Orange Honey; Epilogue

The past four years (2007-2011) much of my energy and capital has been focused into building the Mugtug graphics suite; Sketchpad, Darkroom and Lightbox. The suite has come a long way since I developed Sketchpad during a seven month work binge of Red-Bulls!  Through the collaboration of many developers we’ve moved forwards to create an entire framework that blurs the line between “web-app” and “desktop-app”…

Due to my own budgetary constraints, and differing visions within the corporation, the time has come for me to move on from Orange Honey.  The projects will continue under the direction of my good friend Charles Pritchard.  He is without a doubt the most knowledgable developer I’ve worked with.

The following highlight a few of my final contributions to Mugtug; made possible with HTML5;

Layer Styles; this module creates effects such as InnerShadow, OuterShadow, InnerGlow, and OuterGlow.  These are similar to what Photoshop achieves—the difference is, my version has the ability to do what I’ve coined “style stacking”.  Style stacking allows the designer to add multiple fills (solid, gradient, pattern) to, for instance, InnerShadow;

SVG Parser; this module converts .svg files into <canvas> commands, accepting complex examples, supporting features from <gaussianblur> to the <use> element;

This is an image from OpenClipart rendered in the SVG->Canvas parser;

There is no Gaussian blur in HTML5’s <canvas>, and to do a “true” Gaussian blur takes a lot of processing, and computational time.  I ended up using Mario Klingemann’s StackBlur to polyfill the support in the SVG parser, the results are pretty good;  I think some of the blurring wasn’t turned up enough do to my own Matrix scaling issues. Canvas left, SVG right;

Radial Gradient; this demo shows how fun radials and gradients can be ?

Composite Erase; this module creates a new composite mode, allowing you to erase colors based on the color of a brush;

– Brushes; this module adds some fun new brushes to play with, like galaxy (left);

Marquee; this module creates a “magick wand” with marching ants, for selecting portions of an image, and modifying them with fills, filters, and other effects;


adminOrange Honey; Epilogue

HTML5: Typographic Effects

HTML5 Rocks is a website that helps inspire developers and teach how to implement those shiny new HTML5 features in real world examples. They recently asked me to write an article for their website. Working on a project for Google was inspiring (even if there was no pay involved)!

… Neon Rainbow Jitter;

… Sleek Zebra (inspired by WebDesignerWall);

… Spaceage;

… and many other fun effects.  The following ones are using a CSS->Canvas converter; the CSS used in the converter was sourced from Line25, and Stereoscopic, and Shadow 3D;


adminHTML5: Typographic Effects

HTML5: Typographic Metrics

Typography support between browsers has a history of being spotty.  One of the major hurdles in creating the (yet-to-be-released) next-incarnation of Sketchpad was typographic support; between browsers and even within the <canvas> specs.  One pitfall of the <canvas> tag is the lack of text-metrics support (past calculating the width via ctx.measureText).  This prevents us from emulating how text works in DOM for other elements such as <div> or even <textarea>—fortunately, these metrics can be measured through simple css-hacks.The solutions to the following are provided:
  • ctx.textBaseline=”alphabetic”; // alphabetic, top, bottom, middle
  • ctx. textAlign=”left”; // left, right, middle, start, end
  • measureText(‘hello’).width // px width of typeface

The solution for finding these metrics is compatible with browsers as far back as 2004 (such as Firefox 1.0), before <canvas> was introduced.  Out of 64 browsers tested on BrowserShots and the 30 tested on CrossBrowserShots, four browsers failed to generate proper metrics for Arial with these CSS solutions.  There was an error margin of +/- 2px, which may accounted for in the differences between anti-aliasing in <canvas> vs. rendering in <div>/<span>/<ect>.

Table of contents

Bounding boxes

Calculating the bounding-box of unicode

I’ve used a simple brute-force formula to calculate the bounding-box of unicode characters for the following text experiments called “getTextMetrics”.  Once you know the text metrics, you can do most anything ?

It’s been very useful to me for many purposes:  Unicode Profiling Project, Font-Family Profiling Project, calculating the “Ascent, Descent, and x-height”, and finding the bounding-box of unicode glyphs to create stamps, amongst other things.

See getBitmapBounds demo in action.

