PyArrow Cython封装：Arrow:Table转pyarrow.Table未定义符号错误求助

将pyarrow.Table传入C++的arrow::Table后再传回Python时执行失败，报错信息：

trade.cpython-39-x86_64-linux-gnu.so: undefined symbol: _ZNK5arrow6Status8ToStringB5cxx11Ev

排查后怀疑问题出在pyarrow_wrap_table，但不清楚具体修复方法，相关源码文件如下：

table_func.h #

#ifndef TABLE_FUNC_H
#define TABLE_FUNC_H
#include <vector>
#include <exception>
#include <arrow/api.h>
#include <arrow/result.h>
#include <memory>

namespace MktData
{
    std::shared_ptr<arrow::Table> aggregateTradeTable(const std::shared_ptr<arrow::Table> &);
};
#endif

table_func.cpp #

#include "table_func.h"

std::shared_ptr<arrow::Table> MktData::aggregateTradeTable(const std::shared_ptr<arrow::Table> &trades)
{
    return trades;
}

decl.pxd #

# distutils: language = c++
from pyarrow.lib cimport *
cdef extern from "table_func.h" namespace "MktData":
    cdef shared_ptr[CTable] aggregateTradeTable(shared_ptr[CTable] &);

trade.pyx #

from libcpp cimport bool, vector, memory
from pyarrow.lib cimport *
from decl cimport aggregateTradeTable

def aggregate_trades(obj):
    cdef shared_ptr[CTable] trades = pyarrow_unwrap_table(obj)
    if trades.get() == NULL:
        raise TypeError("not an table")
    cdef shared_ptr[CTable] result = aggregateTradeTable(trades)
    if result.get() == NULL:
        raise TypeError("result incorrect")
    return pyarrow_wrap_table(result)

setup.py #

from setuptools import setup
from Cython.Build import cythonize
from setuptools import Extension, setup
import pyarrow as pa
import numpy as np

sourcefiles = ['trade.pyx', 'table_func.cpp']

extensions = [Extension("trade", sourcefiles)]

ext_modules = cythonize(extensions, gdb_debug=True)
for ext in ext_modules:
    ext.include_dirs.append(pa.get_include())
    ext.include_dirs.append(np.get_include())
    ext.libraries.extend(pa.get_libraries())
    ext.library_dirs.extend(pa.get_library_dirs())

setup(ext_modules=ext_modules, language="c++")

这个错误源于C++编译ABI不匹配（尤其是std::string的cxx11版本）或链接时未正确关联Arrow依赖，按以下步骤修复：

1. 修正setup.py的编译与链接参数 #

确保编译标准、链接参数与pyarrow完全匹配，添加运行时库路径：

from setuptools import setup, Extension
from Cython.Build import cythonize
import pyarrow as pa
import numpy as np

sourcefiles = ['trade.pyx', 'table_func.cpp']

extensions = [
    Extension(
        "trade",
        sourcefiles,
        language='c++',
        extra_compile_args=['-std=c++17'],  # 匹配pyarrow默认编译标准
        extra_link_args=pa.get_link_args()  # 直接复用pyarrow的完整链接参数
    )
]

ext_modules = cythonize(extensions, gdb_debug=True)
for ext in ext_modules:
    ext.include_dirs.append(pa.get_include())
    ext.include_dirs.append(np.get_include())
    ext.libraries.extend(pa.get_libraries())
    ext.library_dirs.extend(pa.get_library_dirs())
    # 添加rpath确保运行时能找到Arrow动态库
    ext.extra_link_args.append(f'-Wl,-rpath,{pa.get_library_dirs()[0]}')

setup(ext_modules=ext_modules)

2. 修正Cython声明的函数参数签名 #

在decl.pxd中匹配C++函数的const引用参数：

# distutils: language = c++
from pyarrow.lib cimport *
cdef extern from "table_func.h" namespace "MktData":
    cdef shared_ptr[CTable] aggregateTradeTable(const shared_ptr[CTable] &);

3. 重新编译扩展 #

清理旧编译产物后重新构建：

rm -rf build/ trade.cpython-39-x86_64-linux-gnu.so
python setup.py build_ext --inplace

内容的提问来源于stack exchange，提问作者Rehon