Christian Heilmann

Giving “image swivel” the vanilla web diet treatment

Friday, May 17th, 2013 at 12:05 am

With my chapter for the upcoming Smashing Book 4 done and talking about “The Vanilla Web Diet” and some workshops in the making I thought it is a good idea to show in an example how to make a seemingly good solution better by rethinking it with the principles of the vanilla web diet in mind.

The solution I am targeting is the image swivel effect that was posted a few days ago on CSS tricks. I sent my solution to the author and he was very interested in getting more information as to the why. The effect as it stands works and does the job to inspire people to play with it. The comments showed that where lots of developers had a go at creating their own solutions fixing some of the issues I am about to cover here. My favourite probably is Eduardo García Sanz’s Pure CSS solution which to me is a good plan to achieve an effect like that if you don’t need touch support or keyboard access.

Before I get accused of “hating” – the code is OK as a demo to get creative juices flowing but it has many issues that would show up when used in production. It is also mouse-dependent and doesn’t work on touch devices. Therefore I’ll start by stating what I find troublesome about the published solution and then explain how to approach the problem to make it as simple and maintainable as possible. I also add support for mouse, keyboard and touch access and try to achieve the effect whilst not blocking people out.

If you prefer to see this as a screencast, here is one I recorded explaining the ideas, the issues I found with the original code and how my solution works. This is live and unscripted.

How much markup do we need?

Let’s start by looking at the solution that is shown in the CSS tricks article.
The HTML is the following:

<div id="faces">
  <div id="face-area">
    <div id="image-1" style="display: none;">
      <img src="/images/look-left-3.jpg">
    <div id="image-2" style="display: none;">
      <img src="/images/look-left-2.jpg">
    <div id="image-3" style="display: none;">
      <img src="/images/look-left-1.jpg">
    <div id="image-4" style="display: none;">
      <img src="/images/look-center.jpg">
    <div id="image-5" style="display: none;">
      <img src="/images/look-right-1.jpg">
    <div id="image-6" style="display: none;">
      <img src="/images/look-right-2.jpg">
    <div id="image-7" style="display: none;">
      <img src="/images/look-right-3.jpg">
    <div id="the_faces_overlay">
      <div class="the_faces" data-number="1"></div>
      <div class="the_faces" data-number="2"> </div>
      <div class="the_faces" data-number="3"></div>
      <div class="the_faces" data-number="4"></div>
      <div class="the_faces" data-number="5"></div>
      <div class="the_faces" data-number="6"></div>
      <div class="the_faces" data-number="7"></div>
  </div><!-- END #face-area -->
</div> <!-- END #faces -->

Warning sign : Repeated HTML structures without logical connections

This is a very common mistake I see when starting an effect like that. We create something that contains the content and then something that gets acted upon to show the effect. Thus we lose the benefit of already interacting with the parent element and using event handling to find out where we are. I guess historically this is based on “CSS only solutions” that needed that kind of separation as you could not calculate or detect mouse position in CSS.

As we create two separate sets of markup for an effect we need to find a way to connect the element that was interacted with to the one we want to show. This means adding IDs to all the elements, classes to all the elements we interact with and a data attribute to tell which of the images to show. This is bad for maintainability. If we add an image, we need to also add an element to interact with. Great code is catered to making maintenance as easy as possible. Here we have a lot of dependencies to deal with when adding or removing an image.

Warning sign : Lack of semantic markup

The other gripe I have with this HTML is that it means nothing at all. As the order of the images is important, the right HTML construct to use here is an ordered list.

Warning sign : Lack of alternative content

Another big HTML mistake is adding images without an alt attribute. This means that screenreader users would get the file path of the images read out to them. Either provide a sensible alternative text or add a alt=”” to hide the image from screenreaders.

Warning sign : Dependency on the number of elements

One big warning sign to me here is that our effect is dependent on the number of images in the widget and that we need to have the data attributes and the IDs maintained together although they are on different elements.

The more elements we add means the more IDs we need to maintain. This is not what coding is about. Computers are good at calculating things for us.

Warning sign : Empty elements and inline styles

Whenever I see inline styles I know something went wrong. There is no point in them and if ever they should only be generated by code, not by humans. The same with empty HTML elements: you probably did some extra work that is not needed. HTML is there to contain content or provide interaction. If you have a lot of empty DIVs without an obvious templating use case, something went wrong.

Giving no power to CSS

The CSS of the solution is not much, and it doesn’t do much either. This is a shame seeing how much easier it is for a visual maintainer to change CSS rather than changing JavaScript.

body {
  background: #333 
#faces {
  height: 333px;
  width: 500px;
  margin: 0 auto;
  border: 8px solid white;
#face-area {
  height: 500px;
  width: 333px;
  position: relative;
#the_faces_overlay {
  position: absolute;
  width: 500px;
  top: 0;
  left: 0;
#faces .the_faces {
  height: 333px;
  width: 14.2857143%;
  float: left;
  margin: 0;
  padding: 0;

Warning sign : CSS dependent on the amount of elements

The glaring issue here is the “width: 14.2857143%;” which is calculated by dividing 100% into seven parts. This means that if you delete an image from the HTML, you also need to change the CSS width here. You should never be dependent on the amount of elements in your CSS, as those are prone to change. In this case especially there is no logical way to find out why this is the width. CSS calc() can at least make that obvious but in general it is a bad idea to create look and feel that is tied to a certain amount of elements.

