Compose is a (experimental, work-in-progress) vector graphics library for Julia. It is forming the basis for the statistical graphics system Gadfly.

Unilke most vector graphics libraries, compose is thoroughly declarative. Rather than issue a sequence of drawing commands, graphics are formed by sticking various things together and then letting the library figure out how to draw it. The "things" in this case fall one of three types: Property, Form, and Canves. "Sticking together" is primary achieved with the compose function.

The semantics of composition are fairly simple, and once grasped provide a consistent and powerful means of building vector graphics.

The easiest way to get a sense of how Compose works is with an example. So, here's how to draw a sierpinski triangle.

load("Compose")
using Compose

function sierpinski(n)
    if n == 0
        compose(canvas(), polygon((1,1), (0,1), (1/2, 0)))
    else
        t = sierpinski(n - 1)
        compose(canvas(),
                (canvas(1/4,   0, 1/2, 1/2), t),
                (canvas(  0, 1/2, 1/2, 1/2), t),
                (canvas(1/2, 1/2, 1/2, 1/2), t))
    end
end

img = SVG("sierpinski.svg", 4inch, 4(sqrt(3)/2)inch)
draw(img, compose(sierpinski(8), linewidth(0.1mm), fill(nothing)))
finish(img)

A graphic in Compose is a tree of Canvas object, each specifying a coordinate system relative to its parent canvas. One canvas is made a child of another with a call to compose(a::Canvas, b::Canvas).

Canvases may also have children of type Form which are rectangles, ellipses, text, and other things that end up being drawn. Properties, which affect how forms are drawn, may be children of Canvas or Form objects.

There are fancier forms of the compose function. In particular, variadic compose, which is roughly defined as:

compose(a, b, cs...) = compose(compose(a, b), cs...)

Compose over tuples or arrays:

compose((as...)) = compose(as...)

In effect, this let's one write a complex series of compose operations as an S-expression. For example:

compose(a, b, ((c, d), (e, f), g))

Since all we are doing is building trees, this syntax tends to be pretty convenient.

There is a lesser operation in Compose called combine, defined over Properties and Forms. It acts essentially as a union operation. The combinations of two forms is a new form that draws both. The combination of two properties is a new property that has the effect of both.

Besides coordinate transformations, Compose also handles mixtures of relative and abosule coordinates. For example, 1w - 10mm is a well formed expression, giving the width of the parent canvas minus ten millimeters.

Compose is indended as a futuristic version of the R library grid, and so takes a few ideas from grid. The Compose canvas is roughly equivalent to a viewport in grid, for example. The Haskell library Diagrams was another starting point with many admirable notions I hope to steal.