CSS Containment Module Level 2

1. Introduction

Efficiently rendering a website relies on the User Agent being able to detect what parts of the page are being displayed, which parts might affect the currently-displayed section, and what can be ignored.

There are various heuristics that can be used to guess when a given sub-tree is independent of the rest of the page in some manner, but they’re fragile, so innocuous changes to a page may inadvertently make it fail such heuristic tests, causing rendering to fall into a slow code path. There are also many things that would be good to isolate which are difficult or impossible to detect in a heuristic manner.

To alleviate these problems and allow strong, predictable isolation of a subtree from the rest of the page, this specification defines a contain property.

To allow even further optimization of off-screen contents, this spec also defines a content-visibility property, enabling the user agent to skip an element’s layout and painting entirely when not needed.

1.1. Module Interactions

This document defines new features not present in earlier specifications. In addition, it aims to replace and supersede [CSS-CONTAIN-1] once stable.

1.2. Value Definitions

This specification follows the CSS property definition conventions from [CSS2] using the value definition syntax from [CSS-VALUES-3]. Value types not defined in this specification are defined in CSS Values & Units [CSS-VALUES-3]. Combination with other CSS modules may expand the definitions of these value types.

In addition to the property-specific values listed in their definitions, all properties defined in this specification also accept the CSS-wide keywords keywords as their property value. For readability they have not been repeated explicitly.

2. Strong Containment: the contain property

User Agents are expected to support this property on all media, including non-visual ones.

The contain property allows an author to indicate that an element and its contents are, as much as possible, independent of the rest of the document tree. This allows user agents to utilize much stronger optimizations when rendering a page using contain properly, and allows authors to be confident that their page won’t accidentally fall into a slow code path due to an innocuous change.

none

This value indicates that the property has no effect. The element renders as normal, with no containment effects applied.

strict

This value turns on all forms of containment except style containment for the element. In other words, it behaves the same as contain: size layout paint;.

content

This value turns on all forms of containment except size containment and style containment for the element. In other words, it behaves the same as contain: layout paint;.

Note: contain: content is reasonably "safe" to apply widely; its effects are fairly minor in practice, and most content won’t run afoul of its restrictions. However, because it doesn’t apply size containment, the element can still respond to the size of its contents, which can cause layout-invalidation to percolate further up the tree than desired. Use contain: strict or contain: strict style when possible, to gain as much containment as you can.

size

The value turns on size containment for the element. This ensures that the containing box can be laid out without needing to examine its descendants.

layout

This value turns on layout containment for the element. This ensures that the containing box is totally opaque for layout purposes; nothing outside can affect its internal layout, and vice versa.

style

This value turns on style containment for the element. This ensures that, for properties which can have effects on more than just an element and its descendants, those effects don’t escape the element.

Note: This value is at-risk.

paint

This value turns on paint containment for the element. This ensures that the descendants of the containing box don’t display outside its bounds, so if an element is off-screen or otherwise not visible, its descendants are also guaranteed to be not visible.

This property generally applies to all elements (including CSS Pseudo-Elements 4 §4.1 Generated Content Pseudo-elements: ::before and ::after), although some types of containment have no effect on some elements, as detailed in § 3 Types of Containment. In addition, in the case of [SVG2], the contain property only applies to svg elements that have an associated CSS layout box.

<body>
  <aside>...</aside>
  <section>
    <h2>Messages</h2>
    <article>
      Lol, check out this dog: images.example.com/jsK3jkl
    </article>
    <article>
      I had a ham sandwich today. #goodtimes
    </article>
    <article>
      I have political opinions that you need to hear!
    </article>
    …
  </section>
</body>

3. Types of Containment

There are several varieties of containment that an element can be subject to, restricting the effects that its descendants can have on the rest of the page in various ways. Containment enables much more powerful optimizations by user agents, and helps authors compose their page out of functional units, as it limits how widely a given change can affect a document.

Specification authors introducing new properties or mechanisms need to consider whether and how the various types of containment affect what they are introducing, and include in their specification any effect not described here.

3.1. Size Containment

If the element does not generate a principal box (as is the case with display: contents or display: none), or its inner display type is table, or its principal box is an internal table box, or an internal ruby box, or a non-atomic inline-level box, size containment has no effect. Otherwise, giving an element size containment makes its principal box a containing box for size containment and has the following effects:

Note: Internal table boxes, which do not include table captions, are excluded, because the table layout algorithm does not allow boxes to become smaller than their inflow content. Sizing a table cell as if it was empty and then layout out its content inside without changing the size is effectively an undefined operation. Manually setting the width or height properties to 0 cannot make it smaller than its content. This concern does not apply to table captions, which are perfectly capable of having a fixed size that is independent of their content.

