While watching Nicholas Zakas’ Maintainable JavaScript talk at the Fluent 2012 conference, there was a very informative section in there about keeping JavaScript separate from the HTML, and other similar concerns of separation.
You can see it from the 25:40 section of the video.
Hi Paul,
Thanks for posting the link to that talk. I was reminded of a lot of things that I should be doing more often 
I do however have one question:
At 27:02 in the video, he tells us to “Keep your HTML out of JavaScript”, citing the reason that if there is something wrong with the layout, you shouldn’t have to find the problem in a JS file.
E.g. this is bad:
var element = document.getElementById("container");
element.innerHTML = "<div class=\\"popup\\"></div>"
By way of a solution he proposes using some form of template (e.g. Moustache or [URL=“http://handlebarsjs.com/”]Handlebars), or pulling back markup from the server and injecting it into the page.
So far, so good, right?
Well, the other day I saw you post this code, with which you can generate random numbers for a lottery.
I noticed that if you pick nine or more numbers, the numbers overflow their container. So, I thought I’d change the markup and make the box expand to fit the numbers.
As you might already guess, I couldn’t find the markup in the HTML section of the fiddle, but instead found it buried deep in the JS. I had a quick play with it, but soon gave up owing to the confusing nature of the escaped backslashes, the inline CSS and the string concatenation.
container.innerHTML = '<div id="container' + idx + '" style="' + 'position:relative;' + 'width:160px;height:50px;' + 'font-family:verdana,arial,sans-serif;' + 'font-size:12px;' + 'color:#000000;' + 'background-color:#fffff0;' + 'text-align:center;' + 'border : 1px solid #000000">' + '<input type="button" id="play' + idx + '"' + 'value="' + buttonText + '" style="margin:5px">' + '<div id="result' + idx + '" style="' + 'width:150px;' + 'font-family:verdana,arial,sans-serif;' + 'font-size:12px;' + 'color:#000000">' + initialText + '<\\/div><\\/div>';
document.body.appendChild(container);
I know this code was only intended to demonstrate a specific functionality, but this got me to wondering, how we might apply what Nicholas Zakas tells us in his talk.
I would be interested in your thoughts on this.
Removing the CSS from the JavaScript that’s intended to be HTML, would be a good first step to things.
Then, the remaining HTML could be placed in to the web page as hidden content, or in to a template file of some kind for later use.
How do you feel about refactoring that code, as an example of how to apply such better practices?
Sure!
Ok, I’ve done this: http://jsfiddle.net/hibbard_eu/8R42e/
As you can see, I’ve created a template div element, which I’ve hidden using CSS.
I can then clone this with JS and manipulate it as required before inserting it into the page.
The code could probably do with some refactoring, but does this seem like a good start?
I’ve also used classes for the styling, as the dynamically generated elements are assigned varying ids, e.g.
container.innerHTML = '<div id="container' + idx + ...
Here’s an interesting trick that gets used far less often than it should.
<script id="list-template" type="text/x-handlebars-template">
<p>YUI is brought to you by:</p>
<ul>
{{#items}}
<li><a href="{{url}}">{{name}}</a></li>
{{/items}}
</ul>
</script>
This way, your markup stays in your HTML files; there’s no need for any backslash escaping; and the browser won’t render this template because it’s wrapped in a script tag.
Then somewhere in your JavaScript (this uses YUI, but you get the gist):
// Extract the template string and compile it into a reusable function.
var template = Y.Handlebars.compile(Y.one('#list-template').getHTML());
// Render the template to HTML using the specified data.
var html = template({
items: [
{name: 'pie', url: 'http://pieisgood.org/'},
{name: 'mountain dew', url: 'http://www.mountaindew.com/'},
{name: 'kittens', url: 'http://www.flickr.com/search/?q=kittens'},
{name: 'rainbows', url: 'http://www.youtube.com/watch?v=OQSNhk5ICTI'}
]
});
Yes, that’s a good start.
The following can be removed entirely:
template.style.display = 'block';
because it is only the template section that is hidden. When it’s copied out using cloneNode and renamed, it’s no longer hidden.
