编译的boost_python扩展无法在python 2.7中导入

姆纳达尼

我正在尝试导入带有Boost的C ++中编码的python扩展。当我在使用cmake编译扩展时遇到一些问题时,我设法将其链接到boost_python27库。然后,我使用pythons distutils将扩展安装到python框架中。

但是,当我尝试导入模块时,出现以下错误:

Traceback (most recent call last):
  File "<stdin>", line 1, in <module>
ImportError: dlopen(./tools.so, 2): Library not loaded: @rpath/libboost_python.dylib
  Referenced from: /Users/DaniBook/CLionProjects/Uebung1/cmake-build-debug/tools.so
  Reason: image not found

我尝试了在互联网上找到的所有内容,包括重新安装boost和使用distutils扩展等进行重新编译,以使其正常工作。有人可以对此提供一些帮助,也可以回答那不祥的形象是什么吗?

工具

#ifndef UEBUNG1_TOOLS_H
#define UEBUNG1_TOOLS_H

#include <string>
#include <vector>
#include <map>
#include <boost/python.hpp>
#include <boost/python/suite/indexing/vector_indexing_suite.hpp>

using namespace std;

vector<string> *product(string alphabet, int repeats);

vector<string> *product(vector<string> pools);

vector<string>* hammdist(string &pattern, int distance);

#endif //UEBUNG1_TOOLS_H

tools.cpp

#include <boost/python.hpp>
#include <boost/python/suite/indexing/vector_indexing_suite.hpp>
#include "tools.h"
#include <vector>
#include <string>
#include <map>

using namespace std;

vector<string> *product(string alphabet, int repeats) {
    //initializing vector
    auto *results = new vector<string>();
    for(auto character : alphabet) {
        string tmpstr;
        tmpstr = character;
        results->push_back(tmpstr);
    }

    //cartesian product generation
    for(int i = 1; i < repeats; i++) {
        vector<string> tmp = *results;
        results->clear();

        //iterating over temporary list adding elements from pool to each contained string
        for(auto &it : tmp) {
            for(auto &character : alphabet) {
                results->push_back(it + character);
            }
        }
    }
    return results;
}

vector<string> product(vector<string> pools) {
    //initializing vector
    auto *results = new vector<string>();
    for(auto character : pools[0]) {
        string tmpstr;
        tmpstr = character;
        results->push_back(tmpstr);
    }

    //removing the first pool container
    pools.erase(pools.begin());

    //cartesian product generation
    for(const auto &pool : pools) {
        vector<string> tmp = *results;
        results->clear();

        //iterating over temporary list adding elements from pool to each contained string
        for(auto &it : tmp) {
            for(auto character : pool) {
                results->push_back(it + character);
            }
        }
    }
    return results;
}


vector<string>* hammdist(string &pattern, int distance) {
    map<char, string> possibles = {
            {'A', "CGT"},
            {'C', "AGT"},
            {'G', "ACT"},
            {'T', "ACG"}
    };
    auto *results = new vector<string>();
    vector<string> *masks = product("01", pattern.size());
    for(auto &mask : *masks) {
        auto *permute = new vector<string>();
        auto *tmp = new vector<string>();
        if(count(mask.begin(), mask.end(), '1') == distance) {
            for(int i = 0; i < pattern.size(); i++) {
                if(mask[i] != '1') {
                    string tmpstr;
                    tmpstr = pattern[i];
                    tmp->push_back(tmpstr);
                }
                else {
                    tmp->push_back(possibles[pattern[i]]);
                }
            }
            permute = product(*tmp);
            results->insert(results->end(), permute->begin(), permute->end());
        }
        delete permute;
        delete tmp;
    }
    return results;
}

/* 
   initializing function pointers in order to tell boost that we have 
   overloaded functions to expose to python
*/
vector<string> *(*product1)(string, int) = &product;
vector<string> *(*product2)(vector<string> pools) = &product;

//include all functions that are used by the function to expose to the BOOST_PYTHON_MODULE call

using namespace boost::python;

