In CSS, we're given a tool to explicitly control the stacking order of HTML elements:
z-index. Elements with a higher value will appear on top:
.first.box has a larger z-index than
.second.box, it stacks in front. If we remove that z-index declaration, it falls to the back. The code above is editable—give it a shot!
Things aren't always so simple, however. Sometimes, the larger z-index value doesn't win.
Check out what's going on here:
.tooltip has a much larger z-index than
header! So why on earth is the header on top?
To unravel this mystery, we'll need to learn about stacking contexts, an obscure-yet-fundamental CSS mechanism. In this article, we'll explore what they are, how they work, and how we can use them to our advantage.
Link to this headingLayers and groups
If you've ever used image-editing software like Photoshop or Figma, you're probably familiar with the concept of layers:
Our image has 3 separate canvases, stacked like pancakes. The bottom layer is a cat photo, with 2 layers on top that add silly details. By flattening these layers, we wind up with a final composition:
In these programs, we can also group layers:
Like files in a folder, a group allows us to segment our layers. In terms of stacking order, layers aren't allowed to “intermingle” between groups: All of
dog's layers will appear on top of all of
When we export the composition, we don't see the cat at all, since it's behind the dog:
When it comes to CSS, things work in a similar way: elements are grouped into stacking contexts. When we give an element a z-index, that value is only compared against other elements in the same context. z-index values are not global.
By default, a plain HTML document will have a single stacking context that encompasses all nodes. But we can create additional contexts!
There are many ways to create stacking contexts, but here's the most common:
By combining these two declarations, a secret mechanism is triggered: a stacking context is created, forming a group around this element and all of its children.
Let's take another look at our problem from above:
We can map out the stacking contexts being created in this snippet:
- The root context
.tooltip element has a z-index of 999999, but that value is only relevant within the
<main> stacking context. It controls whether the tooltip shows up above or below the adjacent
<p> tag, nothing more.
Meanwhile, in the parent context,
<main> are compared. Because
<main> has a smaller z-index, it shows up underneath
<header>. All of its children come along for the ride.
Link to this headingFixing our example
How do we solve our tooltip problem? Well, in this case, we don't actually need to create a stacking context on our
Without a z-index,
<main> won't create a stacking context. Our hierarchy, then, looks like this:
- The root context
Because the header and our tooltip are now in the same context, their z-index values face off, and the tooltip emerges as the victor.
An important distinction: we're not talking about parent/child relationships here. It doesn't matter that the tooltip is more deeply nested than the header. The browser only cares about stacking contexts.
Link to this headingCreating stacking contexts
We've seen how we can create a stacking context by combining relative or absolute positioning with
z-index, but it's not the only way! Here are some others:
opacityto a value less than
sticky(No z-index needed for these values!)
- Applying a
- Adding a
z-indexto a child inside a
will-changewith a value like
- Explicitly creating a context with
isolation: isolate(More on this soon!)
There are a few other ways as well. You can find the full list on MDN.
This can lead to some surprising situations. Check out what's happening here:
main doesn't set a z-index anymore, but it uses
will-change, a property that can create a stacking context all on its own.
Link to this headingA common misconception about z-index
In order for z-index to work, we need to set
position to something like
Not quite. Check out what's happening here:
The second box is lifted above its siblings using
z-index. There are no
position declarations anywhere in the snippet, though!
z-index only works with "positioned" elements (elements that set
position to something other than the default “static”). But the Flexbox specification adds an exception: flex children can use
z-index even if they're statically-positioned.
An earlier version of this post said that all elements that create a stacking context can use
z-index, but that was incorrect. 😬
Link to this headingHold on a minute…
There's a Weird Thing here, and I think it's worth pondering about for a minute or two.
In our Photoshop analogy, there is a clear distinction between groups and layers. All of the visual elements are layers, and groups can be conjured as structural helpers to contain them. They are distinct ideas.
On the web, however, the distinction is a bit less clear. Every element that uses z-index must also create a stacking context.
When we decide to give an element a z-index, our goal is typically to lift or lower that element above/below some other element in the parent stacking context. We aren't intending to produce a stacking context on that element! But it's important that we consider it.
When a stacking context is created, it “flattens” all of its descendants. Those children can still be rearranged internally, but we've essentially locked those children in.