The template doesn’t need to be appended to the page immediately after it has been read too.
container.innerHTML = '';
var template = document.getElementById('template').cloneNode(true);
container.appendChild(template);
template.id = "container" + idx;
template.className = "container";
Instead, changes are better off being made before it has been added to the page, which helps to avoid performance issues too.
var template = document.getElementById('template').cloneNode(true);
template.id = "container" + idx;
template.className = "container";
...
container.appendChild(template);
There’s not much performance-wise to currently worry about, but I plan to take that appendChild down below the element updates and the event attachments.
The section with the roll button can be updated now.
var button = template.getElementsByTagName("input")[0];
button.id = "play" + idx;
button.value = buttonText;
...
var controlButton = document.getElementById("play" + idx);
if (window.addEventListener) {
controlButton.addEventListener("click", justOnce, false);
} else if (window.attachEvent) {
controlButton.attachEvent("onclick", justOnce);
}
The button variable can be renamed to something a but more expressive, to playButton.
And that event code is too complex for its own good. We can either move the cross-compatibility addEvent code out to a separate function, or we can use a more simple convention instead. Let’s go with the simple and use onclick.
We can also entirely get rid of the controlButton variable too.
var playButton = template.getElementsByTagName("input")[0];
playButton.id = "play" + idx;
playButton.value = buttonText;
playButton.onclick = justOnce;
That’s much better.
There’s a minor conflict after that in the justOnce function because it is expecting the controlButton global variable, but that’s easily fixed, by using the this keyword instead.
And why is the b variable there in the justOnce function? Normally it would be evt or some other similar event keyword, but the variable is not used at all in this function, so that variable can be removed too.
Before:
function justOnce(b) {
controlButton.blur();
...
}
After:
function justOnce() {
var button = this;
button.blur();
...
}
After these changes, we now are left with a structure where the template is obtained, changes are made to the content of that template, and at the end of things the template is placed inside of the container.
function lotto() {
// functions
...
var container = document.getElementById('container');
// updates to the template
...
container.innerHTML = '';
appendChild(template);
}
Now because I’m lazy developer, I’m going to have JSLint help me with tidying up the rest of this code, which results in the following code: http://jsfiddle.net/pmw57/8R42e/15/
Now that the code is structured and safe, there are many other improvements that should be made to it as well.
[list][]The large list of variables means that we should break the main lotto function up in to separate parts
[]I want to create a single global lotto object, which we can use to initialize button text values and to act as the single interface for our script
[]The numsort function can be much simpler, and achieve the same job
[]Most of the code should be broken up in to separate functions, dedicated to just one task
[/list]
And the list of improvements goes on
I have to head off, but as something to look forward to, here’s what the roll function looks like after having made a few improvements:
function roll(button) {
if (!validateRollingOptions(opts.pick, opts.from, opts.to)) {
return false;
}
var draw = drawNumbers(opts.pick, opts.from, opts.to);
draw.sort(numericSort);
updateResult(draw.join(' '));
setupNextRoll({
target: button,
totalRolls: opts.totalRolls,
rollDelay: opts.rollDelay,
callback: roll
});
}
Thanks for pointing that out, Jeff.
I’ve been wanting to have a look at Handlebars for a while now (along with so much other cool stuff - where does the time go??).
Maybe when we’ve worked through refactoring the code here, it would be possible to convert it to use Handlebars (and/or Moustache) as a final step.
Hi Paul,
I’ve just finished reading through your first refactoring of the code.
Thank you very much for this, as writing maintainable code, which reflects current best practices is definitely something I could be better at. It is certainly very educational to see how somebody else would go about things.
I should also get into the habit of using JS lint more often. I also ran the code through it so that I could better comprehend your final Fiddle and was quite surprised at the amount of warnings it spat out.
Most of them were logical really, but these are all things which can potentially make your code less error prone in the long run.
Allow me:
function numsort(n1, n2) {
return (n1 - n2);
}
var draw = [19, 32, 18, 36, 4, 43, 5] ;
console.log(draw.sort(numsort));
// [4, 5, 18, 19, 32, 36, 43]
I am looking forward to the second instalment. Things look as though they are shaping up quite nicely.
