ctypes pointer to numpy array with custom dtype

Question

I have a numpy array with a custom dtype:

a = np.zeros(100, dtype=np.dtype([('one',np.double),('two',np.int)]))
a['one']=np.arange(100)
a['two']=np.arange(100)*-1

I want to create a ctypes pointer that I can pass to a C library. The problem is that the C library expects just a pointer to a double array, the 'one' field.

I tried with: a['one'].ctypes.data_as(ctypes.POINTER(ctypes.c_double))

but it does not work, I believe because the C routine does not know what is the correct stride to go trough the array.

Would you have any suggestion, possibly without copying the array?

Answer 1

You are going to have to copy the data to a contiguous array.

one = np.ascontiguousarray(a['one'])
one.ctypes.data_as(ctypes.POINTER(ctypes.c_double))

• Karl Knechtel: This is because there is no way to tell the C function what the correct stride is, and no way to change the stride to what it's expecting (i.e., contiguous doubles) except to create data that's formatted as it expects.
• Andrea Zonca: I was afraid of that...thank you, in principle if the C function would expect insted a structure of double+int would it work?