@taldcroft - not sure this is the way to go for #3011 (will comment there about that), but in principle I don't mind the idea of an insert method. But for Quantity specifically, I would make it as close as possible to np.insert, and indeed use that function, which could be as short as:

def insert(self, obj, values, axis=None):
    out_array = np.insert(self.value, obj, self._to_own_unit(values), axis)
    return self._new_view(out_array)

p.s. EDITED to correct above. It does work:

from astropy.units import Quantity; from astropy.time import TimeDelta
def insert(self, obj, values, axis=None):
    out_array = np.insert(self.value, obj, self._to_own_unit(values), axis)
    return self._new_view(out_array)

q = u.Quantity([2., 3., 4.], 's')
insert(q, 1, 5*u.ms)
# -> <Quantity [ 2.   , 0.005, 3.   , 4.   ] s>

D'oohh! Yes, that is much better, thanks.

This was driven by thinking about getting add_row working for #3011. I'm interested to hear any suggestions before I tackle this.

@mhvk - implemented your idea and added tests. This is ready for final review (assuming that travis passes).

Apart from my small comment, this looks good. Indeed, better than good -- nice to see the test that checks that conversion of float to int leads to an exception!

@mhvk - it turns out there are more issues related to older versions of numpy and integer Quantities. In particular:

>>> q = u.Quantity([1, 2], unit='m', dtype=int)
>>> q2 = q.to(u.km)
>>> q2.dtype
dtype('float64')

I think this is reasonable behavior because in general unit conversions can turn things into float. But this made it kind of silly/annoying to write tests that start from integer quantities.

@taldcroft - the build failure seems fake -- I restarted it.

Me too, maybe at the exact same moment...