Programming Sass to Create Accessible Color Combinations

Avatar of Jason Hogue
Jason Hogue on

We’re all looking for low-hanging fruit to make our sites and apps more accessible. One of the easier things we can do is make sure the colors we use are easy on the eyes. High color contrast is something that benefits everyone. It not only reduces eye strain in general, but is crucial for folks who deal with reduced vision.

So let’s not only use better color combinations in our designs but find a way to make it easier for us to implement high contrasts. There’s one specific strategy we use over at Oomph that lets a Sass function do all the heavy lifting for us. I’ll walk you through how we put that together.

Want to jump right to the code because you already understand everything there is to know about color accessibility? Here you go.

What we mean by “accessible color combinations”

Color contrast is also one of those things we may think we have handled. But there’s more to high color contrasts than eyeballing a design. There are different levels of acceptable criteria that the WCAG has defined as being accessible. It’s actually humbling to crack open the WebAIM Contrast Checker and run a site’s color combinations through it.

My team adheres to WCAG’s Level AA guidelines by default. This means that: 

  • Text that is 24px and larger, or 19px and larger if bold, should have a Color Contrast Ratio (CCR) of 3.0:1.
  • Text that is smaller than 24px should have a CCR of 4.5:1.

If a site needs to adhere to the enhanced guidelines for Level AAA, the requirements are a little higher:

  • Text that is 24px and larger, or 19px and larger if bold, should have a CCR of 4.5:1.
  • Text that is smaller than 24px should have a CCR of 7:1.

Ratios? Huh? Yeah, there’s some math involved here. But the good news is that we don’t need to do it ourselves or even have the same thorough understanding about how they’re calculated the way Stacie Arellano recently shared (which is a must read if you’re into the science of color accessibility).

That’s where Sass comes in. We can leverage it to run difficult mathematical computations that would otherwise fly over many of our heads. But first, I think it’s worth dealing with accessible colors at the design level.

Accessible color palettes start with the designs

That’s correct. The core of the work of creating an accessible color palette starts with the designs. Ideally, any web design ought to consult a tool to verify that any color combinations in use pass the established guidelines — and then tweak the colors that don’t. When our design team does this, they use a tool that we developed internally. It works on a list of colors, testing them over a dark and a light color, as well as providing a way to test other combinations. 

ColorCube provides an overview of an entire color palette, showing how each color performs when paired with white, black, and even each other. It even displays results for WCAG Levels AA and AAA next to each result. The tool was designed to throw a lot of information at the user all at once when evaluating a list of colors.

This is the first thing our team does. I’d venture to guess that many brand colors aren’t chosen with accessibility at the forefront. I often find that those colors need to change when they get translated to a web design. Through education, conversation, and visual samples, we get the client to sign off on the new color palette. I’ll admit: that part can be harder than the actual work of implementing accessible colors combinations.

The Color Contrast Audit: A typical design delivery when working with an existing brand’s color palette. Here, we suggest to stop using the brand color Emerald with white, but use an “Alt” version that is slightly darker instead. 

The problem that I wanted to solve with automation are the edge cases. You can’t fault a designer for missing some instance where two colors combine in an unintended way — it just happens. And those edge cases will come up, whether it is during the build or even a year later when new colors are added to the system.

Developing for accessibility while keeping true to the intent of a color system

The trick when changing colors to meet accessibility requirements is not changing them so much that they don’t look like the same color anymore. A brand that loves its emerald green color is going to want to maintain the intent of that color — it’s “emerald-ness.” To make it pass for accessibility when it is used as text over a white background, we might have to darken the green and increase its saturation. But we still want the color to “read” the same as the original color.

To achieve this, we use the Hue Saturation Lightness (HSL) color model. HSL gives us the ability to keep the hue as it is but adjust the saturation (i.e. increase or decrease color) and lightness (i.e. add more black or more white). The hue is what makes a green that green, or a blue that blue. It is the “soul” of the color, to get a little mystical about it. 

Hue is represented as a color wheel with a value between 0° and 360° — yellow at 60°, green at 120°, cyan at 180°, etc. Saturation is a percentage ranging from 0% (no saturation) to 100% (full saturation). Lightness is also a value that goes from 0% to 100%, where no lightness is at 0%, no black and no white is at 50%, and 100% is all lightness, or very light.

A quick visual of what tweaking a color looks like in our tool:

With HSL, changing the low-contrast green to a higher contrast meant changing the saturation from 63 to 95 and the lightness from 45 to 26 on the left. That’s when the color gets a green check mark in the middle when used with white. The new green still feels like it is in the same family, though, because the Hue remained at 136, which is like the color’s “soul.”

To learn more, play around with the fun HSL visualizer mothereffinghsl.com. But for a more in-depth description of color blindness, WCAG color contrast levels, and the HSL color space, we wrote an in-depth blog post about it.

