Can CSS Be Too Modular?

Written by Harry Roberts on CSS Wizardry.

N.B. All code can now be licensed under the permissive MIT license. Read more about licensing CSS Wizardry code samples…

I frequently get asked questions along the lines of Can CSS be too modular? I recently got asked about this in an email, which I’ve published below:

Harry,

I wanted to first thank you for creating inuitcss which I point my students to in my Object-Oriented CSS class I teach. I do have a philosophical question. From experience, I noticed that my websites have become abstraction heavy, and more often than not code was being duplicated. It is really because abstractions are owned by an object.

For example, I created a simple object called .box with abstractions to change the padding. Later I created an object for alert messages, inline labels, and inputs and noticed that the sizing abstractions were the same. Since then, I moved away from creating abstractions unless they need to target a child element or are SPECIFIC to that object, and began working with atoms.

So:

.box          { }
.box--xs      { padding: 6px; }
.box--s       { padding: 12px; }
.box--m       { padding: 24px; }
.box--l       { padding: 36px; }
.box--xl      { padding: 48px; }

.alert        { }
.alert--xs    { padding: 6px; }
.alert--s     { padding: 12px; }
.alert--m     { padding: 24px; }
.alert--l     { padding: 36px; }
.alert--xl    { padding: 48px; }

.label        { }
.label--xs    { padding: 6px; }
.label--s     { padding: 12px; }
.label--m     { padding: 24px; }
.label--l     { padding: 36px; }
.label--xl    { padding: 48px; }

became an atom called .pa (padding-all):

.pa--xs       { padding: 6px; }
.pa--s        { padding: 12px; }
.pa--m        { padding: 24px; }
.pa-l         { padding: 36px; }
.pa--xl       { padding: 48px; }

Now it could be applied to any element and is not bound to any particular object.

I read a fantastic article on Smashing Magazine that goes over this technique. Yahoo has implemeneted it. Atoms really are low-level utility classes that cache well. I think the only reason to keep abstractions are:

1. They target a specific object, which atoms can not do
2. Too many atoms would be needed to make up the abstraction (like your .btn class)
3. An abstraction is SPECIFIC to an object. Atoms are not.

This is no way replaces modular design, but rather compliments. Everyone uses helper classes, (inuitcss, Bootstrap, SUIT) but creating an atom pool seems to be the next evolution in creating scalable frameworks.

Here is the article (the authors naming conventions for font-sizes are awful it would be better to do .fz-s, .fz-m, .fz-l in case they change):

smashingmagazine.com/…/challenging-css-best-practices-atomic-approach/

I was wondering how you felt about this? Is my thinking on this wrong? When I tell my students about this, I always warn them first I have only seen a handful of websites (besides my own) that use it.

Thanks,
[Name]

To which my response was:

Hey!

Thanks for getting in touch. This is a really interesting one, and something I came up against in my own work a couple of years ago.

The problem, I think, is that with any new tool or technique we learn, we try and take it to its fullest extreme. We then see how that led us slightly astray, and then we begin to correct. Almost any dev problem/solution I’ve ever seen looks like a pendulum: one extreme; extreme overcorrection; finding the sweet spot.

With that in mind, I think that old-style CSS (no classes, very convoluted selectors, etc.) was one peak of the pendulum, and something like Atomic CSS (or anything that begins to replicate/emulate inline styles) is the peak on the opposite side: vast overcorrection to take us as far from the other option as possible. I think we need to find the sweet spot.

With Atomic CSS (this is a pseudo-code example; it doesn’t use any actual Atomic CSS names) we might have really, really, really DRY CSS, but some DOM like this:

<div class="d-bk  p-md  bg-w">
    <a href="#" class="p-sm  fs-m  d-bk  c-gn">Log in</a>
</div>

I know very little about this HTML. Sure, I can tell you which declarations are acting on it (it’s pretty much inline styles) but I couldn’t tell you:

• Where does the ‘scope’ of this section of DOM start and stop?
• Are the div and a part of the same UI component?
• Could the div or a exist without one another?
• Are any of the classes thematically related?
• Do these classes appear side by side purely coincidentally?
• Do we need all classes to remain intact if we were to reuse this HTML?
• Do all classes when combined create a specific UI component?
• Are any of the classes optional at this point?
• If one of those classes changes in this bit of DOM, will they need to be changed in every similar piece of DOM?

Sure, we have tiny CSS, but our HTML has completely lost its ability to be read, or to self-document. This is too extreme.

We could get around this by having some sort of meta templating language that sits above the HTML, so something like (again, pseudo-code):

<% ui-component('promo-button') %>

…might spit out a series of Atomic classes that, when combined, create that piece of UI, but then we’re just designing more complexity into the system in order to solve a problem that we invented in the first place, which seems quite counterproductive.

(Relatedly, we could wrap up the Atomic classes in our Sass, e.g.:

.btn-promo {
  @extend .p-sm;
  @extend .fs-m;
  @extend .d-bk;
  @extend .c-gn;
}

…but this brings the same readability problems as before, as well as being complete overkill—exclusively @extending single declarations will lead to a very unusual project, as well as very tightly coupled dependencies.)

…more often than not code was being duplicated.

Back on the subject of DRYness. To quote myself from this article: Repetition in a compiled system is not a bad thing: repetition in source is a bad thing.

That is to say, having padding: 6px; appear in your CSS 50 times is no big deal. You will feel no performance impact. Gzip will crunch the bejeezus out of it.

However, manually typing out padding: 6px; 50 times in your Sass (or LESS, or whatever) is a problem. DRYness is the pursuit of a Single Source of Truth, meaning that any key data exists (and therefore needs to be changed) only once in the codebase. This ensures consistency (updated values are picked up throughout the entire project) and, more importantly, ease of maintenance. DRY is about repetition in source, not repetition-in-general.

To this end, I would just ensure that the values we’re interested in (6px, 12px, 24px, etc.) are only ever written once and then are simply recycled throughout the project. This is just as DRY (in terms of the Single Source of Truth) but also allows for more readability and context/knowledge in the HTML.

Here’s a very crude proof of concept: sassmeister.com/…/87b2da6cbd80d3f1f993. We can see from this that we only actually have one number involved at all, so we have achieved the Single Source of Truth, and we can just reuse it and its spawned size-variants very easily.

And now our HTML is more like this:

<div class="box  box--promo">
    <a href="#" class="btn  btn--small  btn--positive">Log in</a>
</div>

This tells us a heck of a lot about what’s going on in the DOM, it tells us that these classes are thematically related; it tells us which bits are optional, and it tells us what are candidates for change (i.e. anything with a -- prefix).

There is absolutely no reason you can’t have utility classes—I even do this myself—but they should be the exception rather than the rule. Avoid building entire sections of UI out of them.

Abstracting classes out as far as Atomic CSS (or any other similar approach—this isn’t a criticism of Atomic CSS specifically) suggests will sacrifice (at the very least) context, readability, and maintainability, in favour or smaller filesizes. I feel this is overkill, especially considering that gzip would negate any difference in filesize anyway. We need to find that sweet spot, and I feel Atomic CSS (and paradigms like it) overshoot the mark.

Wow, that was a very long answer. I hope it helps.

Best,
Harry

N.B. All code can now be licensed under the permissive MIT license. Read more about licensing CSS Wizardry code samples…

By Harry Roberts

Harry Roberts is an independent consultant web performance engineer. He helps companies of all shapes and sizes find and fix site speed issues.