The Raven Technique: One Step Closer to Container Queries

Avatar of Mathias Hülsbusch
Mathias Hülsbusch on (Updated on )

DigitalOcean provides cloud products for every stage of your journey. Get started with $200 in free credit!

For the millionth time: We need container queries in CSS! And guess what, it looks like we’re heading in that direction.

When building components for a website, you don’t always know how that component will be used. Maybe it will be render as wide as the browser window is. Maybe two of them will sit side by side. Maybe it will be in some narrow column. The width of it doesn’t always correlate with the width of the browser window.

It’s common to reach a point where having container based queries for the CSS of the component would be super handy. If you search around the web for solution to this, you’ll probably find several JavaScript-based solutions. But those come at a price: extra dependencies, styling that requires JavaScript, and polluted application logic and design logic.

I am a strong believer in separation of concerns, and layout is a CSS concern. For example, as nice of an API as IntersectionObserver is, I want things like :in-viewport in CSS! So I continued searching for a CSS-only solution and I came across Heydon Pickering’s The Flexbox Holy Albatross. It is a nice solution for columns, but I wanted more. There are some refinements of the original albatross (like The Unholy Albatross), but still, they are a little hacky and all that is happening is a rows-to-columns switch.

I still want more! I want to get closer to actual container queries! So, what does CSS have offer that I could tap into? I have a mathematical background, so functions like calc(), min(), max() and clamp() are things I like and understand.

Next step: build a container-query-like solution with them.

Want to see what is possible before reading on? Here is a CodePen collection showing off what can be done with the ideas discussed in this article.


Why “Raven”?

This work is inspired by Heydon’s albatross, but the technique can do more tricks, so I picked a raven, since ravens are very clever birds.

Recap: Math functions in CSS

The calc() function allows mathematical operations in CSS. As a bonus, one can combine units, so things like calc(100vw - 300px) are possible.

The min() and max() functions take two or more arguments and return the smallest or biggest argument (respectively).

The clamp() function is like a combination of min() and max() in a very useful way. The function clamp(a, x, b) will return:

  • a if x is smaller than a
  • b if x is bigger than b and
  • x if x is in between a and b

So it’s a bit like clamp(smallest, relative, largest). One may think of it as a shorthand for min(max(a,x),b). Here’s more info on all that if you’d like to read more.

We’re also going to use another CSS tool pretty heavily in this article: CSS custom properties. Those are the things like --color: red; or --distance: 20px. Variables, essentially. We’ll be using them to keep the CSS cleaner, like not repeating ourselves too much.

Let’s get started with this Raven Technique.

Step 1: Create configuration variables

Let’s create some CSS custom properties to set things up.

What is the base size we want our queries to be based on? Since we’re shooting for container query behavior, this would be 100% — using 100vw would make this behave like a media query, because that’s the width of the browser window, not the container!

--base_size: 100%;

Now we think about the breakpoints. Literally container widths where we want a break in order to apply new styles.

--breakpoint_wide: 1500px; 
/* Wider than 1500px will be considered wide */
--breakpoint_medium: 800px;
/* From 801px to 1500px will be considered medium */
/* Smaller than or exact 800px will be small */

In the running example, we will use three intervals, but there is no limit with this technique.

Now let’s define some (CSS length) values we would like to be returned for the intervals defined by the breakpoints. These are literal values:

--length_4_small: calc((100% / 1) - 10px); /* Change to your needs */
--length_4_medium: calc((100% / 2) - 10px); /* Change to your needs */
--length_4_wide: calc((100% / 3) - 10px); /* Change to your needs */

This is the config. Let’s use it!

Step 2: Create indicator variables

We will create some indicator variables for the intervals. They act a bit like boolean values, but with a length unit (0px and 1px). If we clamp those lengths as minimum and maximum values, then they serve as a sort of “true” and “false” indicator.

