Invoker Commands (Explainer)
- Authors
- @keithamus, @lukewarlow
- Created
- Last Updated
- WHATWG Issue
- https://github.com/whatwg/html/issues/9625
- WHATWG PR
- https://github.com/whatwg/html/pull/9841
- Specification
- https://html.spec.whatwg.org/multipage/form-elements.html#attr-button-command
- MDN
- https://developer.mozilla.org/en-US/docs/Web/API/Invoker_Commands_API
NOTE: This explainer covers the graduated parts of this proposal. While we try to keep it roughly in line with the actual shipped features, it might be more informative to look the specification and MDN documentation links.
See also the future proposed commands.
Invoker Buttons
Pitch in Code
<button command="show-modal" commandfor="my-dialog">This opens a dialog</button>
<dialog id="my-dialog">This is the dialog</dialog>
Introduction
Adding commandfor and command attributes to <button> and elements would
allow authors to assign behaviour to buttons in a more accessible and
declarative way, while reducing bugs and simplifying the amount of JavaScript
pages are required to ship for interactivity. Buttons with command will -
when clicked, touched, or enacted via keypress - dispatch a CommandEvent
on the element referenced by commandfor, with some default behaviours.
Background
All elements within the DOM are capable of having interactions added to them. A
long while ago this took the form of adding inline JavaScript to an event
attribute, such as <button onclick="other.open()"></button>. Inline JavaScript
has (rightly so) fallen out of favour due to the security and maintainability
concerns. Newer pages may instead introduce more JavaScript to imperatively
discover elements and call addEventListener('click', ...) to invoke the same
behaviour. These patterns reduce developer experience and introduce more
boilerplate and friction, while remediating security and maintainability
concerns. Some frameworks attempt to reintroduce the developer experience of
inline handlers by introducing new JavaScript or HTML shorthands, such as
React’s onClick={...}, Vue’s @click=".." or HTMX’s hx-trigger="click".
Handling clicking is only half of the problem though, and this proposal doesn’t seek to re-invent click handlers. Dispatching an action to a related element on the click can also become chore-like depending on the framework of choice. Many frameworks recommend associating a click event with a state change and having the element to be invoked react to that state change. For example the React docs on Adding Interactivity attach click handlers which set state.
There has also been a rise in the desire to customise controls for components. Many sites, for example, introduce custom controls for File Uploads or dropdown menus. These often require a large amount of work to reintroduce the built in functionality of those controls, and often unintentionally sacrifice accessibility in doing so.
With the new popover attribute, we saw a straightforward declarative way to
tell the DOM that a button was interested in being a participant of the popover
interaction. popovertarget would indicate to a browser that if that button was
interacted with, it should open the element the popovertarget value pointed
to. This allows for popovers to be created and interacted with - in an
accessible and reliable way - without writing any additional JavaScript, which
is a very welcome addition. While popovertarget sufficiently captured the use
case for popovers, it fell short of providing the same developer & user
experience for other interactive elements such as <dialog>.
This proposal attempts to redress the balance.
Terms
- Invoke/Invoked/Invoking: The action of Invoking refers to a complete press
action of a button, using a Human Input Device (HID). If the HID is a mouse,
this would be a
clickevent. If the HID is a touch screen, this would be a press using a stylus or finger. If the HID is a keyboard this would be theSpaceorEnter(Carriage Return) key on the keyboard. For other HIDs such as eye tracking or pedals or game controllers, the equivalent expected “click” action should be used to invoke the element. - Invoker: An invoker is a button control (that is a
<button type="button">) that has acommandforandcommandattribute set. - Invoker Target: An element which is referenced to by an Invoker, via the
commandforattribute.
Proposed Plan
In the style of popovertarget, this document proposes we add
commandForElement, and command as available attributes to <button>.
interface mixin CommandElement {
[CEReactions, Reflect="commandfor"] attribute Element? commandForElement;
[CEReactions, ReflectSetter] attribute DOMString command;
};
HTMLButtonElement extends InvokerElement
The commandfor value should be an IDREF pointing to an element within the
document. .commandForElement also exists on the element to imperatively
assign a node to be the invoker target, allowing for cross-root invokers (in
some cases, see
the popovertarget attr-asociated element steps
for more).
