A custom type contains a vector of matrices. Each two-column matrix defines a continuous line segment.

To plot this type, a plot!() function needs to be called as many times as there are line segments.
I am not sure which macro from RecipesBase I need to use so as not to introduce the
dependency on Plots.

Would that be a “user recipe” where the @series macro is called in a loop?

struct mytype
x::Vector{Matrix{Float64}}
end
using RecipesBase
@recipe function f(m::mytype)
@series begin
Tuple.(collect.(eachcol.(m.x)))
end
end
m = mytype([[sort(rand(5)) rand(5)] for _ in 1:10])
using Plots
plot(m)

I think you raise a very good point. The Plots.jl documentation could be greatly improved in the Input Data section.

I learn by poking around.
Maybe the following simple example could be added to the docs:

x1, x2 = [1, 0], [2, 3] # vectors
y1, y2 = [4, 5], [6, 7] # vectors
m1, m2 = [x1 y1], [x2 y2] # matrices
plot([m1, m2]) # array of matrices -> 4 series, plots each matrix column, x assumed to be integer count
plot([[x1,y1], [x2,y2]]) # array of array of arrays -> 4 series, plots each individual array, x assumed to be integer count
plot([(x1,y1), (x2,y2)]) # array of tuples of arrays -> 2 series, plots each tuple as new series

It probably should. At present, this method is undiscoverable except by reading code or trial and error. I tried a tuple of tuples and array of arrays, whereas the internet search only brought the suggestion to insert NaNs into a vector.