Cloud Four Blog

Technical notes, War stories and anecdotes

The Forgotten Responsive Images Spec: image-set()

Now that responsive images have landed in Chrome and Opera, I’ve started working on a flowchart to help people decide how to use these new features.

This work led me to wonder what ever happened to the image-set() specification.

For those who haven’t heard of the image-set() specification, it was a precursor to srcset which is now part of the responsive images specification. It was originally proposed in February 2012, and WebKit-based browsers shipped prefixed support for it in August of the same year.

The are a few differences between srcset and image-set(), but the biggest difference is that image-set() deals with CSS images whereas srcset is an attribute on the <img> element.

How we forgot about image-set()

In 2012, the image-set() spec was still under development and we were cautioned against using it at the time. Because media queries were available in CSS, handling CSS images in responsive designs wasn’t as difficult as handling responsive images in HTML itself.

So the Responsive Images Community Group focused on how to solve the <img> problem. And I gradually forgot about image-set() thinking that it was moving forward in the CSS Working Group and browsers.

It seems that I may not have been the only one who forgot about image-set() because despite being two years older than <picture>, it is still only supported under prefixing in Chrome and Safari.1 Worse, it isn’t on the roadmap for either Internet Explorer or Firefox.

Why we need image-set()

We need image-set() for the exact same reasons we need srcset and sizes. Whenever we are dealing with a CSS image that fits the resolution switching use case instead of the art direction use case, we should be using image-set() instead of media queries.

In fact, you can take nearly everything I wrote in Don’t Use <picture> (most of the time) and substitute image-set() for srcset and media queries for <picture> and the same logic applies.

We need image-set() for resolution switching because it gives browsers the choice of what size image source will work best. And in cases where all we are dealing with is resolution switching, the browser is better informed than we are as web authors.

Help get image-set() into browsers

We need your help to make sure that image-set() is implemented in browsers. Help us by voicing your support and ask browsers to prioritize this feature:

Image-set() is a key piece of having a comprehensive responsive images solution in browsers. It has been languishing too long. Let’s get it back on track.


  1. Eugeny Vlasenko built a nice test for image-set() as well as publishing the results of testing a bunch of browsers and their support of the spec.

Image Resizing Services

Responsive images are landing soon and many organizations are looking for ways to resize images. Thankfully, there are a number of startups, established companies, and open source solutions for image resizing.

I’ve pulled together a spreadsheet of the image resizing services to make it easy for people to explore their options.

Figuring out what to include and what to ignore was difficult.

For example, GD and ImageMagick are available on many platforms and you could build your own service using them. Many CMS tools have built-in image resizing tools.

I focused on services and software that provided a level of abstraction (e.g., you can ask for image resizing via a URL) or were specifically focused on responsive images.

Does the image resizer try to detect the correct image size and would that be desired?

The spreadsheet has a column for detection to indicate whether or not the image resizer tries to automatically detect what size of image to request or deliver.

Detection comes in two forms. One is using the user agent string to look up information in a device database. The other uses JavaScript to either find the same information or find the exact size of the image element in the page.

With the responsive images specification landing in browsers soon, I’m not sure if detection is desirable. The benefit of srcset and sizes is letting the browser choose what source is best.

What other image manipulation is offered?

Many of the image services offer services that can crop, filter, and otherwise manipulate images. Some even add things like focal point or facial recognition.

Performance is the missing column

As long as you’re centralizing images processing, you should try to find a service that will compress them as much as possible and provide other performance benefits such as caching and CDN support.

If I were evaluating services, I would be looking for these things, but comparing image compression is tough because it is a balance of raw file size and image quality the requires a judgment call.

Cost is a column I didn’t want to add

When I published this spreadsheet originally, I got a lot of people asking for column to compare price. I resisted adding it until recently. And I only report on whether the service is free or paid.

Comparing prices in the spreadsheet is pointless. Every service that charges varies pricing based on the volume of images. The prices will change in the future, and I don’t want to maintain that information. And the services are different.

I encourage you to explore the services themselves. Some are quite reasonably priced and the expensive ones offer a ton of features.

Anything missing?

So take a look at the list of image resizing services. If you see something that is missing, please let me know.

I hope you find this useful.

Fixed and inflexible

Even before smartphones came along and dashed any hope for a 960-pixel-wide web, designers and organizations have struggled with the challenge of prioritizing and composing content that scrolls. Our screens act like windows to content of variable size and scale, demanding an awful lot of abstract thinking to design for. Sometimes we’re successful, revising content, designing modern day deliverables and embracing compromise like we know in our hearts we should. Other times, we convince ourselves that we can predict this inherently unpredictable medium, making decisions that age quickly and poorly by prioritizing the window instead of the content.

Most recently, I’ve noticed a sharp uptick in the number of requests I receive to explore fixed-position (aka “sticky”) interface elements. Fixed elements are positioned relative to the viewport instead of the page, allowing them to maintain position even as the document scrolls. Some of the most popular sites on the web employ “sticky” menus, and with good reason… when applied thoughtfully, they can yield substantial usability improvements.

But when fixed positioning is used without care, restraint or precision, it can have disastrous consequences. Here are some of the reasons why.

We can’t predict how much space we have.

Our industry has a nasty habit of quietly embracing display resolution “standards” that are mostly fantasy… design decisions are a lot easier if you assume all devices are 320 pixels wide by 480 pixels tall (and users never turn them sideways). The inconvenient truth is that display resolutions vary wildly. Each additional “sticky” element increases the risk of obscuring the page content mostly (or entirely) for some users… in which case the page might as well not exist at all.

We can’t get those pixels back.

Fixed elements aren’t just “prominent,” they photobomb the interface, robbing focus and attention from what really matters — the content! Before you make something “sticky,” consider reevaluating the element’s importance relative to the entire page (instead of on its own merits).

It makes scrolling tougher.

As evidenced by the years-long debate over “The Fold,” the fear that users won’t know (or lack willingness) to scroll persists to this day. When that fear is real, then fixed positioning is a godsend, ensuring the visibility of content regardless of scroll position.

But those fears don’t mesh with reality very well.

Today’s users are so familiar with scrolling that most mobile browsers will hide the scrollbar entirely. “Sticky” elements complicate matters by reducing or obscuring the scrollable area, forcing the user to swipe more carefully to avoid accidental actions.

Ironically, our desire to alleviate the supposed “difficulty” of scrolling may make scrolling that much more difficult!

It can slow everything down.

Users care about speed. But speed isn’t all about navigation… there’s also the overall speed of the experience. Fixed positioning can result in strange browser-specific quirks or even costly repaints, potentially counteracting any efficiency you might have gained.

It may not actually work.

Mobile websites share a lot of their design vocabulary with native mobile apps, where fixed headers, menus and tab bars are commonplace. This makes it easy to forget that fixed positioning as we know it is relatively new to the web, and often unreliable.

“Sticky” headers, footers and navigation flourished in the 1990s, often implemented using frames. When frames fell out of fashion in the early 2000s, most browsers did not support fixed positioning using CSS. In the absence of frames, position: fixed or consistent, intuitive JavaScript, fixed positioning became just another discarded Web 1.0 trend… at least until CSS support arrived in IE7.

But smartphone browsers have historically not supported position: fixed as predictably as their desktop counterparts. It was entirely absent from mobile Safari prior to iOS5, and largely unusable in Android browsers prior to Honeycomb. To this day, behavior can be inconsistent across platforms. To quote Brad Frost in his excellent post on Fixed Positioning in Mobile Browsers, “‘support’ isn’t exactly binary.”

Since you can’t rely on support for fixed positioning, you’ll need to make sure your experience works without it anyway. Which begs the question…

Why position: fixed at all?

I believe there are plenty of interface elements that benefit from fixed positioning, provided they follow a few best practices:

  • The “sticky” element is clearly more important than everything else on the page.
  • The footprint of the element is modest enough that it does not obscure too much of the page content (even in landscape).
  • Any efficiency gained from the element’s consistent availability is significantly greater than any lost as a result of the element’s inclusion (due to performance, obscuring of page content, etc.).
  • There should only be one “global” (navigation, tab bar, etc.) and one “temporary” (modal, dialog, etc.) fixed-position element on-screen at any one time.
  • Fixed positioning should always be an enhancement. Your interfaces should never rely on it.

If any of these considerations completely upend your design aspirations, you may want to rethink your user experience with less rigidity. As John Allsopp so aptly put it fourteen years ago, “The journey begins by letting go of control, and becoming flexible.”

Beyond “Hello, World” with Gulp

(This post is for: web developers who have used, or thought about using, a build tool before, are JS-comfy and maybe have noodled around in node.js but wouldn’t, say, feel up to submitting a talk to a node conference.)

It would be hackneyed to call the last twelve months the “Year of Build and Workflow” for web developers, but there’s an element of reality to the contrivance. In 2013, there was grunt (heck, there still is; it didn’t die or anything) and a whole lot of blog posts about streamlining workflow around SASS and templating and build package management, etc.: topics that were—poof!—suddenly an imperative for general people working on the web. I wrote a thing about workflow for A List Apart earlier this year, a few weary weeks after the whole “Oh, you learned grunt? Sucker! Now you need to learn gulp, because whatever!” revolution happened.

Summarized as this: a whole lot of people (increasingly non-nerdcore) learned to use grunt to wrangle previously-tedious workflow tasks for building web sites. This represented a big jump in things we all needed to know. And then, kind of literally overnight, there was gulp. For me, this felt like a burden as well as an opportunity.

I wasn’t exactly a holdout (as you can see), but I was initially resistant to throw the grunt baby out with the 2013 bathwater simply because a new tool existed. And, like maybe some of you (?), I found the initial sell a bit baffling: “streaming build system?” I didn’t know I wanted that? Awesome syntax? I didn’t really have a gripe with grunt on that front. “Speed?” Maybe? I didn’t exactly have to go brew a cup of coffee while my grunt tasks ran.

Grieving and Surmounting

But smart people I know kept extolling its highlights and I decided to give it a whirl. The familiar stages of dev-new-tool excitement-grief happened:

  1. Find and install new tool that people crow about.
  2. Get satisfying example configuration/program running using new tool.
  3. Extend satisfying example configuration/program in a tiny way and find success in making it do something useless, but cool. HELLS YEAH.
  4. Extend awesome slightly-extended configuration/program in another way to make it do something you actually need it to do. Spend 3 hours/days progressively assuring yourself you are a moron—this should really be so easy! Two possible outcomes: rage quit or eventual, sweat-soaked breakthrough.
  5. Assuming you’ve made it this far, progress with a slight wariness, possibly even progressively assuring yourself that the tool devs or Stack Overflow or Google or a particular plugin or, hell, the entire language you’re working in is the real moron here. Also revel in the fact that extant examples of what you’ve now managed to do exist on the Internet at all—you’ve birthed something truly novel. This is often a necessary step to repair the psyche.
  6. If you are kind, you then help other people avoid step 4 (or, at least, put a reassuring hand on their shoulder).

Gulp: The Good

Here’s some of why you might want to use gulp instead of another alternative:

  • Pretty syntax and intuitive structure: There is nothing wrong with grunt’s syntax, but gulp’s is kinda pretty. Using Streams means you’ll be able to do a lot of chaining, which always feels pleasing to the code-scanning eye. I was never a huge fan of the double-step grunt convention of grunt.loadNpmTasks() and grunt.registerTask(). In gulp you can require() and define gulp.task()—that feels more natural to me. I find I don’t have to look up core API stuff for gulp; with grunt, I never memorized it.
  • Speed: There’s no doubt. gulp feels noticeably fast. That is not exactly a scientific observation, but I stand by it.
  • Streams: If you know node already, gulp’s Streams architecture will make you feel right at home.
  • Gulp has a nice clarity that it is about working with streams of files. As such, there is more symmetry between any two given gulp tasks than most given grunt tasks. gulp.src globs and gulp.dest paths become second nature. Stream in, though, out.
  • Works with lots of things, easily: Like grunt, gulp has plugins (many of them). But the first edict in the guidelines for creating a gulp plugin is: “Your plugin should not do something that can be done easily with an existing node module,” meaning that gulp encourages the use of non-plugin-i-fied node packages wherever possible (versus “[w]rapping every possible thing just for the sake of wrapping it”). The plugin landscape feels correspondingly more serene.

Avoiding Step 4: Tips for Getting along with Gulp

  • Make sure your gulp tasks finish correctly. If not, it will wreak havoc with dependencies and other tasks. Be sure to read the section of the docs about valid things to return from gulp tasks and how to handle async tasks. Trust me on this one.
  • In fact, read the entire API documentation. If this sounds like a tall order, take note that it is shorter than this blog post. The core API for gulp is outstanding in its simplicity.
  • Get an understanding of Streams, a core node construct, because this is the way that gulp works. Take a few minutes/a while and read a primer on Streams. It’s a bit to get your head around, but if you’re doing any heavy lifting with gulp (or other node, for that matter), you’ll be glad you learned.
  • Watch out for plugin quality. The plugin landscape may be straightforward, but it’s young, and there are some stinkers. I lost some serious time to a couple.
  • Dig hard for real-world (read: complex) examples. This situation has gotten better in the past few months. Dan Tello’s excellent post on Gulp + Browserify: The Everything Post is a must-read-right-now. When I first encountered his repo full of organized gulp tasks, it changed my everything. I ended up extending my fork to handle dependencies effectively, but his gulp-starter repo has since been updated to do that, also, so, win! There is more documentation and examples out there now than there was a few months ago. But documentation about gulp things can sometimes presuppose that you know more about various concepts than—gasp—perhaps you do. Don’t fret. That jargon sounds weird to other people, too. Read slowly and give yourself some time!
  • Watch out for plugins getting between you and what you’re doing. Like any build tool with plugins, gulp puts a layer between you and (in gulp and grunt’s case) node. Sometimes it’s hard to debug whether you’re doing it wrong, or whether the plugin between you and the wrapped node package is doing something wrong (or at least making things real complicated). If you don’t understand the syntax of a plugin or what it’s doing, consider that a red flag and slow down until you do. Or: see if you can accomplish what you’re doing with a node module instead.
  • Complexity explodes as you add more tools to the mix. We are using browserify for our application code, but it got outrageously onerous for managing vendor code, much of which was not modular in any sensical way (this was serious step-four territory). We ended up throttling down and using something a bit simpler. Don’t be afraid to over-simplify if it gets the job done. I find you can usually kick up complexity later, when needed, but that the opposite—dialing it down—is not so easy.

We do some really cool things at present with gulp. We use it to parse YAML front-matter out of templates (which get piped through underscore and have partials inserted and come out as static HTML files) to create an auto-index of project assets, adding automated screenshots (thanks, phantom.js!) at different widths to show responsive layout. This of course along with SASS and auto-prefixer and browserify and concatenation and and and…whew. Because each task in the works is totally modular, this ends up not feeling rickety and spaghetti, believe it or not—and that alone is worth the price of admission.

Go out and build! Let me know how it goes!

My thanks to Cloud Fourian Erik Jung for leading us out of step 4 hell vis-a-vis the whole vendor JS thing I mentioned above.


Slow and ugly

Many years ago at a previous job, I was responsible for pushing our organization to adopt a new ticket tracking system. We decided to use an open source project called Request Tracker (RT).

This was before the days of Zen Desk and similar services. RT matched our requirements best. I worked with our sysadmins and engineers to get the software running and rolled it out to the organization.

It was a failure. People were openly swearing at RT, and I’m sure they were privately cursing me for forcing them to use it.

While I sympathized with the complaints, I felt hamstrung because RT was deployed on an underpowered server and we had not been allowed to modify the look and feel. Back then RT was a bit of a dog both from a performance and aesthetics perspective.

As I worked with engineering to secure better hardware and get someone to assist with the design, the number of grievances continued to pile up.

One conversation stands out to me. A member of the management team told me that the fact that customers couldn’t reset their password in RT was unacceptable and as long as that was true, we couldn’t use RT.

I listened to the objections and kept a list of the issues that were raised, but I kept my focus on making RT faster and more attractive.

A few months later we relaunched RT. It was met with skepticism, but to the credit of my colleagues, they gave me and it a second chance.

It wasn’t too long after the relaunch that previous skeptics were swearing by the system. One of the team members became the master of RT and made a bunch of process improvements that I would have never considered.

What happened to that list of complaints that I gathered? We didn’t address a single one of them. Passwords worked exactly as they did before.

We simply made RT fast and attractive.

I have a simple rule when it comes to user experience: when something is slow and ugly, nothing else matters.

Until you address those two issues, you’re not going to be tell whether other complaints are real or red herrings.