Document loading and load events can be a bit confusing for newcomers. Multiple names for the same basic things, incompletely documented, ambiguously explained and those ever-present browser inconsistencies don’t exactly help. I want to try and remedy this by providing a rundown of document loading lifecycle events.
Document start
The document begins to load; HTML is tokenized and parsed. Scripts are usually downloaded, parsed and executed as they are encountered, and hold up the parsing of the document, unless they have the async
or defer
properties. Async and deferred scripts are given to another thread to handle whilst the ‘main’ thread goes through the HTML and all other scripts in source order. Newer browsers also perform speculative parsing, racing ahead in the HTML to see if there are any resources it can start downloading whilst the current script is being obtained and run.
Once the HTML is parsed the DOM has been built, the browser waits for all deferred scripts to finish downloading, then executes them in their source order.
As they’ll be executing against a document structure that isn’t ready, it’s best for any scripts running in this phase to just perform data setup and attach callbacks to events that will fire when the document is more complete, either window.onLoad or, more commonly, DOMContentLoaded.
DOMContentLoaded, aka ‘jQuery ready()’
Once the HTML is parsed, the document tree is built, and all non-async scripts have been executed (including scripts with defer
[note 1]), the browser will typically fire the first lifecycle event, DOMContentLoaded. This can happen before any resources (images, iframes, stylesheets, fonts, frames) are loaded. Most developers know this event through jQuery’s ready()
, which lets you attach callbacks to be executed once the document is ready to read and manipulate.
DOMContentLoaded is a relatively recent standard, however, and as you might expect, legacy IE doesn’t implement it. Internet Explorer 5-8 set document.readyState="interactive"
and fire readyStateChange
. The timing isn’t perfect, though, and can sometimes fire late, so jQuery uses a trick where it constantly attempts to scroll the document until that action no longer returns an exception - at this point, we know IE’s DOM is stable. You can check this out in the appendix below, showing how ready()
does its stuff.
There also seems to be a bug in Firefox currently where DOMContentLoaded can fire before all scripts with defer
have been run. Hopefully it should be fixed soon.
When the event fires, its callback gets executed as you’d expect. If a lot of JS is attached, this can take a few hundred milliseconds, but threads downloading async scripts and images will continue doing their thing whilst this happens (though JavaScript won’t be run until the callback finishes). If you’re interested in timings, WebPageTest exposes it with the domContentLoadedEventStart
and domContentLoadedEventEnd
metrics.
Resources loaded in this phase will fire their own load
events as they pop into place.
window.onLoad, aka ‘body.onLoad’
Once all resources are downloaded (including async scripts), we fire window.onLoad [note 2]. This is the granddaddy of load events, and was implemented as far back as the Netscape days. It’s implemented pretty consistently across browsers, and is a fairly common metric for “loadedness” (because the UI’s responsive, the images are loaded, and the browser busy indicator usually switches off). Note that ‘all resources’ doesn’t include AJAX requests, and that dynamically added resources don’t trigger re-firings of the onLoad event.
Window.onLoad and body.onLoad are aliases for the same event, and will overwrite each other. There seems to be some misinformation circulating that body.onLoad is an alias for DOMContentLoaded, but this is incorrect.
If a dependent resource does not load (e.g. an image request returns a 404), this will not stop the event firing.
Summary
- If you just need the DOM to be in a legal state for your JS to manipulate, you can target DOMContentLoaded rather than window.onLoad. jQuery’s “ready” makes it quite easy to bind a callback to this event, or some suitable alternative. If you don’t want to use jQuery, you can implement a fallback that uses readyStateChange.
- DOMContentLoaded will not wait for async scripts, so stuff attached to this event shouldn’t have any async dependencies. You should be able to rely on dependencies from scripts with ‘defer’, Firefox’s current bugs notwithstanding.
- If you have an async dependency, wait for window.onLoad
- If you’re interested in the loading of a particular element (including a script or stylesheet), you can listen to load events on those elements alone.
Notes
- There is apparently a bug in Firefox where DOMContentLoaded can fire before all scripts with
defer
have been run. - Caveat: Unless we’re IE6-9, in which case it turns out we might fire the onLoad event before async scripts are downloaded. Typical.
Appendix: jQuery’s cross-browser ‘ready’ implementation
1 | // Catch cases where $(document).ready() is called after the |