BOOST_PYTHON_MODULE(tools) {
    //telling boost_python that we have overloaded functions which need to be called in the respective situations
    //return_value_policy<manage_new_objects> is required in order for the interface to handle the new invocation and the returned pointer

    def("product", product1, return_value_policy<manage_new_object>());
    def("product", product2, return_value_policy<manage_new_object>());

    def("hammdist", hammdist, return_value_policy<manage_new_object>());

    //vector_indexing_suite handles the wrapping of vector member functions
    //enables handling vector in a pythonic way when using in python

    class_<std::vector<string>>("string_vector")
        .def(vector_indexing_suite<std::vector<string>>());
}

CMakeLists.txt

make_minimum_required(VERSION 3.12)
project(tools)

set(CMAKE_CXX_STANDARD 17)
set(CMAKE_INCLUDE_CURRENT_DIR ON)

if(APPLE)
    set(CMAKE_SHARED_LIBRARY_SUFFIX ".so")
endif(APPLE)

find_package(PythonLibs 2.7 REQUIRED)
include_directories(${PYTHON_INCLUDE_DIRS})

set(PROJECT_SOURCE_DIR src/)
include_directories(${PROJECT_SOURCE_DIR})

set(BOOST_ROOT "/Users/DaniBook/miniconda3/pkgs/boost-1.66.0-py27_1")
set(BOOST_LIBRARYDIR "/Users/DaniBook/miniconda3/pkgs/boost-1.66.0-27_1/lib")

find_package(Boost COMPONENTS python REQUIRED)
include_directories(${Boost_INCLUDE_DIRS})

add_library(tools SHARED src/tools.cpp src/tools.h)
target_link_libraries(tools ${Boost_LIBRARIES} ${PYTHON_LIBRARIES})
set_target_properties(tools PROPERTIES PREFIX "")

setup.py

from distutils.core import setup

setup(
    name = 'tools',
    version = '0.1',
    py_modules = ['tools'])
姆纳达尼

好吧,经过一些额外的研究,我得出了一个非常有趣的结论,当您查看错误消息时,这一结论非常明显。

问题不是共享库本身不存在,而是在编译时(链接程序阶段)对boost库的引用的未定义相对路径。这可能是由于使用了安装在miniconda中而不是在pythonpath或其他目录中的/ usr / local / lib中的boost安装。但是,这可以使用otool手动解决install_name_tool(可以在Mac上通过安装XCode开发工具来获取)。

因此,您可以在命令行上执行以下操作:

otools -L ${PATH_TO_PYTHONEXTENSION}/someextension.so

这样列出了终端上的所有库及其系统路径

./build/lib.macosx-10.13-x86_64-2.7/tools.so:
    @rpath/libboost_python.dylib (compatibility version 0.0.0, current version 0.0.0)
    /usr/lib/libc++.1.dylib (compatibility version 1.0.0, current version 400.9.0)
    /usr/lib/libSystem.B.dylib (compatibility version 1.0.0, current version 1252.50.4)

可以很容易地看到,我们在编译时就链接了boost库的缺少路径,可以使用install_name_tool以下方法解决

install_name_tool -change @rpath/libboost_python.dylib ${ACTUAL_PATH_TO_libbost_python.dylib} ${PATH_TO_PYTHONEXTENSION}/someextension.so

setup.py install现在,重新运行可以解决问题,并且可以成功导入扩展:

import tools
dir(tools)

给出以下输出:

['__doc__', '__file__', '__name__', '__package__', 'hammdist', 'product', 'string_vector']

链接到外部资源:制作boost.python helloword演示时不安全地使用相对rpath libboost.dylib吗?

我认为它对于其他基于Unix的操作系统(如Linux)也能发挥相同的作用。希望这可以帮助。

本文收集自互联网,转载请注明来源。

如有侵权,请联系 [email protected] 删除。

编辑于
0

我来说两句

0 条评论
登录 后参与评论

相关文章