Python写入np.float64二进制文件，C++以char读零值为何非零？

我用Python生成np.float64类型的数据并写入二进制文件，用C++读取时，以double格式能得到原始数据，但以char格式读取时，原数据中的零值却显示大量非零值。这是什么原因？操作有误吗？

Python写入代码（原版本） #

import numpy as np
a = np.array([1,3,4,5,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,5,4,3,1], dtype=np.float64)
with open("temp.txt", "wb") as binary_file:
    # Write bytes to file
    for values in a:
       binary_file.write(values)

C++读取代码（原版本） #

#include <fstream>
#include <iterator>
#include <iostream>
#include "compression.h"
#include <vector>
#include <iomanip>
#include <chrono>
using namespace std;
int main(int argc, char ** argv)
{
    if (argc != 2) {
        std::cerr << "Need filename" << std::endl;
        return -1;
    }
    
    // enable check against decompress
    bool CHECK_ERR = true;
    std::string file(argv[1]);
    std::ifstream input(file, std::ios::in | std::ios::binary | std::ios::ate);
    std::streampos size;
    size = input.tellg();
    input.seekg(0, std::ios::beg);
      
    // Reading data as double 
    std::vector<double> fileData(size/sizeof(double));
    input.read((char*) &fileData[0], size);
    // This will generate the correct values, which I need
    for (int i=0; i<fileData.size(); i++ )
     cout<< fileData.at(i)<<" ";
    cout << endl;
    
    // Reading data in character format
    std::ifstream inputChr(file, std::ios::in | std::ios::binary | std::ios::ate);
   inputChr.seekg(0, std::ios::beg);
    while(size >= 64)
    {
      char memblock[64];
      inputChr.read((char*) &memblock, 64);
      cout << "Character values "<< endl;
      for (int i=0; i<64; i++)
          cout  << +(memblock[i]) << "  ";
      cout << endl;
      size--;
    }
return 0;
}

原因分析与解决方案 #

1. Python写入代码的核心错误 #

binary_file.write(values)的写法是无效的，file.write()要求传入字节对象，而直接传入np.float64标量会触发类型错误。错误的写入方式会导致文件内容并非正确的double二进制表示，这是读取异常的根源之一。

正确的写入方式应该将numpy标量转换为字节流，或者直接写入整个数组：

import numpy as np
a = np.array([1,3,4,5,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,5,4,3,1], dtype=np.float64)
with open("temp.bin", "wb") as binary_file:
    # 高效写法：一次性写入整个数组的字节流
    a.tofile(binary_file)
    # 或循环写入每个元素：
    # for val in a:
    #     binary_file.write(val.tobytes())

2. C++读取代码的循环逻辑错误 #

原代码中size--的逻辑完全错误：每次读取64字节后，剩余字节数应该减去64，而非1。当前逻辑会导致循环执行数百次，反复读取文件内容，甚至读取文件末尾外的操作系统垃圾数据，这些垃圾值就是你看到的“非零值”。

修正后的C++读取逻辑：

#include <fstream>
#include <iterator>
#include <iostream>
#include "compression.h"
#include <vector>
#include <iomanip>
#include <chrono>
using namespace std;
int main(int argc, char ** argv)
{
    if (argc != 2) {
        std::cerr << "Need filename" << std::endl;
        return -1;
    }
    
    bool CHECK_ERR = true;
    std::string file(argv[1]);
    std::ifstream input(file, std::ios::in | std::ios::binary | std::ios::ate);
    std::streampos totalSize = input.tellg();
    input.seekg(0, std::ios::beg);
      
    // 以double读取（原逻辑正确，保留）
    std::vector<double> fileData(totalSize/sizeof(double));
    input.read(reinterpret_cast<char*>(&fileData[0]), totalSize);
    for (double val : fileData)
        cout << val << " ";
    cout << endl;
    
    // 以char读取，修正循环逻辑
    std::ifstream inputChr(file, std::ios::in | std::ios::binary);
    std::streampos remainingSize = totalSize;
    inputChr.seekg(0, std::ios::beg);
    
    while(remainingSize >= 64)
    {
        char memblock[64];
        inputChr.read(memblock, 64);
        cout << "Character values:" << endl;
        for (int i=0; i<64; i++)
            cout << +static_cast<signed char>(memblock[i]) << "  ";
        cout << endl;
        remainingSize -= 64;
    }
    
    // 处理剩余不足64字节的部分
    if (remainingSize > 0)
    {
        char memblock[64];
        inputChr.read(memblock, remainingSize);
        cout << "Remaining character values:" << endl;
        for (int i=0; i<remainingSize; i++)
            cout << +static_cast<signed char>(memblock[i]) << "  ";
        cout << endl;
    }
    return 0;
}

3. 浮点数二进制存储的补充说明 #

np.float64对应C++的double类型，占8字节。IEEE 754标准中，数值0.0的double二进制表示是全8字节0，修正代码后，以char读取零值时会显示连续8个0，符合预期。

内容的提问来源于stack exchange，提问作者nivedita