Contents
htmx is a library that allows you to access modern browser features directly from HTML, rather than using javascript.
To understand htmx, first let’s take a look at an anchor tag:
<a href="/blog">Blog</a>
This anchor tag tells a browser:
“When a user clicks on this link, issue an HTTP GET request to ‘/blog’ and load the response content into the browser window”.
With that in mind, consider the following bit of HTML:
<button hx-post="/clicked"
hx-trigger="click"
hx-target="#parent-div"
hx-swap="outerHTML">
Click Me!
</button>
This tells htmx:
“When a user clicks on this button, issue an HTTP POST request to ‘/clicked’ and use the content from the response to replace the element with the id
parent-div
in the DOM”
htmx extends and generalizes the core idea of HTML as a hypertext, opening up many more possibilities directly within the language:
GET
and POST
, can be usedNote that when you are using htmx, on the server side you typically respond with HTML, not JSON. This keeps you firmly within the original web programming model, using Hypertext As The Engine Of Application State without even needing to really understand that concept.
It’s worth mentioning that, if you prefer, you can use the data-
prefix when using htmx:
<a data-hx-post="/click">Click Me!</a>
Finally, Version 1 of htmx is still supported and supports IE11.
If you are migrating to htmx 2.x from htmx 1.x, please see the htmx 1.x migration guide.
If you are migrating to htmx from intercooler.js, please see the intercooler migration guide.
Htmx is a dependency-free, browser-oriented javascript library. This means that using it is as simple as adding a <script>
tag to your document head. There is no need for a build system to use it.
The fastest way to get going with htmx is to load it via a CDN. You can simply add this to your head tag and get going:
<script src="https://unpkg.com/[email protected]" integrity="sha384-0895/pl2MU10Hqc6jd4RvrthNlDiE9U1tWmX7WRESftEDRosgxNsQG/Ze9YMRzHq" crossorigin="anonymous"></script>
An unminified version is also available as well:
<script src="https://unpkg.com/[email protected]/dist/htmx.js" integrity="sha384-BBDmZzVt6vjz5YbQqZPtFZW82o8QotoM7RUp5xOxV3nSJ8u2pSdtzFAbGKzTlKtg" crossorigin="anonymous"></script>
While the CDN approach is extremely simple, you may want to consider not using CDNs in production.
The next easiest way to install htmx is to simply copy it into your project.
Download htmx.min.js
from unpkg.com and add it to the appropriate directory in your project
and include it where necessary with a <script>
tag:
<script src="/path/to/htmx.min.js"></script>
For npm-style build systems, you can install htmx via npm:
npm install [email protected]
After installing, you’ll need to use appropriate tooling to use node_modules/htmx.org/dist/htmx.js
(or .min.js
).
For example, you might bundle htmx with some extensions and project-specific code.
If you are using webpack to manage your javascript:
htmx
via your favourite package manager (like npm or yarn)index.js
import 'htmx.org';
If you want to use the global htmx
variable (recommended), you need to inject it to the window scope:
index.js
(below the import from step 2)import 'path/to/my_custom.js';
window.htmx = require('htmx.org');
The core of htmx is a set of attributes that allow you to issue AJAX requests directly from HTML:
Attribute | Description |
---|---|
hx-get | Issues a GET request to the given URL |
hx-post | Issues a POST request to the given URL |
hx-put | Issues a PUT request to the given URL |
hx-patch | Issues a PATCH request to the given URL |
hx-delete | Issues a DELETE request to the given URL |
Each of these attributes takes a URL to issue an AJAX request to. The element will issue a request of the specified type to the given URL when the element is triggered:
<button hx-put="/messages">
Put To Messages
</button>
This tells the browser:
When a user clicks on this button, issue a PUT request to the URL /messages and load the response into the button
By default, AJAX requests are triggered by the “natural” event of an element:
input
, textarea
& select
are triggered on the change
eventform
is triggered on the submit
eventclick
eventIf you want different behavior you can use the hx-trigger attribute to specify which event will cause the request.
Here is a div
that posts to /mouse_entered
when a mouse enters it:
<div hx-post="/mouse_entered" hx-trigger="mouseenter">
[Here Mouse, Mouse!]
</div>
A trigger can also have a few additional modifiers that change its behavior. For example, if you want a request to only
happen once, you can use the once
modifier for the trigger:
<div hx-post="/mouse_entered" hx-trigger="mouseenter once">
[Here Mouse, Mouse!]
</div>
Other modifiers you can use for triggers are:
changed
- only issue a request if the value of the element has changeddelay:<time interval>
- wait the given amount of time (e.g. 1s
) before
issuing the request. If the event triggers again, the countdown is reset.throttle:<time interval>
- wait the given amount of time (e.g. 1s
) before
issuing the request. Unlike delay
if a new event occurs before the time limit is hit the event will be discarded,
so the request will trigger at the end of the time period.from:<CSS Selector>
- listen for the event on a different element. This can be used for things like keyboard shortcuts. Note that this CSS selector is not re-evaluated if the page changes.You can use these attributes to implement many common UX patterns, such as Active Search:
<input type="text" name="q"
hx-get="/trigger_delay"
hx-trigger="keyup changed delay:500ms"
hx-target="#search-results"
placeholder="Search...">
<div id="search-results"></div>
This input will issue a request 500 milliseconds after a key up event if the input has been changed and inserts the results
into the div
with the id search-results
.
Multiple triggers can be specified in the hx-trigger attribute, separated by commas.
You may also apply trigger filters by using square brackets after the event name, enclosing a javascript expression that
will be evaluated. If the expression evaluates to true
the event will trigger, otherwise it will not.
Here is an example that triggers only on a Control-Click of the element
<div hx-get="/clicked" hx-trigger="click[ctrlKey]">
Control Click Me
</div>
Properties like ctrlKey
will be resolved against the triggering event first, then against the global scope. The
this
symbol will be set to the current element.
htmx provides a few special events for use in hx-trigger:
load
- fires once when the element is first loadedrevealed
- fires once when an element first scrolls into the viewportintersect
- fires once when an element first intersects the viewport. This supports two additional options:
root:<selector>
- a CSS selector of the root element for intersectionthreshold:<float>
- a floating point number between 0.0 and 1.0, indicating what amount of intersection to fire the event onYou can also use custom events to trigger requests if you have an advanced use case.