1. When calculating the size of the containing box, including when computing its intrinsic size, it must be treated as having no contents.

   Note: Even when the element’s sizing properties are auto this does not necessarily make the element zero-sized: properties set on the element itself, such as the columns property or the grid property, continue to be taken into account.

   Then, its contents must then be laid out into the containing box's resolved size.

   Note: size containment does not suppress baseline alignment. See layout containment for that.

   Replaced elements must be treated as having an intrinsic width and height of 0.

2. Containing boxes for size containment are monolithic (See CSS Fragmentation 3 §4.1 Possible Break Points).

By itself, size containment does not offer much optimization opportunity. Its primary benefit on its own is that tools which want to lay out the containing box’s contents based on the containing box’s size (such as a JS library implementing the "container query" concept) can do so without fear of "infinite loops", where having a child’s size respond to the size of the containing box causes the containing box’s size to change as well, possibly triggering further changes in how the child sizes itself and possibly thus more changes to the containing box’s size, ad infinitum.

When paired with layout containment, though, possible optimizations that can be enabled include (but are not limited to):

1. When the style or contents of a descendant of the containing box is changed, calculating what part of the DOM tree is "dirtied" and might need to be re-laid out can stop at the containing box.

2. When laying out the page, if the containing box is off-screen or obscured, the layout of its contents can be delayed or done at a lower priority.

3.2. Layout Containment

If the element does not generate a principal box (as is the case with display values of contents or none), or its principal box is an internal table box other than table-cell, or an internal ruby box, or a non-atomic inline-level box, layout containment has no effect. Otherwise, giving an element layout containment makes its principal box a containing box for layout containment and has the following effects:

1. The containing box establishes an independent formatting context.

2. If at least one fragmentation container of a fragmentation context has layout containment, or if at least one fragmentation container of a fragmentation context is a descendant of containing box for layout containment and at least one subsequent fragmentation container of the same fragmentation context is not a descendant of that same element with layout containment, then the first containing box for layout containment which is either a fragmentation container itself or is an ancestor of a fragmentation container must “trap” the remainder of the fragmented flow: fragmentation must not continue past the layout containment boundary, and the last fragmentation container within the first layout containment boundary is treated as if it is the last fragmentation container in its fragmentation context.

   If subsequent fragmentation containers in the fragmentation context are only generated when more content remains in the fragmented flow, then they are not generated. If they would exist regardless, they remain part of the fragmentation context, but do not receive any content from the fragmented flow.

   Note: At the time of writing, no stable specification is affected by this point. Only specifications that would enable some (but not all) fragmentation containers of a fragmentation context to be layout-contained (or descendants of a layout contained element) are concerned. This is not the case of [CSS-PAGE-3] nor of [CSS-MULTICOL-1]. This requirement is nonetheless included because several mechanisms that would make this a possibility have been considered (e.g.: [CSS-REGIONS-1], ::nth-fragment(), a hypothetical selector for individual columns of a multicol…), and the guarantees that layout containment is intended to offer would not be realized if such mechanisms did not abide by this rule. [CSS-REGIONS-1] has details over how layout containment affects regions.

   <article>Lorem ipsum…</article>
   <div id=a></div>
   <aside>
     <div id=b></div>
     <div id=c></div>
   </aside>
   <aside>
     <div id=d></div>
     <div id=e></div>
   </aside>
   <div id=f></div>
   

   article {flow-into: foo;}
   #a, #b, #c, #d, #e, #f {flow-from: foo;}
   aside {contain: layout}
   

   In this [CSS-REGIONS-1] example, content can flow from #a to #b, from #b to #c. However as #c is the last fragment container in the first containing box for layout containment, it traps all the remaining content, and nothing gets flowed into #d, #e, or #f.

3. If the computed value of the overflow property is either visible or clip or a combination thereof, any overflow must be treated as ink overflow.

4. The containing box establishes an absolute positioning containing block and a fixed positioning containing block.

5. The containing box creates a stacking context.

6. Forced breaks are allowed within containing boxes for layout containment, but do not propagate to the parent as otherwise described in CSS Fragmentation 3 §3.1 Breaks Between Boxes: the break-before and break-after properties.

   Note: This introduces the previously non-existent possibility that forced breaks may occur between a box and its container (See CSS Fragmentation 3 §4.1 Possible Break Points).

7. For the purpose of the vertical-align property, or any other property whose effects need to relate the position of the containing box's baseline to something other than its descendants, the containing box is treated as having no baseline.

