在cython中使用迭代器

Question

I am trying to learn cython and I modified the example found here .

#pyx file

import numpy as np
cimport numpy as np
import cython

np.import_array()

def test1(a):
    out = np.empty(a.shape, np.double)

    cdef np.flatiter ita = np.PyArray_IterNew(a)
    cdef np.flatiter ito = np.PyArray_IterNew(out)

    cdef double value

    cdef double i
    i = 0.0

    while np.PyArray_ITER_NOTDONE(ita):

        value = (<double*>np.PyArray_ITER_DATA(ita))[0]

        print(ita)  # for debugging
        print(value,i)  # for debugging

        value = value + i

        (<double*>np.PyArray_ITER_DATA(ito))[0] = value

        i += 1.0

        np.PyArray_ITER_NEXT(ita)
        np.PyArray_ITER_NEXT(ito)

    return out

So I expect the function to add each element of the input array by i , where the value of i increases by one each time the iterator passes to the next element. When I run the function with a=np.arange(10) , the print statement shows something like:

(2.121995791e-314, 0.0)
(<numpy.flatiter object at 0x0000024A2CE27B20>)

(4.2439915824e-314, 1.0)
(<numpy.flatiter object at 0x0000024A2CE27B20>)

(6.365987374e-314, 2.0)
(<numpy.flatiter object at 0x0000024A2CE27B20>)

(8.4879831653e-314, 3.0)
(<numpy.flatiter object at 0x0000024A2CE27B20>)
# rest not shown here

which isn't what I expect as it seems that the iterator is still pointing to the same element of the array, and the value returned by value is essentially zero, but not the values of a (number 0-9). But the iterator does seem to stop at the end as the function terminates successfully and the returned array has the same shape as a .

So my question is:

1. What does the line value = (<double*>np.PyArray_ITER_DATA(ita))[0] actually mean? <double*> declares the pointer type? np.PyArray_ITER_DATA is just what the API says. Then what about [0] ?

2. What did I get wrong and how to modify the code? If a=np.arange(2,8) , then test1(a) should give array([ 2, 4, 6, 8, 10, 12]) .

Answer 1

The problem was that arange gives you an int array by default, which you're then interpreting as if it was a double array (small integers misinterpreted as doubles are often around 1e-300 ish, so this is usually a good clue). Ideally you should make sure you check the dtype in the function.

To answer your further questions:

print(ita) is printing the address of the iterator object, not the data it's currently accessing. Therefore it always shows the same address.

np.PyArray_ITER_DATA(ita) gets a pointer to the address of the current bit of data. Since it's designed as a generic interface for any numpy array we don't know the type so it's a void* . <double*> is you telling Cython that this pointer actually points to a double . [0] looks up the value stored at that pointer