HTML5 javascript sudoku solver.


This article describes an HTML5 Sudoku solver. The entire program is in one HTML file and it uses Javascript and the new HTML5 Canvas element. This is my third article about HTML5/Javascript each article being more complicated than the previous and really just a series of learning steps for myself.


I don’t think I really need to say much about Sudoku except that it is a very popular and addictive game and there have been many articles published about Sudoku solvers in various languages. My wife actually got me interested in the game – I learned the basic rules for solving and quickly decided that I was less interested in mechanically working through various solving techniques and more interested in coding up a solver to do it for me – particulary when I hit a hard problem!

A screen shot of the application below shows the main elements. The main element is the board which shows the 9 x 9 grid divided into 3 x 3 “squares”. The “givens” are in a darker font while user entered digits are lighter. Each cell on the board has the allowed values shown in the background in a very light font. “Singles” are shown in a red font – note singles is s standard Sudoku term and refers to a case where an llowed value is the only possible one.

The current cell has a pink background and it can be changed by using the mouse to click in a cell or using the keyboard arrow keys. The user enters a digit by just typing the digit and clears it by typing the 0 digit.

Below the board is a text box containing the serial format of the board data. This is a common format used by Sudoku programs and is simply each digit on the board in order from top to bottom and left to right with empty cells represented by a “.”. As the user enters new digits the serial format is automatically updated and if the Load button is pressed, then whatever text is in the text box is loaded into the board.

All the buttons are located to the right of the board. The Load button was already mentioned but one other point to make is that if you are manually entering a problem the digits will appear as user entered but by clicking the Load button it will “convert” them to givens. Its important to distinguish user values from givens because givens can’t be modified, also when the Reset button is pressed all cells except the givens are cleared. The other buttons we haven’t covered are the Clear button which basically clears every cell including givens; Accept will convert a “single” into an actual value and Solve solves the entire problem. After solving the status is displayed below the buttons and the time taken to solve is displayed below the serial format text box.

Below the buttons are two check boxes, one to enable display of the allowed values and another to enable highlighting singles in red. Note some Sudoku players prefer not to get assistance when solving which is why you would use these options.

Using the code

The code for the solver uses the most basic techniques for solving – first scan all the “givens” and determine which digits are allowed in which cells. If only one allowed value exists then it is a “naked” single. Because each digit must occur exactly once in each row, column or square, we then scan each of these and if an allowed value only occurs once then it is a “hidden” single. Singles values are not necessarily correct – the board can get to a state where a given digit is the only digit in two cells in the same row col etc, so we test whether the value is valid before applying it.

Applying the above techniques will only get you so far. There are in fact many other manual techniques, e.g. naked and hidden doubles, triples, quads, X-wing etc. We don’t use any of these but simply adopt a simple trial and error or decision tree approach. So at a given point we find a cell with the least number of alloweds then try each one in turn. If succesful with that cell then we advance further up the decision tree, regressing when the board becomes invalid and we have run out of alternatives.

Well that’s the algorithm in basic terms. The model code was implemented in four classes as follows:

  • AllowedValues – this stores the allowed values for a cell. It is implemented in a bit mask, i.e. if the digit N is allowed then bit N in an integer is 1 while 0 is not allowed. This provides efficient storage of alloweds (which speeds up making copies of the board as required by the decision tree code) and simplifies combining alloweds using simple bitwise OR operations or removing them by masking.
  • Cell – this represents one of the 9 x 9 locations on the board. It contains the AllowedValues for the cell, the given value and the user entered value.
  • Location – this has two fields: the row and column. It is used to simplify calculations and provides useful functions like a list of locatons of sibling cells in rows, columns and squares.
  • Board – contains 9 x 9 Cell instances. It implements most of the solving related code such as calculting allowed values, singles and solving.

There is a lot of code to cover so I will just mention one key piece of code which is the calculation of the allowed values shown below.

	Board.prototype.updateAllowed = function () {
		// Called whenever the user sets a value or via auto solve
		// Updates the allowed values for each cell based on existing digits
		// entered in a cell's row, col or square
		var cols = new Array(BoardSize);
		var rows = new Array(BoardSize);
		var squares = new Array(BoardSize);

		// First aggregate assigned values to rows, cols, squares
		var locs = Location.grid();
		for (var i = 0; i < locs.length; i++) {
			var loc = locs[i];
			// Disallow for all cells in this row
			var contains = this.getCell(loc).valueMask();
			rows[loc.row] |= contains;
			cols[loc.col] |= contains;
			squares[loc.getSquare()] |= contains;

		// For each cell, aggregate the values already set in that row, col and square.
		// Since the aggregate is a bitmask, the bitwise inverse of that is therefore the allowed values.
		this._isValid = true;
		this._isSolved = true;
		for (var i = 0; i < locs.length; i++) {
			var loc = locs[i];
			// Set allowed values
			var contains = rows[loc.row] | cols[loc.col] | squares[loc.getSquare()];
			var cell = this.getCell(loc);
			cell.setAllowed(~contains); // set allowed values to what values are not already set in this row, col or square
			cell.setAnswer(0); //clear any previous answers
			// As an extra step look for "naked singles", i.e. cells that have only one allowed value, and use
			// that to set the answer (note this is different from the "value" as this can only be assigned
			// by the user or any auto solve functions like "accept singles"
			if (!cell.isAssigned()) {
				this._isSolved = false;
				var mask = new AllowedValues(~contains);
				var count = mask.count();
				if (count == 0)
					this._isValid = false;
				else if (count == 1)

		// Step 2: Look for "hidden singles".
		// For each row, col, square, count number of times each digit appears.
		// If any appear once then set that as the answer for that cell.
		// Count in rows
		for (var i = 0; i < locs.length; i++) {
			var loc = locs[i];
			if (!this.checkForHiddenSingles(loc, SibType.Row))// first check row sibs for a hiddne single
				if (!this.checkForHiddenSingles(loc, SibType.Col))// then check cols
					this.checkForHiddenSingles(loc, SibType.Square); // then check square

		// TO DO: Add code here to detect naked/hidden doubles/triples/quads


The first step is to create an array representing each row, column and square on the board. Using the Location.grid() method to provide a list of all locations on the board we get each cell, obtain its value bit mask (sets bit N in the integer all others 0) and OR them together to obtain a mask of all digits used in that row, column or square, e.g. rows[loc.row] |= contains.

Next we step through each cell again and combine the digits used in that cell’s row, column and square using var contains = rows[loc.row] | cols[loc.col] | squares[loc.getSquare()]. If the cell has not been assigned we set the allowed values as being the bitwise inverse. If only one value is in the allowed digits we set the cells “answer” as that digit – this is a naked single. I use the term answer but in fact it just a good possible.

In the final step of the updateAllowed() function we use the allowed values for each cell to look for hidden singles, e.g. we scan a cell’s row and if it has an allowed value that does not occur in any of its sibling cells then its a hiddne single.

Points of Interest

I learned a lot more about Javascript in doing this article. It is suprisingly powerful but performance varies between browsers. For difficult problems (those with 18 or less givens) I found that Chrome was clearly the fastest.


It would be interesting to try other Sudoku variants such as 25 x 25 😉 when I have some spare time. Thanks for the comments to date. I have made several fixes to the current version based on those comments.


This article, along with any associated source code and files, is licensed under The Code Project Open License (CPOL)

HTML5 animation tutorial

We are going to animate a moving car with two different HTML5 techniques.


HTML5 is becoming more and more popular. With the increasing popularity of mobile devices such as tables and smartphones, the need for alternatives to the popular Flash plugin from Adobe has also been grown. Just recently, Adobe announced that Flash will no longer be supported for mobile devices. This means that Adobe itself will focus on HTML5 as a key technology for those devices – and desktop systems sooner or later.

One disadvantage of HTML was the lack of multimedia techniques. In HTML, you could not display a video or draw on the screen. With HTML5, new elements such as <video> and <canvas> have been introduced. Those elements give developers the possibility to use multimedia technology in “native” HTML, just by writing some JavaScript in combination with HTML. A basic action that should be provided by multimedia technologies is animation. In HTML5, there are some ways to create such actions.

In this article, I will only compare the new <canvas> element with the upcoming CSS3 animation technique. Other possibilities would include the creation and animation of DOM elements or SVG elements. Those possibilities will not be included in this discussion. It should be noted from the beginning that the canvas-technology is supported in the current releases of all major browsers, while CSS3 animations are only possible in the latest editions of Firefox and Chrome. The next IE will also provide CSS3 animations.


I am currently giving a lecture on creating WebApplications using HTML5, CSS3 and JavaScript. This is a lecture with tutorials. For one of the tutorials, I picked a sample canvas animation – just showing in which direction we are heading to with a technology like this. Then I introduced the CSS3 animation in the lecture (everyone was very excited about it) and wanted to create a simple homework task using the CSS3 animations. What came to my mind was: how easy or hard would it be to actually transform the canvas animation into a complete CSS3 animation?

This involved several parts:

  • Creating all the different <div>-elements in order to “box” everything
  • Draw everything using styles on those elements with style rules like borders, background-gradients and rounded corners
  • Actually animating the elements

The reason for using CSS3 animation over the <canvas>-element is quite important: While browsers can optimize their elements performance (regarding their style, i.e., CSS), they cannot optimize our custom drawing routines used in a canvas. The reason for this lies in the browser’s ability to use hardware mainly the graphics card. Currently the browser does not give us direct access to the graphics card, i.e., every drawing operation has to go over some function in the browser first.