Possible optimizations that can be enabled by layout containment include (but are not limited to):

1. When laying out the page, the contents of separate containing boxes can be laid out in parallel, as they’re guaranteed not to affect each other.

2. When laying out the page, if the containing box is off-screen or obscured and the layout of the visible parts of the screen do not depend on the size of the containing box (for example, if the containing box is near the end of a block container, and you’re viewing the beginning of the block container), the layout of the containing box' contents can be delayed or done at a lower priority.

   (When paired with size containment, this optimization can be applied more liberally.)

3.3. Style Containment

Note: Style Containment is at-risk.

Giving an element style containment and has the following effects:

1. The counter-increment and counter-set properties must be scoped to the element’s sub-tree and create a new counter.

2. The effects of the content property’s open-quote, close-quote, no-open-quote and no-close-quote must be scoped to the element’s sub-tree.

   Note: This implies that the depth of quote nesting in the subtree is unchanged and starts at the value that its context normally implies, but that changes to the depth of quote nesting by these values inside the subtree do not affect the depth of quote nesting outside the subtree.

Note: [CSS-REGIONS-1] has normative requirements on how style containment affects regions.

A scoped property has its effects scoped to a particular element or subtree.

• If scoped to an element, it must act as if the scoping element was the root of the document for the purpose of evaluating the property’s effects: any uses of the property outside the scoping element must have no effect on the uses of the property on or in the scoping element, and vice versa.

  Note: “Scoping to an element” is currently unused. It is defined as an extension point for future specifications to use.

• If scoped to a sub-tree, it’s the same, except the scoping element itself is counted as "outside" the tree, like the rest of the document, and the effects of the property on that element are unaffected by scoping. When considering the effects of the scoped property on elements inside the subtree, the element at the base of the subtree is treated as if it was the root of the document.

<div></div>

div {
  contain: style;
  counter-increment: n;
}
div::before, div::after {
  content: counters(n, '.') " ";
}
div::after {
  counter-increment: n 2;
}

Possible optimizations that can be enabled by style containment include (but are not limited to):

1. Whenever a property is changed on a descendant of an element with style containment, calculating what part of the DOM tree is "dirtied" and might need to have its style recalculated can stop at the containing box.

3.4. Paint Containment

If the element does not generate a principal box (as is the case with display values of contents or none), or its principal box is an internal table box other than table-cell, or an internal ruby box, or a non-atomic inline-level box, paint containment has no effect. Otherwise, giving an element paint containment makes its principal box a containing box for paint containment and has the following effects:

1. The contents of the element including any ink or scrollable overflow must be clipped to the overflow clip edge of the containing box, taking corner clipping into account. This does not include the creation of any mechanism to access or indicate the presence of the clipped content; nor does it inhibit the creation of any such mechanism through other properties, such as overflow, resize, or text-overflow.

   Note: This clipping shape respects overflow-clip-margin, allowing an element with paint containment to still slightly overflow its normal bounds.

   Note: The behavior is described in this paragraph is equivalent to changing overflow-x: visible into overflow-x: clip and overflow-y: visible into overflow-y: clip at used value time, while leaving other values of overflow-x and overflow-y unchanged.

2. The containing box establishes an absolute positioning containing block and a fixed positioning containing block.

3. The containing box creates a stacking context.

4. The containing box establishes an independent formatting context.

4. Suppressing An Element’s Contents Entirely: the content-visibility property

The content-visibility property controls whether or not an element renders its contents at all, along with forcing a strong set of containments, allowing user agents to potentially omit large swathes of layout and rendering work until it becomes needed. It has the following values:

visible

No effect. The element’s contents are laid out and rendered as normal.

hidden

The element skips its contents.

The skipped contents must not be accessible to user-agent features, such as find-in-page, tab-order navigation, etc., nor be selectable or focusable.

Note: This is similar to giving the contents display: none.

auto

The element turns on layout containment, style containment, and paint containment.

If the element is not relevant to the user, it also skips its contents.

Unlike hidden, the skipped contents must still be available as normal to user-agent features such as find-in-page, tab order navigation, etc., and must be focusable and selectable as normal.

4.1. Using content-visibility: hidden

This section is non-normative.

content-visibility: hidden lays powerful restrictions onto an element, and so should be used with caution. It also enables some very useful scenarios, often improving on existing techniques, a few of which are outlined here.

1. If a page needs to take some measurements of elements or text which aren’t themselves going to be rendered, commonly this is done by positioning the stuff-to-be-measured off-screen, using something like position: absolute; left: -100000px;, then calling an API like getBoundingClientRect().