Thanks.
Going through the code at http://jsfiddle.net/pmw57/8R42e/15/ there are more issues with it that was dealing with now.
[list][]It is tempting to define a global object for the code, by that is not required yet at this stage
[]the “choose your lottery format” piece needs to be moved out as configuration options
[]the configuration options will need to be provided through a global object for the code, so we now have a good reason to define the code as a global object
[]the playing variable can be renamed to something a bit more expressive, such as isRolling
[]timer doesn’t seem to be needed, as we are not stopping the timed event at any stage
[]counter is for the number of times the numbers are rolled, which should be moved out to a configurable options object
[]idx and individually numbered identifier should not be required. Experiments with having multiple lotto rollers will need to be done in order to troubleshoot any remaining issues with that
[]container and template need to be configurable options
[]the play button and result sections should be not be retrieved from the page, but instead be assigned when the lotto section is being created
[]numsort should be replaced with much simpler code, and be renamed to numberSort
[]the roll function needs a much simpler structure, which should result inonly three functions being called. One for validation, one to get the drawn numbers and one to show them
[]the template modifications should be moved to a separate function
[]the check that getDocumentById exists should throw an error if it doesn’t
[]the firstChild check shouldn’t be needed due to the refactoring, but testing will confirm that
[*]the onsubmit function should just pass the form element to an init function[/list]
Those are the broad strokes of what needs to be done, which I’ll temporarily attempt to work on from the ipad for a while.
The timer that is declared in the lotto function is used I only one place, and that is where the decision on whether to keep the rolling going on occurs.
timer = setTimeout(roll, 50);
if (counter > 50) {
clearTimeout(timer);
isRolling = false;
counter = 0;
}
By reorganizing that, we can remove the need for clearing the timer.
But, I cannot do this with my ipad because when I head off to edit some code, I’m forced to reload this comment page when I return, which forces me to lose all that I have typed up before.
The third time is the charm, time to admit defeat and do something else until at get back to a real computer.
Okay now - removing the need for the timer.
We can check if we need the timer before setting it. That will allow us to remove the need for the separate timer variable.
if (counter <= 50) {
setTimeout(roll, 50);
else {
isRolling = false;
counter = 0;
}
Those 50’s each have different meanings, so we should place those in a config options area.
var lotto = function (pick, from, to) {
var opts = {
totalRolls: 50,
rollDelay: 50
},
...
After other refactorings have occurred to simplify things, we’ll replace those function parameters with a single options object instead.
The playing variable can be renamed to isRolling, and we can come back to that one later to decide if it’s better as a variable, or a property on the button.
buttonText and initalText can go in to the opts object:
var opts = {
buttonText: "Lotto Lucky Dip",
initialText: "Your Lucky Numbers",
...
playButton.value = opts.buttonText;
...
resultDiv.innerHTML = opts.initialText;
Let us now split up the roll function in to separate parts.
First there is the validation code, which uses pick, from and to.
if (from >= to) {
window.alert("from value must be less than to value");
return false;
}
if ((to + 1) - from < pick) {
window.alert("Error - You want " + pick + " numbers.\
\
" + "The range you have entered is from " + from + " to " + to + ".\
" + "This leaves " + (rng + 1) + " available random numbers.");
return false;
}
Because both parts return false, we can use a standard validation technique of having an isValid variable at the start of the function, which gets set to false if anything fails, and then return isValid.
function validateChoices(pick, from, to) {
var isValid = true;
if (from >= to) {
window.alert("from value must be less than to value");
isValid = false;
}
if ((to + 1) - from < pick) {
window.alert("Error - You want " + pick + " numbers.\
\
" + "The range you have entered is from " + from + " to " + to + ".\
" + "This leaves " + (rng + 1) + " available random numbers.");
isValid = false;
}
return isValid;
}
...
