CairoMakie: polygon with hole example not working

[fatteneder]

See the 2nd example at the docs page: https://docs.makie.org/v0.18.3/examples/plotting_functions/poly/index.html

Expected output:

[jaakkor2]

This polygon works in v0.18.3, i.e., with reversed winding of the hole

p = Polygon(
    Point2f[(0, 0), (2, 0), (3, 1), (1, 1)],
    [Point2f[(0.75, 0.25), (1.25, 0.75), (2.25, 0.75), (1.75, 0.25)]]
)

[jkrumbiegel]

Yes exactly, it's just that I didn't notice before that the winding order was incorrect because earcut doesn't care, on the mesh conversion route.

We could check winding order and error or flip the polygons in case it's wrong, not sure what the better option is.

[jaakkor2]

Checking the winding order is not trivial

p = Polygon(
    Point2f[(0, 0), (5, 0), (5, 5), (0, 5)],
    [Point2f[(1, 1), (1, 4), (4, 4), (4, 1)],
     Point2f[(2, 2), (3, 2), (3, 3), (2, 3)]]
)

should produce (as it does on CairoMakie + Makie 0.18.3)

[fatteneder]

Checking the winding order is not trivial
p = Polygon(
Point2f[(0, 0), (5, 0), (5, 5), (0, 5)],
[Point2f[(1, 1), (1, 4), (4, 4), (4, 1)],
Point2f[(2, 2), (3, 2), (3, 3), (2, 3)]]
)

To me this doesn't look like a correct usage of poly!.
Also if you try to add another rectangular cutout inside the small rectangle you would have to type (not tested):

p = Polygon(
    Point2f[(0, 0), (5, 0), (5, 5), (0, 5)],
    [Point2f[(1, 1), (1, 4), (4, 4), (4, 1)],
     Point2f[(2, 2), (3, 2), (3, 3), (2, 3)],
     Point2f[(2.2,2.2), (2.2,2.8), (2.8,2.8), (2.8,2.2)]]
)
poly!(ax, p)

Whereas there is a (logically) much simpler way of doing things:

p1 = Polygon(
    Point2f[(0, 0), (5, 0), (5, 5), (0, 5)],
    [Point2f[(1, 1), (1, 4), (4, 4), (4, 1)]]
)
poly!(ax, p1)
p2 = Polygon(
    Point2f[(2, 2), (3, 2), (3, 3), (2, 3)],
    [Point2f[(2.2,2.2), (2.2,2.8), (2.8,2.8), (2.8,2.2)]]
)
poly!(ax, p2)

[fatteneder]

Yes exactly, it's just that I didn't notice before that the winding order was incorrect because earcut doesn't care, on the mesh conversion route.

We could check winding order and error or flip the polygons in case it's wrong, not sure what the better option is.

Is there anything mentioned about winding order in the docs?
But even if there were, the API still seems not intuitive to me.
E.g. there are only two fields of Polygon which are called exterior and interiors (singular and plural). As it is coded in CairoMakie, I think it is always assumed that cutouts are the ones in interiors. After all, you cannot even nest a Polygon inside the interiors, or can you?

To me the signature poly(exterior, interiors) suggests that this should draw a polygon that is build from the shape given by exterior and every shape given in interiors is then subtracted from that.

[fatteneder]

To me the signature poly(exterior, interiors) suggests that this should draw a polygon that is build from the shape given by exterior and every shape given in interiors is then subtracted from that.

If we were to define its usage like that, then there would be no confusion about winding order.

[jkrumbiegel]

The reason the nested polygon works in CairoMakie, is that the outlines are just drawn as a multi-segment path, and then filling is determined via the even-odd rule. But I would also argue that this is not canonical polygon behavior. I concur with Florian's opinion that we are talking about one exterior and multiple interior holes.

In order to match CairoMakie and GLMakie we should just clearly define somewhere what data we expect, so far it has been to loosely defined. We can probably use whatever definitions the leading GIS packages use, as they probably have the most experience with polygons.

[jaakkor2]

Sorry for my confusion, I agree my example is malformed

p = Polygon(
    Point2f[(0, 0), (5, 0), (5, 5), (0, 5)],
    [Point2f[(1, 1), (1, 4), (4, 4), (4, 1)],
     Point2f[(2, 2), (3, 2), (3, 3), (2, 3)]]
)

[jaakkor2]

The reason the nested polygon works in CairoMakie, is that the outlines are just drawn as a multi-segment path, and then filling is determined via the even-odd rule.

The fill rule seems to be nonzero

using CairoMakie
using Makie.GeometryBasics

f = Figure()
Axis(f[1, 1])

# polygon with hole
p = Polygon(
    Point2f[(0, 0), (2, 0), (3, 1), (1, 1)],
    [Point2f[(0.75, 0.25), (1.75, 0.25), (2.25, 0.75), (1.25, 0.75)]]
)

poly!(p, color = :blue)

fn = "cm.svg"
save(fn, f)
collect(eachmatch(r"fill-rule:(.*?);", read(fn, String)))

gives

2-element Vector{RegexMatch}:
 RegexMatch("fill-rule:nonzero;", 1="nonzero")
 RegexMatch("fill-rule:nonzero;", 1="nonzero")

With evenodd fill rule, holes seem to work regardless of winding.

[ffreyer]

The docs have been fixed and I think the Polygon discussion is something for GeometryBasics instead... Can this be closed?

Ref JuliaGeometry/GeometryBasics.jl#173

[asinghvi17]

IMO this is still something we have to figure out in CairoMakie - whether it should respect this kind of "malformed polygon" or somehow try and correct it before plotting (either assert winding order for the rings, or change the fill rule).

The correct GIS thing in this case would be a MultiPolygon that can contain the inner filled rectangle as the second polygon.