   Unfortunately, even tho the page never intends to display this content, the user agent will still have to do full styling, layout, and rendering for the content, just in case it affects what’s shown on screen. The author also can’t, without further work, guarantee that the content won’t accidentally show up on-screen; even a very negative left value (like above) might not be enough, depending on the content.

   Wrapping this content in a content-visibility: hidden container solves all of these problems. If the wrapper has no border, background, etc, then it and its skipped contents are guaranteed to never render anything to the screen, no matter how big they get. Because the contents are skipped, the user agent can also avoid styling or laying them out until absolutely necessary, when script finally asks for it.

2. A "single-page app" often consists of several independent panes or "views", of which only one is displayed at a time.

   If the author wants to avoid paying styling/layout/rendering/etc cost for the inactive views, they can remove them from the document entirely, or at minimum apply display:none to them. Unfortunately, this means that when the view does need to be displayed, all of the styling/layout/rendering/etc work needs to be done all at once, potentially causing a noticeable delay before the view actually shows up.

   Alternately, the view can just be positioned off-screen. This means it’ll be immediately ready when it’s time to be used, but it incurs the cost of styling/layout/rendering all the time, which might be significant, especially if there are a number of inactive views. The inactive views also might still show up to accessibility tooling, confusing users of screen-readers, people using Ctrl-F to find-in-page, etc.

   content-visibility: hidden improves on both of these options. Because the contents are skipped, the user agent isn’t spending time on them when they’re not active. They’re also not visible to screen readers, find-in-page, and other tools. And because user agents should preserve previous styling/layout work if possible, if the view was displayed before, re-rendering it might be very fast.

3. If an author wants to make an element "invisible", but still show up in the page for layout purposes, one option is visibility: hidden. However, descendants of a visibility: hidden element can set visibility: visible and start showing up again, which isn’t always intuitive or expected.

   content-visibility: hidden performs a very similar purpose, but descendants can’t turn it "off" and start displaying; they stay "hidden" until the ancestor turns it off.

   Because content-visibility: hidden also applies many containment values to the container, it’s not always quite as usable as visibility: hidden would be, but when its restrictions are acceptable, it can be a more reliable, more consistent way to hide an element’s contents.

4.2. Using content-visibility: auto

This section is non-normative.

content-visibility: auto is a more complex value than hidden; rather than being similar to display: none, it adaptively hides/displays an element’s contents as they become relevant to the user. It also doesn’t hide its skipped contents from the user agent, so screen readers, find-in-page, and other tools can still interact with it.

It is best to think of it as an upgrade to containment: if an author has a large amount of content to display that will often be off-screen (such as a long scrollable list), and that content is okay with heavy containment, they should consider using content-visibility: auto to apply all of the containments at once. This also strongly hints to the user agent that it’s acceptable to skip work on the contents (possibly causing a small delay when they do come on-screen) because it’s more important to have a large amount of content in the document and most of it won’t be seen anyway.

Note: content-visibility: auto can thus be used instead of complicated "virtual list" techniques, at least in many cases.

Because content-visibility: auto only causes the element to skip its contents when none of it is relevant to the user, it’s best to use at a reasonably fine granularity.

Because content-visibility: auto imposes size containment when the element skips its contents, if the element depends on its contents to determine its size the layout of the page (or at least, the scrollbar position) can "jump around" as elements go off-screen and start skipping.

4.3. Restrictions and Clarifications

1. From the perspective of an IntersectionObserver, the skipped contents of an element are never intersecting the intersection root. This is true even if both the root and the target elements are in the skipped contents.

2. From the perpsective of a ResizeObserver, the skipped contents of an element never change their size. If these elements become non-skipped later, the resize observation will be delivered if the new size differs from the last size used to notify the resize observer.

3. If an element starts or stops skipping its contents, this change happens after the requestAnimationFrame callbacks of the frame that renders the effects of the change have run. Specifically, such changes will take effect right after step 11 of Update the Rendering step of the Processing Model.

   Determining the viewport intersection of the element can be done with an internal version of an IntersectionObserver. However, since the observations from this are dispatched at step 12 of Update the Rendering, any changes to the skipped (and thus painted) state will not be visible to the user until the next frame’s processing. For this reason, updating the skipped state, including containment adjustments, is deferred to that frame as well. This ensures that script accessing, for example, the containment value of the element between these two events (internal intersection observation and skipped state update) will retrieve values consistent with current painted state and not cause any forced layouts.

4. For the purposes of scrolling operations, such as scrollIntoView(), an element with content-visibility: auto that is skipping its contents has its size and location determined with size containment still active.

   Note: Once it’s scrolled into view, the element will no longer skip its contents, and so might not have size containment; if this changes the element’s size, it might not align in the viewport exactly as requested.

