Columnar vectors / NumPy difference

[wolfv]

This:

>>> b
array([[ 1.],
       [ 1.],
       [ 1.]])
>>> b[0] = 10
>>> b
array([[ 10.],
       [  1.],
       [  1.]])
>>> b[2] = 100
>>> b
array([[  10.],
       [   1.],
       [ 100.]])

works in numpy, but not in xtensor (because the stride of the first dimension is 0). We should either fix this, or document it somewhere.

[JohanMabille]

I think this is the intended behavior, b[i] is equivalent to b(i), and the behavior when there are less arguments than the number of dimension is to prepend them with 0.
This is required so that (a+b)(i0, ...., iN) = a(i0, ...., iN) + b(i0, ...., iN) is always true, whatever the shapes of a and b are.

But I agree that we should document it. Maybe something symmetric to the cheatsheet, with the differences between numpy and xtensor.