So, if, and only if --base_size is bigger than --breakpoint_wide, we want a variable that’s 1px. Otherwise, we want 0px. This can be done with clamp():

--is_wide: clamp(0px,
  var(--base_size) - var(--breakpoint_wide),
  1px
);

If var(--base_size) - var(--breakpoint_wide) is negative, then --base_size is smaller than --breakpoint_wide, so clamp() will return 0px in this case.

Conversely, if --base_size is bigger than --breakpoint_wide, the calculation will give a positive length, which is bigger than or equal to 1px. That means clamp() will return 1px.

Bingo! We got an indicator variable for “wide.”

Let’s do this for the “medium” interval:

--is_medium: clamp(0px,
  var(--base_size) - var(--breakpoint_medium),
  1px
); /*  DO NOT USE, SEE BELOW! */

This will give us 0px for the small interval, but 1px for the medium and the wide interval. What we want, however, is 0px for the wide interval and 1px for the medium interval exclusively.

We can solve this by subtracting --is_wide value. In the wide interval, 1px - 1px is 0px; in the medium interval 1px - 0px is 1px; and for the small interval 0px - 0px gives 0px. Perfect.

So we get:

--is_medium: calc(
  clamp(0px, 
  var(--base_size) - var(--breakpoint_medium), 
  1px) 
  - var(--is_wide)
); 

See the idea? To calculate an indicator variable, use clamp() with 0px and 1px as borders and the difference of --base_width and --breakpoint_whatever as the clamped value. Then subtract the sum of all indicators for bigger intervals. This logic produces the following for the smallest interval indicator:

--is_small: calc(
  clamp(0px,
    (var(--base_size) - 0px,
    1px)
  - (var(--is_medium) + var(--is_wide))
); 

We can skip the clamp here because the breakpoint for small is 0px and --base_size is positive, so --base_size - 0px is alway bigger than 1px and clamp() will always return 1px. Therefore, the calculation of --is_small can be simplified to:

--is_small: calc(1px - (var(--is_medium) + var(--is_wide))); 

Step 3: Use indicator variables to select interval values

Now we need to go from these “indicator variables” to something useful. Let’s assume we’re working with a pixel-based layout. Don’t panic, we will handle other units later.

Here’s a question. What does this return?

calc(var(--is_small) * 100);

If --is_small is 1px, it will return 100px and if --is_small is 0px, it will return 0px.

How is this useful? See this:

calc(
  (var(--is_small) * 100) 
  +
  (var(--is_medium) * 200) 
);

This will return 100px + 0px = 100px in the small interval (where --is_small is 1px and --is_medium is 0px). In the medium interval (where --is_medium is 1px and --is_small is 0px), it will return 0px + 200px = 200px.

Do you get the idea? See Roman Komarov’s article for a deeper look at what is going on here because it can be complex to grasp.

You multiply a pixel value (without a unit) by the corresponding indicator variable and sum up all these terms. So, for a pixel based layout, something like this is sufficient:

width: calc(
    (var(--is_small)  * 100) 
  + (var(--is_medium) * 200) 
  + (var(--is_wide)   * 500) 
  );

But most of the time, we don’t want pixel-based values. We want concepts, like “full width” or “third width” or maybe even other units, like 2rem, 65ch, and the like. We’ll have to keep going here for those.

Step 4: Use min() and an absurdly large integer to select arbitrary-length values

In the first step, we defined something like this instead of a static pixel value:

--length_4_medium: calc((100% / 2) - 10px);

How can we use them then? The min() function to the rescue!

Let’s define one helper variable:

--very_big_int: 9999; 
/* Pure, unitless number. Must be bigger than any length appearing elsewhere. */

Multiplying this value by an indicator variable gives either 0px or 9999px. How large this value should be depends on your browser. Chrome will take 999999, but Firefox will not accept that high of a number, so 9999 is a value that will work in both. There are very few viewports larger than 9999px around, so we should be OK.