(function() {
var fontFamily = "Arial, san-serif";
var fontSize = 14;
getFontMetrics = function(props) {
	var ctx = props.ctx;
	var text = props.text;
	var bboxHeight = props.bboxHeight;
	var canvasHeight = props.canvasHeight;
	var baseline = props.baseline || "alphabetic";
	var flip = props.flip || false;
	var drawBaseline = props.drawBaseline || false;
	if (props.fontFamily) fontFamily = props.fontFamily;
	if (props.fontSize) fontSize = props.fontSize;
	// setting up the canvas; // create canvas to use as buffer
	ctx.font = fontSize + "px " + fontFamily;
	var textWidth = ctx.measureText(text).width;
	// This keeps font in-screen, measureText().width doesn't
	// quite do it in some cases. For instance "j", or the letter "f"
	// in the font "Zapfino".
	var offsetx = fontSize * 2;
	var offsety = fontSize * 2;
	var cwidth = ctx.canvas.width = Math.round(textWidth + offsetx * 2);
	var cheight = ctx.canvas.height = canvasHeight ? canvasHeight : Math.round(offsety * 2);
	if (typeof(baseline) == "string") {
		offsety = 0; // using <canvas> baseline
 ctx.textBaseline = baseline;
 // ctx.font has to be called twice because resetting the size resets the state
 if (flip) ctx.scale(1, -1);
 ctx.font = fontSize + "px " + fontFamily
 ctx.fillText(text, offsetx, (typeof(bboxHeight)=="number" ? bboxHeight : offsety));
 // drawing baseline
 if (drawBaseline) {
 ctx.fillRect(0, canvasHeight/2, ctx.canvas.width, 1);
 // grabbing image data
 var imageData = ctx.getImageData(0, 0, cwidth, cheight);
 var data =;
 // calculating top
 var top = 0;
 var pos = 0;
 while (pos < data.length) { if (data[pos + 3]) { pos -= pos % (cwidth * 4); // back to beginning of the line top = (pos / 4) / cwidth; // calculate pixel position top -= offsety - fontSize; pos = data.length; // exit loop } pos += 4; } // calculating bottom var bottom = 0; var pos = data.length; while (pos > 0) {
 if (data[pos + 3]) {
 pos -= pos % (cwidth * 4); // back to beginning of the line
 bottom = (pos / 4) / cwidth;
 bottom -= offsety - fontSize;
 pos = 0; // exit loop
 pos -= 4;
 // calculating left
 var left = 0;
 var col = 0, row = 0; // left bounds
 while (row < cheight && col < cwidth) {
 var px = data[(row * cwidth * 4) + (col * 4) + 3];
 if (px) {
 left = col - offsetx;
 row = cheight;
 col = cwidth;
 row ++;
 if (row % cheight == 0) {
 row = 0;
 // calculating right
 var right = 0;
 var col = cwidth, row = 0; // right bounds
 while (row < cheight && col > 0) {
 if (data[(row * cwidth * 4) + (col * 4) + 3]) {
 right = col - offsetx;
 row = cheight;
 col = cwidth;
 row ++;
 if (row % cheight == 0) {
 row = 0;
 col --;
 // calculating real-bottom
 var realBottom = 0;
 var pos = data.length;
 while (pos > 0) {
 if (data[pos + 3]) {
 pos -= pos % (cwidth * 4); // back to beginning of the line
 realBottom = (pos / 4) / cwidth;
 pos = 0; // exit loop
 pos -= 4;
 // restoring state
 // returning raw-metrics
 return {
 "left": (-left),
 "top": (fontSize - top),
 "width": (right - left),
 "height": (bottom - top),
 "bottom": realBottom


Mimicking the way text is supported in other DOM elements

By default <canvas> text is aligned to the “alphabetic” baseline.  When drawing to a canvas, using the default baseline, and a y-position of 0, only the descenders can be seen:

ctx.fillText(“Hello world!”, 0, 0);

To mimic how text is viewed in <div>/<span>/<input> other html elements we can use the “top” baseline.  This works wonderfully, however, not all browsers support the “top” baseline. If baseline=”top” works in your browser “Hello world!” should become visible in the following example:

ctx.textBaseline = “top”;
ctx.fillText(“Hello world!”, 0, 0);

Opera, Safari and Chrome work identically with “top”, “bottom”, “middle”, and “alphabetic”… however, Firefox seems to remove the “buffer-spacing” (the invisible spacing that goes above the ascenders, and below the descenders on each line of text) which prevents it from mimicking how Firefox draws other DOM elements.  Older browsers wont support baseline=”top” or other baseline variants at all.

This illustrates our problem… lets find a solution!

<canvas> -> “Ascent”, “Descent” and “x-height”

According the Apache FOP, the “top” baseline is equal to the “ascent”, the “bottom” basline is equal to the “descent”, and the “middle” baseline is half of the “x-height”.  Given this information, it’s possible to measure these values in <canvas> when ctx.getImageData() and ctx.textBaseline are supported.

To measure these baselines, the following code was used:

See <canvas> baseline demo in action.

// finding portion that protrudes past bottom of alphabetic baseline
var Descent = getFontMetrics({
	ctx: ctx,
	text: "YourText",
	bboxHeight: 0,
	canvasHeight: bboxHeight * 3,
	baseline: "alphabetic",
	fontFamily: fontFamily + ", " + defaultFont,
	fontSize: fontSize
// calculating top-baseline
var TopBaseline = getFontMetrics({
	ctx: ctx,
	text: "YourText",
	bboxHeight: 0,
	canvasHeight: bboxHeight * 3,
	baseline: "top"
}).bottom - Descent;
// calculating bottom-baseline
var BottomBaseline = getFontMetrics({
	ctx: ctx,
	text: "YourText",
	bboxHeight: bboxHeight,
	canvasHeight: bboxHeight * 3,
	baseline: "bottom"
}).bottom - bboxHeight - Descent;
// calculating middle-baseline
var MiddleBaseline = getFontMetrics({
	ctx: ctx,
	text: "YourText",
	bboxHeight: bboxHeight,
	canvasHeight: bboxHeight * 3,
	baseline: "middle"
}).bottom - bboxHeight - Descent;

Unfortunately, the only reliable baseline identically supported between all browsers in <canvas> is “alphabetic”, and there is no way to measure the “ascent” and “descent” when the only working baseline is “alphabetic”.  The reason is the invisible “buffer-spacing” above and below the descenders changes the expected output, and there is no way to measure “invisible” space.

To recap: at this point, we still don’t have the ability to measure the “ascent”, “descent” and “x-height” when textBaseline isn’t supporting “top” or “bottom”… but, we do now know what values we’re searching for.  Next, let’s look into getting the values from CSS.

CSS -> “Ascent”, “Descent” and “x-height”

After searching for a solution for hours in <canvas> under the crazy pursuit of measuring something invisible, I came home and found a simple solution in CSS. By using an <img> (or any inline-block element) and the vertical-align property inside of a container element (such as a <div>), the values <canvas> provides for ctx.textBaseline can be matched with an error margin of +/- 2px—many fonts are matched exactly.  Likely, these discrepancies are due to anti-aliasing in <canvas> vs. DOM.

  1. The “top” baseline is equal to the image.offsetTop since an <img> element is automatically aligned to the baseline of a font.
  2. The “bottom” baseline can be found by subtracting the height of the text (see the measureText section) from the “top” baseline.
  3. The “middle” baseline can be found using “line-height: 0” on the container element and measuring the image.offsetTop—this works, because “line-height: 0” aligns the image to the center of the text, as the text now has a height of 0.  In order to get the proper values the whole experiment must be offset vertically (so the text isn’t hidden off-screen).

See CSS text-metrics demo in action.
See text baseline demo in action.

// setting up html used for measuring text-metrics
var container = document.getElementById("container");
var parent = document.createElement("div");
var image = document.createElement("img");
image.width = 42;
image.height = 1;
image.src = "./media/1x1.png";
// getting css equivalent of ctx.measureText() = "none"; = "inline";
var measureHeight = parent.offsetHeight;
var measureWidth = parent.offsetWidth;
// making sure super-wide text stays in-bounds = "inline";
var forceWidth = measureWidth + image.offsetWidth;
// capturing the "top" and "bottom" baseline = "margin: 50px 0; display: block; width: " + forceWidth + "px";
var TopCSS = image.offsetTop - 49;
var HeightCSS = parent.offsetHeight;
var BottomCSS = TopCSS - HeightCSS;
// capturing the "middle" baseline = "line-height: 0; display: block; width: " + forceWidth + "px";
var MiddleCSS = image.offsetTop + 1;


Compensating for “overhanging” and “clipping”

Some font-faces protrude past the outside of their em-box, clipping their x or y axis.  This protrusion, in typography, is called an overhang (or overshoot).  Fortunately, this is something that can be compensated for.

To fix the clipping on the top portion of the text:

  1. Measure the entire height of the font in <canvas> using a lot of padding to make sure the entire text-string is visible.
  2. Resize the <canvas> to the height of the em-box, and align the baseline of your text to “top”.  Measure the entire height again.
  3. Subject the first value from the second, this difference is the amount the text protrudes past the top of the em-box.
  4. Using this value, the top portion of the font-face that was clipped can become visible.  This can be applied to <div>’s using padding-top, or to <canvas>’s by adjusting the y-offset of a text drawing command such as ctx.fillText() and adding the same value to the height of the <canvas> (assuming we want a perfect bounding-box).

Here’s an example of finding the clipping on left and top:

// compensating for text-clipping using padding-left
var leftClipping = getFontMetrics({
	ctx: ctx,
	text: "YourText"
if (leftClipping < 0) { // is padding, not clipping
	leftClipping = 0;
// compensating for text-clipping using padding-top
var topClipping = getFontMetrics({
	ctx: ctx,
	text: "YourText",
	bboxHeight: 25,
	canvasHeight: bboxHeight,
	baseline: "top"
}).top + 25;
if (topClipping < 0) { // is padding, not clipping
	topClipping = 0;

Native support between browsers

On a side-note, it looks like at this time no browser supports the HTML5 standards “ideographic” or “hanging”.  Opera/Safari/Chrome default to “top” and “bottom” respectively, whereas Firefox defaults to “alphabetic” in when these options are chosen.  It looks impossible to fix “ideographic” and “hanging” without these values being provided via an external means (i.e. Python)—please prove me wrong ?


Fixing older browsers

The only align that works in all browsers, old and new, is ctx.textAlign=”left”.  We can fix “right” and “center” in older browsers by calculating the text-width.  This can be done simply by creating a <span> with the same font-properties, and measuring the span.offsetWidth (DOM level 0).

Align to “start” and “end” require the additional knowledge of the directionality of the font to distinguish RTL from LTR languages.  This poses a problem if the website doesn’t specify this information in the

See textAlign demo in action.

<span id="control">Hello world!</span>


function getAlign(text, type, offsetx) {
	var direction = window.getComputedStyle(document.body)["direction"];
	control.textContent = text;
	switch(type) {
		case "left": break;
		case "start": offsetx -= (direction == 'ltr') ? 0 : control.offsetWidth; break;
		case "end": offsetx -= (direction == 'ltr') ? control.offsetWidth : 0; break;
		case "right": offsetx -= control.offsetWidth; break;
		case "center": offsetx -= control.offsetWidth / 2; break;
	return offsetx;

Native support between browsers

Opera, Safari, Chrome and Firefox work identically with “left”, “start”, “end”, “right” and “center” in their latest branches.  Older browsers such as Firefox 2.x, and Opera 9.x require a fallback in order for ctx.textAlign to work properly, such as this CSS solution.


Fixing older browsers

We can fix old browsers with broken ctx.measureText support in CSS by calculating the span.offsetWidth using the same methods we used to fix ctx.textAlign:

See measureText demo in action.

<span id="control">Hello world!</span>


function measureText(text) { = "inline";
	control.textContent = text;
	return {
		height: control.offsetHeight,
		width: control.offsetWidth

Measurements provided this CSS solution produces identical results as ctx.measureText in Safari, Chrome and Opera. Firefox results are identical most of the time, with the occasional erroneous result within +/- 1px.  Zapfino is an example of a font-face that produces this deviance.

Native support between browsers

Opera, Safari, Chrome and Firefox work nearly identically across systems within a difference of +/- 1px. Older browsers such as Firefox 2.x, and Opera 9.x require a fallback in order for ctx.measureText() to work at all.

Line-breaks in <canvas>

Finding the “em-height”

The em-height of a font can be found in Opera, Safari, Chrome and Firefox through measuring CSS-positioning in a similar method to the ctx.textAlign and ctx.measureText() demos.

Create a <span> element with the text you want to measure, with the font-properties set how you want them, and run span.offsetHeight—this measures the em-height.  The em-height is the value we’re going to use to offset our line-breaks.

Calculating line & letter breaks

Calculating word-wrapping entails looping through the text.split(“ “) into singular words, and measuring each word individually until the edge of the bounding-box is hit, at which point a break is inserted, and the process continues.   This works great for words that don’t extend past the bounding-box on their own.

When the word is so large it extends past the bounding box, we need to add in “letter-wrapping”.  This is similar to hyphenation, but without the hyphen ? Calculating letter-wrapping entails looping through the text.length of the long word in question, until the edge of the bounding-box is hit, at which point a line-break is added.