5. If an element with content-visibility: auto that is skipping its contents is focused (or its contents are), it becomes relevant to the user (and thus stops skipping its contents) before it is scrolled into view due to the focusing.

   Note: Thus, unlike the previous point, the element will be correctly sized and aligned in the viewport. This is consistent with the order of the steps for the focus() method.

6. If an iframe skips its contents or is part of an element’s skipped contents, the user agent should entirely skip the Update The Rendering step in the iframe’s event loop, if possible.

   Note: At the moment the iframe starts being skipped, it needs to run that step at least once to remove the painted output.

7. Skipped contents do not contribute to the result of innerText.

4.4. Accessibility Implications

If a user agent exposes some form of "accessibility tree", akin to the DOM tree but specialized for accessibility use-cases such as screen-readers (thus providing the positions/etc of elements relevant to accessibility APIs, such as focusable elements), then the skipped contents of content-visibility: hidden elements must similarly be "skipped" (omitted) in the accessibility tree (similar to how display: none elements are omitted in all views of the document).

Skipped contents of content-visibility: auto elements must not expose the fact that a user is interacting with the page via the accessibility tree, rather than via rendering visually to the screen. In particular, if a user agent uses content-visibility: auto to avoid doing layout and painting work on off-screen content for displaying to the screen, it must similarly avoid doing that work on off-screen content for representing in the accessibility tree. If this is not possible (for example, if the user agent’s representation of a focusable element in the accessibility tree requires knowledge of its exact position, and thus requires a full layout to be done on it and surrounding contents), then the user agent must omit the skipped contents from the accessibility tree entirely.

Note: This requirement is intended to protect users utilizing accessibility tooling from being identified and profiled as such via observation of timing channels; if a user agent can skip significant amounts of work when rendering visually, but has to do all of the work when rendering to an accessibility tree, then an author can tell how a user is interacting with the page by observing the timing of layout operations.

4.5. Examples

<style>
.sv {
  content-visibility: auto;
  min-height: 50px;
}
</style>

<div class=sv>
  ... some content goes here ...
</div>

<style>
.sv {
  content-visibility: hidden;
}
</style>

<div class=sv>
  ... some content goes here ...
</div>

<style>
body {
  margin: 0;
}
.sv {
  content-visibility: hidden;
  position: relative;
  left: 10px;
  top: 20px;
}
#child {
  position: relative;
  left: 1px;
  top: 2px;
  width: 100px;
  height: 200px;
}
</style>

<div id=target class=sv>
  <div id=child></div>
  ... some other content goes here ...
</div>
<script>
  ...
  // This will force rendering work, including layout,
  // if the UA previously avoided it.
  target.firstElementChild.getBoundingClientRect();
  ...
</script>

5. Privacy and Security Considerations

This specification introduces no new privacy or security considerations.

Like any other CSS specification, it affects the rendering of the document, but does not introduce any special ability to present content in a misleading way that was not previously available through other CSS modules and that isn’t inherent to the act of formatting the document.

The TAG has developed a self-review questionnaire to help editors and Working Groups evaluate the risks introduced by their specifications. Answers are provided below.

Does this specification deal with personally-identifiable information?

No.

Does this specification deal with high-value data?

No.

Does this specification introduce new state for an origin that persists across browsing sessions?

No.

Does this specification expose persistent, cross-origin state to the web?

No.

Does this specification expose any other data to an origin that it doesn’t currently have access to?

No.

Does this specification enable new script execution/loading mechanisms?

No.

Does this specification allow an origin access to a user’s location?

No.

Does this specification allow an origin access to sensors on a user’s device?

No.

Does this specification allow an origin access to aspects of a user’s local computing environment?

No.

Does this specification allow an origin access to other devices?

No.

Does this specification allow an origin some measure of control over a user agent’s native UI?

No.

Does this specification expose temporary identifiers to the web?

No.

Does this specification distinguish between behavior in first-party and third-party contexts?

No.

How should this specification work in the context of a user agent’s "incognito" mode?

No difference in behavior is needed.

Does this specification persist data to a user’s local device?

No.

Does this specification have a "Security Considerations" and "Privacy Considerations" section?

Yes, this is the section you are currently reading.

Does this specification allow downgrading default security characteristics?

No.

Appendix A. Changes

This appendix is informative.

Changes from 2019-11-11 Working Draft

• Define the interaction between paint containment and overflow-clip-margin

• Add the content-visibility property

• Editorial improvements and clarifications to size and paint containment (in sync with the same improvements made to Level 1)

Changes from CSS Containment Level 1

• Restored style containment, which had been dropped from Level 1