Goldfish jQuery

The jQuery code to make the effect work is very short:

// Reveal the "center" image
var centerImage = $("#image-4").show();
// Bind hovers to each column
$(".the_faces").each(function() {
  $(this).on("mouseover", function() {
    $("#image-" + $(this).attr("data-number")).show();
  }).on("mouseout",function() {
    $("#image-" + $(this).attr("data-number")).hide();
// Reset center image
$("#face-area").on("mouseleave", function() {;
}).on("mouseenter", function() {

It is, however, very demanding to the browser. Slowing down a browser can be done in many ways – the most damaging ones are heavy computation, accessing the DOM and lots and lots of event handling. The latter two is what we do here.

Warning sign : lack of data caching

We loop over all the elements with the class “.the_faces” and add a mouseover and mouseout handler to each of them. Every time these get fired, we read an attribute, create a string with it and access an element that has the ID of the string and show or hide it. Showing and hiding using the jQuery methods is another access to the DOM as it manipulates the display style property. We show and hide the “center image” upfront and also on another event hander on the overall parent element.

If we were to add touch and keyboard handlers we’d triple the amount of assigned event handlers as we apply them on each image.

I call this Goldfish code – we keep asking the browser for things we should already know. The widget interface we have here is static HTML - there is no loading of content, no changes in it. Therefore there is no point in continuously reading out what the data-number attribute is and ask the browser to find the element with a certain ID. Caching results is a very simple thing to do and the performance benefits are amazing.

Rethinking the solution the vanilla web diet way

I approached the solution by looking at what we need to do here:

  • We have a widget of a certain size with images in it
  • We have an unknown amount of images – it should be dead easy to remove or add one or replace them all
  • Moving the mouse over the widget should loop through the images, also touching the widget should do so and it would be nice to be able to flip forward and backward with the keyboard
  • Should things not work out it would be nice to have an display that still makes sense

The HTML - we don’t need IDs or data attributes

<a href="/productpage" id="rollover">
  <li><img src="pics/IMG_0518.jpg" alt""></li>
  <li><img src="pics/IMG_0517.jpg" alt=""></li>
  <li><img src="pics/IMG_0516.jpg" alt=""></li>
  <li><img class="current" src="pics/IMG_0515.jpg" alt=""></li>
  <li><img src="pics/IMG_0519.jpg" alt=""></li>
  <li><img src="pics/IMG_0520.jpg" alt=""></li>
  <li><img src="pics/IMG_0521.jpg" alt=""></li>

We should link this to something – after all a beautiful effect like that should react in some sale or deep-dive, right? Good thing is that in HTML5 links can contain other elements. So all we add is a link. This automatically gives us keyboard access to the widget – something otherwise we’d have to create by using roaming tabIndex.

As the order of the images is important, an OL is the right element to use. Each image has an empty alt attribute to ensure there is no hassle with screenreaders. Instead of defining which image to show as the first one in our JavaScript we keep this maintained in HTML, too, by adding a class of “current” to the image.

Showing and hiding with CSS

body {
  font-family: arial, sans-serif;
#rollover.js {
  display: block;
  margin: 2em;
  z-index: 3;
  position: relative;
  height: 270px;
  width: 200px;
  cursor: none;
#rollover.js img {
  width: 100%;
  position: absolute;
  top: 0;
  left: 0;
  visibility: hidden;
#rollover.js img.current {
  visibility: visible;

As our functionality is dependent on JavaScript, our styling should be, too. This we can achieve by adding a “js” class to the element when JavaScript is available and only apply the styles when needed. This CSS gives the widget a fixed size and positions all the images stacked inside it.

Instead of doing a hide() and show() in JavaScript, all we need to do is to apply a class of “current” to the element we need to show. All the hiding and showing is thus done in CSS which means that in the future we’d want to do other visual things with the “shown” and “hidden” images, all we need to change is the CSS.