If you want an element to poll the given URL rather than wait for an event, you can use the every
syntax
with the hx-trigger
attribute:
<div hx-get="/news" hx-trigger="every 2s"></div>
This tells htmx
Every 2 seconds, issue a GET to /news and load the response into the div
If you want to stop polling from a server response you can respond with the HTTP response code 286
and the element will cancel the polling.
Another technique that can be used to achieve polling in htmx is “load polling”, where an element specifies
a load
trigger along with a delay, and replaces itself with the response:
<div hx-get="/messages"
hx-trigger="load delay:1s"
hx-swap="outerHTML">
</div>
If the /messages
end point keeps returning a div set up this way, it will keep “polling” back to the URL every
second.
Load polling can be useful in situations where a poll has an end point at which point the polling terminates, such as when you are showing the user a progress bar.
When an AJAX request is issued it is often good to let the user know that something is happening since the browser
will not give them any feedback. You can accomplish this in htmx by using htmx-indicator
class.
The htmx-indicator
class is defined so that the opacity of any element with this class is 0 by default, making it invisible
but present in the DOM.
When htmx issues a request, it will put a htmx-request
class onto an element (either the requesting element or
another element, if specified). The htmx-request
class will cause a child element with the htmx-indicator
class
on it to transition to an opacity of 1, showing the indicator.
<button hx-get="/click">
Click Me!
<img class="htmx-indicator" src="/spinner.gif">
</button>
Here we have a button. When it is clicked the htmx-request
class will be added to it, which will reveal the spinner
gif element. (I like SVG spinners these days.)
While the htmx-indicator
class uses opacity to hide and show the progress indicator, if you would prefer another mechanism
you can create your own CSS transition like so:
.htmx-indicator{
display:none;
}
.htmx-request .htmx-indicator{
display:inline;
}
.htmx-request.htmx-indicator{
display:inline;
}
If you want the htmx-request
class added to a different element, you can use the hx-indicator
attribute with a CSS selector to do so:
<div>
<button hx-get="/click" hx-indicator="#indicator">
Click Me!
</button>
<img id="indicator" class="htmx-indicator" src="/spinner.gif"/>
</div>
Here we call out the indicator explicitly by id. Note that we could have placed the class on the parent div
as well
and had the same effect.
You can also add the disabled
attribute to
elements for the duration of a request by using the hx-disabled-elt attribute.
If you want the response to be loaded into a different element other than the one that made the request, you can use the hx-target attribute, which takes a CSS selector. Looking back at our Live Search example:
<input type="text" name="q"
hx-get="/trigger_delay"
hx-trigger="keyup delay:500ms changed"
hx-target="#search-results"
placeholder="Search...">
<div id="search-results"></div>
You can see that the results from the search are going to be loaded into div#search-results
, rather than into the
input tag.
hx-target
, and most attributes that take a CSS selector, support an “extended” CSS syntax:
this
keyword, which indicates that the element that the hx-target
attribute is on is the targetclosest <CSS selector>
syntax will find the closest
ancestor element or itself, that matches the given CSS selector.
(e.g. closest tr
will target the closest table row to the element)next <CSS selector>
syntax will find the next element in the DOM matching the given CSS selector.previous <CSS selector>
syntax will find the previous element in the DOM matching the given CSS selector.find <CSS selector>
which will find the first child descendant element that matches the given CSS selector.
(e.g find tr
would target the first child descendant row to the element)In addition, a CSS selector may be wrapped in <
and />
characters, mimicking the
query literal syntax of hyperscript.
Relative targets like this can be useful for creating flexible user interfaces without peppering your DOM with lots
of id
attributes.
htmx offers a few different ways to swap the HTML returned into the DOM. By default, the content replaces the
innerHTML
of the target element. You can modify this by using the hx-swap attribute
with any of the following values:
Name | Description |
---|---|
innerHTML | the default, puts the content inside the target element |
outerHTML | replaces the entire target element with the returned content |
afterbegin | prepends the content before the first child inside the target |
beforebegin | prepends the content before the target in the target’s parent element |
beforeend | appends the content after the last child inside the target |
afterend | appends the content after the target in the target’s parent element |
delete | deletes the target element regardless of the response |
none | does not append content from response (Out of Band Swaps and Response Headers will still be processed) |
In addition to the standard swap mechanisms above, htmx also supports morphing swaps, via extensions. Morphing swaps attempt to merge new content into the existing DOM, rather than simply replacing it. They often do a better job preserving things like focus, video state, etc. by mutating existing nodes in-place during the swap operation, at the cost of more CPU.
The following extensions are available for morph-style swaps:
The new, experimental View Transitions API gives developers a way to create an animated transition between different DOM states. It is still in active development and is not available in all browsers, but htmx provides a way to work with this new API that falls back to the non-transition mechanism if the API is not available in a given browser.
You can experiment with this new API using the following approaches:
htmx.config.globalViewTransitions
config variable to true
to use transitions for all swapstransition:true
option in the hx-swap
attributehtmx:beforeTransition
event and call preventDefault()
on it to cancel the transition.View Transitions can be configured using CSS, as outlined in the Chrome documentation for the feature.
You can see a view transition example on the Animation Examples page.
The hx-swap attribute supports many options for tuning the swapping behavior of htmx. For
example, by default htmx will swap in the title of a title tag found anywhere in the new content. You can turn this
behavior off by setting the ignoreTitle
modifier to true:
<button hx-post="/like" hx-swap="outerHTML ignoreTitle:true">Like</button>
The modifiers available on hx-swap
are:
Option | Description |
---|---|
transition | true or false , whether to use the view transition API for this swap |
swap | The swap delay to use (e.g. 100ms ) between when old content is cleared and the new content is inserted |
settle | The settle delay to use (e.g. 100ms ) between when new content is inserted and when it is settled |
ignoreTitle | If set to true , any title found in the new content will be ignored and not update the document title |
scroll | top or bottom , will scroll the target element to its top or bottom |
show | top or bottom , will scroll the target element’s top or bottom into view |
All swap modifiers appear after the swap style is specified, and are colon-separated.
See the hx-swap documentation for more details on these options.
Often you want to coordinate the requests between two elements. For example, you may want a request from one element to supersede the request of another element, or to wait until the other element’s request has finished.
htmx offers a hx-sync
attribute to help you accomplish this.
Consider a race condition between a form submission and an individual input’s validation request in this HTML:
<form hx-post="/store">
<input id="title" name="title" type="text"
hx-post="/validate"
hx-trigger="change">
<button type="submit">Submit</button>
</form>
Without using hx-sync
, filling out the input and immediately submitting the form triggers two parallel requests to
/validate
and /store
.
