Marten van Kerkwijk

Indeed, it is the `try` part that goes from being cheap to being essentially free, so it pays to put the common path inside a `try/except` and have all the...

@taldcroft - I had only a quick look but think this approach makes the most sense. In principle, it should be possible to use the `np.array2string` machinery, but I don't...

No, I think it should work. I mostly wondered about creating the smaller array, but thinking more, with `info._new_like` this really is not that difficult!

@deepchatterjeeligo - sounds like a plan!

Yes, definitely what's-new worthy!

It is slightly worse than that (and I can say that as I wrote the code...), `u.mag` is not the same as `u.mag()` - only the latter is a proper...

I haven't looked in detail, but echo @eerovaher's point that it is not obvious this is needed, especially if we go the route of subclassing metaclasses instead of having methods...

@byrdie - one has to take real care with rounding errors for `Time`. The suggestion of doing `jd1` and `jd2` separately is good, but the mean of the former will...

@maxnoe - Thanks for that link! For addition and subtraction, what we have is most similar to the first entry in the link (`msum`). Need to think a bit about...

@byrdie - I have a PR in the works which will let `Time` use our own `Masked` internally, which will remove this limitation. But for now, I think your suggestion...