Ctypes arrays have invalid buffer interface information (on Python 3.1.2):

>>> import ctypes
>>> x = (ctypes.c_double*2)()
>>> y = memoryview(x)
>>> y.shape
(2,)
>>> y.format
'(2)<d'

This implies that the array contains 2 items, each consisting of 2 floats, which is incorrect -- the shape information is duplicated.

A suggested fix is attached.

(From http://projects.scipy.org/numpy/ticket/1699)

Thanks for the patch. I agree with the interpretation of the format string.
One thing is unclear though: Using this interpretation the multi-dimensional 
array notation in format strings only seems useful for pointers to arrays.

The PEP isn't so clear on that, would you agree?


I'm not done reviewing the patch, just a couple of nitpicks:

  - We need a function declaration of _ctypes_alloc_format_string_with_shape()
    in ctypes.h.

  - prefix_len = 32*(ndim+1) + 3: This is surely sufficient, but (ndim+1)
    is not obvious to me. I think we need (20 + 1) * ndim + 3.

  - I'd use "%zd" for Py_ssize_t (I know that in other parts of the
    code "%ld" is used, too).

The array notation is useful for arrays inside structs, such as "T{(4)i(2,3)f}".