_.defer and process.nextTick under node.js

[matehat]

Really simple improvement to use process.nextTick instead of setTimeout under Node.js. Quick benchmarks show that the incurred overhead of the current implementation (using setTimeout(fn, 0)) varies between 600 µs to 1.2 ms. With the current improvement, it varies between 200 µs to 450 µs.

After a few years in production, that could potentially save us 1 or 2 seconds! :)

[jdalton]

setTimeout(fn, 0), and low ms in general, varies by environment. I know IE 10 and Node v0.10 have setImmediate which may be a better choice or at least covers a few more enviros (also there are shims for setImmediate).

[matehat]

How about this decision flow:

1. If process.nextTick is available, use that
2. Otherwise, if setImmediate is available, use that
3. Otherwise, use setTimeout

[ghost]

Hmm, I'd prefer setImmediate to nextTick—since the latter now executes at the end of the current tick, instead of at the beginning of the next one, it's no longer recursion-safe (so this pattern wouldn't work):

_.defer(function next() {
  // ...
  _.defer(next);
});

[matehat]

setImmediate being available and nextTick being executed at the end of the current tick will be effective when Node.js v0.10 comes out, and when that happens we should definitely use setImmediate. But until v0.10 is out and viable for most people nextTick is the only option better than setTimeout. So how about reversing 1 and 2:

1. If setImmediate is available (latest browsers and Node v0.10), use that
2. Otherwise, if process.nextTick is available (Node pre-v.10), use that
3. Otherwise, use setTimeout (everyone else)

[matehat]

Is there something holding back this PR?

[jashkenas]

Just general slowness. Since the nextTick and setTimeout semantics are now different -- and they can't be guaranteed to work interchangeably, it wouldn't be wise to conflate them, no? Using setTimeout for a recursive call is a pretty common pattern.

[matehat]

I agree. Would it be more acceptable if we just tried to detect if setImmediate is available and use that if it is?

[jashkenas]

Probably not, given the MDN note:

This method is not expected to become standard, and is only implemented by recent builds of
Internet Explorer. It meets resistance both from Gecko (Firefox) and Webkit (Google/Apple)

I think Underscore would be wise to just stay out of it, and let you optimize further if you wish.