[Numpy-discussion] Struct alignment

[Dag Sverre Seljebotn]

I'm looking at fixing up support for NumPy record arrays in Cython, and 
realize my knowledge of NumPy is a bit too limited. Searching the docs 
didn't enlighten me, so:

Does NumPy always use packed alignment for nested dtypes?

In [35]: T
Out[35]: dtype([('f0', '<f4'), ('f1', '|i1'), ('f2', '<i4')])

In [36]: T.itemsize
Out[36]: 9

In [37]: T.alignment
Out[37]: 1

In [48]: np.zeros(3,T).itemsize
Out[48]: 9

Is there a way of making an array with the above dtype but 12 bytes per 
element? Or can you guarantee that NumPy arrays with records are always 
packed?

-- 
Dag Sverre