This problem could be prevented with techniques such as webgl, where the browser does give the developer direct access to the graphics card. However, this is treated as a security problem and will not become standardized. One important rule for developing WebApplications is standardization – since this is our portal to a huge customer base. If we excluded some of the most popular browsers, we would certainly lose a lot of potential visitors.

Starting with Canvas

Our basic HTML document looks like the following:

 Collapse | Copy Code
<!DOCTYPE html>
<html lang="en">
<meta charset="UTF-8"/>
<title>Animations in HTML5 using the canvas element</title>
<canvas id="canvas" width="1000" height="600">Your browser does not support the 
<code>&lt;canvas&gt;</code>-element. Please think about updating your browser!</canvas>
<div id="controls">
<button type="button" onclick="speed(-0.1);">Slower</button>
<button type="button" onclick="play(this);">Play</button>
<button type="button" onclick="speed(+0.1);">Faster</button>

Here we set the HTML5-Doctype and build a page containing a <canvas> for drawing the animation and some buttons contained in a panel (<div>). We could have shortened the document by omitting certain tags. One of the advantages of HTML5 is that each browser has to implement certain fallbacks, e.g., if a tag is not closed or if a certain tag is missing. The shown (more complete and verbose) form is my personal favorite.

A canvas alone would not make a great painting – nor does it give us some animation. What does the actual drawing? It’s some JavaScript! The required <script>-tag is already placed in the <head>-section. In this case, we will not use external JavaScript files to get the required lines of code. This makes only sense if you consider using not many lines of code (small size) and non-repeating JavaScript. Otherwise you could benefit from the browser’s caching as well as multiple (download) connections. Another important point by using external JavaScript files is to consider moving the script-tags to the bottom of the document in order to prevent performance problems.

Let’s start by declaring some variables:

 Collapse | Copy Code
var	dx = 5,				// Velocity at rate 1
	rate = 1, 			// The current playback rate
	ani,				// The current animation loop
	c,				// (Drawing) Canvas Context
	w,				// (Car, hidden) Canvas Context
	grassHeight = 130,	// Height of the background
	carAlpha = 0,		// Rotation-angle of the tires
	carX = -400,		// (x-)Position of the car (will move)
	carY = 300, 		// (y-)Position of the car (will stay constant)
	carWidth = 400, 	// Width of the car
	carHeight = 130,	// Height of the car
	tiresDelta = 15,	// Distance from one tire to the closest edge of the chassis 
			// of the car
	axisDelta = 20,	// Distance from the bottom of the chassis of the car 
			// to the axis with the tires being attached
	radius = 60;	// Radius of one tire

	// In order to init the car canvas (hidden) we use this anonymous function
(function() {
	var car = document.createElement('canvas');	// Creating the element
	car.height = carHeight + axisDelta + radius;	// Setting the appropriate 
						// attributes
	car.width = carWidth;
	w = car.getContext('2d');			// Now we can set the car canvas
})(); // Executed directly

As you can see, most of the variables are actually used as constants. This means the code will give you some freedom in order to adjust the dimensions of the car as well as some other parameters like the speed of the car. The change in the rotation of the tire can actually be calculated by using the ratio of dx to the radius.

Another interesting property arises by looking at the bottom of the code. Here I use an anonymous function. This can be very useful in some cases. This means that variables that are declared within the scope of the anonymous function like car will be deleted by the browser and most importantly will not cause any conflicts with other existing variables.

Why is there a second (invisible) canvas set up? First of all, this will be useful in the transition to CSS3. However, this approach is also very useful in general. By using a canvas for each major object we are animating, we will be able to keep their states as they are, i.e., if we do not have to redraw them we do not have to as supposed to the approach where we will wipe does drawings away with a clean of the main canvas. Another reason is that this model allows us to think more physical. We do not have to animate the chassis (moving) and the tires (moving and rotating) separately, but just as a whole thing meaning the tires (rotating) within the car (moving). This states the physical model more closely, where the tires are rotating which makes the car moving.

Starting the loop:

 Collapse | Copy Code
function play(s) {				// The referenced function, s is the button
	if(ani) {				// If ani is not NULL we have an animation
		clearInterval(ani);	// So we clear (stop) the animation
		ani = null;		// Explicitly setting the variable to NULL
		s.innerHTML = 'Play';	// Renaming the button
	} else  {
		ani = setInterval(drawCanvas, 40);	// We are starting the animation 
						// with 25 fps
		s.innerHTML = 'Pause';	// Renaming the button

This function has already been referenced by the HTML we’ve written. Here we just start or stop the animation depending on the current state that is displayed using the ani variable. The framerate has a maximum of 25 frames per second – this might not be the best choice. jQuery is using 77 fps (13 ms) as a default for its DOM object animation. In case of this simple animation, it should give a good insight. An important issue is that our logic (dx = 5) will be bound to those 25 fps. This is something to be careful about when building professional animations or even games.

