Nick Desaulniers

The enemy's gate is down

Commandments of a Mobile Web

| Comments

Over the past few years, there’s been certain paradigm shifts in web development. When you think of milestones that really changed how development on the web was done, the two biggest were Ajax and HTML5. Development was identifiably different before and after such technological advancements. There were some who initially doubted the technologies, but I’m sure such doubters eventually saw the light. After spending time working on applications for Mozilla’s upcoming mobile operating system, Firefox OS, and talking with my fellow employees, I feel that the mobile web is another one of those shifts in how we approach web development that looking back will be an identifiable point in time where we can say that we did things differently before and after. So in that sense, I want to share some of the insights I’ve found to help other developers wrap their heads around how developing for the mobile web isn’t their traditional cup of tea.

Internet connectivity is not guaranteed

This is a fundamental divorce from the World Wide Web and the Internet. I feel that a lot of people having trouble differentiating the Web from the Internet; where you’ve had one, you’ve always had the other. Don’t assume your application will always have a valid connection. When a user is on a wifi connection, or hardwired, it’s so obvious that if they’re on your website, then they must be connected to the Internet. Right? But what happens now when one of your users loads up your site or app on a mobile device, then enters a tunnel? What does it do when offline? Does it work? Maybe it doesn’t make sense if you’re offering a service that requires data from the back end, but does that mean that the front end should totally look like crap or break when the back end can’t be reached? Do your Ajax requests have error callbacks? Do you try and reconnect your WebSocket connections after they’ve failed? Do you cache previous requests or use one of the many forms of offline storage to show something to the user? Developing on the loopback is literally the best case for connectivity, but it gives the developer a false sense of how latency and connectivity issues affect their application. It’s literally shocking the first time you encounter your site/application offline if you didn’t think about the user experience up front.

Bandwidth is not free

The advent of broadband made it acceptable for sites to utilize massive amounts of assets. Seeing sites that load up so much stuff makes my heart sink when I wonder about viewing such a site on my mobile device. I only pay for so much mobile data per month, then I’m billed ridiculous amounts as a “Data Overage”. Users in other countries frequently have a more pay as you go style plan, so they’re being billed for each bit across the wire. In countries without the infrastructure necessary to get broadband Internet to every household, mobile is more prolific in getting users connected.

Things like minification and gzip certainly help, but deferred loading of assets until they’re necessary is frequently overlooked. Massive libraries are nice for devs, but may end up giving mobile users more problems than they are worth. There exist more advanced techniques such as WebSocket compression.

How much does your site weigh?

An expensive website does not make an awesome app

Why doesn’t my million dollar website make a kick ass web app?

This point is brought to you by Matt Basta. The point is that there is something fundamentally different between a web “site” and a web “app”. In a web site the usual flow involves loading up a page and maybe navigating between pages. An app more often than not will be a single page that loads data asynchronously and dynamically modifies the content displayed to the user. Some web sites make great use of things like Ajax, and are great example of single page sites. But not all are. Some still haven’t come around to loading all data asynchronously on request. And it’s not that they necessarily even have to, they can be a website and never have to be anything more. But you’ll find that some sites make better apps than others.

No chrome

Not the browser made by Google, Google Chrome; the actual controls on the top of your browser such as the back and reload buttons, and the url bar. This point comes from Chris Van Wiemeersch. One of the things you’ll find when evaluating whether a site makes a good app is whether it relies on chrome to navigate. If it does, then it’s not really going to cut it as an app. If you’re going to make a single page app, try making it fullscreen in your browser and try navigating around. Are you hindered without the chrome? One thing I found recently was an error condition where I notified the user that they had to be online to check for an update. But when I was developing, I would just refresh the page to clear the error message and try again. On a device, as an app, where there was no chrome, this meant closing and restarting the app. That sucked, so I made it so the user could click/touch, after figuring out their connectivity issues, to retry the fetch. Even better may have been to listen for online events!

Traditional forms of input are not fun

Here’s some tips from Matthew “Potch” Claypotch.

Fingers aren’t as precise as cursors

Having tiny click targets is really frustrating. How many times have you clicked the wrong thing on a mobile device? Tiny buttons are not the easiest thing for users to specify, especially when groups of them are clustered nearby. Custom buttons enabled by a little CSS can go a long way.

Typing on little keyboards in tedious

This is an effect of tiny buttons, but requiring the user to type in large strings gets annoying fast. Sometimes this can’t be avoided, but it should be when it can. Just as typing in a long, complex url to a mobile browser is not enjoyable, neither is doing so in an app.

Detect features not browser engine (User agent sniffing is a sin)

Print this and staple it to your wall above your workspace. This should be old news at this point.

Vendor prefixes are meant for browser vendors to test, not production code

I place a majority of the blame for this on vendors; prefixes should never see the light of day in release builds. It’s ridiculous to have to write four copies of the same rule, when you’re trying to express one thing. I should rewrite this sentence four different ways to prove a point. -o-Do you understand what I’m getting at? -ms-It’s almost like rambling. -moz-Repetitive department of repetition. -webkit-This is ridiculous.

But developers need to recognize the habit as an addiction, and not feed it. If you have, I forgive you. Now stop doing it.

