The WHATWG Blog

Focusing on focus

Wednesday, October 16th, 2019

Focus behavior in HTML had been under-specified for the past few years, and it was also quite confusing due to a variety of subtle differences between focusing methods, UA-specific behaviors, relation to the tabindex attribute, relations to shadow DOM, etc.

A few months ago Domenic filed a meta-bug that contains a list of things to fix so that we would have a good foundation for further additions to focus-related stuff, and to hopefully make focus in HTML make sense to browser engineers and web authors!

Types of focus

You may know that you can focus on stuff by clicking on them, tabbing, or calling focus() on it — but did you know that things might be focusable with one method but not with others? Domenic and Mu-An made a giant interactive list of HTML elements to showcase how they react to different methods of focusing, and can be tested from various browsers to show the difference between them.

To reflect the varying behaviors of focus in various UAs in the spec, we classified focusability into three types: “programmatically focusable”, “click focusable”, and “sequentially focusable”.

If an element is programmatically focusable, it will get focused when calling the focus() method on it or when putting an autofocus attribute on the element. In all platforms, all elements that are either click focusable or sequentially focusable are also programmatically focusable, so “programmatically focusable” is interchangeable with “focusable”.

If an element is click focusable, the element will get focused when it’s clicked. This has the same set of elements as “programmatically focusable” in most UAs/platforms. A notable exception is Safari where non-editable form controls (checkboxes, etc.) are not click focusable by default.

If an element is sequentially focusable, the element can be focused through “tabbing” — in most UAs this means pressing Tab/Shift+Tab (and in Safari, Option+Tab too!).

Previously the spec didn’t clearly differentiate “programmatically focusable” and “sequentially focusable”, didn’t even mention “click focusable”, and used the “tabindex focus flag” concept which was slightly confusing due to its relations with the tabindex attribute. So we updated it in this PR.

tabindex

As you may know already, built-in elements like <button>, <input>, <a>, etc. all have a “default” focus behavior — they are focusable (or not) by default. When the tabIndex property getter is run on an element whose tabindex attribute is not explicitly set already, it will return 0 sometimes and -1 other times. Previously, the spec said to return 0 if the element is “focusable” by default (which type of focusable?), and -1 otherwise. But this wasn’t implemented anywhere, because of possible differences in which elements are focusable by default in various UAs. So now the spec actually checks if the tag name is one of the tag names included in a pre-defined list. This is quite awkward, but at least it’s interoperable! (PR is here.)

Now, what is the use of tabindex exactly? You can make an element focusable by setting its tabindex attribute to an integer. This will set the tabindex value, and thus impact the focusability of the element. If the integer is non-negative, the element is also sequentially focusable. You can’t, however, make an element not focusable at all through this attribute — there is no value you can set the tabindex attribute to on a <button> that will stop it from being focusable.

You can also modify the order of elements traversed with sequential navigation/tabbing — elements with a positive tabindex value will be traversed first, in ascending order (and in tree order in case of a tie), and then elements with a tabindex value of zero (or unspecified but the element is sequentially focusable by default), in tree order.

autofocus

The autofocus content attribute is useful if you want to set focus on a form control element on page load. However, it had some issues such as:

• The autofocus attribute was available only for some form control elements, and unavailable for other focusable elements like <summary>, <div contenteditable="true">, and <span tabindex="0">.
• The behavior with an autofocused element in a page accessed with a fragment identifier (e.g. https://example.com/#foo) was not interoperable.
• What should happen if there are multiple autofocused elements, and one of them is not focusable?

The HTML specification and the SVG specification were updated to resolve these issues by changes 1, 2, and 3. Now the autofocus content attribute and the autofocus IDL attribute are available on all HTML and SVG elements. The autofocus content attribute is not processed if its document has a fragment identifier, or has a focused element. If an autofocus element is not focusable, the element is skipped and another autofocus element is handled.

Shadow DOM and delegatesFocus

All of the previously mentioned concepts were already specced in the HTML spec in some way, albeit a bit unclear, not reflecting the actual behavior, etc. Focus behavior with shadow DOM, though, had not been upstreamed from the old Shadow DOM spec at all, so this part of the effort took the most time overall. Since some parts of the Shadow DOM spec on focus were unclear and needed more explanation, we also took a look at how it’s implemented in Blink to get the exact behavior down, and discussed with other browser vendors and web developers on whether we want to keep the implemented behavior or not in some cases.

Sequential focus navigation got some significant additions (PR) with Shadow DOM. We now have a concept of “tabindex-ordered focus navigation scope”, “focus navigation scope owners”, etc. Essentially, elements are put in different focus navigation scopes, where those that belong to the same shadow tree/root, or are slotted to the same slot, are put in the same focus navigation scope. The tabindex order explained earlier now only applies to elements in the same scope, and then finally we flatten all of the scopes to get the final sequential focus navigation order.