Using hx-sync="closest form:abort"
on the input will watch for requests on the form and abort the input’s request if
a form request is present or starts while the input request is in flight:
<form hx-post="/store">
<input id="title" name="title" type="text"
hx-post="/validate"
hx-trigger="change"
hx-sync="closest form:abort">
<button type="submit">Submit</button>
</form>
This resolves the synchronization between the two elements in a declarative way.
htmx also supports a programmatic way to cancel requests: you can send the htmx:abort
event to an element to
cancel any in-flight requests:
<button id="request-button" hx-post="/example">
Issue Request
</button>
<button onclick="htmx.trigger('#request-button', 'htmx:abort')">
Cancel Request
</button>
More examples and details can be found on the hx-sync
attribute page.
htmx makes it easy to use CSS Transitions without javascript. Consider this HTML content:
<div id="div1">Original Content</div>
Imagine this content is replaced by htmx via an ajax request with this new content:
<div id="div1" class="red">New Content</div>
Note two things:
red
class has been added to the new contentGiven this situation, we can write a CSS transition from the old state to the new state:
.red {
color: red;
transition: all ease-in 1s ;
}
When htmx swaps in this new content, it will do so in such a way that the CSS transition will apply to the new content, giving you a nice, smooth transition to the new state.
So, in summary, all you need to do to use CSS transitions for an element is keep its id
stable across requests!
You can see the Animation Examples for more details and live demonstrations.
To understand how CSS transitions actually work in htmx, you must understand the underlying swap & settle model that htmx uses.
When new content is received from a server, before the content is swapped in, the existing
content of the page is examined for elements that match by the id
attribute. If a match
is found for an element in the new content, the attributes of the old content are copied
onto the new element before the swap occurs. The new content is then swapped in, but with the
old attribute values. Finally, the new attribute values are swapped in, after a “settle” delay
(20ms by default). A little crazy, but this is what allows CSS transitions to work without any javascript by
the developer.
If you want to swap content from a response directly into the DOM by using the id
attribute you can use the
hx-swap-oob attribute in the response html:
<div id="message" hx-swap-oob="true">Swap me directly!</div>
Additional Content
In this response, div#message
would be swapped directly into the matching DOM element, while the additional content
would be swapped into the target in the normal manner.
You can use this technique to “piggy-back” updates on other requests.
Table elements can be problematic when combined with out of band swaps, because, by the HTML spec, many can’t stand on
their own in the DOM (e.g. <tr>
or <td>
).
To avoid this issue you can use a template
tag to encapsulate these elements:
<template>
<tr id="message" hx-swap-oob="true"><td>Joe</td><td>Smith</td></tr>
</template>
If you want to select a subset of the response HTML to swap into the target, you can use the hx-select attribute, which takes a CSS selector and selects the matching elements from the response.
You can also pick out pieces of content for an out-of-band swap by using the hx-select-oob attribute, which takes a list of element IDs to pick out and swap.
If there is content that you wish to be preserved across swaps (e.g. a video player that you wish to remain playing even if a swap occurs) you can use the hx-preserve attribute on the elements you wish to be preserved.
By default, an element that causes a request will include its value if it has one. If the element is a form it will include the values of all inputs within it.
As with HTML forms, the name
attribute of the input is used as the parameter name in the request that htmx sends.
Additionally, if the element causes a non-GET
request, the values of all the inputs of the nearest enclosing form
will be included.
If you wish to include the values of other elements, you can use the hx-include attribute with a CSS selector of all the elements whose values you want to include in the request.
If you wish to filter out some parameters you can use the hx-params attribute.
Finally, if you want to programmatically modify the parameters, you can use the htmx:configRequest event.
If you wish to upload files via an htmx request, you can set the hx-encoding attribute to
multipart/form-data
. This will use a FormData
object to submit the request, which will properly include the file
in the request.
Note that depending on your server-side technology, you may have to handle requests with this type of body content very differently.
Note that htmx fires a htmx:xhr:progress
event periodically based on the standard progress
event during upload,
which you can hook into to show the progress of the upload.
See the examples section for more advanced form patterns, including progress bars and error handling.
You can include extra values in a request using the hx-vals (name-expression pairs in JSON format) and hx-vars attributes (comma-separated name-expression pairs that are dynamically computed).
Often you will want to confirm an action before issuing a request. htmx supports the hx-confirm
attribute, which allows you to confirm an action using a simple javascript dialog:
<button hx-delete="/account" hx-confirm="Are you sure you wish to delete your account?">
Delete My Account
</button>
Using events you can implement more sophisticated confirmation dialogs. The confirm example shows how to use sweetalert2 library for confirmation of htmx actions.
Another option to do confirmation with is via the htmx:confirm
event. This event
is fired on every trigger for a request (not just on elements that have a hx-confirm
attribute) and can be used
to implement asynchronous confirmation of the request.
Here is an example using sweet alert on any element with a confirm-with-sweet-alert='true'
attribute on it:
document.body.addEventListener('htmx:confirm', function(evt) {
if (evt.target.matches("[confirm-with-sweet-alert='true']")) {
evt.preventDefault();
swal({
title: "Are you sure?",
text: "Are you sure you are sure?",
icon: "warning",
buttons: true,
dangerMode: true,
}).then((confirmed) => {
if (confirmed) {
evt.detail.issueRequest();
}
});
}
});
Most attributes in htmx are inherited: they apply to the element they are on as well as any children elements. This allows you to “hoist” attributes up the DOM to avoid code duplication. Consider the following htmx:
<button hx-delete="/account" hx-confirm="Are you sure?">
Delete My Account
</button>
<button hx-put="/account" hx-confirm="Are you sure?">
Update My Account
</button>
Here we have a duplicate hx-confirm
attribute. We can hoist this attribute to a parent element:
<div hx-confirm="Are you sure?">
<button hx-delete="/account">
Delete My Account
</button>
<button hx-put="/account">
Update My Account
</button>
</div>
This hx-confirm
attribute will now apply to all htmx-powered elements within it.
Sometimes you wish to undo this inheritance. Consider if we had a cancel button to this group, but didn’t want it to
be confirmed. We could add an unset
directive on it like so:
<div hx-confirm="Are you sure?">
<button hx-delete="/account">
Delete My Account
</button>
<button hx-put="/account">
Update My Account
</button>
<button hx-confirm="unset" hx-get="/">
Cancel
</button>
</div>
The top two buttons would then show a confirm dialog, but the bottom cancel button would not.
