SWIG接口适配问题：Python传递int64类型numpy数组至C++ long函数失败

Hey there! Let's work through why your SWIG setup isn't properly handling that int64 numpy array as a C++ long—I’ve run into similar type-matching headaches before, so here’s what to check and fix:

First, Let’s Clear Up Type Compatibility #

The biggest gotcha here is platform-dependent long sizes:

• On Linux/macOS, long is 64-bit (matches numpy’s int64).
• On Windows, long is only 32-bit—this will cause a type mismatch immediately with int64 arrays.

For cross-platform safety, I’d recommend using int64_t (from <cstdint>) in your C++ code instead of raw long—it guarantees a 64-bit integer everywhere. But if you need to stick with long, adjust based on your OS.

Fix Your SWIG Interface File #

You need to explicitly tell SWIG how to map numpy’s int64 arrays to your C++ type, and enable numpy support. Here’s a corrected example interface file:

%module my_array_module
%{
// Include your C++ header here
#include "array_processor.h"
%}

// Critical: Import numpy's SWIG support and initialize it
%include "numpy.i"
%init %{
import_array(); // Required to handle numpy arrays in SWIG
%}

// Map numpy int64 to C++ long (or int64_t for cross-platform)
%numpy_typemaps(int64_t, NPY_INT64, long)

// If your C++ function uses a pointer + length, apply these typemaps
%apply long* INPLACE_ARRAY1 { long* arr }
%apply int DIM1 { int arr_length }

// Declare your C++ functions to SWIG
%include "array_processor.h"

Example C++ Code #

Let’s say your array_processor.h looks like this (adjust to your actual function):

#include <cstdint>

// Using long for your original use case, or int64_t for cross-platform
void multiply_array(long* arr, int arr_length) {
    for (int i = 0; i < arr_length; ++i) {
        arr[i] *= 2; // Simple demo operation
    }
}

Python Test Code #

Once you compile the SWIG module, call it like this to verify:

import numpy as np
import my_array_module

# Create an int64 numpy array
test_arr = np.array([10, 20, 30, 40], dtype=np.int64)
print("Before processing:", test_arr)

# Call the C++ function
my_array_module.multiply_array(test_arr)
print("After processing:", test_arr)

Common Error Fixes #

1. Windows Type Mismatch:
   If you’re on Windows, swap long in your C++ code and SWIG interface with long long (since Windows long is 32-bit). Update the typemap line to:

   %numpy_typemaps(int64_t, NPY_INT64, long long)
   %apply long long* INPLACE_ARRAY1 { long long* arr }
   

2. Forgot numpy.i Setup:
   Skipping %include "numpy.i" or import_array() will cause SWIG to reject numpy arrays entirely—don’t skip these lines!

3. Using std::vector<long> Instead of Pointers:
   If your C++ function uses a vector instead of a raw pointer, add SWIG’s std_vector support:

   %include "std_vector.i"
   %template(LongVector) std::vector<long>;
   

   You can then pass numpy arrays directly to functions accepting std::vector<long> (SWIG handles the conversion).

4. Compilation Issues:
   When building your SWIG module (e.g., with setup.py), make sure to include numpy’s header directory:

   from setuptools import setup, Extension
   import numpy as np
   
   my_module = Extension('_my_array_module',
                         sources=['my_array_module_wrap.cxx', 'array_processor.cpp'],
                         include_dirs=[np.get_include()])
   
   setup(name='my_array_module',
         ext_modules=[my_module],
         py_modules=["my_array_module"])
   

内容的提问来源于stack exchange，提问作者ElenaZhebel