Let’s have a look at the main drawing function that will be called in the loop:

 Collapse | Copy Code
function drawCanvas() {
	c.clearRect(0, 0, c.canvas.width, c.canvas.height);	// Clear the (displayed) 
							// canvas to avoid errors;		// Saves the current state of properties (coordinates!)
	drawGrass();		// Draws the background (grass)
	c.translate(carX, 0);	// Does some coordinate transformation
	drawCar();		// Draws the car (redraws hidden canvas)
	c.drawImage(w.canvas, 0, carY);	// Draws the car actually to visible canvas
	c.restore();	// Restores to last saved state (original coordinates!)
	carX += dx;		// Increases the position by the dx we set per time
	carAlpha += dx / radius;	// Increases the angle of the tires by the ratio

	if(carX > c.canvas.width)	// We keep some periodic boundary conditions
		carX = -carWidth - 10;	// We could also reverse the speed dx *= -1;

Basically we just redraw the image. The animation is actually coming from two little methods inside. First we use a translation of coordinates to always draw to the same coordinates but being placed on a different location. The second one is the incrementation of the car’s current position. Without one of those two calls, the car would not move at all! We also outsourced as much code as possible, making it more maintainable (unfortunately, this can also decrease JavaScript code performance).

Why are coordinate transformations so important? The offer us some nice possibilities: Rotating something without writing any mathematical function at all! Or as we have seen: We can just draw to the same coordinates but receive different results. This is the power of coordinate transformation. The translation method is called by using context.translate(dx, dy) and allows us to set a new center (0, 0). The usual center is the upper left corner. If we would rotate without translating at all, we would always just rotate around the upper left corner. In order to rotate around the center of a canvas, we could use something like:

 Collapse | Copy Code
context.translate(context.canvas.width / 2, context.canvas.height / 2)

Now we look at the function for drawing the car itself:

 Collapse | Copy Code
function drawCar() {
	w.clearRect(0, 0, w.canvas.width, w.canvas.height);	// Now we have to 
							// clear the hidden canvas
	w.strokeStyle = '#FF6600'; 		// We set the color for the border
	w.lineWidth = 2;			// And the width of the border (in pixels)
	w.fillStyle = '#FF9900';		// We also do set the color of the background
	w.beginPath(); 			// Now we begin some drawing
	w.rect(0, 0, carWidth, carHeight);	// By sketching a rectangle
	w.stroke();			// This should be drawn (border!)
	w.fill();				// And filled (background!)
	w.closePath();			// Now the close the drawing
	drawTire(tiresDelta + radius, carHeight + axisDelta);		// And we draw 
								// tire #1
	drawTire(carWidth - tiresDelta - radius, carHeight + axisDelta);// Same routine, 
							// different coordinates

Here we used some of the possibilities that the <canvas>-element offers us. On the hand we do use paths (very simple – and in this case more or less obsolete since we could have used the drawRect() and fillRect() method. Again we outsourced some code in order to obtain a better maintainability. In this case, this is totally justified since we just have one method to take care of (drawTire()) instead of two identical code blocks that form one big mess.

Finally, let’s have a look at the method for drawing one of the tires:

 Collapse | Copy Code
function drawTire(x, y) {;				// Again we save the state
	w.translate(x, y); 		// And we perform a translation 
					// (middle of the tire should be centered)
	w.rotate(carAlpha);		// Now we rotate (around that center)
	w.strokeStyle = '#3300FF'; 		// Setting the draw color
	w.lineWidth = 1;			// The width of the border (drawline)
	w.fillStyle = '#0099FF';		// The filling / background
	w.beginPath();				// Another sketching is started
	w.arc(0, 0, radius, 0, 2 * Math.PI, false); 	// With a full circle around 
					// the center
	w.fill(); 			// We fill this one
	w.closePath(); 			// And close the figure
	w.beginPath();			// Start a new one
	w.moveTo(radius, 0);		// Where we move to the left center
	w.lineTo(-radius, 0); 		// Sketch a line to the right center
	w.stroke();			// Draw the line
	w.closePath();			// Close the path
	w.beginPath();			// Start another path
	w.moveTo(0, radius);		// Move to the top center
	w.lineTo(0, -radius); 		// And sketch a line to the bottom center
	w.stroke();			// Draw the line
	w.closePath();			// Close it
	w.restore();			// And restoring the initial state 
					// (coordinates, ...)

Additionally, I’ve added a preview image by using the onload-event of the <body>-element. I also included a method to increase the speed of the animation by changing the value of the dx variable. The framerate will not be changed in order to increase or decrease the speed.

This is the result of our coding. A car moving around in a canvas!

This example can be viewed live at

Moving to CSS3

The basic HTML document now has the following format:

 Collapse | Copy Code
<!DOCTYPE html>
<html lang="en">
<meta charset="UTF-8"/>
<title>Animations in HTML5 using CSS3 animations</title>
<div id="container">
<div id="car">
<div id="chassis"></div>
<div id="backtire" class="tire"><div class="hr"></div><div class="vr"></div></div>
<div id="fronttire" class="tire"><div class="hr"></div><div class="vr"></div></div>
<div id="grass"></div>

The document looks quite similar to the one before. However, some things can be noted right away:

  • There is no <script>-element prepared. Instead, we will use a <style>-declaration.
  • There are no controls for setting the speed or play and pause. This is actually one of the key differences I will explain later (interaction).
  • The whole scene is already presented in HTML-elements. They are not styled right now, but they do represent the scenario we want to animate.

While animating something in a canvas is mostly a programming job, we clearly shifted towards a design problem here. We can only manage to animate the scene appropriately if we build a correct model of the scene using HTML-elements (well, <div>s). In this logical scenario, the <div id="container"> represents the canvas from before, <div id="car"> is the HTML-element that is equivalent to the hidden canvas of the previous example and <div id="grass"></div> is exactly the element that has been drawn using drawGrass() before.

The basic CSS outline is the following:

 Collapse | Copy Code
	position: relative;	/* Relative pos. - just that we can place 
				absolute divs in there */
	width: 100%;		/* Yes this will get the whole page width */
	height: 600px;		/* and 600px height */
	overflow: hidden;	/* Really important */

	position: absolute;	/* We position the car absolute in the container */
	width: 400px;		/* The total width */
	height: 210px;		/* The total height incl. tires and chassis */
	z-index: 1;		/* car is in front of the grass */
	top: 300px;		/* distance from the top (y-coordinate) */
	left: 50px;		/* distance to the left (x-coordinate) */

	position: absolute;	/* This defines the space of our car w/o tires */
	width: 400px;		/* The total width */
	height: 130px;		/* The height of the chassis */
	background: #FF9900;	/* Some color */
	border: 2px solid #FF6600;	/* Some thick border */

	z-index: 1;		/* Always in front of the chassis */
	position: absolute;	/* Absolute positioned */
	bottom: 0;		/* Will be placed at the bottom of the car */
	border-radius: 60px;	/* And there is our radius ! */
	height: 120px;		/* 2 * radius = height */
	width: 120px;		/* 2 * radius = width */
	background: #0099FF;	/* The filling color */
	border: 1px solid #3300FF;	/* And the border color and width */

	right: 20px; /* Positions the right tire with some distance to the edge */

	left: 20px; /* Positions the left tire with some distance to the edge */

	position: absolute;	/* Grass is absolute positioned in the container */
	width: 100%;		/* Takes all the width */
	height: 130px;		/* And some height */
	bottom: 0;		/* 0 distance to the bottom */
	background: -webkit-linear-gradient(bottom, #33CC00, #66FF22);
	background: -moz-linear-gradient(bottom, #33CC00, #66FF22);
	background: -ms-linear-gradient(bottom, #33CC00, #66FF22);
	background: linear-gradient(bottom, #33CC00, #66FF22); /* Currently we need 
							all of them */

.hr, .vr	/* Rules for both: hr and vr */
	position: absolute;	/* Want to position them absolutely */
	background: #3300FF;	/* The border color */

	height: 1px;		/* Linewidth of 1 Pixel */
	width: 100%;		/* Just a straight line (horizontal) */
	left: 0;
	top: 60px;		/* Remember 60px was the radius ! */

	width: 1px;		/* Linewidth of 1 Pixel */
	height: 100%;		/* Just a straight line (vertical) */
	left: 60px;		/* Remember 60px was the radius ! */
	top: 0;

Here we use classes and IDs to attach rules one or multiple times. This can be seen best by looking at the code for creating a tire. We do use two different IDs (fronttire and backtire) for distinguishing between different positions. Since both share the rest of their properties (and a rotation!), it was just useful to give them also a class (tire) which attaches some CSS rules to the corresponding elements.

The horizontal and vertical lines will be drawn using the classes hr and vr. We can use the position: absolute; andposition: relative; rules to set positions for each child-element within the mother-element. For the grass, we required a gradient to be drawn. This is no problem in CSS3. Well, maybe it is a small problem since the browser vendors are currently using vendor specific prefixes for those rules. This is an annoying fact and will hopefully vanish within the next year.

A really important rule can be found in the container itself. By setting overflow: hidden; we prevent the car from escaping from our beautiful container-world. This means that the car will behave the same way as in the previous scenario, where it was not possible to draw outside the canvas-area. However, one more really important thing is missing: the animations itself!

Definitions of the keyframes:

 Collapse | Copy Code
@keyframes carAnimation
	0% { left: -400px; }	/* Initial position */
	100% { left: 1600px; }	/* Final position */

@keyframes tyreAnimation
	0% { transform: rotate(0); }	 	/* Initial angle */
	100% { transform: rotate(1800deg); }		/* Final angle */

The first comment on those lines is a really important one: do not try making this at home! I do not show the real code here for a purpose. This is how it should look like. However, it does not (but it looks similar). The problem is that CSS3 keyframes are quite new and therefore (yes, you guess (or know) it) you require those prefixes again. Also CSS3 transforms do require prefixes. So just copy those lines and add -webkit- and -moz- and others (IE is about to come to the CSS3 animations party with version 10, hopefully Opera is about to follow!) in front of keyframes and transform.

The next thing you’ll notice is that the syntax is really clean and smooth. You just specify some keyframes (in this case, just one as an initial frame and one as a final frame) and the browser will calculate the rest. However, we cannot integrate such nice physical considerations like the relation between the angle and the position from the previous example.

Now that we specified some keyframes we need to integrate them! How does that look:

 Collapse | Copy Code
	animation-name: carAnimation;
	animation-duration: 10s;
	animation-iteration-count: infinite;
	animation-timing-function: linear;

	animation-name: tyreAnimation;
	animation-duration: 10s;
	animation-iteration-count: infinite;
	animation-timing-function: linear;

Again – be cautious when trying this at home. You will need some copy and paste plus you’ll have to add the appropriate prefixes in front of those entries. So this tells the browser which set of keyframes to take for the element and what kind of frame distribution within which time to consider. We also tell the browser the number of repetitions to perform. In this case, we will end up with an infinite loop.

This is the result of our styling. A car moving around in a div!

This example can be viewed live at

Comparison of the Two Methods

In my opinion, both of those methods have their benefits. The first one is actually pretty straight forward from a programmer’s perspective (however, it took quite long until this was possible in HTML / JavaScript). The second one does offer some really nice features and will most certainly be used in an amazing way to build websites that will certainly have an impact on the web’s future.

Pro Canvas-Animations

  • Interaction (like pause, slower, faster, …) is easily possible.
  • We can connect values to form one physical unit directly.
  • We do not need to know or have a complete structure before the animation.

Pro CSS3-Animations

  • The performance is definitely better.
  • We can use the full screen (100% does not work with the canvas element).
  • The frames are computed by the browser – we just specify the keyframes.

All in all, it will depend on the project (as usual). Even though I thought about some possible workarounds to actually stop an infinite CSS3-animation (and start again), it would be tedious to actually implement this in a bigger multimedia animation. Here the interaction possibilities are a clear indicator for a canvas-animation. On the other side, some small (non-interactive) animations could be done by the CSS3 technique. It provides a nice clear syntax that can be GPU accelerated by modern browsers and can be styled to fit everywhere.

Points of Interest

While writing a lot of <canvas>-animations, I never had to opportunity to write an actual CSS3 animation. This was certainly related to the propagation of this really new technique. Now that Firefox implemented the keyframes and with Microsoft up to come we will most certainly see more keyframe rules online. I thought that this technique was quite hard to write and got really surprised how easy it is. I actually had more problems with the linear gradient syntax than the keyframes one.

My only hope is that browser vendors will stop rolling out CSS rules with those prefixes. In nightly builds or such they are acceptable, but having something in a final build that is either not the standard (why rolling it out?) or standardized (why prefixing it?) but with a prefix is just a mess for web developers.

Connect to a PHP Server using HTML5 WebSockets

Before HTML5 introduced WebSockets, you had to rely on specialized technologies like Java applets (remember those?!) or, more recently, Flash. I covered some of the challenges of working with WebSockets in this early stage of development in my Making HTML5 WebSockets Work article.

In this article, we’ll be utilizing WebSockets to write a kind of automated web chat, where the other party is a machine of great intellect, called “Multivac”, named after the Chatbot demo of the Google phpwebsocket project.

As mentioned in the conclusion of my last article, we will be using the lemmingzshadow php-websocket library because it supports the newer draft-ietf-hybi-thewebsocketprotocol-10 (HyBi-10) WebSockets protocol. You’ll also need a web server that supports PHP, such as Wamp or Apache, and a fairly up-to-date browser; I would recommend google Chrome 14+ or Firefox 6+ as the best choices.

WebSocket Connections

When you see all of the files involved in managing the WebSocket connections, it quickly becomes apparent why it’s not advisable to write your won server-side code from scratch, unless socket communication protocols and other low level operations really float your boat. Nonetheless you should have some familiarity with the files that you’ll be working with; some of them, you’ll have to modify, others, are better left alone! Here’s a rundown of what’s contained in the lemmingzshadow php-websocket download:

  • server.php: Creates the WebSocket server instance, sets server properties, and registers the application to be called by the client component.
  • SplClassLoader.php: Implements the technical interoperability standards for PHP 5.3 namespaces and class names.
  • Connection.php: Manages the WebSocket Connection. The hybi10Decode() and hybi10Encode() functions are located therein.
  • Server.php (upper case ‘S’): It’s the “real” server which monitors the port for client requests.
  • Socket.php: Creates and maintains references to sockets.
  • Application.php: An abstract class that contains the getInstance() method, which ensures that applications are created as singletons, as well as the onConnect(), onDisconnect(), and onData() abstract methods which all applications must implement.
  • ChatApplication.php: A fully functional simple webchat application.
  • DemoApplication.php: A Websocket-Server demo and test application. It can either echo data back to the client or execute a user function.
  • EchoApplication.php: Even simpler than the demo app, this class just parrots back whatever data is sent to it. This is the class upon which our Multivac app will be built.
  • chat.html: The client component of the chat application.
  • client.php: A basic WebSocket client that runs from a command line.
  • echo.html: The client component for the echo application. This is the client-side component that we will be modifying today.
  • index.html: An extremely basic WebSocket server test console.
  • client.css: A stylesheet to spiffy up the browser UI.
  • client.js: A JavaScript template with everything you need to create and communicate with a WebSocket.
  • jquery.min.js: A local minified jQuery script – useful for local development.
  • json2.js: A Douglas Crockford JSON parser from

Here is how the folders should be structured under your server’s web document root:

Image 1

The Client Page

Open the echo.html page in your favorite HTML editor and you’ll see that most of what we need is already done for us. As it stands, when you enter text in the Message textbox and click the Send button, the data travels to the server.php page, is decoded, re-encoded and sent back to the browser to display in the log


not very exciting to be sure, but it does demonstrate that all of the basic processes are functioning just as they should. What we’ll do is tweak the code so that the server sends back a response based on our request. The object of this exercise is to familiarize ourselves with what code to change to develop an application based on existing code.

With regards to the client page, the first thing we’ll do is rename it to something more instructive. Let’s call it “multivacClient.html”.

Next, we’ll add a couple of new CSS classes so that we can differentiate our requests from the server responses:

.human { color: blue; }
.computer { color: green; }

We can now update the log function to accept a class argument. It’s called clazz because class is a reserved word:

var log = function(msg, clazz){ $('#log').append('' + msg + '
') };

A second way to differentiate the client requests from the server responses is to prefix them with a direction arrow: hence, requests have a right facing arrow (>) while the responses have a left facing one (

log('> ' + msg, 'human');

…and socket.onmessage():

log('< ' +, 'computer');

Another change that I would make is to update the code that creates the WebSocket object. I am not a fan of browser sniffing, so I would replace the existing if statement with the following:

var Socket = window.WebSocket || window.MozWebSocket;

		$('#log, input, button, span').fadeOut("fast");

You need a browser that supports WebSockets such as Google Chrome.

	} else {

The Application File

Now it’s time to work on the EchoApplication.php file. Open it up and save it as “MultivacApplication.php”. Rather than take the received data and spit it right back, we’ll send it to our process() function:

    public function onData($data, $client)
        foreach($this->_clients as $sendto) {
            $this->process($sendto, $data);

The process function evaluates the $action parameter and sends back the appropriate response. I will admit that the AI component of this app is a little weak, but IBM wouldn’t lend me the code for Watson:

  private function process($sendto, $action) {
	    case "hello" : $sendto->send("hello human");                        break;
	    case "hi"    : $sendto->send("greetings human");                    break;
	    case "name"  : $sendto->send("my name is Multivac, super-genius."); break;
	    case "age"   : $sendto->send("I am far too vain to say");           break;
	    case "date"  : $sendto->send("today is ".date("Y.m.d"));            break;
	    case "time"  : $sendto->send("server time is ".date("H:i:s"));      break;
	    case "thanks": $sendto->send("you're welcome");                     break;
	    case "bye"   : $sendto->send("bye");                                break;
	    default      : $sendto->send($action." not understood");            break;