Inheritance can be disabled on a per-element and per-attribute basis using the
hx-disinherit
attribute.
If you wish to disable attribute inheritance entirely, you can set the htmx.config.disableInheritance
configuration
variable to true
. This will disable inheritance as a default, and allow you to specify inheritance explicitly
with the hx-inherit
attribute.
Htmx supports “boosting” regular HTML anchors and forms with the hx-boost attribute. This attribute will convert all anchor tags and forms into AJAX requests that, by default, target the body of the page.
Here is an example:
<div hx-boost="true">
<a href="/blog">Blog</a>
</div>
The anchor tag in this div will issue an AJAX GET
request to /blog
and swap the response into the body
tag.
A feature of hx-boost
is that it degrades gracefully if javascript is not enabled: the links and forms continue
to work, they simply don’t use ajax requests. This is known as
Progressive Enhancement, and it allows
a wider audience to use your site’s functionality.
Other htmx patterns can be adapted to achieve progressive enhancement as well, but they will require more thought.
Consider the active search example. As it is written, it will not degrade gracefully: someone who does not have javascript enabled will not be able to use this feature. This is done for simplicity’s sake, to keep the example as brief as possible.
However, you could wrap the htmx-enhanced input in a form element:
<form action="/search" method="POST">
<input class="form-control" type="search"
name="search" placeholder="Begin typing to search users..."
hx-post="/search"
hx-trigger="keyup changed delay:500ms, search"
hx-target="#search-results"
hx-indicator=".htmx-indicator">
</form>
With this in place, javascript-enabled clients would still get the nice active-search UX, but non-javascript enabled
clients would be able to hit the enter key and still search. Even better, you could add a “Search” button as well.
You would then need to update the form with an hx-post
that mirrored the action
attribute, or perhaps use hx-boost
on it.
You would need to check on the server side for the HX-Request
header to differentiate between an htmx-driven and a
regular request, to determine exactly what to render to the client.
Other patterns can be adapted similarly to achieve the progressive enhancement needs of your application.
As you can see, this requires more thought and more work. It also rules some functionality entirely out of bounds. These tradeoffs must be made by you, the developer, with respect to your projects goals and audience.
Accessibility is a concept
closely related to progressive enhancement. Using progressive enhancement techniques such as hx-boost
will make your
htmx application more accessible to a wide array of users.
htmx-based applications are very similar to normal, non-AJAX driven web applications because htmx is HTML-oriented.
As such, the normal HTML accessibility recommendations apply. For example:
Web Sockets and Server Sent Events (SSE) are supported via extensions. Please see the SSE extension and WebSocket extension pages to learn more.
Htmx provides a simple mechanism for interacting with the browser history API:
If you want a given element to push its request URL into the browser navigation bar and add the current state of the page to the browser’s history, include the hx-push-url attribute:
<a hx-get="/blog" hx-push-url="true">Blog</a>
When a user clicks on this link, htmx will snapshot the current DOM and store it before it makes a request to /blog. It then does the swap and pushes a new location onto the history stack.
When a user hits the back button, htmx will retrieve the old content from storage and swap it back into the target,
simulating “going back” to the previous state. If the location is not found in the cache, htmx will make an ajax
request to the given URL, with the header HX-History-Restore-Request
set to true, and expects back the HTML needed
for the entire page. Alternatively, if the htmx.config.refreshOnHistoryMiss
config variable is set to true, it will
issue a hard browser refresh.
NOTE: If you push a URL into the history, you must be able to navigate to that URL and get a full page back! A user could copy and paste the URL into an email, or new tab. Additionally, htmx will need the entire page when restoring history if the page is not in the history cache.
By default, htmx will use the body
to take and restore the history snapshot from. This is usually the right thing, but
if you want to use a narrower element for snapshotting you can use the hx-history-elt
attribute to specify a different one.
Careful: this element will need to be on all pages or restoring from history won’t work reliably.
If you are using a 3rd party library and want to use the htmx history feature, you will need to clean up the DOM before
a snapshot is taken. Let’s consider the Tom Select library, which makes select elements
a much richer user experience. Let’s set up TomSelect to turn any input element with the .tomselect
class into a rich
select element.
First we need to initialize elements that have the class in new content:
htmx.onLoad(function (target) {
// find all elements in the new content that should be
// an editor and init w/ TomSelect
var editors = target.querySelectorAll(".tomselect")
.forEach(elt => new TomSelect(elt))
});
This will create a rich selector for all input elements that have the .tomselect
class on it. However, it mutates
the DOM and we don’t want that mutation saved to the history cache, since TomSelect will be reinitialized when the
history content is loaded back into the screen.
To deal with this, we need to catch the htmx:beforeHistorySave
event and clean out the TomSelect mutations by calling
destroy()
on them:
htmx.on('htmx:beforeHistorySave', function() {
// find all TomSelect elements
document.querySelectorAll('.tomSelect')
.forEach(elt => elt.tomselect.destroy()) // and call destroy() on them
})
This will revert the DOM to the original HTML, thus allowing for a clean snapshot.
History snapshotting can be disabled for a URL by setting the hx-history attribute to false
on any element in the current document, or any html fragment loaded into the current document by htmx. This can be used
to prevent sensitive data entering the localStorage cache, which can be important for shared-use / public computers.
History navigation will work as expected, but on restoration the URL will be requested from the server instead of the
local history cache.
Htmx expects responses to the AJAX requests it makes to be HTML, typically HTML fragments (although a full HTML document, matched with a hx-select tag can be useful too). Htmx will then swap the returned HTML into the document at the target specified and with the swap strategy specified.
Sometimes you might want to do nothing in the swap, but still perhaps trigger a client side event (see below).
For this situation, by default, you can return a 204 - No Content
response code, and htmx will ignore the content of
the response.
In the event of an error response from the server (e.g. a 404 or a 501), htmx will trigger the htmx:responseError
event, which you can handle.
In the event of a connection error, the htmx:sendError
event will be triggered.
You can configure the above behavior of htmx by mutating or replacing the htmx.config.responseHandling
array. This
object is a collection of JavaScript objects defined like so:
responseHandling: [
{code:"204", swap: false}, // 204 - No Content by default does nothing, but is not an error
{code:"[23]..", swap: true}, // 200 & 300 responses are non-errors and are swapped
{code:"[45]..", swap: false, error:true}, // 400 & 500 responses are not swapped and are errors
{code:"...", swap: false} // catch all for any other response code
]
When htmx receives a response it will iterate in order over the htmx.config.responseHandling
array and test if the
code
property of a given object, when treated as a Regular Expression, matches the current response. If an entry
does match the current response code, it will be used to determine if and how the response will be processed.
