setTimeout(fn) does not work as expected

[ry]

never get timeout.

[bnoordhuis]

Is this about test/pummel/test-timers.js sometimes failing when the system is under load?

[ry]

i'm not sure what this was about. closing.

[freewil]

I think you meant that when you omit the second param, the timeout never actually occurs, instead of it happening immediately like one might think.

[7fe]

Yup @ER88 is right the function is never called and a error is never produced.

[bnoordhuis]

Right, that's a bug. Reopening.

[bnoordhuis]

@ry or @isaacs - can one of you review 73dcd4f?

[isaacs]

Looks mostly ok to me.

1. What does it mean to supply a timeout less than 0? Should it maybe throw for anything negative?
2. Number.POSITIVE_INFINITY is probably much higher than we can reasonably support. I think it should probably not allow any number greater than Math.pow(2,32) - 1 (the MAX_INT in javascript.)

[bnoordhuis]

@isaacs

1. It makes the callback run ASAP. Not that useful but it maintains backward compatibility.
2. I'm not a fan of arbitrary limits. That check for infinities and NaNs is because they confuse libev no end.

[ry]

we should match web browser semantics here. what is the web browser doing?

[freewil]

I know in the latest version of Firefox, anything equal to or less than zero, the timeout just occurs immediately, same if you omit the second param all together

[ghost]

@isaacs: What MAX_INT do you mean? Afaik, there is no MAX_INT nor Number.MAX_INT and maximal integer number is Math.pow(53)-1 (maximal UInt32 is another matter, of course).

EDIT: The maximal possible Date is enough a limit, I think. In specs, it's year 9999, I think; in v8, it is very near to the maximal integer I mentioned above (in numbers of milliseconds from epoch). If an ending Date may be constructed, ok, if not, problem,

[bnoordhuis]

Firefox 3.6.18

1. -1, -1000, NaN, +Infinity and -Infinity all make the callback run on the next tick.
2. Values up to 2^31-1 schedule the callback in the future, 2^31 and up make it run on the next tick.
3. Passing negative values to setInterval() in node makes it throw an exception so in that respect we're not compliant.

[bnoordhuis]

@ER88 Not immediately, at least not in 3.6.18. It runs on the next tick, a couple of milliseconds later.

[bnoordhuis]

Can someone review 8ad75f9? It's a standalone patch that implements the behaviour observed in FF 3.6.

[tshinnic]

Hello.... this is related to

#1877: Large timeouts for setTimeout stop propagation of event loops

please see comments there.

[isaacs]

Landed on 7fc835a. Thanks!

[tshinnic]

So... does this help/apply to

#1877: Large timeouts for setTimeout stop propagation of event loops

I see the limit to (2**31)-1 , so maybe?

[isaacs]

Yes, this ought to fix #1877 as well. I'll check it out to make sure.