What happens, then, when we min() this with any value smaller than 9999px but bigger than 0px?

min(
  var(--length_4_small), 
  var(--is_small) * var(--very_big_int) 
);

If, and only if --is_small is 0px, it will return 0px. If --is_small is 1px, the multiplication will return 9999px (which is bigger than --length_4_small), and min will return: --length_4_small.

This is how we can select any length (that is, smaller than 9999px but bigger than 0px) based on indicator variables.

If you deal with viewports larger than 9999px, then you’ll need to adjust the --very_big_int variable. This is a bit ugly, but we can fix this the moment pure CSS can drop the unit on a value in order to get rid of the units at our indicator variables (and directly multiply it with any length). For now, this works.

We will now combine all the parts and make the Raven fly!

Step 5: Bringing it all together

We can now calculate our dynamic container-width-based, breakpoint-driven value like this:

--dyn_length: calc(
    min(var(--is_wide)   * var(--very_big_int), var(--length_4_wide)) 
  + min(var(--is_medium) * var(--very_big_int), var(--length_4_medium))
  + min(var(--is_small)  * var(--very_big_int), var(--length_4_small))
);

Each line is a min() from Step 4. All lines are added up like in Step 3, the indicator variables are from Step 2 and all is based on the configuration we did in Step 1 — they work all together in one big formula!

Want to try it out? Here is a is a Pen to play with (see the notes in the CSS).

This Pen uses no flexbox, no grid, no floats. Just some divs. This is to show that helpers are unnecessary in this kind of layout. But feel free to use the Raven with these layouts too as it will help you do more complex layouts.

Anything else?

So far, we’ve used fixed pixel values as our breakpoints, but maybe we want to change layout if the container is bigger or smaller than half of the viewport, minus 10px? No problem:

--breakpoint_wide: calc(50vw - 10px);

That just works! Other formulas work as well. To avoid strange behavior, we want to use something like:

--breakpoint_medium: min(var(--breakpoint_wide), 500px);

…to set a second breakpoint at 500px width. The calculations in Step 2 depend on the fact that --breakpoint_wide is not smaller than --breakpoint_medium. Just keep your breakpoints in the right order: min() and/or max() are very useful here!

What about heights?

The evaluations of all the calculations are done lazily. That is, when assigning --dyn_length to any property, the calculation will be based on whatever --base_size evaluates to in this place. So setting a height will base the breakpoints on 100% height, if --base_size is 100%.

I have not (yet) found a way to set a height based on the width of a container. So, you can use padding-top since 100% evaluates to the width for padding.

What about showing and hiding things?

The simplest way to show and hide things the Raven way is to set the width to 100px (or any other suitable width) at the appropriate indicator variable:

.show_if_small {
  width: calc(var(--is_small) * 100);
}
.show_if_medium {
  width: calc(var(--is_medium) * 100);
}
.show_if_wide {
  width: calc(var(--is_wide) * 100);
}

You need to set:

overflow: hidden;
display: inline-block; /* to avoid ugly empty lines */

…or some other way to hide things within a box of width: 0px. Completely hiding the box requires setting additional box model properties, including margin, padding and border-width, to 0px . The Raven can do this for some properties, but it’s just as effective to fix them to 0px.

Another alternative is to use position: absolute; and draw the element off-screen via left: calc(var(--is_???) * 9999);.

Takeaways

We might not need JavaScript at all, even for container query behavior! Certainly, we’d hope that if we actually get container queries in the CSS syntax, it will be a lot easier to use and understand — but it’s also very cool that things are possible in CSS today.

While working on this, I developed some opinions about other things CSS could use:

  • Container-based units like conW and conH to set heights based on width. These units could be based on the root element of the current stacking context.
  • Some sort of “evaluate to value” function, to overcome problems with lazy evaluation. This would work great with a “strip unit” function that works at render time.

Note: In an earlier version, I had used cw and ch for the units but it was pointed out to me that those can easily be confused by with CSS units with the same name. Thanks to Mikko Tapionlinna and Gilson Nunes Filho in the comments for the tip!)