The fields available for response handling configuration on entries in this array are:
code
- a String representing a regular expression that will be tested against response codes.swap
- true
if the response should be swapped into the DOM, false
otherwiseerror
- true
if htmx should treat this response as an errorignoreTitle
- true
if htmx should ignore title tags in the responseselect
- A CSS selector to use to select content from the responsetarget
- A CSS selector specifying an alternative target for the responseswapOverride
- An alternative swap mechanism for the responseAs an example of how to use this configuration, consider a situation when a server-side framework responds with a
422 - Unprocessable Entity
response when validation errors occur. By default, htmx will ignore the response,
since it matches the Regular Expression [45]..
.
Using the meta config mechanism for configuring responseHandling, we could add the following config:
<!--
* 204 No Content by default does nothing, but is not an error
* 2xx, 3xx and 422 responses are non-errors and are swapped
* 4xx & 5xx responses are not swapped and are errors
* all other responses are swapped using "..." as a catch-all
-->
<meta
name="htmx-config"
content='{
"responseHandling":[
{"code":"204", "swap": false},
{"code":"[23]..", "swap": true},
{"code":"422", "swap": true},
{"code":"[45]..", "swap": false, "error":true},
{"code":"...", "swap": true}
]
}'
/>
If you wanted to swap everything, regardless of HTTP response code, you could use this configuration:
<meta name="htmx-config" content='{"responseHandling": [{"code":".*", "swap": true}]}' /> <!--all responses are swapped-->
Finally, it is worth considering using the Response Targets extension, which allows you to configure the behavior of response codes declaratively via attributes.
When using htmx in a cross origin context, remember to configure your web server to set Access-Control headers in order for htmx headers to be visible on the client side.
See all the request and response headers that htmx implements.
htmx includes a number of useful headers in requests:
Header | Description |
---|---|
HX-Boosted | indicates that the request is via an element using hx-boost |
HX-Current-URL | the current URL of the browser |
HX-History-Restore-Request | “true” if the request is for history restoration after a miss in the local history cache |
HX-Prompt | the user response to an hx-prompt |
HX-Request | always “true” |
HX-Target | the id of the target element if it exists |
HX-Trigger-Name | the name of the triggered element if it exists |
HX-Trigger | the id of the triggered element if it exists |
htmx supports some htmx-specific response headers:
HX-Location
- allows you to do a client-side redirect that does not do a full page reloadHX-Push-Url
- pushes a new url into the history stackHX-Redirect
- can be used to do a client-side redirect to a new locationHX-Refresh
- if set to “true” the client-side will do a full refresh of the pageHX-Replace-Url
- replaces the current URL in the location barHX-Reswap
- allows you to specify how the response will be swapped. See hx-swap for possible valuesHX-Retarget
- a CSS selector that updates the target of the content update to a different element on the pageHX-Reselect
- a CSS selector that allows you to choose which part of the response is used to be swapped in. Overrides an existing hx-select
on the triggering elementHX-Trigger
- allows you to trigger client-side eventsHX-Trigger-After-Settle
- allows you to trigger client-side events after the settle stepHX-Trigger-After-Swap
- allows you to trigger client-side events after the swap stepFor more on the HX-Trigger
headers, see HX-Trigger
Response Headers.
Submitting a form via htmx has the benefit of no longer needing the Post/Redirect/Get Pattern. After successfully processing a POST request on the server, you don’t need to return a HTTP 302 (Redirect). You can directly return the new HTML fragment.
Also the response headers above are not provided to htmx for processing with 3xx Redirect response codes like HTTP 302 (Redirect). Instead, the browser will intercept the redirection internally and return the headers and response from the redirected URL. Where possible use alternative response codes like 200 to allow returning of these response headers.
The order of operations in a htmx request are:
htmx-request
class is applied to the appropriate elementshtmx-swapping
classhtmx-swapping
class is removed from the targethtmx-added
class is added to each new piece of contenthtmx-settling
class is applied to the targethtmx-settling
class is removed from the targethtmx-added
class is removed from each new piece of contentYou can use the htmx-swapping
and htmx-settling
classes to create
CSS transitions between pages.
Htmx integrates with the HTML5 Validation API and will not issue a request for a form if a validatable input is invalid. This is true for both AJAX requests as well as WebSocket sends.
Htmx fires events around validation that can be used to hook in custom validation and error handling:
htmx:validation:validate
- called before an element’s checkValidity()
method is called. May be used to add in
custom validation logichtmx:validation:failed
- called when checkValidity()
returns false, indicating an invalid inputhtmx:validation:halted
- called when a request is not issued due to validation errors. Specific errors may be found
in the event.detail.errors
objectNon-form elements do not validate before they make requests by default, but you can enable validation by setting
the hx-validate
attribute to “true”.
Here is an example of an input that uses the hx-on
attribute to catch the
htmx:validation:validate
event and require that the input have the value foo
:
<form id="example-form" hx-post="/test">
<input name="example"
onkeyup="this.setCustomValidity('') // reset the validation on keyup"
hx-on:htmx:validation:validate="if(this.value != 'foo') {
this.setCustomValidity('Please enter the value foo') // set the validation error
htmx.find('#example-form').reportValidity() // report the issue
}">
</form>
Note that all client side validations must be re-done on the server side, as they can always be bypassed.
Htmx allows you to use CSS transitions in many situations using only HTML and CSS.
Please see the Animation Guide for more details on the options available.
htmx provides an extensions mechanism that allows you to customize the libraries’ behavior.
Extensions are defined in javascript and then enabled via
the hx-ext
attribute.
Here is how you would install the idiomorph extension, which allows you to use the Idiomorph DOM morphing algorithms for htmx swaps:
<head>
<script src="https://unpkg.com/[email protected]/dist/idiomorph-ext.min.js"></script>
</head>
<body hx-ext="morph">
...
<button hx-post="/example" hx-swap="morph" hx-target="#content">
Update Content
</button>
...
</body>
First the extension is included (it should be included after htmx.js
is included), then the extension is referenced
by name via the hx-ext
attribute. This enables you to then use the morph
swap.
htmx supports a few “core” extensions, which are supported by the htmx development team:
morph
swap strategy using idiomorph404
)You can see all available extensions on the Extensions page.