The server.php File

This file is where you would set your basic initializing details such as the server name or IP, port number, as well as your application name and class. There’s also an added security feature to check the “Origin” header in WebSockets handshake. Though it could still be potentially spoofed, browsers always add the Origin of the page which initiated WebSockets connection:

$server = new \WebSocket\Server('localhost', 8000);

// server settings:

$server->registerApplication('multivac', \WebSocket\Application\MultivacApplication::getInstance());

Note that the application name (‘multivac’) must match the Socket call in your JavaScript code:

var socket = new Socket('ws://localhost:8000/multivac');

Running the Application

Open a command window and type the following to launch the WebSocket server process and register our application:


Here’s what I entered in my Windows DOS prompt:

C:\Program Files\wamp\bin\php\php5.3.8>php.exe "C:\Program Files\wamp\www\phpwebsocket\server\server.php"

If all went well, you should see a confirmation message that says “Server created”.

Start up your web server and make sure that PHP is enabled and working.

Launch your web browser and enter the URL of the multivacClient.html page:


Look for the green online indicator in the top-right corner to confirm that you are connected to the WebSocket server.

Try some commands, then click the Disconnect button when you’re done.

Image 2


There’s a lot that can go wrong when a technology is made up of so many interconnected parts, but hopefully you’ve learned enough here today to avoid the most common pitfalls. WebSockets is definitely something to watch to see where it leads. I for one can’t wait!

HTML5 and Mobile Web

By forcing Web developers, and ultimately Adobe, out of the Flash business, Apple made HTML5 apps better. That’s good for Safari users, but it’s also good for users on other Web platforms, like Android. If there’s a truly good universal platform for online apps, it stands to reason that the smart developer will build apps for it, since this way the app will be available to the largest number of users. Right?

Furthermore, now that Adobe has HTML5 religion, the company is releasing high-quality HTML5 developer tools, migrating its Flash developers over to the new platform. So we should be about to see a flood of new Web-based mobile apps.

All this appears to be just as Apple intended: Steve Jobs’ campaign to rid the world of Flash is succeeding. The Web is getting better apps and the Web-browsing experience on Apple’s mobile devices is getting better.

But this could be bad news for Apple’s lucrative App Store business. While Apple takes a 30 percent cut of all app sales through its store (still the only way for consumers to get apps), Apple gets 0 percent of “Web apps” loaded up through the browser. The better HTML5 gets, the less developers will write apps, less money Apple will make, and the less unique the iPhone and iPad will be.

I talked with the CEO of a Web company who is excited about the advances in HTML5 because he sees a future in which the devices and operating systems are on the same level. It’s a world where new products, like Microsoft Windows phones, can come along and be instantly competitive. And where companies like his can have a shot at dominance in their market niche after building only one app, the HTML5 version.

In other words, Apple, in pushing the world toward HTML5, is killing its own golden goose.

Or is it?

The counter argument comes from another CEO, one who’s built a successful business around both Web services, and software for mobiles and traditional computers. He says that apps are here to stay, because the “lowest common denominator” strategy (using the Web) doesn’t cut it for developers ultimately, nor for consumers, and most importantly not for the tech megaliths behind mobile operating systems: Apple, Google, and Microsoft.

The challenge for apps developers, my source explains, is getting their apps seen for more than a fleeting moment. For those making Web apps, there’s just no good way. Even a good review of a Web app on a popular site has only a temporary impact. The way to get your app in front of potential customers, time and again, is to get it featured in an app store.

You do that by building an app that highlights its unique hardware capabilities–the features that the hardware company is using to sell the product. These will likely be features that you can’t access today, or in the foreseeable future, with a Web app. This isn’t because HTML5 won’t advance, but because the device and OS manufacturers will always do their best to keep their products somewhat ahead of the lower-common-denominator Web platform. It is how they sell products.

So if you make a mobile app, you want to feed into that trend, because it will feed you.

And what about HTML5? It’s good for apps that don’t depend on the app stores for sales. The means enterprise apps, essentially. HTML5 has its place in consumer apps, too, including inside many successful mobile apps. But not for the core features or the main UI.

Relying on HTML5 to quickly get to broad compatibility across the mobile landscape is a trap, my guy says. To succeed, you need to tackle each platform separately. In fact, he says, you might want to build apps that only work on the latest and greatest version of a phone, and intentionally not on previous models. Yes, fewer people will be able to use it. But everyone who buys the new toy will. The more your app makes the hot new hardware look good, the more it’ll get promoted by the hardware or OS manufacturer. That can give your app presence it could not otherwise get. Once your product is succeeding on the brand-new hardware, you can start adapting it to other platforms.

Try to conquer the entire mobile world at once and you’ll have no marketing partners. Or, put more starkly: It’s hard to win when nobody loses.