The use case I want to solve

Designers can adjust colors with the tools that we just reviewed, but so far, no Sass that I have found could do it with mathematical magic. There had to be a way. 

These are some similar approaches I have seen in the wild:

  • An idea by Josh Bader uses CSS variables and colors split into their RGB values to calculate whether white or black is the best accessible color to use in a given situation.
  • Another idea by Facundo Corradini does something similar with HSL values and a very cool “switch function” in CSS.

I didn’t like these approaches. I didn’t want to fallback to white or black. I wanted colors to be maintained but adjusted to be accessible. Additionally, changing colors to their RGB or HSL components and storing them with CSS variables seemed messy and unsustainable for a large codebase.

I wanted to use a preprocessor like Sass to do this: given two colors, automagically adjust one of them so the pair receives a passing WCAG grade. The rules state a few other things to consider as well — size of the text and whether or not the font is bold. The solution had to take this into account. 

In code terms, I wanted to do this:

// Transform this non-passing color pair:
.example {
  background-color: #444;
  color: #0094c2; // a 2.79 contrast ratio when AA requires 4.5
  font-size: 1.25rem;
  font-weight: normal;
}


// To this passing color pair:
.example {
  background-color: #444;
  color: #00c0fc; // a 4.61 contrast ratio
  font-size: 1.25rem;
  font-weight: normal;
}

A solution that does this would be able to catch and handle those edge cases we mentioned earlier. Maybe the designer accounted for a brand blue to be used over a light blue, but not a light gray. Maybe the red used in error messages needs to be tweaked for this one form that has a one-off background color. Maybe we want to implement a dark mode feature to the UI without having to retest all the colors again. These are the use cases I had in mind going into this. 

With formulas can come automation

The W3C has provided the community with formulas that help analyze two colors used together. The formula multiplies the RGB channels of both colors by magic numbers (a visual weight based on how humans perceive these color channels) and then divides them to come up with a ratio from 0.0 (no contrast) to 21.0 (all the contrast, only possible with white and black). While imperfect, this is the formula we use right now:

If L1 is the relative luminance of a first color 
And L2 is the relative luminance of a second color, then
- Color Contrast Ratio = (L1 + 0.05) / (L2 + 0.05)
Where
- L = 0.2126 * R + 0.7152 * G + 0.0722 * B
And
- if R sRGB <= 0.03928 then R = R sRGB /12.92 else R = ((R sRGB +0.055)/1.055) ^ 2.4
- if G sRGB <= 0.03928 then G = G sRGB /12.92 else G = ((G sRGB +0.055)/1.055) ^ 2.4
- if B sRGB <= 0.03928 then B = B sRGB /12.92 else B = ((B sRGB +0.055)/1.055) ^ 2.4
And
- R sRGB = R 8bit /255
- G sRGB = G 8bit /255
- B sRGB = B 8bit /255

While the formula looks complex, it’s just math right? Well, not so fast. There is a part at the end of a few lines where the value is multiplied by a decimal power — raised to the power of 2.4. Notice that? Turns out that it’s complex math which most programming languages can accomplish — think Javascript’s math.pow() function — but Sass is not powerful enough to do it. 

There’s got to be another way…

Of course there is. It just took some time to find it. 🙂

My first version used a complex series of math calculations that did the work of decimal powers within the limited confines of what Sass can accomplish. Lots of Googling found folks much smarter than me supplying the functions. Unfortunately, calculating only a handful of color contrast combinations increased Sass build times exponentially. So, that means Sass can do it, but that does not mean it should. In production, build times for a large codebase could increase to several minutes. That’s not acceptable. 

After more Googling, I came across a post from someone who was trying to do a similar thing. They also ran into the lack of exponent support in Sass. They wanted to explore “the possibility of using Newtonian approximation for the fractional parts of the exponent.” I totally understand the impulse (not). Instead, they decided to use a “lookup table.” It’s a genius solution. Rather than doing the math from scratch every time, a lookup table provides all the possible answers pre-calculated. The Sass function retrieves the answer from the list and it’s done.

In their words:

The only part [of the Sass that] involves exponentiation is the per-channel color space conversions done as part of the luminance calculation. [T]here are only 256 possible values for each channel. This means that we can easily create a lookup table.

Now we’re cooking. I had found a more performant direction. 

Usage example

Using the function should be easy and flexible. Given a set of two colors, adjust the first color so it passes the correct contrast value for the given WCAG level when used with the second color. Optional parameters will also take the font size or boldness into account.