If you are interested in adding your own extension to htmx, please see the extension docs.
Htmx has an extensive events mechanism, which doubles as the logging system.
If you want to register for a given htmx event you can use
document.body.addEventListener('htmx:load', function(evt) {
myJavascriptLib.init(evt.detail.elt);
});
or, if you would prefer, you can use the following htmx helper:
htmx.on("htmx:load", function(evt) {
myJavascriptLib.init(evt.detail.elt);
});
The htmx:load
event is fired every time an element is loaded into the DOM by htmx, and is effectively the equivalent
to the normal load
event.
Some common uses for htmx events are:
Using the htmx:load
event to initialize content is so common that htmx provides a helper function:
htmx.onLoad(function(target) {
myJavascriptLib.init(target);
});
This does the same thing as the first example, but is a little cleaner.
You can handle the htmx:configRequest
event in order to modify an AJAX request before it is issued:
document.body.addEventListener('htmx:configRequest', function(evt) {
evt.detail.parameters['auth_token'] = getAuthToken(); // add a new parameter into the request
evt.detail.headers['Authentication-Token'] = getAuthToken(); // add a new header into the request
});
Here we add a parameter and header to the request before it is sent.
You can handle the htmx:beforeSwap
event in order to modify the swap behavior of htmx:
document.body.addEventListener('htmx:beforeSwap', function(evt) {
if(evt.detail.xhr.status === 404){
// alert the user when a 404 occurs (maybe use a nicer mechanism than alert())
alert("Error: Could Not Find Resource");
} else if(evt.detail.xhr.status === 422){
// allow 422 responses to swap as we are using this as a signal that
// a form was submitted with bad data and want to rerender with the
// errors
//
// set isError to false to avoid error logging in console
evt.detail.shouldSwap = true;
evt.detail.isError = false;
} else if(evt.detail.xhr.status === 418){
// if the response code 418 (I'm a teapot) is returned, retarget the
// content of the response to the element with the id `teapot`
evt.detail.shouldSwap = true;
evt.detail.target = htmx.find("#teapot");
}
});
Here we handle a few 400-level error response codes that would normally not do a swap in htmx.
Note that all events are fired with two different names
So, for example, you can listen for htmx:afterSwap
or for htmx:after-swap
. This facilitates interoperability
with other libraries. Alpine.js, for example, requires kebab case.
If you set a logger at htmx.logger
, every event will be logged. This can be very useful for troubleshooting:
htmx.logger = function(elt, event, data) {
if(console) {
console.log(event, elt, data);
}
}
Declarative and event driven programming with htmx (or any other declarative language) can be a wonderful and highly productive activity, but one disadvantage when compared with imperative approaches is that it can be trickier to debug.
Figuring out why something isn’t happening, for example, can be difficult if you don’t know the tricks.
Well, here are the tricks:
The first debugging tool you can use is the htmx.logAll()
method. This will log every event that htmx triggers and
will allow you to see exactly what the library is doing.
htmx.logAll();
Of course, that won’t tell you why htmx isn’t doing something. You might also not know what events a DOM
element is firing to use as a trigger. To address this, you can use the
monitorEvents()
method available in the
browser console:
monitorEvents(htmx.find("#theElement"));
This will spit out all events that are occurring on the element with the id theElement
to the console, and allow you
to see exactly what is going on with it.
Note that this only works from the console, you cannot embed it in a script tag on your page.
Finally, push come shove, you might want to just debug htmx.js
by loading up the unminimized version. It’s
about 2500 lines of javascript, so not an insurmountable amount of code. You would most likely want to set a break
point in the issueAjaxRequest()
and handleAjaxResponse()
methods to see what’s going on.
And always feel free to jump on the Discord if you need help.
Sometimes, in order to demonstrate a bug or clarify a usage, it is nice to be able to use a javascript snippet site like jsfiddle. To facilitate easy demo creation, htmx hosts a demo script site that will install:
Simply add the following script tag to your demo/fiddle/whatever:
<script src="https://demo.htmx.org"></script>
This helper allows you to add mock responses by adding template
tags with a url
attribute to indicate which URL.
The response for that url will be the innerHTML of the template, making it easy to construct mock responses. You can
add a delay to the response with a delay
attribute, which should be an integer indicating the number of milliseconds
to delay
You may embed simple expressions in the template with the ${}
syntax.
Note that this should only be used for demos and is in no way guaranteed to work for long periods of time as it will always be grabbing the latest versions htmx and hyperscript!
Here is an example of the code in action:
<!-- load demo environment -->
<script src="https://demo.htmx.org"></script>
<!-- post to /foo -->
<button hx-post="/foo" hx-target="#result">
Count Up
</button>
<output id="result"></output>
<!-- respond to /foo with some dynamic content in a template tag -->
<script>
globalInt = 0;
</script>
<template url="/foo" delay="500"> <!-- note the url and delay attributes -->
${globalInt++}
</template>
While htmx encourages a hypermedia approach to building web applications, it offers many options for client scripting. Scripting is included in the REST-ful description of web architecture, see: Code-On-Demand. As much as is feasible, we recommend a hypermedia-friendly approach to scripting in your web application:
The primary integration point between htmx and scripting solutions is the events that htmx sends and can respond to. See the SortableJS example in the 3rd Party Javascript section for a good template for integrating a JavaScript library with htmx via events.
Scripting solutions that pair well with htmx include:
We have an entire chapter entitled “Client-Side Scripting” in our book that looks at how scripting can be integrated into your htmx-based application.
hx-on*
AttributesHTML allows the embedding of inline scripts via the onevent
properties,
such as onClick
:
<button onclick="alert('You clicked me!')">
Click Me!
</button>
This feature allows scripting logic to be co-located with the HTML elements the logic applies to, giving good
Locality of Behaviour (LoB). Unfortunately, HTML only allows on*
attributes for a fixed
number of specific DOM events (e.g. onclick
) and
doesn’t provide a generalized mechanism for responding to arbitrary events on elements.
In order to address this shortcoming, htmx offers hx-on*
attributes. These attributes allow
you to respond to any event in a manner that preserves the LoB of the standard on*
properties.
If we wanted to respond to the click
event using an hx-on
attribute, we would write this:
<button hx-on:click="alert('You clicked me!')">
Click Me!
</button>
So, the string hx-on
, followed by a colon (or a dash), then by the name of the event.
For a click
event, of course, we would recommend sticking with the standard onclick
attribute. However, consider an
htmx-powered button that wishes to add a parameter to a request using the htmx:config-request
event. This would not
be possible using a standard on*
property, but it can be done using the hx-on:htmx:config-request
attribute:
<button hx-post="/example"
hx-on:htmx:config-request="event.detail.parameters.example = 'Hello Scripting!'">
Post Me!
</button>
Here the example
parameter is added to the POST
request before it is issued, with the value ‘Hello Scripting!’.
The hx-on*
attributes are a very simple mechanism for generalized embedded scripting. It is not a replacement for more
fully developed front-end scripting solutions such as AlpineJS or hyperscript. It can, however, augment a VanillaJS-based
approach to scripting in your htmx-powered application.
Note that HTML attributes are case insensitive. This means that, unfortunately, events that rely on capitalization/ camel casing, cannot be responded to. If you need to support camel case events we recommend using a more fully functional scripting solution such as AlpineJS or hyperscript. htmx dispatches all its events in both camelCase and in kebab-case for this very reason.
Htmx integrates fairly well with third party libraries. If the library fires events on the DOM, you can use those events to trigger requests from htmx.
A good example of this is the SortableJS demo:
<form class="sortable" hx-post="/items" hx-trigger="end">
<div class="htmx-indicator">Updating...</div>
<div><input type='hidden' name='item' value='1'/>Item 1</div>
<div><input type='hidden' name='item' value='2'/>Item 2</div>
<div><input type='hidden' name='item' value='2'/>Item 3</div>
</form>
With Sortable, as with most javascript libraries, you need to initialize content at some point.
In jquery you might do this like so:
$(document).ready(function() {
var sortables = document.body.querySelectorAll(".sortable");
for (var i = 0; i < sortables.length; i++) {
var sortable = sortables[i];
new Sortable(sortable, {
animation: 150,
ghostClass: 'blue-background-class'
});
}
});
In htmx, you would instead use the htmx.onLoad
function, and you would select only from the newly loaded content,
rather than the entire document:
htmx.onLoad(function(content) {
var sortables = content.querySelectorAll(".sortable");
for (var i = 0; i < sortables.length; i++) {
var sortable = sortables[i];
new Sortable(sortable, {
animation: 150,
ghostClass: 'blue-background-class'
});
}
})
This will ensure that as new content is added to the DOM by htmx, sortable elements are properly initialized.
If javascript adds content to the DOM that has htmx attributes on it, you need to make sure that this content
is initialized with the htmx.process()
function.
For example, if you were to fetch some data and put it into a div using the fetch
API, and that HTML had
htmx attributes in it, you would need to add a call to htmx.process()
like this:
let myDiv = document.getElementById('my-div')
fetch('http://example.com/movies.json')
.then(response => response.text())
.then(data => { myDiv.innerHTML = data; htmx.process(myDiv); } );
Some 3rd party libraries create content from HTML template elements. For instance, Alpine JS uses the x-if
attribute on templates to add content conditionally. Such templates are not initially part of the DOM and,
if they contain htmx attributes, will need a call to htmx.process()
after they are loaded. The following
example uses Alpine’s $watch
function to look for a change of value that would trigger conditional content:
<div x-data="{show_new: false}"
x-init="$watch('show_new', value => {
if (show_new) {
htmx.process(document.querySelector('#new_content'))
}
})">
<button @click = "show_new = !show_new">Toggle New Content</button>
<template x-if="show_new">
<div id="new_content">
<a hx-get="/server/newstuff" href="#">New Clickable</a>
</div>
</template>
</div>
Please see the Web Components Examples page for examples on how to integrate htmx with web components.
htmx works with standard HTTP caching mechanisms out of the box.
If your server adds the
Last-Modified
HTTP response header to the response for a given URL, the browser will automatically add the
If-Modified-Since
request HTTP header to the next requests to the same URL. Be mindful that if
your server can render different content for the same URL depending on some other
headers, you need to use the Vary
response HTTP header. For example, if your server renders the full HTML when the
HX-Request
header is missing or false
, and it renders a fragment of that HTML
when HX-Request: true
, you need to add Vary: HX-Request
. That causes the cache to be
keyed based on a composite of the response URL and the HX-Request
request header —
rather than being based just on the response URL.
If you are unable (or unwilling) to use the Vary
header, you can alternatively set the configuration parameter
getCacheBusterParam
to true
. If this configuration variable is set, htmx will include a cache-busting parameter
in GET
requests that it makes, which will prevent browsers from caching htmx-based and non-htmx based responses
in the same cache slot.
htmx also works with ETag
as expected. Be mindful that if your server can render different content for the same
URL (for example, depending on the value of the HX-Request
header), the server needs
to generate a different ETag
for each content.
htmx allows you to define logic directly in your DOM. This has a number of advantages, the largest being Locality of Behavior, which makes your system easier to understand and maintain.
A concern with this approach, however, is security: since htmx increases the expressiveness of HTML, if a malicious user is able to inject HTML into your application, they can leverage this expressiveness of htmx to malicious ends.
The first rule of HTML-based web development has always been: do not trust input from the user. You should escape all 3rd party, untrusted content that is injected into your site. This is to prevent, among other issues, XSS attacks.
There is extensive documentation on XSS and how to prevent it on the excellent OWASP Website, including a Cross Site Scripting Prevention Cheat Sheet.
The good news is that this is a very old and well understood topic, and the vast majority of server-side templating languages support automatic escaping of content to prevent just such an issue.
That being said, there are times people choose to inject HTML more dangerously, often via some sort of raw()
mechanism in their templating language. This can be done for good reasons, but if the content being injected is coming
from a 3rd party then it must be scrubbed, including removing attributes starting with hx-
and data-hx
, as well as
inline <script>
tags, etc.
If you are injecting raw HTML and doing your own escaping, a best practice is to whitelist the attributes and tags you allow, rather than to blacklist the ones you disallow.
Of course, bugs happen and developers are not perfect, so it is good to have a layered approach to security for your web application, and htmx provides tools to help secure your application as well.
