Notes on FRA implementation

Notes on FRA implementation (functional reactive animation)

Inspired by Fran (by Conal Elliott) and "The Haskell School of Expression" (by Paul Hudak).

Github: https://github.com/ekozhura/fra (work in progress).

Transform type

Let's define a type Transform:

type transform =
  | Translate(float, float)
  | RenderImage(imageElement)
  | Stretch(float)

and operations on that type will represent instructions for HTML5 canvas:

let drawPic = imageEl => RenderImage(imageEl);
let moveXY = (x, y) => Translate(x, y);
let stretch = x => Stretch(x);

To apply those operations on canvas we need a function runTransform:

let runTransform = (ctx, transform) => {
  switch (transform) {
    | Translate(x, y) => ctx->translate(x, y)
    | RenderImage(el) => ctx->drawImage(el, 0., 0.)
    | Stretch(f) => ctx->scale(f, f)
  };
};

where ctx is a 2d rendering context of a canvas element.

To combine transforms let's add another type constructor to our transform type:

type transform =
...
  | ComposedTransform(transform, transform)

and function andThen:

let andThen = (transformA, transformB) => ComposedTransform(transformA, transformB);

to run a composed transform, inside runTransform we run both transforms one after another:

...
  | ComposedTransform(transformA, transformB) =>
    runTransform(ctx, transformA);
    runTransform(ctx, transformB);
...

draw function runs a composed transform:

let draw = transforms => {
  let ctx = elem->getContext;
  runTransform(ctx, transforms);
};

An example of graphics:

let transforms = drawPic(imageEl) |> andThen(moveXY(120., 120.,));

draw(transforms);

Create animation

Animation is a change of transform parameters over time. Let's first define a couple of timing functions:

let wiggle = t => 60. *. Js_math.sin(2. *. Js_math._PI *. 2. *. t /. 6000.);
let waggle = t => 60. *. Js_math.cos(2. *. Js_math._PI *. 2. *. t /. 6000.);

Initial transform will be rewritten as a function of time:

let transforms = t => drawPic(imageEl) |> andThen(moveXY(wiggle(t), waggle(t)));

And drawAnimation function:

let rec drawAnimation = (t) => {
  let ctx = elem->getContext;
  ctx->save;
  runTransform(ctx, transforms(t));
  ctx->restore;
  requestAnimationFrame(drawAnimation);
};

Behavior type

The idea of time-varying value can be expressed by the new type Behavior

type time = float;

type behavior('a) =
  | Behavior(time => 'a);

Then we create helper functions, which lift a value to a Behavior:

let lift0 = x => Behavior(_ => x);

let lift1 = (fn, Behavior(a)) => Behavior(t => fn(a(t)));

let lift2 = (fn, Behavior(a), Behavior(b)) => Behavior(t => fn(a(t), b(t)));

let lift3 = (fn, Behavior(a), Behavior(b), Behavior(c)) => Behavior(t => fn(a(t), b(t), c(t)));

In order to use our transforms in animation, we need to lift them to behaviors. As if we created for each transform a corresponding function from t (time) to transform.

So to run transformation over time we can define a new set of operations:

let constB = lift0;
let moveXYB = lift2(moveXY);
let stretchB = lift1(stretch);
let andThenB = lift2(andThen);

For example constB simply lifts a value to a behavior, moveXYB - lifts a function moveXY to a function of two arguments of type behavior etc.

This is how the drawAnimation function will look like:

let wiggleB = Behavior(wiggle);
let waggleB = Behavior(waggle);

let transforms = const(drawPic(imageEl)) |> andThenB(moveXYB(wiggleB, waggleB));

let rec drawAnimation = (t) => {
  let ctx = elem->getContext;
  ctx->save;
  switch (transforms) {
  | Behavior(transform) => runTransform(ctx, transform(t))
  };
  ctx->restore;
  requestAnimationFrame(drawAnimation);
};

ekozhura/notes.md