Let's take another look at the markup from earlier:
By default, HTML elements will be stacked according to their DOM order. Without any CSS interference,
main will render on top of
We can lift
header to the front by giving it a z-index, but not without flattening all of its children. This mechanism is what led to the bug we discussed earlier.
We shouldn't think of
z-index purely as a way to change an element's order. We should also think of it as a way to form a group around that element's children. z-index won't work unless a group is formed.
Link to this headingAirtight abstractions with “isolation”
One of my favourite CSS properties is also one of the most obscure. I'd like to introduce you to the
isolation property, a hidden gem in the language.
Here's how you'd use it:
When we apply this declaration to an element, it does precisely 1 thing: it creates a new stacking context.
With so many different ways to create a stacking context, why do we need another one? Well, with every other method, stacking contexts are created implicitly, as the result of some other change.
isolation creates a stacking context in the purest way possible:
- No need to prescribe a z-index value
- Can be used on statically-positioned elements
- Doesn't affect the child's rendering in any way
This is incredibly useful, since it lets us "seal off" an element's children.
Let's look at an example. Recently, I built this neat envelope component. Hover or focus to see it open:
It consists of several layers:
I packaged this effect up in a React component,
<Envelope>. It looks something like this (inline styles used for brevity):
(If you're wondering why
Flap has a dynamic z-index, it's because it needs to shift behind the letter when the envelope is open.)
A good React component is sealed off from its environment, like a spacesuit. This spacesuit, however, has sprung a leak. Check out what happens when I use it near a
<Envelope> component wraps the 4 layers in a div, but it doesn't create a stacking context. As a result, those layers can become “intertwined” with other components, like the world's most boring game of Twister.
isolation: isolate on the top-level element within
<Envelope>, we guarantee that it'll be positioned as a group:
Why not create a stacking context the old-fashioned way, with
position: relative; z-index: 1? Well, React components are meant to be reusable; is
1 really the right z-index value for this component in all circumstances? The beauty of
isolation is that it keeps our components unopinionated and flexible.
More and more, I'm starting to believe that z-index is an escape hatch, similar to
!important. This is one trick that allows us to control stacking order without pulling the big red z-index lever.
I'm working on a follow-up tutorial where we look at some other tricks to keep z-index inflation down. Watch this space!
Link to this headingDebugging stacking context issues
Unfortunately, I haven't found much tooling to help debug stacking-context issues.
Microsoft Edge has an interesting “3D view” that allows us to view stacking contexts:
This is an ambitious idea, but honestly I find it pretty overwhelming. It's hard to locate a specific element in this view, and I don't really feel like it helps me understand the stacking contexts in my app.
There's one other neat trick you can use sometimes:
offsetParent returns the closest ancestor rendered with a
position value other than
static. It crawls up the tree looking for relative / absolute / fixed / sticky ancestors.
This is not a perfect solution. Not all stacking contexts use positioned layout, and not all positioned elements create a stacking context! That said, in practice, there does tend to be a pretty strong correlation between these two things. At the very least, it's a starting point.
If you know of any tooling that can help here (or if you create one!), let me know on Twitter
Update: Felix Becker reached out to share a VSCode extension that highlights when stacking contexts are created:
This extension works on .css and .scss files (no CSS-in-JS support).
Update 2: Giuseppe Gurgone reached out to let me know about this Chrome extension which adds a new “z-index” pane to the devtools.
Update 3: Andrea Dragotta created an incredible browser extension that adds a bunch of super-important information about z-index and stacking contexts:
This is an awesome tool, and I've been using it regularly. Install CSS Stacking Context Inspector:
Link to this headingGoing deeper
Stacking contexts are a good example of how CSS is built on "hidden mechanisms". You can spend years building interfaces with CSS without knowing that they exist.
Unless you explicitly take the time to learn about these mechanisms, your mental model will always be missing pieces. And if your mental model is even slightly misaligned, it's only a matter of time until that discrepancy causes problems.
CSS doesn't have warnings or error messages. When something surprising happens, there's no clear "next step" to figure out what went wrong. These disruptions take us out of flow state and shake our confidence. I think this is why so many front-end developers don't enjoy writing CSS.
Once you build up an intuition for the language, though, CSS becomes an absolute joy. I love writing CSS nowadays.
My goal for 2021 is to help other developers discover this joy. I'm creating a comprehensive self-paced online course that explains how CSS works at a deeper level, and teaches the practical skills I use every day to build all kinds of user interfaces.