Observables

Interaction and animations in Makie are handled using Observables.jl. An Observable is a container object whose stored value you can update interactively. You can create functions that are executed whenever an observable changes. You can also create observables whose values are updated whenever other observables change. This way you can easily build dynamic and interactive visualizations.

On this page you will learn how the Observables pipeline and the event-based interaction system work. Besides this, there is also a video tutorial on how to make interactive visualizations (or animations) with Makie.jl and the Observable system:

The Observable structure

A Observable is an object that allows its value to be updated interactively. Let's start by creating one:

using GLMakie, Makie

x = Observable(0.0)

Observable(0.0)

Each Observable has a type parameter, which determines what kind of objects it can store. If you create one like we did above, the type parameter will be the type of the argument. Keep in mind that sometimes you want a wider parametric type because you intend to update the Observable later with objects of different types. You could for example write:

x2 = Observable{Real}(0.0)
x3 = Observable{Any}(0.0)

This is often the case when dealing with attributes that can come in different forms. For example, a color could be :red or RGB(1,0,0).

Triggering A Change

You change the value of a Observable with empty index notation:

x[] = 3.34

This was not particularly interesting. But Observables allow you to register functions that are executed whenever the Observable's content is changed.

One such function is on. Let's register something on our Observable x and change x's value:

on(x) do x
    println("New value of x is $x")
end

x[] = 5.0

New value of x is 5.0

Note

If you updated the Observable using in-place syntax (e.g. img[] .= colorant"red"), you need to manually notify(img) to trigger the function.

Note

All registered functions in a Observable are executed synchronously in the order of registration. This means that if you change two Observables after one another, all effects of the first change will happen before the second change.

There are two ways to access the value of a Observable. You can use the indexing syntax or the to_value function:

value = x[]
value = to_value(x)

The advantage of using to_value is that you can use it in situations where you could either be dealing with Observables or normal values. In the latter case, to_value just returns the original value, like identity.

Chaining Observables With lift

You can create a Observable depending on another Observable using lift. The first argument of lift must be a function that computes the value of the output Observable given the values of the input Observables.

f(x) = x^2
y = lift(f, x)

Observable(25.0)

Now, whenever x changes, the derived Observable y will immediately hold the value f(x). In turn, y's change could trigger the update of other observables, if any have been connected. Let's connect one more observable and update x:

z = lift(y) do y
    -y
end

x[] = 10.0

@show x[]
@show y[]
@show z[]

New value of x is 10.0
x[] = 10.0
y[] = 100.0
z[] = -100.0

If x changes, so does y and then z.

Note, though, that changing y does not change x. There is no guarantee that chained Observables are always synchronized, because they can be mutated in different places, even sidestepping the change trigger mechanism.

y[] = 20.0

@show x[]
@show y[]
@show z[]

x[] = 10.0
y[] = 20.0
z[] = -20.0

Shorthand Macro For lift

When using lift, it can be tedious to reference each participating Observable at least three times, once as an argument to lift, once as an argument to the closure that is the first argument, and at least once inside the closure:

x = Observable(rand(100))
y = Observable(rand(100))
z = lift((x, y) -> x .+ y, x, y)

To circumvent this, you can use the @lift macro. You simply write the operation you want to do with the lifted Observables and prepend each Observable variable with a dollar sign $. The macro will lift every Observable variable it finds and wrap the whole expression in a closure. The equivalent to the above statement using @lift is:

z = @lift($x .+ $y)

This also works with multiline statements and tuple or array indexing:

multiline_node = @lift begin
    a = $x[1:50] .* $y[51:100]
    b = sum($z)
    a .- b
end

If the Observable you want to reference is the result of some expression, just use $ with parentheses around that expression.

container = (x = Observable(1), y = Observable(2))

@lift($(container.x) + $(container.y))

Problems With Synchronous Updates

One very common problem with a pipeline based on multiple observables is that you can only change observables one by one. Theoretically, each observable change triggers its listeners immediately. If a function depends on two or more observables, changing one right after the other would trigger it multiple times, which is often not what you want.

Here's an example where we define two Observables and lift a third one from them:

xs = Observable(1:10)
ys = Observable(rand(10))

zs = @lift($xs .+ $ys)

Now let's update both xs and ys:

xs[] = 2:11
ys[] = rand(10)

We just triggered zs twice, even though we really only intended one data update. But this double triggering is only part of the problem.

Both xs and ys in this example had length 10, so they could still be added without a problem. If we want to append values to xs and ys, the moment we change the length of one of them, the function underlying zs will error because of a shape mismatch. Sometimes the only way to fix this situation, is to mutate the content of one observable without triggering its listeners, then triggering the second one.

xs.val = 1:11 # mutate without triggering listeners
ys[] = rand(11) # trigger listeners of ys (in this case the same as xs)

Use this technique sparingly, as it increases the complexity of your code and can make reasoning about it more difficult. It also only works if you can still trigger all listeners correctly. For example, if another observable listened only to xs, we wouldn't have updated it correctly in the above workaround. Often, you can avoid length change problems by using arrays of containers like Point2f or Vec3f instead of synchronizing two or three observables of single element vectors manually.