allow for different line widths in lines

[lazarusA]

At the moment is possible to do several lines at once by NaN separation. Different colours also works, however passing a set of values for different line widths is not supported, as in linesegments. It would be nice to have this feature for lines, which in all my tests is the fastest one to plot thousands of lines at once.

[rveltz]

Yes, here is the MWE I have in mind:

using GLMakie
x=LinRange(0,2pi,100)
lines(x, sin.(x), linewidth = 0.19) # works
lines(x, sin.(x), linewidth = collect(x)) # does not

we get:

julia> lines(x, sin.(x), linewidth = collect(x)) # does not
Error showing value of type Makie.FigureAxisPlot:
ERROR: MethodError: Cannot `convert` an object of type Vector{Float32} to an object of type Float32

[lazarusA]

I was thinking more like:

using GLMakie
points = [Point2f(0.0, 0.0), Point2f(0.0, 1.0), Point2f(NaN, NaN),
    Point2f(1.0, 0.0), Point2f(1.0, 1.0)]
lines(points; color = [1, 1, 1, 2, 2], linewidth = 5)

this one works, but this one

points = [Point2f(0.0, 0.0), Point2f(0.0, 1.0), Point2f(NaN, NaN),
    Point2f(1.0, 0.0), Point2f(1.0, 1.0)]
lines(points; color = [1, 1, 1, 2, 2], linewidth = [2,5])
#lines(points; color = [1, 1, 1, 2, 2], linewidth = [2,2,2,5,5]) # does not work also

fails.

[SimonDanisch]

Different linewidth in connected lines is a bit cumbersome... linesegments supports it though, since it doesn't need to worry about joins.

[lazarusA]

yes linesegments works at some level. In my case it crashes my Julia session for large numbers whereas lines is completely ok and fast. Bottom line, I was just wondering about the possibilities. For now I'm adding a transparency for small line widths which helps hiding a little bit those lines.

[SimonDanisch]

In my case it crashes my Julia session for large numbers whereas lines is completely ok and fast

Oh thats pretty weird...Linesegments should be way more efficient...

[lazarusA]

ahh, plus I forgot to mention my lines are not straight ones (which lines can perfectly handle), which is the use case of line segments.

[rveltz]

Different linewidth in connected lines is a bit cumbersome... linesegments supports it though, since it doesn't need to worry about joins.

How come, Plots.jl provides it

[jkrumbiegel]

Can you show an example of Plots.jl? I can imagine that it works if segments separated by NaNs have constant line widths, otherwise the Cairo/vector graphics limitations should apply.

[rveltz]

https://bifurcationkit.github.io/BifurcationKitDocs.jl/dev/tutorials/Swift-Hohenberg1d/#d-Swift-Hohenberg-equation-(Automatic)

[SimonDanisch]

How come, Plots.jl provides it

I didn't say it's impossible 🤷 And as I said, with linesegments it's pretty easy, wouldn't be surprised if that's what plots does for different linewidths. Maybe there could be a hybrid to just start a new connected line with every linewidth change...

yes linesegments works at some level. In my case it crashes my Julia session for large numbers whereas lines is completely ok and fast

Can you open an issue with an MWE? That must be a bug!

[rveltz]

I see, should I write my own using dispatch on linewidth?

[SimonDanisch]

https://bifurcationkit.github.io/BifurcationKitDocs.jl/dev/tutorials/Swift-Hohenberg1d/#d-Swift-Hohenberg-equation-(Automatic)

Hm, that does look like proper connected joints though... I guess our best bet to support that right now, would be via bands or generating our own mesh...

[SimonDanisch]

Although, when I look at it again, it looks like the scatter markers hide all the joints, so can't really tell if they have proper joints... So, I guess with a scatter like that, one may get away with linesegments

[rveltz]

Plots.plot(x, sin.(x), linewidth = vcat(ones(50), 10 .*ones(50)))

[jkrumbiegel]

Ok so you see that it's actually all linesegments, no joins. Which makes sense

[ffreyer]

Yes, here is the MWE I have in mind:

using GLMakie
x=LinRange(0,2pi,100)
lines(x, sin.(x), linewidth = 0.19) # works
lines(x, sin.(x), linewidth = collect(x)) # does not

This worked for a while but got removed in #3558 because it's pretty complicated to deal with.

I was thinking more like:

points = [Point2f(0.0, 0.0), Point2f(0.0, 1.0), Point2f(NaN, NaN),
    Point2f(1.0, 0.0), Point2f(1.0, 1.0)]
    lines(points; color = [1, 1, 1, 2, 2], linewidth = [2,2,2,5,5])

This was effectively re-introduced in #2536. The reason I'm pointing at this instead is that the shaders consider linewidth to be constant per line segment. So without the NaN you'll end up a discontinuous connection.

I'll consider this done as the original request is fulfilled.