Do not store a direct reference to Date.now

[lfac-pt]

The adition of _.now made it much harder to use fake timers when testing (e.g., http://sinonjs.org/docs/#clock), because we need to stub Date.now before loading Underscore. This is not practical at all and doesn't allow for using fake timers only in some tests.

This patch adds a wrapper function around Date.now.

I understand that this may incur a small performance penalty, but I think that being able to test code that uses _.now (directly or indirectly) is more important.

[jashkenas]

Thanks, but we shouldn't be changing Underscore code to make it easier to use a particular mocking library. That's would be backwards.

[lfac-pt]

I understand your argument, but this is not for a particular library. I was just giving an example. Storing a direct reference won't allow any clock mocking library to work easily.

[davidchambers]

This isn't quite right, since the LHS || operand will now always evaluate true. We'd lose the feature detection, in other words. We'd need something like this, instead:

_.now = Date.now ? function() { return Date.now(); } :
                   function() { return new Date().getTime(); };

[lfac-pt]

Ops! You are right of course. Anyway, it probably won't matter since the PR is closed (although I can fix that if it gets re-opened).

[davidchambers]

The addition of _.now made it much harder to use fake timers when testing […] because we need to stub Date.now before loading Underscore.

Would another solution be to assign your function to _.now as well as Date.now?

_.now = Date.now = ...;

[lfac-pt]

I fixed the problem on my side by adding:

now : function () {
    return Date.now();
}

to my project mixins (we do not support IE8).

I made the PR because I think it is likely that other people with these kinds of tests will have the same problem and that Underscore shouldn't (by default) get in the way of testing.

Well, at least the issue is documented now!

[eschwartz]

To fix compatibility issues with SinonJS, I created a wrapper around the sinon.useFakeTimers method which stubs out _.now:

define('clock', ['sinon', 'underscore'], function(sinon, _) {
  var now_orig = _.now;
  var clock;

  return {
    useFakeTimers: function() {
      // using date.getTime() instead of Date.now
      // allows the sinon fake timers to do their thing
      _.now = function () { return new Date().getTime(); };
      clock = sinon.useFakeTimers();
    },
    restore: function() {
      clock.restore();

      // restore stubbed out _.now
      _.now = now_orig;
    },
    tick: function(int) {
      clock.tick(int);
    }
  };
});

Then in your tests, you can use your sinon timers (almost) like normal:

// I'm using jasmine, but this should work similarly for any testing framework.
define(['sut', 'clock'], function(SUT, clock) {

  beforeEach(function() {
    clock.useFakeTimers();
  });

  afterEach(function() {
    clock.restore();
  });

  it('should do something after 100ms', function() {
    //...
    clock.tick(100);
    expect(SUT).toHaveDone(something);
  });
});

Now I can update to v1.6.0! 😄

I hope this is helpful to those who are still having issues.

[lo1tuma]

This issue is really annoying. There is no clean way to use a fake clock.

[michaelficarra]

That problem is inherent to JavaScript closures, no?

[lfac-pt]

Hum, how is this issue related with closures?