One of the challenges we face when implementing class-based atomic styling is that it often depends on a specific breakpoint for context.

<div class="span-12"></div> <!-- we want this for small screens  -->
<div class="span-6"></div>  <!-- we want this for medium screens -->
<div class="span-4"></div>  <!-- we want this for large screens  -->

It’s common to use a prefix to target each breakpoint:

<div class="sm-span-12 md-span-6 lg-span-4"></div>

This works well until we start adding multiple classes. That’s when it becomes difficult to keep a track what relates to what and where to add, remove. or change stuff.

<div class="
  sm-span-12 
  md-span-6 
  lg-span-4 
  sm-font-size-xl 
  md-font-size-xl 
  lg-font-size-xl 
  md-font-weight-500 
  lg-font-weight-700">
</div>

We can try to make it more readable by re-grouping:

<div class="
  sm-span-12 
  sm-font-size-xl 


  md-span-6 
  md-font-size-xl 
  md-font-weight-500 


  lg-span-4 
  lg-font-size-xl 
  lg-font-weight-700">
</div>

We can add funky separators (invalid class names will be ignored):

<div class="
  [
   sm-span-12 
   sm-font-size-xl 
  ],[
   md-span-6 
   md-font-size-xl 
   md-font-weight-500 
  ],[
   lg-span-4 
   lg-font-size-xl 
   lg-font-weight-700
  ]">
</div>

But this still feels messy and hard to grasp, at least to me.

We can get a better overview and avoid implementation prefixes by grouping attribute selectors instead of actual classes:

<div 
  data-sm="span-12 font-size-lg"
  data-md="span-6 font-size-xl font-weight-500"
  data-lg="span-4 font-size-xl font-weight-700"
>
</div>

These aren’t lost of classes but a whitespace-separated list of attributes we can select using [attribute~="value"], where ~= requires the exact word to be found in the attribute value in order to match.

@media (min-width: 0) {
 [data-sm~="span-1"] { /*...*/ }              
 [data-sm~="span-2"] { /*...*/ }   
 /* etc. */ 
}
@media (min-width: 30rem) {
 [data-md~="span-1"] { /*...*/ }   
 [data-md~="span-2"] { /*...*/ }   
 /* etc. */   
}
@media (min-width: 60rem) {
 [data-lg~="span-1"] { /*...*/ }   
 [data-lg~="span-2"] { /*...*/ }   
 /* etc. */   
}

It may be a bit odd-looking but I think translating atomic classes to  attributes is fairly straightforward (e.g. .sm-span-1 becomes [data-sm~="span-1"]). Plus, attribute selectors have the same specificity as classes, so we lose nothing there. And, unlike classes, attributes can be written without escaping special characters, like /+.:?.

That’s all! Again, this is merely an idea that aims to make switching declarations in media queries easier to write, read and manage. It’s definitely not a proposal to do away with classes or anything like that.