If we had that second one, it would allow us to set colors (in a clean way), borders, box-shadow, flex-grow, background-position, z-index, scale(), and other things with the Raven.

Together with component-based units, setting child dimensions to the same aspect-ratio as the parent would even be possible. Dividing by a value with unit is not possible; otherwise --indicator / 1px would work as “strip unit” for the Raven.

Bonus: Boolean logic

Indicator variables look like boolean values, right? The only difference is they have a “px” unit. What about the logical combination of those? Imagine things like “container is wider than half the screen” and “layout is in two-column mode.” CSS functions to the rescue again!

For the OR operator, we can max() over all of the indicators:

--a_OR_b: max( var(--indicator_a) , var(--indicator_b) );

For the NOT operator, we can subtract the indicator from 1px:

--NOT_a: calc(1px - var(--indicator_a));

Logic purists may stop here, since NOR(a,b) = NOT(OR(a,b)) is complete boolean algebra. But, hey, just for fun, here are some more:

AND:

--a_AND_b: min(var(--indicator_a), var(--indicator_b)); 

This evaluates to 1px if and only if both indicators are 1px.

Note that min() and max() take more than two arguments. They still work as an AND and OR for (more than two) indicator variables.

XOR:

--a_XOR_b: max(
  var(--indicator_a) - var(--indicator_b), 
  var(--indicator_b) - var(--indicator_a)
);

If (and only if) both indicators have the same value, both differences are 0px, and max() will return this. If the indicators have different values, one term will give -1px, the other will give 1px. max() returns 1px in this case.

If anyone is interested in the case where two indicators are equal, use this:

--a_EQ_b: calc(1px - 
  max(
    var(--indicator_a) - var(--indicator_b), 
    var(--indicator_b) - var(--indicator_a)
  )
);

And yes, this is NOT(a XOR b). I was unable to find a “nicer” solution to this.

Equality may be interesting for CSS length variables in general, rather than just being used for indicator variables. By using clamp() once again, this might help:

--a_EQUALS_b_general: calc(
  1px -
  clamp(0px,
        max(
          var(--var_a) - var(--var_b),
          var(--var_b) - var(--var_a)
        ),
        1px)
  );

Remove the px units to get general equality for unit-less variables (integers).

I think this is enough boolean logic for most layouts!

Bonus 2: Set the number of columns in a grid layout

Since the Raven is limited to return CSS length values, it is unable to directly choose the number of columns for a grid (since this is a value without a unit). But there is a way to make it work (assuming we declared the indicator variables like above):

--number_of_cols_4_wide: 4;
--number_of_cols_4_medium: 2;
--number_of_cols_4_small: 1;
--grid_gap: 0px;

--grid_columns_width_4_wide: calc(
(100% - (var(--number_of_cols_4_wide) - 1) * var(--grid_gap) ) / var(--number_of_cols_4_wide));
--grid_columns_width_4_medium: calc(
(100% - (var(--number_of_cols_4_medium) - 1) * var(--grid_gap) ) / var(--number_of_cols_4_medium));
--grid_columns_width_4_small: calc(
(100% - (var(--number_of_cols_4_small) - 1) * var(--grid_gap) ) / var(--number_of_cols_4_small));

--raven_grid_columns_width: calc( /*  use the Raven to combine the values  */
  min(var(--is_wide) * var(--very_big_int),var(--grid_columns_width_4_wide)) 
  + min(var(--is_medium) * var(--very_big_int),var(--grid_columns_width_4_medium))
  + min(var(--is_small) * var(--very_big_int),var(--grid_columns_width_4_small))
  );

And set your grid up with:

.grid_container{
  display: grid;
  grid-template-columns: repeat(auto-fit, var(--raven_grid_columns_width));
  gap: var(--grid_gap)
};

How does this work?

  1. Define the number of columns we want for each interval (lines 1, 2, 3)
  2. Calculate the perfect width of the columns for each interval (lines 5, 6, 7).

