Converting null-terminated numpy structured array member to char* for a C callback function

Numba Support: How do I do ...?
1

I’m trying (unsuccessfully) to pass a pointer to a numpy structured array null-terminated string member to a native library function with a signature like “void funcname(const char*)”.

I’ve mocked this in the example below with the cfunc “c_callback”. Is there any relatively elegant way to pass the “x” member to the c_callback?

I’ve mucked about trying to nab the pointer with a bunch of lower-level things I understand even less than Python, usually ending up with errors like

mismatch of argument types: [char x 48]* vs int8*

It seems like this interoperability should be relatively simple, I’m probably just missing something obvious. :frowning:

import numpy as np
from numba import types, njit, cfunc

@cfunc(types.void(types.CPointer(types.int8)))
def c_callback(arg):
    pass

@njit
def make_ptr(array):
    # c_callback(array[0]["x"])  # error: mismatch of argument types: [char x 48] vs int8*
    pass

struct_type = np.dtype([("x", "S48")])
arr = np.empty(1, dtype=struct_type)
arr[0]["x"] = "hello"
print(arr)
make_ptr(arr)
2

if it’s only one value you are trying to pass (not the whole array), why don’t you just pass the value instead of the pointer?

3

I’m not understanding your question clearly… I’m trying to pass a pointer to the first element of the array, as the native function takes a null-terminated string (pointer to int8)

I’m still probably missing your point… Can you elaborate, maybe with a line of pseudocode?

4

sorry, I missed the part about this being a native library with a given signature. I thought you had control over that, please ignore my previous comment.

5

No worries, thanks for responding! I do wish I had control of the native side :slightly_smiling_face: