Skip to main
Article
Nested semicircles,
from innermost and slowest
to outermost and fastest-moving:
nature, culture,
governance, infrastructure,
commerce, fashion

When Our Tools Hold Us Back

What happens if the ‘pace layers’ get out of sync?

Ask not just: How well does it work? But also: How well does it fail? What happens when something goes wrong?

—Jeremy Keith

The 2022 Web Almanac has been out for a while, and (as always) it’s an interesting read.

There are more than 20 chapters this year, and I haven’t read them all. But I did enjoy the chapters on CSS (by Rachel Andrew) and webfonts (by Bram Stein). Both authors note a similar trend – many usage patterns that refuse to change from year to year can be explained by popular tools.

I’ve noticed that same effect elsewhere, and I want to dig into it a bit – the various ways our tooling can change (and limit) the web platform features we use.

According to the Almanac data, the most common color formats on the web are HEX (#C71585), short-form HEX (#c18), and then old-school rgba(199, 21, 133, 0.8).

But why are we still using the oldest and least legible color formats, when better formats are well-supported? HSL is more human-readable than HEX, and the rgb() function now supports alpha transparency without switching to rgba().

Alternately, if we like HEX so much, why don’t we use the alpha-HEX syntax (#C7158599) that’s been supported for several years. Why aren’t these options gaining popularity?

It’s hard to say for sure, but I have a multi-part guess.

First, colors are design tokens, and often get set as custom properties or build-tool variables at the start of a project. Once the tokens are established, we don’t spend much time looking at the actual color syntax underneath. I don’t ever interact directly with the HEX code #E60073 – I can just reference var(--brand-pink). So we focus our attention on the naming conventions rather than the underlying color syntax, which is hidden from view.

I think there’s also a tooling component here. Many graphic design tools are a step removed from the web platform. While some do try to export HTML/CSS, only a few consider that the core goal of the tool. Even then, there’s been a long-term disconnect between a web that’s responsive, and our almost completely static design tools.

A few tools have started to provide flexbox-like ‘responsive’ layouts, but I haven’t seen anything CSS-grid-like. Most don’t even have basic support for relative or fluid units. Rachel notes in the Almanac that most lengths on the web are defined in px. Is that because authors intentionally avoid em/rem, or because the mockups that we receive from designers are all limited to px by our design tools?

I’ve commented on this before:

The least responsive part of Responsive Web Design is the design tools. This needs to change.

When it comes to color spaces/gamuts, design tools vary in support. Looking at some of the big players:

I don’t know of any popular design tools that provide LCH, or the newer okLAB/okLCH variants, let alone relative colors defined as a function of mixing or adjusting other underlying colors.

And when it comes to copying our colors out of a design tool, the options are generally 6-digit HEX, or copying individual channel values, one at a time. Either way, converting our colors into a more human-friendly syntax would often be an extra step, and may not be worth the effort.

Since all legacy colors on the web rely on a single color model (RGB) and gamut (sRGB), there has been little meaningful difference between formats (until now). Once we used a graphic interface to pick the color we want, it maybe didn’t matter what format we copy-pasted into a variable somewhere.

But what’s the designer path towards using new, extended color spaces in CSS? Can they ever gain traction on the web without first finding support in our design tools? I’m not sure.

Code tools also play a role in this.

Sass, for example, relies on the same interchangeability of legacy web formats. No matter what syntax an author chooses, Sass will store the value as RGB, and then allow manipulation in any other CSS format. But at the end of the day, CSS output is always generated in whatever format has the widest browser support.

It doesn’t matter to Sass what syntax authors use for their input value, the output value is always either HEX or a color name when opaque, with rgba() for handling transparency. Very few people need that level of legacy browser support, but it’s not hurting anything either.

Not yet, at least. This will all change going forward, as we add Sass support for more color spaces and gamuts.

In this case, the Almanac may give us misleading info about what syntax authors prefer. For all we know, authors might be working primarily in HSL, and we wouldn’t see that in the production code at all. In fact, one of the most popular complaints on the Sass issue tracker is a request for HSL output.

