@taldcroft - now also updated the max, min, etc. Missing still is mean, which needs a bit of thought on how to keep full precision.

@taldcroft - could you have a quick look at the travis failures? I think they occur because Time now has min, max, but am a little unsure...

OK, found the problem. Because I had not defined an out keyword, np.min(t) did not work any more. Now it does again, though it meant adding an out keyword that cannot actually be used.

@taldcroft - and also the sequel to the shape-changing is ready for final review...

@mhvk - can you rebase this to get rid of the redundant commits? I think the commits from #3224 would not appear now except that #3224 was rebased.

@taldcroft - ok, rebased. I must have not quite kept track of all the rebasing!

@mhvk - put in a PR on your branch. Otherwise this looks good to me.

@taldcroft - I looked at your and my argmin code and at first couldn't convince myself either was absolutely correct. But then I remembered my original logic, which was that really one wants to do

<some_guess>
dt = (self - self.__class__(<some_guess>, format='jd')).jd
return dt.argmin()

and looking at __sub__ again, it is clear my original logic should be fine. I now added this as comments to argmin and argmax.

While looking at this again, however, I felt somewhat uncomfortable with argsort, as it really presumes the jd1 and jd2 have gone through day_frac correctly. Of course, this should be the case, but just to be sure, I rewrote it in a way that should be fail-safe for that case too (though a little slower, but since argsort is so slow anyway, I don't think it matters too much).

p.s. Included your test additions with a minor fixup to the way the parametrization was set up.

OK, I'm convinced! I should have remembered itertools.product, good catch.