if (!validateChoices(pick, from, to)) {
return false;
}
and the part of the roll function that draws the different numbers, can be extracted out to a drawNumbers function
function drawNumbers(pick, from, to) {
var draw = [],
rng = to - from,
e = (rng + 1),
i,
j,
number;
isRolling = true;
for (i = 0; i < pick; i += 1) {
number = parseInt(from + Math.random() * e, 10);
for (j = 0; j < pick; j) {
if (number !== draw[j]) {
j += 1;
} else {
number = parseInt(from + Math.random() * e, 10);
j = 0;
}
}
draw[i] = number;
}
return draw;
}
...
var i,
dum = "",
disp,
draw = drawNumbers(pick, from, to);
That drawNumbers function can now be refactored, so that instead of rng and e variables, we can just use a delta variable.
var draw = [],
delta = to - from + 1, // inclusive of both ends
and the random number functions we can move out to a separate getRandomInteger function
function getRandomInteger(from, to) {
var delta = to - from + 1; // inclusive of both ends
return parseInt(from + Math.random() * delta, 10);
}
...
number = getRandomInteger(from, to);
but we still need to simplify the nested loop that draws a new random number, which is currently:
for (i = 0; i < pick; i += 1) {
number = getRandomInteger(from, to);
for (j = 0; j < pick; j) {
if (number !== draw[j]) {
j += 1;
} else {
number = getRandomInteger(from, to);
j = 0;
}
}
draw[i] = number;
}
It may be easier to see what we need to do if we switch around the if/else statement
if (number === draw[j]) {
number = getRandomInteger(from, to);
j = 0;
} else {
j += 1;
}
So what we need to do, is to keep on picking numbers until we find one that has not already been picked.
We have a couple of different options here. We could tidy up the existing structure, or we could come up with a completely different solution where we randomize an array of numbers and then pick the first n from that array.
I’ll leave off temporarily here to take care of some things, and consider those options.
Howdy,
So, I’ve just finished working through your latest set of refactorings.
I like how you are applying the single responsibility principle (if you can call it that). Already the code is starting to become a lot more readable.
I have some questions and some observations.
I’ll reply to your posts separately, so as to make things easier to follow.
This was bothering me when I did the initial separation of HTML and JS.
Do you think you are ever likely to need multiple lotto rollers on the same page?
That would be good.
See my previous post
Why would you check for this?
What purpose does this line actually serve?
if ((document.getElementById && document.firstChild) && window.addEventListener || window.attachEvent) {
It’s not ordinarily likely, but I can see a use for it.
[list][]there may be a page may list different pre-defined lotto draws
[]or a demo page that shows how different types of settings work
[*]or people may be able to save lotto settings that they like, as different rollers on a custom page[/list]
It was a common technique with older web browsers, for example, with IE4 and earlier not supporting getElementById. It’s not a check that we tend to use nowdays, but for the sake of completion I’ll leave it in with an appropriately thrown exception.
I followed along with all of the refactorings, however one thing occurred to me.
If you move the validation into its own method, where should the call to validateChoices go?
Ideally, you want it to be called once, when the user presses “Update Lucky Dip” and not, as is currently happening, when the user presses “Lotto Lucky Dip” to start the draw.
However, as validateChoices is defined from within the lotto function, you cannot simply write:
form.onsubmit = function () {
var form = this,
pick = Number(form.elements.pick.value) || 0,
from = Number(form.elements.from.value) || 0,
to = Number(form.elements.to.value) || 0;
if (!validateChoices(pick, from, to)) {
return false;
}
lotto(pick, from, to);
return false;
};
This is where I get a bit confused with JS. In ruby you could write Lotto::validateChoices (Lotto being a class).
Anyway, I moved this:
if (!validateChoices(pick, from, to)) {
return false;
}
to the top of the lotto function after the variable declarations, so at least it only gets called once.
The downside of that, is that the function itself has no concept of the variable rng, so you have to write:
window.alert("Error - You want " + pick + " numbers.\
\
" + "The range you have entered is from " + from + " to " + to + ".\
" + "This leaves " + ((to - from) + 1) + " available random numbers.");
instead of:
window.alert("Error - You want " + pick + " numbers.\
\
" + "The range you have entered is from " + from + " to " + to + ".\
" + "This leaves " + (rng + 1) + " available random numbers.");
What are your thoughts on this?
Good point!