A new concept added (PR 1, 2) by Shadow DOM is “delegates focus”, which is used when you want attempts to focus on a shadow host to not focus on the host itself, and instead delegate the focus to within the shadow tree (like <input type="date">!). In Blink, this delegation uses the sequential focus navigation order, but we think it is a bit weird and started a discussion on how this should actually work — finally changing the delegation to respect the protocol of whatever the focusing method is originally used. (That is, use sequential order if we used tabbing, otherwise use flat-tree order and respect click focusability if needed.)

We also added the DocumentOrShadowRoot’s activeElement property (PR). And we updated the :focus selector (PR), which will now match on shadow hosts if the focused element is a shadow-including inclusive descendant of it. (This is actually different than the behavior in Blink, and is a result of discussion in TPAC — where we also discussed “delegates focus”.)

The future, and outro

Now that we have a good-enough spec for focus, we have a good foundation for future additions to focus. One new relevant proposal is “custom element default focusability”. As we’ve mentioned, built-in elements have a “default” focus behavior — even though they don’t have a tabindex value explicitly, they are still focusable (or not). When you’re making custom elements, though, currently there is no way to make them focusable by default, without setting the tabindex attribute. The proposal listed various ways this might be solved, and it was talked about in TPAC with a relatively positive response from various parties. Do check it out if you’re interested!

In summary, focus was and is still quite complex to understand. But, at least now there’s a clear source of truth for it, and the browser vendors are working to make it interoperable — implementing new changes to the spec as soon as they can. There are still lots of focus-related things that need to be specced (we’ve heard people mention focusin/focusout, more CSS selectors, etc.), so if you’re intrigued by this post, know that you can also contribute to fix more things like these!

Having an old and mostly confusing focus spec, different types of focus, and multiple uses for tabindex made things quite complicated when starting out. However, one of the trickiest parts of focus, in my opinion, is the fact that what is focusable/click focusable/sequentially focusable might differ in different UAs, and it might be dynamic as well! (E.g. in Safari what’s sequentially focusable changes if you hold down the Option key!) This means we need to make sure the spec is written to allow for differences, but is still, um, specific enough to make things not too ambiguous.

Overall, we’re happy with the result of this effort. We’d like to thank all the parties involved that participated in various ways: experimenting, reviewing spec PRs, implementing the changes, commenting/participating in discussions, etc. Special thanks to Kent Tamura (who wrote some parts of this post and did the autofocus specs) and Domenic Denicola (who kickstarted this whole effort, reviewed all the PRs, and suggested + reviewed this post).

Now we can focus on other parts of HTML Standard...

Posted in Browser API, DOM, Forms, Processing Model | 2 Comments »

Validator.nu HTML Parser 1.2.1

Monday, May 25th, 2009

Version 1.2.1 of the Validator.nu HTML Parser is now available. It fixes an incompatibility with the DOM implementation of the latest Xerces.

Posted in DOM, Processing Model, Syntax | Comments Off on Validator.nu HTML Parser 1.2.1

Supporting New Elements in Firefox 2

Wednesday, March 18th, 2009

We have previously talked about how to get Internet Explorer to play ball when using the new HTML5 elements, but today I'm going to talk about Firefox 2.

Firefox 2 (or any other Gecko-based browser with a Gecko version pre 1.9b5) has a parsing bug where it will close an unknown element when it sees the start tag of a "block" element like p, h1, div, and so forth. So if you have:

<body>
 <header>
  <h1>Test</h1>
 </header>
 <article>
  <p>...</p>
  ...
 </article>
 <nav>
  <ul>...</ul>
 </nav>
 <footer>
  <p>...</p>
 </footer>
</body>

...then in Firefox 2 it will be parsed as if it were:

<body>
 <header>
  </header><h1>Test</h1>
 
 <article>
  </article><p>...</p>
  ...
 
 <nav>
  </nav><ul>...</ul>
 
 <footer>
  </footer><p>...</p>
 
</body>

So if you style the new elements with CSS it will probably look completely broken in Firefox 2.

If you care about Firefox 2 then there are some ways to fix this:

1. Go back to using div elements
2. Use content type negotiation between text/html and application/xhtml+xml
3. Fix up the DOM with scripting

(1) is probably wise if your content structure changes between pages or over time. (2) also works but means that users will be exposed to the Yellow Screen of Death should a markup error slip through your system. Otherwise (3) can be worth to consider.

Fixing up Firefox 2's DOM is actually pretty simple if you have a consistent structure. Using the same markup as above it could look something like this:

<body>
 <header>
  <h1>Test</h1>
 </header>
 <article>
  <p>...</p>
  ...
 </article>
 <nav>
  <ul>...</ul>
 </nav>
 <footer>
  <p>...</p>
 </footer>
 <!--[if !IE]>--><script>
  // dom fixup for gecko pre 1.9b5
  var n = document.getElementsByTagName('header')[0];
  if (n.childNodes.length <= 1) { // the element was closed early
    var tags = ['ARTICLE', 'NAV', 'FOOTER', 'SCRIPT'];
    for (var i = 0; i < tags.length; ++i) {
      while (n.nextSibling && n.nextSibling.nodeName != tags[i]) {
        n.appendChild(n.nextSibling);
      }
      n = n.nextSibling;
    }
  }
 </script><!--<![endif]-->
</body>

You might think that this script would work for IE, too, when not using the createElement hack, but apparently IE throws an exception when trying to append a child to an unknown element. So you still have to use the createElement hack for IE.