  if (document.querySelector) {
    var rollover = document.querySelector('#rollover');
    rollover.className = 'js';
    var images = rollover.querySelectorAll('img');
    var all = images.length;
    var width = rollover.offsetWidth;
    var ox = rollover.offsetLeft;
    var boundarywidth = width / all;
    var current = 0;
    var x = 0;
    var index = 0;
    var touched = false;
    var setcurrent = function(index) {
      if (images[index]) {
        images[current].className = '';
        images[index].className = 'current';
        current = index;
    var findindex = function(x) {
      index = parseInt((x - ox) / boundarywidth, 10);
      if (index !== current) {
    rollover.addEventListener('mousemove', function(ev) {
      if (!touched) {
    }, false);
    rollover.addEventListener('touchstart', function(ev) {
      touched = true;
    }, false);
    rollover.addEventListener('touchend', function(ev) {
      touched = false;
    }, false);
    rollover.addEventListener('touchmove', function(ev) {
      if (touched) {
    }, false);
    rollover.addEventListener('keydown', function(ev) {
      var key = ev.char || ev.key || ev.which;
      if (key === 37) { index = index - 1;}
      if (key === 39) { index = index + 1;}
      if (index < 0) {index = 0;}
      if (index > all - 1) {index = all - 1;}
    }, false);
    if (rollover.querySelector('.current')) {
      for (var i = 0; i < all; i++) {
        if (images[i].className === 'current') {
          current = i;
    } else {

Quite longer than the jQuery, but bear with me as this does a lot more and in a much less demanding fashion. We wrap our code in a closure to makes sure we don’t leave any nasty globals behind. That should always be the first step.

Then we test if the browser supports document.querySelector. This is the standard answer to jQuery’s $() and is supported by lots and lots of great browsers. It is not supported by old and outdated browsers, which is why it is a good idea to test for it. This means that old Internet Explorer versions will not get the effect but instead they get the images as a numbered list (as we made the CSS dependent on a class applied with JavaScript). This is good, as it means we don’t need to test on these old browsers, which is hard to do and frankly a waste of our time.

We get a reference to our rollover widget using document.querySelector() and add the “js” class to it. This hides all the images and sets up the look and feel of the widget – all maintained in CSS. No need to loop through a lot of elements or use inline styles to hide them.

Next we get references to all the DOM elements we need and calculate what we need to find out what to show when the mouse cursor or the finger is on a certain part of the widget.

First we get all the images and store them in images. This will not change while the page is open, and querySelectorAll() gives us references to all of them. We store the amount of images in all to compare against later on.

Next we do the thing dynamically that the original solution did by hand – calculate the width of the different strips of the interaction that shows and hides images. We do this by reading out the offsetWidth of the widget as it is defined in CSS, so we don’t know and don’t want to hard-wire any widths in our JavaScript. We find out how far left the widget is in the browser by reading the offsetLeft property and store that in ox. Then we calculate the width of the interaction boundaries by dividing the width of the widget by the amount of images and store that in boundarywidth.

This makes the widget flexible to change any time the CSS changes or the amount of images does. If you remove or add an image, the width of the boundaries is calculated newly. No need to change the CSS to reflect that width. We made this now maintainable simply by adding or removing list items.

We define the current shown image index as 0, preset an x variable as 0, define the current index of the detected movement as 0 and set touched to false.

The index will be the index of the image to be shown at a certain point in the interaction. The index of the currently shown one is be stored in current, x will be detected position of the mouse or finger on the screen and touched defines if the screen is currently being touched or not.

The setcurrent() function hides the last shown image and shows a new one by shifting the “current” class from one to the other. It then stores the new image index in current. This is a very simple way to show a new state in a collection of things that can only show one at a time. No need to ask the browser which one is visible when we can store this in a variable like current.

The findindex() function converts the detected horizontal position of the mouse cursor or the finger of the user into an index of our image array. All you need to do is to subtract the left position of the widget itself and divide the value by the boundarywidth. Convert it to an integer and compare it to the current index and if it differs, call setcurrent().

All that is left is to assign the event handlers to make the magic happen. The first is a mousemove listener on the widget that calls findindex() when no touch happened. The current horizontal mouse position is stored in clientX of the mousemove event.

Touch interaction needs to be initiated (at least in Chrome in my testing here) so we set the Boolean of touched to true when a touchstart happened and to false when touchend was detected.

When a user moves a finger over the widget the browser fires the touchmove event and the current horizontal position is stored in the clientX property of the changedTouches array. We only detect the first finger in this case.

Keyboard detection doesn’t give us a position on the screen, so all we do is manipulate the image index directly. We listen for a keydown event and check the code of the key that was pressed. If it is the left arrow we subtract one from the current index and in the case of the right arrow we add one to it. We ensure that the index stays in the allowed limits and call setcurrent().

The last thing to do is to show the current image. If there is one with the right class in the HTML we need to find out its index and we do that by looping over them until we find the right one. If there isn’t any we just apply the current class to the first image (as defined at the start of the script).

Many solutions to the same idea

I hope this gives you an idea of how to approach an effect like that when you want to put it in production. There are of course other ways of doing it, but I wanted to ensure a few things that get very often forgotten:

  • The whole effect is now generated from the HTML, so all you need to do to create a new swivel is add other images
  • The whole look and feel is defined in CSS and you can resize the widget without having to worry about the size of the different boundaries
  • It can be used with a keyboard, the mouse or on touch devices
  • The DOM interaction is kept to an absolute minimum which means the performance on low spec devices is much, much better
  • There is no jQuery dependency

Share on Mastodon (needs instance)

Share on Twitter

My other work: