-
-
Save bluejava/9b9542d1da2a164d0456 to your computer and use it in GitHub Desktop.
// See http://www.bluejava.com/4NS/Speed-up-your-Websites-with-a-Faster-setTimeout-using-soon | |
// This is a very fast "asynchronous" flow control - i.e. it yields the thread and executes later, | |
// but not much later. It is far faster and lighter than using setTimeout(fn,0) for yielding threads. | |
// Its also faster than other setImmediate shims, as it uses Mutation Observer and "mainlines" successive | |
// calls internally. | |
// WARNING: This does not yield to the browser UI loop, so by using this repeatedly | |
// you can starve the UI and be unresponsive to the user. | |
// Note: For an even faster version, see https://gist.github.com/bluejava/b3eb39911da03a740727 | |
var soon = (function() { | |
var fq = []; // function queue; | |
function callQueue() | |
{ | |
while(fq.length) // this approach allows new yields to pile on during the execution of these | |
{ | |
var fe = fq[0]; | |
fe.f.apply(fe.m,fe.a) // call our fn with the args and preserve context | |
fq.shift(); // remove element just processed... do this after processing so we don't go 0 and trigger soon again | |
} | |
} | |
// run the callQueue function asyncrhonously, as fast as possible | |
var cqYield = (function() { | |
// This is the fastest way browsers have to yield processing | |
if(typeof MutationObserver !== "undefined") | |
{ | |
// first, create a div not attached to DOM to "observe" | |
var dd = document.createElement("div"); | |
var mo = new MutationObserver(callQueue); | |
mo.observe(dd, { attributes: true }); | |
return function(fn) { dd.setAttribute("a",0); } // trigger callback to | |
} | |
// if No MutationObserver - this is the next best thing - handles Node and MSIE | |
if(typeof setImmediate !== "undefined") | |
return function() { setImmediate(callQueue) } | |
// final fallback - shouldn't be used for much except very old browsers | |
return function() { setTimeout(callQueue,0) } | |
})(); | |
// this is the function that will be assigned to soon | |
// it takes the function to call and examines all arguments | |
return function(fn) { | |
// push the function and any remaining arguments along with context | |
fq.push({f:fn,a:[].slice.apply(arguments).splice(1),m:this}); | |
if(fq.length == 1) // upon adding our first entry, kick off the callback | |
cqYield(); | |
}; | |
})(); |
It's cool.But why not use "process.nextTick"
before the "setImmediate"
? @bluejava
Thanks - I think there are some speed improvements in the version included in Zousan. I would recommend including Zousan and calling Zousan.soon rather than pasting this gist (though you are welcome to of course!) Zousan is tiny and maintained. I've been using it in production for many years - and Zouson.soon for a high performance centralized message bus used heavily in production for a few years as well. If I get a chance I will publish the message bus as well.
Happy to see your comment.😁
I am learning how to create a simple Promise implement,and glad to see your articles on your blog.
Let's keep going.
MutationObserver
is queuing a microtask, as per defined here https://dom.spec.whatwg.org/#mutation-observers
So I think this can be done even faster by just directly using the window.queueMicroTask
? @bluejava lmk if you try this out!
The event loop is defined as such if you are interested and shows why the above works well, awesome read! https://html.spec.whatwg.org/multipage/webappapis.html#event-loop-processing-model
Thanks for this Gabriel - will definitely give it a try!
The effect on performance will likely be extremely limited, since the scheduling of the microtask happens only once per event loop iteration - so if you fire off 1000 soon() calls, only the first one utilizes the MutationObserver to schedule a microtask and the rest pile on..
But it's a much more elegant solution (and Node friendly) so I like it! Thanks for pointing it out.
I've made a couple optimizations which make it a bit faster. This currently can loop a million times in about 2500ms on Firefox v35 on a 5 year old iMac. Removing ability to pass context and arguments can speed it up even faster - I have a mind-numbingly fast version that executes 1 million loops in about 100ms - will post as separate gist.