Thanks! I just pushed this to my staging branch to see if the tests pass on the Mac Jenkins instance now.

Out of curiosity, isn't this equivalent to just setting ACCURACY_DECIMAL to 0? (Either way is of course fine.)

Basically, the same kind of idea but not exactly (or at least I don't think so by looking at Numpy docs; feel free to correct me).

The math for numpy.testing.assert_almost_equal is like this:

abs(desired - actual) < 0.5 * 10**(-decimal)

I am doing something like this:

round(desired) - round(actual) == 0

Ah, you're right, not the same then. Thanks for the clarification!

And this looks fine to me.

@astrofrog, if things look good on Mac Jenkins, feel free to merge when you are ready.

There was a delay in running the staging branch. I'm busy all day tomorrow, but will check on Friday.

@astrofrog , I don't think I am making anymore changes unless something else pops up. Rules are pretty relaxed now. Feel free to merge as you see fit. Thanks!

This looks ready to merge to me, but I'll leave it to @astrofrog, as I'm not sure if it's been tested on the Jenkins Mac setup yet (I don't see it on shiningpanda, anyway).

Sorry for the delay - I'll push this to my staging branch to see if there are any issues.

I'm still seeing problems related to timer testing showing up in the Travis tests, e.g. https://travis-ci.org/astropy/astropy/builds/5259347.

This looks like an older version of the code. ACCURACY_DECIMAL not longer exists. Are you sure you grabbed the latest version of this pull request?

=================================== FAILURES ===================================
1056______________________ TestRunTimePredictor.test_fitting _______________________
1057
1058self = <astropy.utils.tests.test_timer.TestRunTimePredictor object at 0x7f887403d250>
1059
1060 def test_fitting(self):
1061 a = self.p.do_fit()
1062 assert self.p._power == 1
1063> np.testing.assert_almost_equal(a, (1, 0), ACCURACY_DECIMAL)

Sorry, I wasn't totally clear: this is coming from astropy master, not this PR. It showed up after merging PR #829. So if this PR will fix the issue then all is OK.

@taldcroft , this PR will change the assertion tests (which currently failed in master) to the following, which will most certainly pass unless timer is behaving very badly:

assert 0.9 <= 1.00415088 <= 1.1
assert -1 <= -0.00102126 <= 1

The tests are passing on Jenkins OS X - merging!