Ah ok. Thanks!
My thoughts on this are that it’s okay for the range variable to be calculated in a few different sections. If a third or more situations occur though, that is a good time to refactor things further.
Alternatively, you can have a getRange() function which accepts to and from, and returns the calculated range to both areas.
Whichever one you choose is a choice on your part as author, about which one you feel will result in a greater understanding for the person reading the code.
It is tempting to randomize an array of values, but we don’t know if people will choose to pick numbers from a large range of potential millions, so I’ll go with the less brittle technique of checking if the chosen number already exists in the array.
I’ve moved the inner loop out to a separate function called arrayHasNumber(), which has allowed me to use a while loop in there, which represents what we are wanting to achieve to a much more accurate degree.
function drawNumbers(pick, from, to) {
var draw = [],
i,
number;
for (i = 0; i < pick; i += 1) {
do {
number = getRandomInteger(from, to);
} while (arrayHasNumber(draw, number));
draw[i] = number;
}
return draw;
}
With the arrayHasNumber function, I don’t know if a persons web browser yet supports the Array.indexOf method, so we can check if that method exists before using it, and if not we can fall back to looping through the array.
In fact, there’s a better way of doing this, so here it is just in a code block, to show what was going to be used:
function arrayHasNumber(array, number) {
// use native technique if available
if (Array.indexOf) {
return (array.indexOf(number) > -1);
}
// otherwise check manually
var i,
hasNumber = false;
for (i = 0; i < array.length; i += 1) {
if (number === array[i]) {
hasNumber = true;
break;
}
}
return hasNumber;
}
A better way, is to provide compatibility code for the Array.indexOf method right from the start, which we can get from https://developer.mozilla.org/en-US/docs/JavaScript/Reference/Global_Objects/Array/indexOf
if (!Array.prototype.indexOf) {
Array.prototype.indexOf = function (searchElement /*, fromIndex */ ) {
"use strict";
if (this == null) {
throw new TypeError();
}
var t = Object(this);
var len = t.length >>> 0;
if (len === 0) {
return -1;
}
var n = 0;
if (arguments.length > 1) {
n = Number(arguments[1]);
if (n != n) { // shortcut for verifying if it's NaN
n = 0;
} else if (n != 0 && n != Infinity && n != -Infinity) {
n = (n > 0 || -1) * Math.floor(Math.abs(n));
}
}
if (n >= len) {
return -1;
}
var k = n >= 0 ? n : Math.max(len - Math.abs(n), 0);
for (; k < len; k++) {
if (k in t && t[k] === searchElement) {
return k;
}
}
return -1;
}
}
Now we can replace that arrayHasNumber function with:
function arrayHasNumber(array, number) {
return array.indexOf(number) > -1;
}
We might even remove the arrayHasNumber function and just put its code directly in the while loop, but that comes down to a choice of which is more expressive.
Either this:
} while (draw.indexOf(number) > -1);
or that:
} while (numberAlreadyExists(number, draw));
I might go for the latter, but it depends too on the potential audience.
The next part to refactor is the display of the drawn numbers.
for (i = 0; i < pick; i += 1) {
disp = dum += (draw[i] + " ");
}
counter += 1;
document.getElementById("result" + idx).firstChild.data = disp;
The loop can be replaced with a simple join statement, so we don’t need the dum variable.
The resultDiv was defined earlier on, so we can use that here too instead of searching for it.
And the disp variable is used only once, so we can place the join at the end of the assignment.
So the above code is brought down to just the following:
resultDiv.innerHTML = draw.join(' ');
The last part of the roll function is about setting up the next roll:
counter += 1;
if (counter <= opts.totalRolls) {
setTimeout(roll, 50);
} else {
isRolling = false;
counter = 0;
}
The isRolling variable being in the same place as resetting the counter helps me to realise that the isRolling variable isn’t even needed. We can just check if the counter is greater than 0, so all isRolling parts can be removed, and the playButton onclick function can instead have this in there instead:
if (counter > 0) {
return false;
}
So let’s move the remaining code to its own function called setupNextRoll()
function setupNextRoll(totalRolls, rollDelay, callback) {
counter += 1;
if (counter <= totalRolls) {
setTimeout(callback, rollDelay);
} else {
counter = 0;
}
}
...
setupNextRoll(opts.totalRolls, opts.rollDelay, roll);
I’m passing the roll function at the end, because if we didn’t do that then we would need the setupNextRoll function to refer to the roll function that is below it, which in turn refers to the function above it. Linting software complains about calls that go upwards, and the callback is a highly effective method to convey what is happening, so it solves a number of related issues.
