setTimeout(function, delay)docs: https://developer.mozilla.org/en-US/docs/Web/API/WindowOrWorkerGlobalScope/setTimeout
I've always thought that setTimeout would call the function when the delay was up,
maybe a little later,
but never before.
So if I set a timeout for 10ms, I wouldn't be surprised if it took between 11 or 12ms to complete - in some cases even longer.
But while investigating a weird race condition bug today, I found evidence to suggest that it could in fact take around 9ms to complete.
I've boiled the "bug" down to this test case:
function run () {
var start = Date.now()
setTimeout(function () {
var end = Date.now()
var diff = end - start
console.log('diff: %dms', diff)
if (diff < 10) return console.log('Aborting - found <10ms delay!')
run()
}, 10)
}
run()This fails both in the latest Chrome and in latest Node.js.
And before you ask,
yes I've also tested this with using process.hrtime() instead of Date.now():
function hrtime () {
var start = process.hrtime()
setTimeout(function () {
var _diff = process.hrtime(start)
var diff = (_diff[0] * 1e9 + _diff[1]) / 1e6
console.log('diff: %dms', diff)
if (diff < 10) return console.log('Aborting - found <10ms delay!')
hrtime()
}, 10)
}
hrtime()Either this is a simpel off-by-one error and I'm just blind,
or I've completely misunderstood how setTimeout works and it in fact can call the function earlier,
or setTimeout uses a different clock for calculating when it should call the function,
or I have discovered a bug in setTimeout (very unlikely).
Here's an example of running the hrtime test on my laptop:
diff: 10.250009ms
diff: 12.856946ms
diff: 12.830893ms
diff: 11.443824ms
diff: 10.190179ms
diff: 12.186512ms
diff: 12.827462ms
diff: 11.329295ms
diff: 12.825859ms
diff: 12.816031ms
diff: 12.409449ms
diff: 12.689547ms
diff: 11.036313ms
diff: 9.249597ms
Aborting - found <10ms delay!
Sometimes it needs to run through 100+ cycles, sometimes it hits the "bug" within 10 cycles. As always, stuff like this is pretty random.
There were a few discussions around this in Node.js and @apapirovski or @Fishrock123 can likely describe the issue best.