CSS Scroll State Container Queries Explainer

Introduction

Level 3 of CSS Containment introduces Container Queries for querying size and style of containers to style their descendants depending on layout and computed styles respectively. There are requests to allow to query other states associated with a container, in particular various scroll based states.

Background and Motivation

Certain scroll-based states are only available for styling through javascript events. These include whether a sticky positioned element is stuck in the sense that it has an position offset applied, whether a scroll-snap-aligned element is snapped to its scroll snap container or not. Exposing these states in CSS would make it simpler for authors to create effects on snapped elements, or have sticky positioned elements change their appearance when stuck.

This article talks about the need for support for styling scroll-snapped elements: https://web.dev/state-of-css-2022/#scroll-snap-features-are-too-limited. In particular, see: https://web.dev/state-of-css-2022/#snap-target. This article also talks about the lack of being able to style snapped elements, using intersection observers as a javascript alternative: https://blog.logrocket.com/style-scroll-snap-points-css/.

These ideas have been explored before. For instance, see this github issue, and this explainer for :snapped pseudo classes. That explainer pre-dates the container queries specification and implementation that has happened recently.

One major argument against exposing this state to style resolution is that it introduces cycles in rendering updates since layout influences scroll positions which influence this state which again can influence layout through style depending on this state (see this comment and onwards).

There is an open github issue for state container queries where it was resolved to defer to the next level of the spec - presumably css-contain-4.

Container Queries vs Pseudo Classes

Container queries are an alternative to using pseudo classes for these states. There are pros and cons of using new pseudo classes. Container queries would naturally limit the style depending on this state to be applied in the stuck or snapped element's subtree, while a pseudo class could arbitrarily be used to style sibling subtrees and even ancestors (using :has()) unless the use in selectors are restricted to certain combinators. One can argue that using pseudo classes would make it more likely to end up in situations where layout cycles are an issue.

Containers already have the container-type property which can be used to explicitly choose which elements can be queried for state, which could also help avoiding cycles issues. It would also allow for more performant implementations in that fewer elements need to be checked for state. When size containers were introduced, there was a conscious choice to introduce an explicit container-type instead of implicitly deduce a size container from the contain property. For a pseudo class approach, we would either have to introduce an explicit state property or deduce that an element is allowed to match based on another property. E.g. :stuck allowed for position:sticky.

In the discussion in: https://github.com/w3c/csswg-drafts/issues/5979, container queries are brought up as an alternative to pseudo classes. That issue is also talking about containment, whether container queries should be used instead of pseudo classes, and the fact that the need for an extra loop after snapshotting could be hooked into the same HTML rendering update as for scroll driven animations.

States to Query

This explainer tries to capture scroll-based state queries and the following sections cover each of them. The commonality is that they all rely on scroll positions and sizes.

Stuck Sticky-positioned Elements

Introduce a container-type sticky to allow sticky positioned elements as query containers for querying whether the sticky positioned has an offset applied to fulfill the constraint for a given inset property. For instance:

<style>
#sticky {
  container-name: my-menu;
  container-type: sticky;
  position: sticky;
  top: 0px;
  height: 100px;
}

#sticky-child {
  background-color: orange;
  color: white;
  height: 100%;
}

@container my-menu scroll-state(stuck: top) {
  #sticky-child { width: 50%; }
}
</style>
<div id="sticky">
  <div id="sticky-child">
    Sticky
  </div>
</div>

There is a question whether the query should match when the applied offset is still 0, but it is exactly aligned with the edge of where an offset would start to take effect.

Use Cases and Author Requests

Chrome Prototype Demo

Scroll-snapped Elements

Introduce a container-type snapped to allow scroll-snap-aligned elements to be queried for whether they are currently snapped in a given direction. For instance:

@container scroll-state(snapped: block) {
  #snap-child {
    outline: 5px solid yellow;
  }
}

Use Cases and Author Requests

In general being able to highlight the scroll snapped element, for instance in a carousel.

Overflowing

Query whether a container has scrollable overflow. Can be used to indicate there is content to scroll to in a given direction.

Needs further exploration.

Anchor position fallback

In earlier exploration of anchor positioning, it was suggested that container queries could be used to style the contents of an anchor positioned element based on which @position-fallback is used.

This needs further exploration.

Proposed Syntax

Add a new scroll-state() function to @container similar to style() where state:value pairs, and logical combinations of them, can be queried. <state-feature> will be similar to <style-feature> in that it takes a state name and value separated by a colon.

Add a new container-type for each state which can be combined with size container types:

Query values for stuck:

Query values for snapped:

A query with only a state name (boolean context) would match if the state is different from none.

Containment Requirements

The proposed approach to handling layout cycles is to have a two-pass rendering update if necessary like scroll-driven animations have. See the Layout Cycles section below for details. There is strictly no need to have containment applied to the new container-types to have a predictable rendering, but it would help authors to avoid flickering issues. The question is whether it makes sense to enforce containment if containment is not enough to fix all flickering issues.

By flickering issues we mean cases where the stuck styling would cause the sticky position to no longer be stuck and the following rendering update would lose that style and cause it to be stuck again. Similarly a scroll-snapped element could be made no longer snapped through style changes affecting layout and scroll position.

Layout Cycles

The scroll based state queries have the same issue with potential layout cycles that scroll-driven animations have. Whether an element has a sticky position offset or is scroll-snapped is not known until after layout, and state queries affect styles inside the containers which may affect whether the sticky element has a sticky position offset or not.

It is possible that a lot of these cycles could be avoided if enough containment is applied, but that would not allow for inline-size containment (require full size containment), and also nesting sticky positioned elements would be a problematic case.

More importantly, there is the problem of applying scroll based state queries on the first frame, which regardless requires finishing layout before the query containers' state can be found which would require another pass regardless of containment requirements.

The scroll-driven animations spec specifies a set of stale timelines which makes it possible to apply the current timeline based on newly created timelines, or timelines that changed because of layout changes, to the same frame that the timeline was created or changed. It is however limited to a single extra update, specified as modifications to the HTML event loop.

It should be possible to do scroll based state queries using the exact same timing for updating state query state and detect if there is a need for a second pass to update the rendering based on the updated state.

Transitions

The two-pass rendering update introduces a side-effect of starting transitions which can be unfortunate, especially for the first rendered frame. This is already an observable effect for scroll-driven animations, which can be seen in the example below.

<!DOCTYPE html>
<style>
  #timeline {
    scroll-timeline: --scroll y;
    width: 400px;
    height: 400px;
    overflow-y: scroll;
  }
  @keyframes w {
    0% {
      width: 300px;
    }
    100% {
      width: 400px;
    }
  }
  #container {
    container-type: inline-size;
    position: fixed;
    animation-timeline: --scroll;
    animation-name: w;
    width: 200px;
    height: 200px;
  }
  #target {
    height: 100%;
    background-color: green;
    transition: background-color 4s;
  }
  #target { background-color: green; }
  @container (width > 200px) {
    #target { background-color: lime; }
  }
</style>
<div id="timeline">
  <div id="container">
    <div id="target"></div>
  </div>
  <div style="height:2000px"></div>
</div>