See lineBreaks demo in action.

function getLines(text, maxWidth) {
	var returns = text.split("n");
	var lines = [];
	var lastPhrase = "";
	function splitWord() {
		var width = measureText(lastPhrase).width;
		var posA = 0;
		var posZ = 0;
		if (width > maxWidth) {
			for (var n = 0, length = lastPhrase.length; n < length; n ++) { 				var width = measureText(lastPhrase.substr(posA, posZ ++)).width; 				if (width > maxWidth) {
					lines.push(lastPhrase.substr(posA, posZ - 2));
					posA = n - 1;
					posZ = 2;
			return lastPhrase.substr(posA, posZ + 2);
	for (var n = 0; n < returns.length; n++) {
		if (lastPhrase) lines.push(lastPhrase);
		var phrase = returns[n];
		var spaces = phrase.split(" ");
		var lastPhrase = "";
		for (var i = 0; i < spaces.length; i++) {
			var measure = measureText(lastPhrase + " " + spaces[i]).width;
			if (measure < maxWidth) { 				lastPhrase += ((lastPhrase ? " " : "") + spaces[i]); 			} else { 				if (measure > maxWidth) {
					var split = splitWord();
					if (split) {
						lastPhrase = split + " " + spaces[i];
					} else {
						lastPhrase = spaces[i];
			if (i == spaces.length - 1) {
				lastPhrase = "";
	return lines;


Using the above methods, we can get text working when ctx.measureText, ctx.textAlign or ctx.textBaseline is malfunctioning.  In a future installment we’ll look into embedded fonts, including adding support of ctx.fillText and ctx.strokeText by parsing SVG fonts and drawing them using the vector primitives ctx.moveTo, ctx.lineTo, ctx.quadradicCurveTo and ctx.bezierCurveTo.

View CSS text-metrics in <canvas>.
View CSS text-measurements.


adminHTML5: Typographic Metrics

HTML5: Unicode Profiling Project

The Unicode Profiling Project was designed to gather statistics on unicode support across systems.  The software checks each symbol in your systems Unicode catalog (65,535 glyphs) to see which are visible on your computer using <canvas> and Javascript.

The data generated from your computer will help profile the state of unicode support on the web. Your computers unicode support, remote address, user agent and processing time will be submitted to the server upon completion of the test — a statistic analysis of the data will be published — no specific information about your computer will be published.

The code behind this project is an extension of isFontSupported (font detection in <canvas>). As with isFontSupported, the code behind Unicode Profiling Project is released under CC0, free to use this code in creative ways in your own projects ?

Running the test:

Once the test is initiated you’ll be able to watch the glyphs as they’re scanned with their related unicode block name.  It typically takes over a minute to scan an entire collection of unicode characters.  This is what the acid test looks like while being processed:

Reading your profile:

Once the processing has completed you will be presented with a string of binary representing what characters are visible, and which ones are unavailable or invisible (65,535 numbers). Here are the results from my Chromium browser running on OSX 10.6.4:

Unicode characters:

Now the fun part, click on “Show Available” — this may take a few seconds as you’re referencing tens of thousands of unicode characters at once:

Unicode #65018

Brail Patterns [0x2800-28FF]

<a href=” sertraline drug.54.12-PM.png”>

Supported Browsers:

The project fully supports Chrome, Firefox, Safari, and Opera.  Some false positives are produced in Internet Explorer as there is a unique “missing symbol” for every unicode block.

Results on my Mac:

  • Safari
    49,493 visible glyphs
  • Firefox 3.6
    49,428 visible glyphs
    NOTE:  Each undefined symbol has a unique hash unless text size is <=11
  • Google Chrome 7.0
    49,493 visible glyphs
  • Chromium 8.0
    49,492 visible glyphs
  • Opera 10.6
    47,672 visible glyphs
    NOTE:  Supports different fonts in <canvas> than regular DOM

Results on my Windows:

  • IE 9.0
    50,826 visible glyphs
    NOTE:  Some false positives… each range has it’s own undefined symbol.
  • Firefox 3.6
    51,208 visible glyphs
    NOTE:  Each undefined symbol has a unique hash unless text size is <=10
  • Google Chrome 7.0
    47,267 visible glyphs
    NOTE:  Textarea can have different unicode support than Div in some cases.  For instance, on my computer ﰿ works in Textarea, but not in Div.
  • Opera 10.6
    56,024 visible glyphs
    NOTE:  Supports different unicode in Canvas than Div and Textarea.  Also, Opera supports more unicode characters than other browsers by far, possibly included in the package?

Further Research:

adminHTML5: Unicode Profiling Project