    What is happening here?

    First, we calculate the available space for our columns. This is 100%, minus the place the gaps will take. For n columns, there are (n-1) gaps. This space is then divided by the number of columns we want.

  3. Use the Raven to calculate the right column’s width for the actual --base_size.

In the grid container, this line:

grid-template-columns: repeat(auto-fit, var(--raven_grid_columns_width));

…then chooses the number of columns to fit the value the Raven provided (which will result in our --number_of_cols_4_??? variables from above).

The Raven may not be able give the number of columns directly, but it can give a length to make repeat and autofit calculate the number we want for us.

But auto-fit with minmax() does the same thing, right? No! The solution above will never give three columns (or five) and the number of columns does not need to increase with the width of the container. Try to set the following values in this Pen to see the Raven take full flight:

--number_of_cols_4_wide: 1;
--number_of_cols_4_medium: 2;
--number_of_cols_4_small: 4;

Bonus 3: Change the background-color with a linear-gradient()

This one is a little more mind-bending. The Raven is all about length values, so how can we get a color out of these? Well, linear gradients deal with both. They define colors in certain areas defined by length values. Let’s go through that concept in more detail before getting to the code.

To work around the actual gradient part, it is a well known technique to double up a color stop, effectively making the gradient part happen within 0px. Look at this code to see how this is done:

background-image:linear-gradient(
  to right,
  red 0%,
  red 50%,
  blue 50%,
  blue 100%
);

This will color your background red on the left half, blue on the right. Note the first argument “to right.” This implies that percentage values are evaluated horizontally, from left to right.

Controlling the values of 50% via Raven variables allows for shifting the color stop at will. And we can add more color stops. In the running example, we need three colors, resulting in two (doubled) inner color stops.

Adding some variables for color and color stops, this is what we get:

background-image: linear-gradient(
  to right,
  var(--color_small) 0px,
  var(--color_small) var(--first_lgbreak_value),
  var(--color_medium) var(--first_lgbreak_value),
  var(--color_medium) var(--second_lgbreak_value),
  var(--color_wide) var(--second_lgbreak_value),
  var(--color_wide) 100%
);

But how do we calculate the values for --first_lgbreak_value and --second_lgbreak_value? Let’s see.

The first value controls where --color_small is visible. On the small interval, it should be 100%, and 0px in the other intervals. We’ve seen how to do this with the raven. The second variable controls the visibility of --color_medium. It should be 100% for the small interval, 100% for the medium interval, but 0px for the wide interval. The corresponding indicator must be 1px if the container width is in the small or the medium interval.

Since we can do boolean logic on indicators, it is:

max(--is_small, --is_medium)

…to get the right indicator. This gives:

--first_lgbreak_value: min(var(--is_small) * var(--very_big_int), 100%);
--second_lgbreak_value: min(
  max(var(--is_small), var(--is_medium)) * var(--very_big_int), 100%);

Putting things together results in this CSS code to change the background-color based on the width (the interval indicators are calculated like shown above):

--first_lgbreak_value: min(
      var(--is_small) * var(--very_big_int), 100%);
--second_lgbreak_value: min(
    max(var(--is_small), var(--is_medium)) * var(--very_big_int), 100%);

--color_wide: red;/* change to your needs*/
--color_medium: green;/* change to your needs*/
--color_small: lightblue;/* change to your needs*/

background-image: linear-gradient(
  to right,
  var(--color_small) 0px,
  var(--color_small) var(--first_lgbreak_value),
  var(--color_medium) var(--first_lgbreak_value),
  var(--color_medium) var(--second_lgbreak_value),
  var(--color_wide) var(--second_lgbreak_value),
  var(--color_wide) 100%
);

Here’s a Pen to see that in action.

Bonus 4: Getting rid of nested variables

While working with the Raven, I came across a strange problem: There is a limit on the number of nested variables that can be used in calc(). This can cause some problems when using too many breakpoints. As far as I understand, this limit is in place to prevent page blocking while calculating the styles and allow for faster circle-reference checks.

In my opinion, something like evaluate to value would be a great way to overcome this. Nevertheless, this limit can give you a headache when pushing the limits of CSS. Hopefully this problem will be tackled in the future.

There is a way to calculate the indicator variables for the Raven without the need of (deeply) nested variables. Let’s look at the original calculation for the --is_medium value:

--is_medium:calc(
  clamp(0px, 
        var(--base_size) - var(--breakpoint_medium), 
        1px) 
        - var(--is_wide)
); 

The problem occurs with the subtraction of --is_wide . This causes the CSS parser to paste in the definition of the complete formula of --is_wide. The calculation of --is_small has even more of these types of references. (The definition for --is_wide will even be pasted twice since it is hidden within the definition of --is_medium and is also used directly.)

Fortunately, there is a way to calculate indicators without referencing indicators for bigger breakpoints.

The indicator is true if, and only if, --base_size is bigger than the lower breakpoint for the interval and smaller or equal than the higher breakpoint for the interval. This definition gives us the following code:

--is_medium: 
  min(
    clamp(0px, var(--base_size) - var(--breakpoint_medium), 1px),
    clamp(0px, 1px + var(--breakpoint_wide) - var(--base_size), 1px)
  );
  • min() is used as a logical AND operator
  • the first clamp() is “--base_size is bigger than --breakpoint_medium
  • the second clamp() means “--base_size is smaller or equal than --breakpoint_wide.”
  • Adding 1px switches from “smaller than” to “smaller or equal than.” This works, because we are dealing with whole (pixel) numbers (a <= b means a < (b+1) for whole numbers).

The complete calculation of the indicator variables can be done this way:

--is_wide: clamp(0px, var(--base_size) - var(--breakpoint_wide), 1px);

--is_medium: min(clamp(0px, var(--base_size) - var(--breakpoint_medium), 1px),
                 clamp(0px, 1px + var(--breakpoint_wide) - var(--base_size), 1px)
             );

--is_small: clamp(0px,1px + var(--breakpoint_medium) - var(--base_size), 1px);

The calculations for --is_wide and --is_small are simpler, because only one given breakpoint needs to be checked for each.

This works with all the things we’ve looked at so far. Here’s a Pen that combines examples.

Final thoughts

The Raven is not capable of all the things that a media query can do. But we don’t need it to do that, as we have media queries in CSS. It is fine to use them for the “big” design changes, like the position of a sidebar or a reconfiguration of a menu. Those things happen within the context of the full viewport (the size of the browser window).

But for components, media queries are kind of wrong, since we never know how components will be sized.

Heydon Pickering demonstrated this problem with this image:

Three boxes representing browsers from left-to-right. The first is a wide viewport with three boxes in a single row. The second is a narrow viewport with the boxes stacked vertically. The third is a wide viewport, but with a dashed vertical line down the middle representing a container and the three boxes are to the right of it in a single row.

I hope that the Raven helps you to overcome the problems of creating responsive layouts for components and pushes the limits of “what can be done with CSS” a little bit further.

By showing what is possible today, maybe “real” container queries can be done by adding some syntax sugar and some very small new functions (like conW, conH, “strip-unit” or “evaluate-to-pixels”). If there was a function in CSS that allows to rewrite “1px” to a whitespace, and “0px” to “initial“, the Raven could be combined with the Custom Property Toggle Trick and change every CSS property, not just length values.

By avoiding JavaScript for this, your layouts will render faster because it’s not dependent on JavaScript downloading or running. It doesn’t even matter if JavaScript is disabled. These calculations will not block your main thread and your application logic isn’t cluttered with design logic.


Thanks to Chris, Andrés Galante, Cathy Dutton, Marko Ilic, and David Atanda for their great CSS-Tricks articles. They really helped me explore what can be done with the Raven.