But this is a distinction with very little at stake. So long as authors can use their favorite format – and the output really does provide the same meaning as the input – it’s not a barrier for authors using the latest color formats. We’re just making that harder to see.

In Sass, we have a strict policy that valid CSS is always valid Sass. If authors use syntax that we don’t recognize, we pass that along to CSS. We can check for obvious parsing errors, but only the browser can determine if it’s a supported feature. While we provide extra Sass support for colors that we understand, authors can use new CSS color formats in Sass without us releasing any new features. Sass may not understand those colors for a time, but it will leave them untouched in the CSS output.

Similarly, Sass hasn’t added support for either @layer or @container – and there may not be any reason to. I’m already using both in my Sass, without any issue.

There have been unfortunate exceptions on occasion (see min() and max() for example), but we’ve worked very hard to remove those limitations when they crop up. The tool should never stand between authors and the underlying language.

Last month, I spoke about @layer for An Event Apart Denver. During the Q&A session after my talk, someone asked, “How can I use this with CSS-in-JS tools?” And the truth is… I don’t know. It depends entirely on the tool.

Single-file components in tools like Svelte & Vue are technically CSS-in-JS, but allow writing arbitrary/plain CSS. I would expect (hope?) that @layer should just work in that context (similar to using Sass).

But many CSS-in-JS tools and utility frameworks are more invasive – replacing some or most CSS with a proprietary syntax. They don’t provide additions to the language, but a whole new language that stands directly between us and the basic CSS functionality we need. There’s no option to write CSS directly, we have to rely on the tool to do it for us. If the tool can’t do something, neither can the authors using it. All we can do is wait.

I even ran into this issue by simply hosting my personal site on Netlify, which provides minification for deployed assets like CSS. That seems like a great feature! But the specific CSS minifier that they use is somewhat overeager, and began stripping out all my new @layer rules. I had to turn off the minification step in order to publish my entirely-valid CSS.

That shouldn’t be the case. Minifiers shouldn’t need to understand all the new features that land in CSS. This is similar to the issue with removing units from zero, or Harry Roberts’ recent warnings around Critical CSS. In a language like CSS – where order is meaningful, and duplicate properties are a useful feature – it’s risky for automated build tools to try and be clever about minification. I don’t know that a CSS minifier should do much more than white-space management.

In regard to web fonts, Bram points out that authors are not taking full advantage of variable fonts – we’re mostly changing the font weights, and selecting from standard values.

As someone who has spent hours searching through lists of variable fonts, that doesn’t surprise me much. Most variable fonts are either experimental display fonts without a lot of practical use, or they only provide a single weight axis. While I love having all my weights in a single file and smoothly transitioning from one weight to another, specialty weights are not a high priority for me. I’ll start using more powerful font variables once there are more fonts available with more interesting variables to manipulate.

Most of us are also using the over-complicated font-feature-settings property, instead of the simpler font-variant. I’ve been doing that on my site as well!

Why did I think that was the best approach? Likely because that’s the generated output recommended by Wakamai Fondue – a popular tool for inspecting font capabilities. It’s a great tool, filling a much-needed gap, but I wonder how the usage stats would change if they updated their output CSS?

Maybe even more common, much of the web is built using CSS libraries like Bootstrap, or third-party themes provided by a site-builder or CMS. The influence of both WordPress and Bootstrap is clear throughout the Almanac data. When Rachel notes the popularity of flexbox (very popular) vs grid (not so much), I wonder how much of that can be traced back to Bootstrap ‘grids’ being built in flexbox.

I don’t think that’s the only factor. It’s possible authors have settled on flexbox as good-enough, and they don’t think it’s worth their time learning another technique. I can understand that, but it’s unfortunate. I think people are missing out on one of the coolest features added to CSS ever.

When we talk about web trends, I often hear people refer back to the idea of ‘pace layers’ on the web:

Described by Jeremy Keith in his Building talk a few years back.

In context, Jeremy is not just referring to the web languages themselves – JS isn’t necessarily changing faster than CSS – but also the ecosystems of these languages. The closer we get to the core functionality of the web, the slower each language/ecosystem is likely to change. Even within a given language, we expect our tools and conventions (fashion) to move faster than the languages themselves.

But that can only happen if the tools are designed to sit on top of their underlying languages. Once the tools stand between us and the language, we become entirely reliant on tool-builders to determine what features are available.

When our primary tools intervene so strongly between us and our ‘materials’ – the core web languages – we can accidentally flip the pace layers of the web on their head. Suddenly CSS is able to move faster than the ecosystem, and we’re stuck waiting on our tools to catch up with well-supported platform features.

That’s not how a healthy ecosystem should behave.

I am asked quite often what I think about CSS-in-JS or utility-first frameworks, or even CSS libraries & conventions more generally. And I try to stay out of any spicy flame wars, or gate-keeping nonsense.

For myself? I don’t find many of these tools useful. They don’t solve a problem that I have. But different projects and different teams have different needs, and might find different tools useful. There’s no single ‘correct’ approach for writing all CSS everywhere.

As long as it results in an accessible, resilient, and performant experience for users: use whatever tools work best for you.

(Anyone marketing their product as The Only Solution™️ to some Essential Problem with CSS™️ is selling you a religion, not a tool.)

It’s not my job to judge what tools are useful for other web developers. But it is my job to help people think about the CSS choices they make. So in that spirit, please consider:

⚠️🚨⚠️ When tools intervene between you and your access to the web platform, proceed with caution. Ask not only: How well does it work? But also: How well does it fail? Not only: What features do they provide? But also: What features do they prevent? ⚠️🚨⚠️

Webmentions

Carter Baxter

from mastodon.social

@matthiasott @mia I wonder if the tooling has followed the front-end community, which has largely undervalued CSS for a long time now. And now that's come home to roost.

Matthias Ott

from mastodon.social

@mia This is so down my alley. There are growing gaps anywhere atm, and one of the greatest “great divides” is the one between what is possible today with the material(s) of the Web and what designers know about it and are able to create in design tools.

Miriam Suzanne

from front-end.social

@matthiasott I also get the problem there, it's not easy to solve. How do you provide the full power of CSS in a graphic interface? Reproducing the language as a series of form fields is not exactly elegant.

On the code tooling side, I think people also leaned into…

🦑 meduz'

from twitter.com

“Suddenly CSS is able to move faster than the ecosystem, and we’re stuck waiting on our tools to catch up with well-supported platform features. That’s not how a healthy ecosystem should behave.” Fully agree!

Egor Kloos

from twitter.com

Back in the day, I'd sometimes use a hack when I had to make IE6 work. I've always insisted that any hack has to be valid CSS. It shouldn't be our fault when a site breaks because we ignored non-conforming code. Apparently, it's okay for tools to mess it up…

Jonathan Neal

from twitter.com

That’s when I broke a sweat, waiting for Mia to start naming names. The part about design tools being a step removed from the web also rang true with a heavy heart. Even popular tools designed explicitly for export on the web are priming to hold it back.

Mia

from twitter.com

Yeah, the first draft ended with "In summary: everything is Jon's fault. Blame Jon." – but the editors thought that was a bit much (and not really supported by the data). 😅

Recent Articles

  1. Article post type

    Use the Right Container Query Syntax

    Size queries are stable, and shipping in browsers

    Since we got a first look at a Container Queries prototype back in April 2021, the syntax has changed a few times. But now the spec is stable, browsers are getting ready to ship, and it’s time to make sure you’re using the same syntax they are!

    see all Article posts
  2. Article post type

    Not All Zeros are Equal

    And every ‘best practice’ comes with caveats

    There’s a well-established ‘best practice’ that CSS authors (as well as linters and minifiers) should remove units from any 0 value. It’s a fine rule in most cases, but there are a few common situations where it will break your code.

    see all Article posts
  3. Article post type

    Every Transition is a Page Transition?

    Experimenting with the shared element transitions API

    There’s a new web API proposal for transitioning shared-elements across pages. It’s great for making smooth page transitions, but what if we apply it to individual elements with changing styles on a single page?

    see all Article posts