If you want to move the script to head and run it on load and you don't have anything after the footer then you would replace 'SCRIPT' in the tags array with undefined to make it work.

(If you want to do content type negotiation and want to just serve XHTML to Gecko-based browsers with this bug then you should look for the substrings "Gecko/" and "rv:1.x" where x is less than 9, or "rv:1.9pre" or "rv:1.9a" or "rv:1.9bx" where x is less than 5.)

Posted in Browsers, DOM, Elements | 17 Comments »

Styling HTML5 markup in IE without script

Saturday, January 17th, 2009

The previous post discussed how to enable styling of the new HTML5 elements in IE by using a simple script. However, if the user has scripting disabled, the layout would probably fall apart badly.

So that means if you care about IE users with scripting disabled, you can't use the new elements, right?

Not necessarily.

There are some tricks to work around the broken DOM and limited styling in IE:

1. Know what the DOM looks like and target other elements than the new elements for styling.
2. Use the universal selector (*) to target the right element.
3. Use noscript.

What does this mean?

Target other elements for styling

Consider you have the following markup:

<body>
 <article>
  ...
 </article>
 <nav>
  <ul>
   ...
  </ul>
 </nav>
</body>

Instead of doing this:

* { margin:0; padding:0 }
body { background:silver }
article { border:solid; background:white; margin-left:10em }
nav { position:absolute; top:0; left:0; width:10em }

...do this:

* { margin:0; padding:0 }
html { background:silver }
body { border:solid; background:white; margin-left:10em }
ul { position:absolute; top:0; left:0; width:10em }

Now of course you're going to use other ul elements than the navigation, so how do we get more specific on which element to target? The obvious solution is to set a class or id on the ul element, but there's another solution which might be more convenient in some cases, which brings me to...

Using the universal selector

Depending on the situation, and whether you care about IE6 or not, you can use the universal selector to target the element you want.

Consider you have the following markup:

<body>
 <article>
  <header>
   <h1>...</h1>
   <p>...</p>
  </header>
  ...

...and you want to style the p that is in header, you would do this in the normal case:

article header p { font-weight:bold }

But in IE, the article, header, h1 and p elements are all siblings, so the selector wouldn't match.

So then one would expect this to match, but it doesn't (IE doesn't allow selecting unknown elements using type selectors):

article + header + h1 + p { font-weight:bold }

However, this matches:

body > * + * + h1 + p { font-weight:bold }

Using noscript

The above techniques shouldn't mess up other browsers (or IE when scripting is enabled), however if you prefer (or if something would screw up) you can use a separate style sheet for IE when scripting is disabled by just using the following markup:

<head>
 <!--[if IE]>
  <noscript><link rel="stylesheet" href="ie-noscript.css"></noscript>
 <![endif]-->
 ...

Conclusion

The above techniques might not be very scalable or might well impact maintanence, but the point of this article is to show that it is possible to use the new elements while still supporting IE with scripting disabled.

Posted in Browsers, DOM, Elements | 11 Comments »

Supporting New Elements in IE

Wednesday, January 7th, 2009

Internet Explorer poses a small challenge when it comes to making use of the new elements introduced in HTML5. Among others, these include elements like section, article, header and footer. The problem is that due to the way parsing works in IE, these elements are not recognised properly and result in an anomalous DOM representation.

To illustrate, consider this simple document fragment:

<body>
  <section>
    <p>This is an example</p>
  </section>
</body>

Strangely, IE 6, 7 and 8 all fail to parse the section element properly and the resulting DOM looks like this.

• BODY
  • SECTION
  • P
    • #text: This is an example
  • /SECTION

Notice how IE actually creates 2 empty elements. One named SECTION and the other named /SECTION. Yes, it really is parsing the end tag as a start tag for an unknown empty element.

There is a handy workaround available to address this problem, which was first revealed in a comment by Sjoerd Visscher. The basic concept is that by using document.createElement(tagName) to create each of the unknown elements, the parser in IE then recognises those elements and parses them in a more reasonable and useful way. e.g. By using the following script:

document.createElement("section");

The resulting DOM for the fragment given above looks like this:

• BODY
  • section
    • P
      • #text: This is an example

This same technique works for all unknown elements in IE 6, 7 and 8. Note that there is a known bug that prevented this from working in IE 8 beta 2, but this has since been resolved in the latest non-public technical preview.

For convenience, Remy Sharp has written and published a simple script that provides this enhancement for all new elements in the current draft of HTML5, which you can download and use.

This script is not needed for other browsers. Opera 9, Firefox 3 and Safari 3 all parse unknown elements in a more reasonable way by default. Note, however, that Firefox 2 does suffer from some related problems, for which there is unfortunately no known solution; but it is hoped that given the faster upgrade cycle for users of Firefox, relatively speaking compared with IE, Firefox 2 won't pose too much of a problem in the future.

Posted in Browsers, DOM, Elements, Events, Syntax | 25 Comments »