Skip to content

Latest commit

 

History

History
274 lines (203 loc) · 8.27 KB

EXPLAINER.md

File metadata and controls

274 lines (203 loc) · 8.27 KB

CSS Paint API Explained

The CSS Paint API is being developed to improve the extensibility of CSS.

Specifically this allows developers to write a paint function which allows us to draw directly into an elements background, border, or content.

This work was motivated for a couple of reasons:

Reduction of DOM

We noticed that developers are using an increasing amount of DOM to create visual effects. As an example the <paper-button> uses multiple divs to achieve the "ripple" effect.

Instead of using addition divs the developer could just draw directly into the background-image of the element instead.

This means that the memory and cpu usage of the page would go down, the rendering engine doesn't have to keep in memory the additional DOM nodes, in addition to performing style-recalc, layout, painting for all these additional divs.

Efficiency Gains

Providing a "hook" into the rendering engine allows for efficiency gains which are difficult to achieve with current APIs.

Invalidation

As the CSS paint API Invalidation is based off style changes, this check can occur in the same pass as the rest of the box tree. For example:

my-button {
    --property-which-invalidates-paint: no-hover;
}

my-button:hover {
    --property-which-invalidates-paint: hover;
}

To achieve the same effect with current APIs you have to rebuild the engines invalidation logic which is potentially less efficient.

Painting

Once a box has been invalidated, a rendering engine isn't required to paint it that frame. For example some rendering engines just paint what is visible within the "visual viewport". This means that only a smaller amount of work is needed to be done.

A naive implementation with existing APIs may try and paint everything within the document.

Extensibility of CSS

We believe that allowing developers to extend CSS is good for the ecosystem. As an example if a developer wanted an additional feature they could implement it themselves. E.g. if the developer wanted a new type of dashed border, they shouldn't have to wait for browsers to implement this.

They should have the power to implement this themselves with the same capability as the rendering engine.

Getting Started

First you'll need to add a module script into the paint worklet.

if ('paintWorklet' in CSS) {
  await CSS.paintWorklet.addModule('my-paint-script.js');
  console.log('paint script installed!');
}

See the worklets explainer for a more involved explaination of worklets. In short worklets:

  • Are similar to workers in that the script runs in a separate global script context.
  • A script inside a worklet has no DOM, Network, Database, etc access.
  • The global script context lifetime is not defined (you should expect the script context to be killed at any point).
  • May have multiple copies of the script context spawned on multiple CPU cores.

As a couple of concrete example as to how the user agent may use these capabilities:

  • When a "tab" is backgrounded the script context(s) of the paint worklet may be killed to free up memory.
  • A multi-threaded user-engine may spawn multiple backing script contexts to perform the paint phase of the rendering engine in parallel.

Painting a circle

The global script context of the paint worklet has exactly one method exposed to developers: registerPaint.

registerPaint('circle', class {
  static inputProperties = ['--circle-color'];
  
  paint(ctx, size, style) {
    // Change the fill color.
    const color = style.get('--circle-color');
    ctx.fillStyle = color;

    // Determine the center point and radius.
    const x = size.width / 2;
    const y = size.height / 2;
    const radius = Math.min(x, y);

    // Draw the circle \o/
    ctx.beginPath();
    ctx.arc(x, y, radius, 0, 2 * Math.PI, false);
    ctx.fill();
  }
});

There are a few things going on in this example so lets step through them one-by-one.

The paint function is your callback into the browsers paint phase in the rendering engine. You are given:

  • ctx, a rendering context, similar to a CanvasRenderingContext2D.
  • size, the size of the area in which you should paint.
  • style, the computed style of the element which are you currently painting.

The style is a Typed-OM style map. It will only contain the CSS properties that you listed under the static inputProperties accessor.

This is to allow the engine to cache results of your paint class. For example if --some-other-property changes the user-agent knows that this doesn't affect your paint class, and can re-use the stored result.

The only time when your paint class must be called is when the element it is painting into is within the viewport, and the size or CSS input properties have changed.

Why classes?

Classes allow composition of paint handlers. As an example:

class RectPainter {
  static inputProperties = ['--rect-color'];
  
  paint(ctx, size, style) {
    // Change the fill color.
    ctx.fillStyle = style.get('--circle-color');

    // Draw the solid rect.
    ctx.fillRect(0, 0, size.width, size.height);
  }
}

class BorderRectPainter extends RectPainter {
  static inputProperties = ['--border-color', ...super.inputProperties];
  
  paint(ctx, size, style) {
    super.paint(ctx, size, style);
    
    ctx.strokeStyle = style.get('--border-color');
    ctx.lineWidth = 4;
    
    ctx.strokeRect(0, 0, size.width, size.height);
  }
}

registerPaint('border-rect', BorderRectPainter);

Classes also give the developer a specific point in time to perform pre-initialization work. As an example:

registerPaint('lots-of-paths', class {

  constructor() {
    this.paths = performPathPreInit();
  }
  
  performPathPreInit() {
    // Lots of work here to produce list of Path2D object to be reused.
  }
  
  paint(ctx, size, style) {
    ctx.stroke(this.paths[i]); 
  }
});

In this example performPathPreInit is doing CPU intensive work which should only be done once. Without classes this would typically be done when the script is first run, instead this is performed when the class instance is first created (which may be much later in time).

Drawing Images

The API works in conjunction with the CSS Properties and Values proposal and the CSS Typed OM proposal.

Using the Properties and Values registerProperty method, developers can create a custom CSS property with the <image> type. E.g.

registerProperty({
  name: '--profile-image',
  syntax: '<image>'
});

This tells the rendering engine to treat the CSS property --profile-image as an image; and as a result the style engine will parse and begin loading that image.

You can then directly draw this image from within your paint method:

registerPaint('avatar', class {
  static inputProperties = ['--profile-image'];
  
  paint(ctx, size, styleMap) {
    // Determine the center point and radius.
    const x = size.width / 2;
    const y = size.height / 2;
    const radius = Math.min(x, y);
    
    ctx.save();
    // Set up a round clipping path for the profile image.
    ctx.beginPath();
    ctx.arc(x, y, radius, 0, 2 * Math.PI, false);
    ctx.clip();
    
    // Draw the image inside the round clip.
    ctx.drawImage(styleMap.get('--profile-image'));
    ctx.restore();
    
    // Draw a badge or something on top of the image.
    drawBadge(ctx);
  }
});

The above example would be used in CSS by:

.avatar-img {
  background: paint(avatar);
  --profile-image: url("profile-image.png");
}

Paint Arguments

It is also possible with this API to have additional arguments to the paint() function, for example:

registerPaint('circle-args', class {
  static inputArguments = ['<color>'];

  paint(ctx, size, _, args) {
    const color = args[0].cssText;
    ctx.fillStyle = color;

    const x = size.width / 2;
    const y = size.height / 2;
    const radius = Math.min(x, y);

    ctx.beginPath();
    ctx.arc(x, y, radius, 0, 2 * Math.PI, false);
    ctx.fill();
  }
});
my-element {
  background:
    paint(circle-args, red) center/50% no-repeat,
    paint(cirlce-args, blue);
}