These two particular articles are just so good, it wouldn’t do them justice to try and summarize them. Please go read them, I’ll wait. this and this

Developing towards WebKit and not HTML5 is a sin

I understand that Google Chrome is your favorite browser, and I am so happy for you; but it is not mine. I’ll be the first to admit that Google caught all of the other browser vendors with their pants down, but when I see pages that look flawless in Chrome and not so hot in others, it reminds me of days when sites only worked in IE6. Surely you remember those days. I hate when content publishers try and dictate which browser I should use to view their content. I understand WebKit based browsers dominant in mobile, but so did IE6 in desktop share at one point. It’s not unreasonable to expect users to use the most updated version of their browser, but empower your users to choose their browser. Don’t take that choice away from them.

A neat point by Robert Nyman is that WebKit itself already has forks. Can you imagine if there were eventually vendor prefixes for forks of WebKit? Continuing with vendor prefixes means that we’ll now have seven vendor prefixes: unprefixed, -moz-, -o-, -ms-, -webkit-o-, -webkit-chrome-, -webkit-safari-. Awesome! Maybe I should rewrite this sentence seven different ways to make a point!

I’m also curious if Google, Apple, and the WebKit maintainers are turning a blind eye to this, or what their opinions are? Being a vendor and wanting an open web are almost conflicts of interest; you want your browser to “win” or dominate in marketshare, but at the same time you don’t want any one browser having too much marketshare.

Design Responsively

What is responsive design? Responsive design is making a site look great on any size screen, without duplicating assets or sniffing a user agent string. Firefox has a neat web dev tool called “Responsive Design View”. It’s great for testing out your site on various screen sizes. Twitter Bootstrap is an excellent example of a framework for developing a single app that looks great on any screen size. Even if you don’t want to use a whole big framework, simple things like using CSS rules in terms of percentages instead of hard coded pixels can go a long way.

Sites that are trying to become more app like have trouble with conforming to responsive design. Instead of starting with a site and trying to figure out how to hide or not display information as the screen gets smaller, you’ll find it much easier to start small, and dynamically add content as the screen gets bigger.

High performance code respects battery life

In the end of the day, all of the code you write that runs in the browser is a set of instructions the processor can decode, not necessarily the high level JavaScript you wrote. While it’s important to avoid premature optimizations, having a few rules of thumb up front will help you write better, faster code. If the same overall action can be expressed in one instruction or one hundred, which do you think will use less power? Keep in mind that transistors leak current while switching.

Native methods over library methods, CSS over JS

What do I mean by “native methods?” Like C++? Well, yes. You see under the hood the DOM bindings that you’re calling are probably written in C++. Call toSource() on document.getElementById(). The [native code] statement in the returned string refers to the implementation. While type specialized code emitted by the JIT can match or even beat native code, you can only count on that for hot loops. In the same vein, library code is going to be slower than any code written natively. I’m not saying that you shouldn’t use libraries, just know that the use of libraries can incur some overhead. Things like animations will also be faster when handled by natively implemented CSS over JS. You can use JS to dynamically add classes to elements to get finer resolution over events, but then get the performance of CSS.

Avoid JIT bailout

The Just In Time (JIT) interpreter is a new breed of VM that most browser vendors are now using. The JIT can analyze running code, and emit native code that is faster than reinterpreting code again, and even higher-optimized code for type stable JavaScript, as long as certain “guard” conditions are met. When a guard fails, the JIT has to bailout the emitted native code and start reinterpreting code again.

Keep up to date on new APIs

HTML5 is a big spec, and is getting bigger. So big that some recommendations are being spun off from the original HTML5 spec. As an engineer, I’m painfully aware that you need to keep up in whatever industry you work in in order to stay relevant. The complacent are the most vulnerable. There’s a lot to keep track of with HTML5 and CSS3, but many new features offer higher performance methods of skinning the cat.

requestAnimationFrame

window.requestAnimation frame is a godsend for animation. Not too long ago, I wrote up a quick example of various ways of implementing animation loops and their issues; you should check it out.

indexedDB over localstorage

The indexedDB api might not be as simple as localstorage’s is, but localstorage is synchronous and is noticeably slow on mobile. If you can bite the bullet and use indexedDB, you’ll find you’re JS isn’t blocking on serializing/deserializing objects for storage. Fabrice DesrĂ© shared this with me.

WebWorkers

Webworkers can’t modify the DOM, but they can do heavy lifting without blocking the main thread.

CSS translate over absolute top and left rules

Harald Kirschner recommends CSS translates over top and left rules for absolutely positioning some elements.

Gradients are expensive

Dan Buchner notes that without beefy graphics processing units of their desktop counterparts to enable hardware acceleration, things like gradients will give you noticeable performance hits.

createDocumentFragment

Dan also suggests queuing up DOM changes. Whenever you manipulate items in the DOM, you’re going to trigger a reflow, which may consist of an update to the layout and/or a repaint. Minimizing these makes for a faster update to the DOM. For example, it can be faster to use document.createDocumentFragment and append child nodes to it, and then append that to the DOM, instead of appending lots of child nodes in between timer calls. Surprise, this isn’t actually a new DOM binding from HTML5.

Conclusion

These are just some tips I have for application developers. I am by no means an expert; I’m sure if you dig deep enough, you can find plenty of examples of my past work that contradicts some of my recommendations from this article. But I’m a little smarter today than I was yesterday, and now so are you! What are some tips that you have to share that you’ve found helpful developing for the mobile web?

Comments