That should do us for this part of things in the roll function. It’s now looking nice and expressive, where validate before drawing, sorting, and showing the numbers, thengo to the next roll.
function roll() {
if (!validateChoices(pick, from, to)) {
return false;
}
var draw = drawNumbers(pick, from, to);
draw.sort(numericSort);
resultDiv.innerHTML = draw.join(' ');
setupNextRoll(opts.totalRolls, opts.rollDelay, roll);
}
The resultDiv part could be moved out to a separate function so that we can more easily separate the action of rolling from the act of showing the content, but there are bigger fish to fry for now.
Now that the roll function has been simplified we can move the function parameters to the opts object, which will help us to keep our configuration information all in the same consistent location.
var lotto = function (options) {
var opts = {
buttonText: "Lotto Lucky Dip",
initialText: "Your Lucky Numbers",
pick: options.pick,
from: options.from,
to: options.to,
totalRolls: 50,
rollDelay: 50
},
...
if (!validateChoices(opts.pick, opts.from, opts.to)) {
...
var draw = drawNumbers(opts.pick, opts.from, opts.to);
...
var form = this;
lotto({
pick: Number(form.elements.pick.value) || 0,
from: Number(form.elements.from.value) || 0,
to: Number(form.elements.to.value) || 0
});
What we end up with is shown at http://jsfiddle.net/pmw57/8R42e/18/
We are nearly ready now to init the lotto object using custom parameters, which will be delved in to about tomorrow.
Wow! Good stuff!
That is really well explained and easy to follow.
I have a couple of questions:
indexof() method, as opposed to building this functionality into the arrayHasNumber function?roll as a callback to setupNextRoll.javascript setTimeout(roll, rollDelay); Look forward to the next instalment.
That can be something that different people disagree with, for myself, I am hesitant to fiddle with any pre-existing objects. In this case with Array.indexOf, it’s a method that is built in to many web browsers and the behavior is well defined, so I am happy to help bring less capable web browsers along for the ride. Such As IE.
It’s something that you have to be careful of though. The Prototype library monkey-patched getElementsByClassName which ended up causing all sorts of grief when web browsers eventually decided to do things slightly different.
A good explanation of what happened ther can be found in John Resig’s getElementsByClassName pre-prototype 1.6 article.
The main thing that I was worrying about there was in terms of the linter that I use to help keep the code clean and tidy. JSLint can be a harsh mistress at times, but it helps to ensure that your code has a minimum of problematic issues. One of those is in regard to placing calls to functions further down that may not exist yet. This can be a complex issue, but here’s a basic rundown of the situation.
You may have some code, that goes:
function foo() {
bar(); // this works
}
foo();
function bar() {
...
}
The problem comes when someone changes the second function to a function expression, for which there may be several reasons to do so.
function foo() {
bar(); // doesn't work
}
foo();
var bar = function () {
...
}
That doesn’t work because the second function doesn’t exist until it is actually defined. So, placing calls to lower areas of code can be problematic. So coders help to prevent that problem by placing calls to code that is above them.
So what happens when you have two functions that call each other?
function foo() {
setTimeout(bar);
}
function bar() {
alert('called');
if (Math.random() > 0.3) {
foo();
}
}
bar();
The first function is placing a call to the second one, which can cause problems due to what we saw above.
If you are going to live by rule that you don’t call functions below you, something has to occur to stop the first function from calling further down. We can achieve that by passing it in to the first function as a callback, which also makes the function more flexible in how it behaves too.
function foo(callback) {
setTimeout(callback);
}
function bar() {
alert('called');
if (Math.random() > 0.3) {
foo(bar);
}
}
bar();