// @function a11y-color(
//   $color-to-adjust,
//   $color-that-will-stay-the-same,
//   $wcag-level: 'AA',
//   $font-size: 16,
//   $bold: false
// );


// Sass sample usage declaring only what is required
.example {
  background-color: #444;
  color: a11y-color(#0094c2, #444); // a 2.79 contrast ratio when AA requires 4.5 for small text that is not bold
}


// Compiled CSS results:
.example {
  background-color: #444;
  color: #00c0fc; // which is a 4.61 contrast ratio
}

I used a function instead of a mixin because I preferred the output of a single value independent from a CSS rule. With a function, the author can determine which color should change. 

An example with more parameters in place looks like this:

// Sass
.example-2 {
  background-color: a11y-color(#0094c2, #f0f0f0, 'AAA', 1.25rem, true); // a 3.06 contrast ratio when AAA requires 4.5 for text 19px or larger that is also bold
  color: #f0f0f0;
  font-size: 1.25rem;
  font-weight: bold;
}


// Compiled CSS results:
.example-2 {
  background-color: #087597; // a 4.6 contrast ratio
  color: #f0f0f0;
  font-size: 1.25rem;
  font-weight: bold;
}

A deeper dive into the heart of the Sass function

To explain the approach, let’s walk through what the final function is doing, line by line. There are lots of helper functions along the way, but the comments and logic in the core function explain the approach:

// Expected:
// $fg as a color that will change
// $bg as a color that will be static and not change
// Optional:
// $level, default 'AA'. 'AAA' also accepted
// $size, default 16. PX expected, EM and REM allowed
// $bold, boolean, default false. Whether or not the font is currently bold
//
@function a11y-color($fg, $bg, $level: 'AA', $size: 16, $bold: false) {
  // Helper: make sure the font size value is acceptable
  $font-size: validate-font-size($size);
  // Helper: With the level, font size, and bold boolean, return the proper target ratio. 3.0, 4.5, or 7.0 results expected
  $ratio: get-ratio($level, $font-size, $bold);
  // Calculate the first contrast ratio of the given pair
  $original-contrast: color-contrast($fg, $bg);
  
  @if $original-contrast >= $ratio {
    // If we pass the ratio already, return the original color
    @return $fg;
  } @else {
    // Doesn't pass. Time to get to work
    // Should the color be lightened or darkened?
    // Helper: Single color input, 'light' or 'dark' as output
    $fg-lod: light-or-dark($fg);
    $bg-lod: light-or-dark($bg);


    // Set a "step" value to lighten or darken a color
    // Note: Higher percentage steps means faster compile time, but we might overstep the required threshold too far with something higher than 5%
    $step: 2%;
    
    // Run through some cases where we want to darken, or use a negative step value
    @if $fg-lod == 'light' and $bg-lod == 'light' {
      // Both are light colors, darken the fg (make the step value negative)
      $step: - $step;
    } @else if $fg-lod == 'dark' and $bg-lod == 'light' {
      // bg is light, fg is dark but does not pass, darken more
      $step: - $step;
    }
    // Keeping the rest of the logic here, but our default values do not change, so this logic is not needed
    //@else if $fg-lod == 'light' and $bg-lod == 'dark' {
    //  // bg is dark, fg is light but does not pass, lighten further
    //  $step: $step;
    //} @else if $fg-lod == 'dark' and $bg-lod == 'dark' {
    //  // Both are dark, so lighten the fg
    //  $step: $step;
    //}
    
    // The magic happens here
    // Loop through with a @while statement until the color combination passes our required ratio. Scale the color by our step value until the expression is false
    // This might loop 100 times or more depending on the colors
    @while color-contrast($fg, $bg) < $ratio {
      // Moving the lightness is most effective, but also moving the saturation by a little bit is nice and helps maintain the "power" of the color
      $fg: scale-color($fg, $lightness: $step, $saturation: $step/2);
    }
    @return $fg;
  }
}

The final Sass file

Here’s the entire set of functions! Open this in CodePen to edit the color variables at the top of the file and see the adjustments that the Sass makes:

All helper functions are there as well as the 256-line lookup table. Lots of comments should help folks understand what is going on. 

When an edge case has been encountered, a version in SassMeister with debug output was helpful while I was developing it to see what might be happening. (I changed the main function to a mixin so I can debug the output.) Feel free to poke around at this as well.

Play with this gist on SassMeister.

And finally, the functions have been stripped out of CodePen and put into a GitHub repo. Drop issues into the queue if you run into problems. 

Cool code! But can I use this in production?

Maybe. 

I’d like to say yes, but I’ve been iterating on this thorny problem for a while now. I feel confident in this code but would love more input. Use it on a small project and kick the tires. Let me know how the build time performs. Let me know if you come across edge cases where passing color values are not being supplied. Submit issues to the GutHub repo. Suggest improvements based on other code you’ve seen in the wild. 

I’d love to say that I have Automated All the A11y Things, but I also know it needs to be road-tested before it can be called Production Ready™. I’m excited to introduce it to the world. Thanks for reading and I hope to hear how you are using it real soon.