C++调用含Numpy的Python函数报错：ImportError: cannot import name _remove_dead_weakref

Let's break down why you're hitting this error and how to fix it:

Root Cause #

This error almost always comes down to version incompatibility between Python 2.7 and numpy, or missing proper initialization of the Python environment in your C++ code. Python 2.7 only supports numpy versions up to 1.16.6—any newer numpy release will try to access functions that don't exist in Python 2.7's standard library. Additionally, if you don't explicitly set up the Python path in your C++ code, numpy might not be discovered or initialized correctly.

Step-by-Step Solutions #

1. Downgrade numpy to a Python 2.7-compatible version
   First, replace any incompatible numpy installation with the last official release that supports Python 2.7. Run this command:

   pip2 install numpy==1.16.6 --force-reinstall
   

   This ensures numpy won't try to call unsupported functions from Python 2.7.

2. Properly Initialize the Python Environment in C++
   Your existing code likely only calls Py_Initialize(), but you need to explicitly set the Python path to make numpy and your custom module discoverable. Here's an updated code example:

   #include <Python.h>
   #include <stdio.h>
   
   int main() {
       // Initialize the Python interpreter
       Py_Initialize();
   
       // Set Python path to include numpy's site-packages and your script directory
       // Replace the paths below with your actual system paths for Python 2.7
       const char* python_path = "/usr/lib/python2.7/site-packages:/path/to/your/python/scripts";
       PySys_SetPath(python_path);
   
       // Import your Python module that uses numpy
       PyObject* module = PyImport_ImportModule("your_numpy_module");
       if (!module) {
           PyErr_Print();
           return 1;
       }
   
       // Get the target function from the module
       PyObject* func = PyObject_GetAttrString(module, "your_numpy_function");
       if (!func || !PyCallable_Check(func)) {
           PyErr_Print();
           return 1;
       }
   
       // Prepare arguments and call the function
       PyObject* args = PyTuple_New(2);
       PyTuple_SetItem(args, 0, PyFloat_FromDouble(5.0));
       PyTuple_SetItem(args, 1, PyFloat_FromDouble(10.0));
       PyObject* result = PyObject_CallObject(func, args);
   
       if (result) {
           double output = PyFloat_AsDouble(result);
           printf("Function result: %.2f\n", output);
           Py_DECREF(result);
       } else {
           PyErr_Print();
       }
   
       // Clean up Python objects to avoid leaks
       Py_DECREF(args);
       Py_DECREF(func);
       Py_DECREF(module);
       Py_Finalize();
   
       return 0;
   }
   

3. Verify Your Compilation Command
   Keep using your existing compile command, but double-check that the Python 2.7 headers and library are correctly referenced:

   g++ test.cpp -I/usr/include/python2.7 -lpython2.7 -o test_numpy_call
   

   If you installed numpy via pip2, use pip2 show numpy | grep Location to find the correct site-packages path for the C++ code.

Bonus: A Quick Note About Python 2.7 #

Python 2.7 reached end-of-life in 2020, so if possible, consider migrating your code to Python 3 (which has full, active numpy support). But if you're stuck on Python 2.7 for legacy reasons, the above steps should resolve your error.

内容的提问来源于stack exchange，提问作者Steven T