{Point{2, Float32}}}}
 ┣━ Lines{Tuple{Vector{Point{2, Float32}}}}
 ┣━ Lines{Tuple{Vector{Point{2, Float32}}}}
 ┗━ Poly{Tuple{Vector{MultiPolygon{2, Float32, Polygon{2, Float32, Point{2, Float32}, LineString{2, Float32, Point{2, Float32}, Base.ReinterpretArray{Line{2, Float32}, 1, Tuple{Point{2, Float32}, Point{2, Float32}}, TupleView{Tuple{Point{2, Float32}, Point{2, Float32}}, 2, 1, Vector{Point{2, Float32}}}, false}}, Vector{LineString{2, Float32, Point{2, Float32}, Base.ReinterpretArray{Line{2, Float32}, 1, Tuple{Point{2, Float32}, Point{2, Float32}}, TupleView{Tuple{Point{2, Float32}, Point{2, Float32}}, 2, 1, Vector{Point{2, Float32}}}, false}}}}, Vector{Polygon{2, Float32, Point{2, Float32}, LineString{2, Float32, Point{2, Float32}, Base.ReinterpretArray{Line{2, Float32}, 1, Tuple{Point{2, Float32}, Point{2, Float32}}, TupleView{Tuple{Point{2, Float32}, Point{2, Float32}}, 2, 1, Vector{Point{2, Float32}}}, false}}, Vector{LineString{2, Float32, Point{2, Float32}, Base.ReinterpretArray{Line{2, Float32}, 1, Tuple{Point{2, Float32}, Point{2, Float32}}, TupleView{Tuple{Point{2, Float32}, Point{2, Float32}}, 2, 
1, Vector{Point{2, Float32}}}, false}}}}}}}}}

Using to_value should presumably return something that is ::AbstractArray{<: MultiPolygon}

allplots = Makie.collect_atomic_plots(fig.scene; is_atomic_plot = CairoMakie.is_cairomakie_atomic_plot)
poly_plot = allplots[39]
to_value.(poly_plot.input_args) # returns `(FeatureCollection with 127 Features,)`

Saving/visualizing this figure will use a mesh

save("test.svg", fig)  # creates an embedded png in an svg because of the mesh
save("test.pdf", fig) # creates a vector format, but very large file size because of the mesh

[asinghvi17]

yeah CairoMakie has regressed a bit unfortunately. This has been a consistent issue specifically because CairoMakie checks the input types of the call to poly...we are looking into this!