Let’s take a look at them.
hx-disable
The first tool htmx provides to help further secure your application is the hx-disable
attribute. This attribute will prevent processing of all htmx attributes on a given element, and on all elements within
it. So, for example, if you were including raw HTML content in a template (again, this is not recommended!) then you
could place a div around the content with the hx-disable
attribute on it:
<div hx-disable>
<%= raw(user_content) %>
</div>
And htmx will not process any htmx-related attributes or features found in that content. This attribute cannot be
disabled by injecting further content: if an hx-disable
attribute is found anywhere in the parent hierarchy of an
element, it will not be processed by htmx.
hx-history
Another security consideration is htmx history cache. You may have pages that have sensitive data that you do not
want stored in the users localStorage
cache. You can omit a given page from the history cache by including the
hx-history
attribute anywhere on the page, and setting its value to false
.
htmx also provides configuration options related to security:
htmx.config.selfRequestsOnly
- if set to true
, only requests to the same domain as the current document will be allowedhtmx.config.allowScriptTags
- htmx will process <script>
tags found in new content it loads. If you wish to disable
this behavior you can set this configuration variable to false
htmx.config.historyCacheSize
- can be set to 0
to avoid storing any HTML in the localStorage
cachehtmx.config.allowEval
- can be set to false
to disable all features of htmx that rely on eval:
hx-on:
attributeshx-vals
with the js:
prefixhx-headers
with the js:
prefixNote that all features removed by disabling eval()
can be reimplemented using your own custom javascript and the
htmx event model.
If you want to allow requests to some domains beyond the current host, but not leave things totally open, you can
use the htmx:validateUrl
event. This event will have the request URL available in the detail.url
slot, as well
as a sameHost
property.
You can inspect these values and, if the request is not valid, invoke preventDefault()
on the event to prevent the
request from being issued.
document.body.addEventListener('htmx:validateUrl', function (evt) {
// only allow requests to the current server as well as myserver.com
if (!evt.detail.sameHost && evt.detail.url.hostname !== "myserver.com") {
evt.preventDefault();
}
});
Browsers also provide tools for further securing your web application. The most powerful tool available is a Content Security Policy. Using a CSP you can tell the browser to, for example, not issue requests to non-origin hosts, to not evaluate inline script tags, etc.
Here is an example CSP in a meta
tag:
<meta http-equiv="Content-Security-Policy" content="default-src 'self';">
This tells the browser “Only allow connections to the original (source) domain”. This would be redundant with the
htmx.config.selfRequestsOnly
, but a layered approach to security is warranted and, in fact, ideal, when dealing
with application security.
A full discussion of CSPs is beyond the scope of this document, but the MDN Article provide a good jumping off point for exploring this topic.
Htmx has some configuration options that can be accessed either programmatically or declaratively. They are listed below:
Config Variable | Info |
---|---|
htmx.config.historyEnabled | defaults to true , really only useful for testing |
htmx.config.historyCacheSize | defaults to 10 |
htmx.config.refreshOnHistoryMiss | defaults to false , if set to true htmx will issue a full page refresh on history misses rather than use an AJAX request |
htmx.config.defaultSwapStyle | defaults to innerHTML |
htmx.config.defaultSwapDelay | defaults to 0 |
htmx.config.defaultSettleDelay | defaults to 20 |
htmx.config.includeIndicatorStyles | defaults to true (determines if the indicator styles are loaded) |
htmx.config.indicatorClass | defaults to htmx-indicator |
htmx.config.requestClass | defaults to htmx-request |
htmx.config.addedClass | defaults to htmx-added |
htmx.config.settlingClass | defaults to htmx-settling |
htmx.config.swappingClass | defaults to htmx-swapping |
htmx.config.allowEval | defaults to true , can be used to disable htmx’s use of eval for certain features (e.g. trigger filters) |
htmx.config.allowScriptTags | defaults to true , determines if htmx will process script tags found in new content |
htmx.config.inlineScriptNonce | defaults to '' , meaning that no nonce will be added to inline scripts |
htmx.config.attributesToSettle | defaults to ["class", "style", "width", "height"] , the attributes to settle during the settling phase |
htmx.config.inlineStyleNonce | defaults to '' , meaning that no nonce will be added to inline styles |
htmx.config.useTemplateFragments | defaults to false , HTML template tags for parsing content from the server (not IE11 compatible!) |
htmx.config.wsReconnectDelay | defaults to full-jitter |
htmx.config.wsBinaryType | defaults to blob , the type of binary data being received over the WebSocket connection |
htmx.config.disableSelector | defaults to [hx-disable], [data-hx-disable] , htmx will not process elements with this attribute on it or a parent |
htmx.config.withCredentials | defaults to false , allow cross-site Access-Control requests using credentials such as cookies, authorization headers or TLS client certificates |
htmx.config.timeout | defaults to 0, the number of milliseconds a request can take before automatically being terminated |
htmx.config.scrollBehavior | defaults to ‘instant’, the scroll behavior when using the show modifier with hx-swap . The allowed values are instant (scrolling should happen instantly in a single jump), smooth (scrolling should animate smoothly) and auto (scroll behavior is determined by the computed value of scroll-behavior). |
htmx.config.defaultFocusScroll | if the focused element should be scrolled into view, defaults to false and can be overridden using the focus-scroll swap modifier. |
htmx.config.getCacheBusterParam | defaults to false, if set to true htmx will append the target element to the GET request in the format org.htmx.cache-buster=targetElementId |
htmx.config.globalViewTransitions | if set to true , htmx will use the View Transition API when swapping in new content. |
htmx.config.methodsThatUseUrlParams | defaults to ["get", "delete"] , htmx will format requests with these methods by encoding their parameters in the URL, not the request body |
htmx.config.selfRequestsOnly | defaults to true , whether to only allow AJAX requests to the same domain as the current document |
htmx.config.ignoreTitle | defaults to false , if set to true htmx will not update the title of the document when a title tag is found in new content |
htmx.config.disableInheritance | disables attribute inheritance in htmx, which can then be overridden by the hx-inherit attribute |
htmx.config.scrollIntoViewOnBoost | defaults to true , whether or not the target of a boosted element is scrolled into the viewport. If hx-target is omitted on a boosted element, the target defaults to body , causing the page to scroll to the top. |
htmx.config.triggerSpecsCache | defaults to null , the cache to store evaluated trigger specifications into, improving parsing performance at the cost of more memory usage. You may define a simple object to use a never-clearing cache, or implement your own system using a proxy object |
htmx.config.responseHandling | the default Response Handling behavior for response status codes can be configured here to either swap or error |
htmx.config.allowNestedOobSwaps | defaults to true , whether to process OOB swaps on elements that are nested within the main response element. See Nested OOB Swaps. |
You can set them directly in javascript, or you can use a meta
tag:
<meta name="htmx-config" content='{"defaultSwapStyle":"outerHTML"}'>
And that’s it!
Have fun with htmx! You can accomplish quite a bit without writing a lot of code!