The command attribute (and the .command reflected property) is a freeform
hint to the invoker target. Command can be a “built-in” action or a “custom”
action. If command is not supplied then it is considered invalid and no
action will take place, and no event will be fired.
Custom command values must start with a double dash (--), similar to the CSS
“dashed-ident”
syntax. This is a safe namespace to use, and browsers guarantee to never have
built-in values starting with --. The only browser behaviour for those values
will be dispatching a CommandEvent on the invokee. This allows web developers
to create custom Invoke actions for their components.
Built-in interactive elements have built-in behaviours (detailed below) which
are determined by the command attribue/property. The built-in names must not
start with a -.
Valid commands (that is: custom commands or a valid built-in) will
dispatch CommandEvents, allowing custom code to take control of invocations (for
example calling .preventDefault() or executing custom side-effects).
Invokers will dispatch an CommandEvent on the Invokee (the element referenced
by commandfor) when the element is Invoked. The CommandEvent’s type is
always command. The event also contains an source property that will reference
button that was enacted causing the CommandEvent to fire. The dispatched
CommandEvents are always non-bubbling, composed, trusted, cancellable events.
[Exposed=Window]
interface CommandEvent : Event {
constructor(CommandEventInit invokeEventInit);
readonly attribute Element source;
readonly attribute DOMString type = "command";
readonly attribute DOMString command;
};
dictionary CommandEventInit : EventInit {
DOMString command = "";
Element source;
};
If an element also has both a popovertarget and the commandfor/command
attributes set, then popovertarget must be ignored: command takes
precedence.
If a <button> is a form participant, or has type=submit, then command
must be ignored. It also must not submit the form it is attached to.
Buttons like this are “author errors”, and browsers may issue warnings in the
developer console explaining to the developer why this button doesn’t work.
This means a <button> with command/commandfor that is inside a form
must also have type=button.
Example Code
Popovers
When pointing to a popover, commandfor acts much like popovertarget,
allowing the toggling of popovers.
<button type="button" commandfor="my-popover" command="toggle-popover">Open Popover</button>
<!-- Effectively the same as popovertarget="my-popover" -->
<div id="my-popover" popover="auto">Hello world</div>
Dialogs
When pointing to a <dialog>, command can toggle a <dialog>’s
openness.
<button type="button" commandfor="my-dialog" command="show-modal">Open Dialog</button>
<dialog id="my-dialog">
Hello world!
<button type="button" commandfor="my-dialog" command="close">Close</button>
</dialog>
Custom behaviour
Invokers will dispatch events on the Invokee element. Using a dash in the command allows for custom JavaScript to be triggered without having to wire up manual event handlers to the Invokers.
<!-- Custom commands must contain a double dash! -->
<button type="button" commandfor="my-custom" command="--my-frobulate">Frobulate</button>
<div id="my-custom"></div>
<script>
document.getElementById("my-custom").addEventListener("command", (e) => {
if (e.command === "--my-frobulate") {
alert("Successfully frobulated the div");
}
});
</script>
Defaults
Depending on the target set by commandfor, invoking the button will trigger
additional behaviours alongside the event dispatch, depending on the value of
command. The following table represents how built-in invocations on specific
element types are handled.
When the command attribute is missing it will default to an invalid state.
| Invokee Element | command hint | Behaviour |
|---|---|---|
<* popover> | 'toggle-popover' | Shows the popover if closed, otherwise hides. Similar to .togglePopover() |
<* popover> | 'hide-popover' | Hides the popover if open, otherwise does nothing. Similar to .hidePopover() |
<* popover> | 'show-popover' | Shows the popover if closed, otherwise does nothing. Similar to .showPopover() |
<dialog> | 'show-modal' | If the <dialog> is not open, shows it as modal. Similar to .showModal() |
<dialog> | 'close' | If the <dialog> is open, close and use the button value for returnValue. Similar to `.close(value) |
<dialog> | 'request-close' | If the <dialog> is open, request to close and use the button value for returnValue. Similar to `.requestClose(value) |
Accessibility
For built-in behaviours command attribute maps to specific accessibility
mappings which are placed on the button. The button may also use the
commandfor referenced element to gather other details (for example the state
of the element) to reflect that state on the button.
These mappings will happen implicitly on the browsers Accessible Nodes, and so
while (for simplicity) this section refers to various aria- attributes,
buttons will not sprout these attributes in the DOM, but the effective
equivalent will be exposed to Assistive Technologies.
Buttons with command=toggle-popover, command=hide-popover, or command=show-popover
Buttons with these popover commands will implicitly receive
aria-details=IDREF, where IDREF matches that of the commandfor attribute,
under certain conditions while the popover is in the showing state.
Why?
Some Assistive Technologes allow users to press a keyboard shortcut to navigate
to the element pointed to by aria-details. This is useful in scenarios where
the popover is not the invoking element’s next accessibility sibling, such as if there are buttons
or other controls between the invoking button and the popover.
Browsers are expected to omit aria-details based on heuristics where it is redundant,
such as the contents of the popover being the next accessibility sibling to the invoking element, or while
the popover’s contents are not traversable, such as when it is hidden, or not
in the DOM.
Browsers should also omit aria-details when the invoking button is a descendant
of the popover it refers to, for example a close button inside the popover.
Buttons will also implicitly receive an aria-expanded value. The state of the
aria-expanded value will map to the state of the popover. When the popover is
open the button will have aria-expanded="true", when closed,
aria-expanded="false". If the button is an descendant of the popover then it
will be explictly set to aria-expanded="undefined", indicating to assistive
technology that the expanded state is not applicable to this button.
This attribute will be recomputed whenever the popover changes state, such that
the button always reflects the state of expansion.
Why?
Some Assistive Technologies will announce the state of expansion while the
button is focussed, for example a <button aria-expanded=false>File</button>
may be announced by a screen reader as “File, button, collapsed”, while
<button aria-expanded=true>File</button> may be announced by a screen reader
as File, button, expanded. This is helpful information for users who require
an alternative of the visual confirmation of the popover being rendered on
screen.
If the popover element is not in the DOM, browsers should ensure the button has
the aria-expanded=false state. Some websites may chose to render such
elements only when they are open, often referred to as “conditional rendering”.
Browsers should aim to do their best to cater for this, and as such provide the
state under the assumpsion that a popover element may appear in the DOM on
press.
Buttons with these popover commands will also have focus restored to the button when the popover closes. This will require the popover keeping a reference to the opening button so that it can return the focus to that button.
Why?
It is important for keyboard users (or other users who rely on focus) to not have their focus state lost when performing actions on the page. If the page shows a new piece of navigable UI, and shifts the user’s focus to that UI, then it is important to restore focus when that UI closes, so that the user’s journey is maintained.
Other considerations not explicitly proposed.
aria-pressed
Given elements will have aria-expanded, adding aria-pressed would be
confusing or redundant, and as such won’t be proposed for these buttons.
aria-controls
While aria-controls attempts to establish a similar style of relationship to
aria-details, aria-details sees broader support among various assistive
technologies, and it would be redundant to add both.
aria-haspopup=dialog
When the commandfor= element is a <dialog>, adding aria-haspopup=dialog is
a potentially viable attribute to add. Some Assistive Technologies will announce
that the button will open a dialog, for example
<button aria-haspopup=dialog>Delete</button> may be announced by a screen
reader as “Delete, button, opens dialog”. However, buttons that open dialogs
rarely come with additional visual treatment, and so the lack of a visual
analog means that using this attribute likely provides surplus information
to users of assistive technologies that isn’t otherwise provided. Additionally,
the intent for aria-haspopup is more focused towards combobox popovers,
where the type of popover may be ambigious. As command buttons can be used for
more than just combobox controls, aria-haspopup is not present by default.
Buttons with command=show-modal
Buttons with this dialog command will also have focus restored to the button when the dialog closes. This will require the dialog keeping a reference to the opening button so that it can return the focus to that button.
Why?
It is important for keyboard users (or other users who rely on focus) to not have their focus state lost when performing actions on the page. If the page shows a new piece of navigable UI, and shifts the user’s focus to that UI, then it is important to restore focus when that UI closes, so that the user’s journey is maintained.
Other considerations not explicitly proposed.
aria-details
Exposing an aria-details relationship between the button and the dialog is
redundant. These relationships are useful only when the related element is
open and present on the page and the invoking button is still navigable. A
modal dialog renders the rest of the page inert meaning that it won’t be
possible to navigate to the invoking button.
aria-controls
Similar to aria-details, aria-controls is not useful for a dialog that
is closed, and non-navigable for a dialog that is open as modal, so it would
provide no benefit to add.
aria-pressed
The aria-pressed state is the incorrect type of association for elements
which conditionally appear on the page.
aria-expanded
Similar to aria-details, the aria-expanded state redundant, given the
dialog is opened as modal. For the user to have navigated to a button that
opens a modal dialog, the dialog must not be open. If the dialog is open then
the button will be inert, therefore non-navigable, therefore it is redundant
to provide it an implicit aria-expanded=true state also.
aria-haspopup=dialog
When the commandfor= element is a <dialog> it might seem aria-haspopup=dialog
would be a potentially viable attribute to add. Some Assistive Technologies will announce
that the button will open a dialog, for example
<button aria-haspopup=dialog>Delete</button> may be announced by a screen
reader as “Delete, button, opens dialog”. However, buttons that open dialogs
rarely come with additional visual treatment, and so the lack of a visual
analog means that using this attribute likely provides surplus information
to users of assistive technologies that isn’t otherwise provided. Additionally,
the intent for aria-haspopup is more focused towards combobox popovers,
where the type of popover may be ambigious. As command buttons can be used for
more than just combobox controls, aria-haspopup is not present by default.
Buttons with command=close
Buttons with this dialog command will implicitly receive aria-details=IDREF,
where IDREF matches that of the commandfor attribute, while the dialog is
in the showing state, and the button is not a descendant of the dialog.
Why?
A button that closes a dialog is very typically found inside of the dialog, but in some cases it may be found outside, perhaps as a “toggle” style button which opens and closes a dialog as non-modal. This button may be used to close an open dialog that is shown as non-modal. It may be useful for a user to traverse into the dialog before closing it, for example to check if they have unsaved changes within the dialog.
Buttons will also implicitly receive an aria-expanded value, if they are not a
descendant of the commandfor= referenced element. The state of the
aria-expanded value will map to the state of the dialog’s openness. When the
dialog is open the button will have aria-expanded="true", when closed,
aria-expanded="false". This will be recomputed whenever the dialog changes
state, such that the button always reflects the state of openness.
Why?
A button that closes a dialog is very typically found inside of the dialog, but in some cases it may be found outside, perhaps as a “toggle” style button which opens and closes a dialog as non-modal.
Buttons outside of the dialog may be used to close an open dialog that is shown
as non-modal. It may be useful for a user to traverse into the dialog before
closing it, for example to check if they have unsaved changes within the dialog.
It may also be useful to know if the dialog is already closed (as in its
aria-expanded state is false), as this may help the user make a decision to
whether or not they action the close button.
Other considerations not explicitly proposed.
aria-pressed
Given elements will have aria-expanded, adding aria-pressed would be
confusing or redundant, and as such won’t be proposed for these buttons.
aria-controls
While aria-controls attempts to establish a similar style of relationship to
aria-details, aria-details sees broader support among various assistive
technologies, and it would be redundant to add both.
Other built-in command= types
The above commands are the core built-in command types which will be initially shipping. Further built-in commands will be proposed on a case-per-case basis, and additional aria or other logic will be considered with those at the time.
Custom command= types
Custom command types have accessibility requirements which will be hard to determine based on the presence of the button and invokee alone. Consequently, it is left to the implementer or user of the custom command to assign appropriate aria markup to these buttons.
Elements with commandfor= but no command=
In this current proposal, elements without a command= attribute are
considered “author errors”; they will do nothing when invoked. These buttons
will recieve no additional implicit aria states.
Invokers and Custom Elements
As the underlying system for invoke elements uses event dispatch, Custom Elements
can make use of CommandEvent for their own behaviours. Consider the following:
<button type="button" commandfor="my-element" command="--spin-widget">
Spin the widget
</button>
<spin-widget id="my-element"></spin-widget>
<script>
customElements.define(
"spin-widget",
class extends HTMLElement {
connectedCallback() {
this.addEventListener("command", (e) => {
if (e.command === "--spin-widget") {
this.spin();
}
});
}
},
);
</script>
Security
This proposal adds new capabilities in that elements can be interacted with using HTML where prior these required scripting (read: JS). Care has been taken within this proposal to ensure there are no substantially new security implications, and that the risk is kept low.
In general, we should not grant significant new capabilities with or without scripting, and seek other ways to mitigate security concerns.
Invokers, by design, have two security affordances:
- Without scripting, they require user interaction - the invoker button needs
to be clicked for the browser to fire an Command Event. Scripts are able to
synthesise the
CommandEvent, or theclickevent on the button - but scripts are already capable of causing the same interactions without Invokers. - The
commandfortakes an IDREF which must reference an ID within the same document.commandForElementcan be assigned a cross-root element, but this requires scripting which means all actions are already possible.
To go into more detail, below is a brief summary of the security considerations per interaction:
Popovers
This proposal allows opening and closing popovers with invoketarget. This is
already possible with popovertarget, and so there is no new capability here.
Dialogs
This proposal allows opening and closing <dialog> elements with
invoketarget. While this is new capability (prior to this proposal scripting
is required), it poses the same security considerations as popovers in that it
toggles the visiblity of a new element. Modal dialog elements are also placed
on the “top layer” which means they occlude other content. This is already
possible with popovers, which are also placed on the top layer, so while
buttons to open <dialog> elements is a new capability, it is not a novel new
security threat considering attackers could already craft a popover.
Modal dialogs also trap focus, such that focus can only be moved within
elements inside the dialog or the browser native controls (such as the URL
bar). This is considered to not pose a significant threat as the invoking
button requires user activation, the dialog cannot render outside of the
document frame (iframe or browser window) and so cannot occlude the browsers
Chrome, and there are built in gestures to close a dialog (such as the Esc key
via a CloseWatcher) that cannot be opted out of with or without invokers.
PAQ (Potentially Asked Questions)
Why the name command? Why not click?
While click is a fairly well established name in the world of the web, it is
quite specific to the action from a user gesture, whereas invoke events notify
a related element that an interactive control is requesting action. In other
words, a click happens on the element being interacted with - the invoker,
while an invoke happens on the element the interactive control points to -
the invokee.
If the invokee were to recieve click events from the interactive control, this would make it difficult to differentiate user interaction on the element vs on invoking elements. Take for example a video player; clicking on the area which renders the video may play/pause the video, but clicking on a video control may cause different behaviours on the video. The extra controls invoke an action on the video but the user gesture was performed on the controls themselves.
Invokers rely on click events for their behavior to work, and they are not
intended to replace them. Instead, invokers enhance click events to drive
behaviour on a different element. Consequently they need a new conceptual name
for this. Given the opportunity to supply a new name, invoke was settled on.
Given the new conceptual model of “command” it stands to reason that all related
concepts share the same name, as such commandfor and command were chosen
rather than, say, clickfor/clickaction. Similarly the event class is named
CommandEvent and the event name command.
Didn’t this used to be called invoke? What happened?
The original proposal for Invoke Commands was to use the term invoke
exhaustively through the API, and so the attributes were originally named
invoketarget/invokeaction and the event named InvokeEvent with a type of
invoke. This was deemed too abstract, and standards groups and web developers
desired a simpler name. For those interested in the history, some conversation
that lead to changing invoke to command starts around here:
https://github.com/whatwg/html/issues/9625#issuecomment-2115718679
Do we have to always supply both? Can’t we make command or commandfor implicit?
The original proposal had the concept of an “auto” command value which would
determine an explicit command based on various heuristics, such as the target
element. This has been deferred for the initial ship, but may be explored
further. This is considered out of scope for the initial ship, however.
We may also explore the possibility of making commandfor implicit, for example
if a button is a descendant of a dialog, omitting commandfor may make sense.
This is also considered out of scope for the initial ship.
What about adding Invoker defaults for <form>?
Defaults for <form> are intentionally omitted as this proposal does not aim to
replace Reset or Submit buttons. If you want to control forms, use those.
What about adding Invoker defaults for <a>?
Defaults for <a> are intentionally omitted as this proposal does not aim to
replace anchors. If you intend to produce a page navigation, use an <a> tag.
Why is command/commandfor limited to buttons?
This is by design, to allow for a “pit of success”; invoking actions on
non-button elements such as <div>s or <a>s creates many problems, especially
for non-interactive elements. While <a>s are interactive, they should only
be used for page navigation and not for invoking other behaviours, and so
command/commandfor should not be allowed.
Why isn’t button[type=submit] included?
This is not added by design. Submit inputs already have a default action:
submitting forms. If you want a button to control the submission of a form, use
button[type=submit], if you want a button to control invocation of something
other than a form then you should use button[type=button].
Why isn’t button[type=reset] included?
This is not added by design. Reset inputs already have a default action:
resetting forms. If you want a button to control the resetting of a form, use
button[type=reset], if you want a button to control invocation of something
other than a form then you should use button[type=button].
Why isn’t input[type=button] included?
This is not added by design. input[type=button] is a legacy element,
use of this element is discouraged in favour of <button>. So we only support
<button> elements.
What does this mean for popovertarget?
Whilst command/commandfor does replicate popovertarget’s functionality,
the two will co-exist side by side for some while to enable a smooth transition.
It is, however, intended that commandfor will replace popovertarget leading
to popovertarget’s eventual deprecation.
Why doesn’t command just call the JS API method?
HTML attribute values have different casing to method names, so for example an
attribute value of show-modal does not match the method name showModal().
This requires manual mapping to the respective methods, but browsers - under the
hood - effectively map from these values to the respective methods. Similarly,
custom commands must be prefixed with a double dash (--). It is possible to
convert the dashed value to camel case, and use computed properties to call
methods, if that’s desired:
<button type="button" commandfor="my-element" command="--spin-widget">
Spin the widget
</button>
<spin-widget id="my-element"></spin-widget>
<script>
customElements.define(
"spin-widget",
class extends HTMLElement {
connectedCallback() {
this.addEventListener("command", (e) => {
const method = e.command
.slice(2) // remove the double dash
.replace(/(-\w)/g, c => c[1].toUpperCase()); // dash-to-camel case
if (method in this) {
this[method](event);
}
});
}
spinWidget(event) {
// ...
}
},
);
</script>
Command seems limited, what if I wanted to add arguments?
Custom commands can be used in a freeform way, for your own elements. If
you feel it necessary you can invent your own DSLs for passing extra data using
this hint. For example:
<button type="button" commandfor="my-counter" command="--add-1">Add 1</button>
<button type="button" commandfor="my-counter" command="--add-2">Add 2</button>
<button type="button" commandfor="my-counter" command="--add-10">Add 10</button>
<input readonly id="my-counter" value="0" />
<script>
const counter = document.getElementById("my-counter");
counter.addEventListener("command", (e) => {
let addMatch = /^--add-(\d+)$/.match(e.command);
if (addMatch) {
counter.value = Number(counter.value) + Number(addMatch[1]);
}
});
</script>
You can also leverage the buttons value attribute to effectively parameterise
certain commands:
<button type="button" commandfor="my-counter" command="--add-num" value="1">Add 1</button>
<button type="button" commandfor="my-counter" command="--add-num" value="2">Add 2</button>
<button type="button" commandfor="my-counter" command="--add-num" value="10">Add 10</button>
<input readonly id="my-counter" value="0" />
<script>
const counter = document.getElementById("my-counter");
counter.addEventListener("command", (e) => {
if (e.command == '--add-num') {
counter.value = Number(counter.value) + Number(e.invoker.value);
}
});
</script>
You could also use data-* attributes for even more granular control.
Polyfilling
A basic polyfill has been written, which implements the behaviors described by this explainer.
The polyfill is available at Github:
Patches are welcome!
Open UI