Convert `int *` to a Python or Numpy object in a Cythonized function

Question

(I think this question can easily be answered by an expert without an actual copy-paste-working-example, so I did not spent extra time on it…)

I have a C++ method, which returns an array of integers:

int* Narf::foo() {
    int bar[10];
    for (int i = 0; i < 10; i++) {
        bar[i] = i;
    }
    return bar;
}

I created the Cython stuff for its class:

cdef extern from "Narf" namespace "narf":
    cdef cppclass Narf:
        Narf() except +
        int* foo()

And these are my Python wrappers:

cdef class PyNarf:
    cdef Narf c_narf
    def __cinit__(self):
        self.c_narf = Narf()
    def foo(self):
        return self.c_narf.foo()

The problem is the foo method, with its int* return type (other methods which I did not list in this example work perfectly fine!). It does not compile and gives the following error:

    def foo(self):
        return self.c_narf.foo()
                             ^
------------------------------------------------------------

narf.pyx:39:37: Cannot convert 'int *' to Python object

Of course, it obviously does not accept int * as return type. How do I solve this problem? Is there an easy way of wrapping this int * into a numpy array (I'd prefer numpy), or how is this supposed to be handled?

I'm also not sure how to handle the memory here, since I'm reading in large files etc.

I should go for an instance variable or maybe even for a pointer pass instead of… — tamasgal, May 24 '16 at 12:40
you should accept the buffer instead: `foo(int*)` i.e., create a numpy array (that will manage the memory for you) and pass it to `foo()`. See [calling c from python with ctypes: passing vectors](http://stackoverflow.com/a/22685515/4279). Otherwise, see [How to convert pointer to c array to python array (`PyArray_SimpleNewFromData()`)](http://stackoverflow.com/q/7543675/4279) — jfs, May 24 '16 at 12:43
I guess the buffer version is the better one, since numpy will manage malloc. — tamasgal, May 24 '16 at 12:45
btw, if you are working with Cython, make sure to study [Typed Memoryviews](http://docs.cython.org/src/userguide/memoryviews.html) — jfs, May 24 '16 at 12:51
Thanks, I just need to understand how I can pass a memory view of a numpy array to the C-function. — tamasgal, May 24 '16 at 12:57
If you mean a Cython function that is declared to accept a memory view then it is handled automatically (if the shape, etc are compatible). — jfs, May 24 '16 at 12:59

score 2 · Answer 1 · answered May 24 '16 at 12:22

2

To wrap it a numpy array, you need to know the size, then you can do it like this:

def foo(self):
    cdef int[::1] view = <int[:self.c_narf.size()]> self.c_narf.foo()
    return np.asarray(view)

The above code assumes that there exists a function self.c_narf.size() that returns the size of the array.

answered May 24 '16 at 12:22

aldanor

3,371
2
26
26

This is basically what I want, now as J.F. Sebastian pointed out, I'm returning a pointer to local data, so even the first call works, I get a segmentation fault in the next calls… – tamasgal May 24 '16 at 12:37
@septi why not return an `std::vector`? – aldanor May 24 '16 at 12:58

score 1 · Answer 2 · edited May 23 '17 at 12:14

1

This looks like it can be solved using the solution to this question: Have pointer to data. Need Numpy array in Fortran order. Want to use Cython

edited May 23 '17 at 12:14

Community

1
1

answered May 24 '16 at 12:13

mobiusklein

1,403
9
12

Convert `int *` to a Python or Numpy object in a Cythonized function

2 Answers2