如何提升::序列化为sqlite :: blob？

Question

我正在开展一项需要多项计划能力的科学项目.在浏览了可用的工具之后,我决定使用Boost库,它为我提供了C++标准库不提供的所需功能,例如日期/时间管理等.

我的项目是一组命令行,它处理来自旧的,自制的,基于文本的纯文本数据库的一堆数据:导入,转换,分析,报告.

现在我达到了我需要持久性的程度.所以我包含了我发现非常有用的boost :: serialization.我能够存储和恢复"中等"数据集(不是那么大但不是那么小),它们大约是(7000,48,15,10)-dataset.

我还使用SQLite C API来存储和管理命令默认值,输出设置和变量元信息(单位,比例,限制).

我想到了一些东西:序列化为blob字段而不是单独的文件.可能有一些我尚未见过的缺点(总是存在),但我认为它可以是一个适合我需要的好解决方案.

我能够将文本序列化为std :: string,所以我可以这样做:没有困难,因为它只使用普通字符.但我想二进制序列化为一个blob.

填写INSERT查询时,如何继续使用标准流？

Answer 1

哈.我之前从未使用过sqlite3 C API.我从来没有写过输出streambuf实现.但看到我将来可能会在c ++代码库中使用sqlite3,我想我会花一些时间与

• http://www.sqlite.org/c3ref/funclist.html

• cppreference http://en.cppreference.com/w/cpp/io/basic_streambuf

  

所以事实证明你可以 为增量IO打开一个blob字段.但是,尽管您可以读取/写入BLOB,但您无法更改大小(通过单独的UPDATE语句除外).

所以,我演示的步骤变为:

1. 将记录插入表中,绑定某个(固定)大小的"零点"
2. 在新插入的记录中打开blob字段
3. 将blob句柄包装在一个自定义blob_buf对象中,该对象派生自std::basic_streambuf<>并可用于std::ostream写入该blob
4. 序列化一些数据到 ostream
5. 红晕
6. 自毁/清理

有用 :)

代码main:

int main()
{
    sqlite3 *db = NULL;
    int rc = sqlite3_open_v2("test.sqlite3", &db, SQLITE_OPEN_READWRITE, NULL);
    if (rc != SQLITE_OK) {
        std::cerr << "database open failed: " << sqlite3_errmsg(db) << "\n";
        exit(255);
    }

    // 1. insert a record into a table, binding a "zero-blob" of a certain (fixed) size
    sqlite3_int64 inserted = InsertRecord(db);

    {
        // 2. open the blob field in the newly inserted record
        // 3. wrap the blob handle in a custom `blob_buf` object that derives from `std::basic_streambuf<>` and can be used with `std::ostream` to write to that blob
        blob_buf buf(OpenBlobByRowId(db, inserted));
        std::ostream writer(&buf); // this stream now writes to the blob!

        // 4. serialize some data into the `ostream`
        auto payload = CanBeSerialized { "hello world", { 1, 2, 3.4, 1e7, -42.42 } };

        boost::archive::text_oarchive oa(writer);
        oa << payload;

#if 0   // used for testing with larger data
        std::ifstream ifs("test.cpp");
        writer << ifs.rdbuf();
#endif

        // 5. flush
        writer.flush();

        // 6. destruct/cleanup 
    }

    sqlite3_close(db);
    // ==7653== HEAP SUMMARY:
    // ==7653==     in use at exit: 0 bytes in 0 blocks
    // ==7653==   total heap usage: 227 allocs, 227 frees, 123,540 bytes allocated
    // ==7653== 
    // ==7653== All heap blocks were freed -- no leaks are possible
}

您将认识到所概述的步骤.

要测试它,假设您创建一个新的sqlite数据库:

sqlite3 test.sqlite3 <<< "CREATE TABLE DEMO(ID INTEGER PRIMARY KEY AUTOINCREMENT, FILE BLOB);"

现在,一旦运行程序,就可以查询它:

sqlite3 test.sqlite3 <<< "SELECT * FROM DEMO;"
1|22 serialization::archive 10 0 0 11 hello world 5 0 1 2 3.3999999999999999 10000000 -42.420000000000002

如果启用测试代码(放置的数据多于blob_size允许的数据),您将看到blob被截断:

contents truncated at 256 bytes

完整计划

#include <sqlite3.h>
#include <string>
#include <iostream>
#include <ostream>
#include <fstream>
#include <boost/serialization/vector.hpp>
#include <boost/archive/text_oarchive.hpp>

template<typename CharT, typename TraitsT = std::char_traits<CharT> >
class basic_blob_buf : public std::basic_streambuf<CharT, TraitsT> 
{
    sqlite3_blob* _blob; // owned
    int max_blob_size;

    typedef std::basic_streambuf<CharT, TraitsT> base_type;
    enum { BUFSIZE = 10 }; // Block size - tuning?
    char buf[BUFSIZE+1/*for the overflow character*/];

    size_t cur_offset;
    std::ostream debug;

    // no copying
    basic_blob_buf(basic_blob_buf const&)             = delete;
    basic_blob_buf& operator= (basic_blob_buf const&) = delete;
public:
    basic_blob_buf(sqlite3_blob* blob, int max_size = -1) 
        : _blob(blob), 
        max_blob_size(max_size), 
        buf {0}, 
        cur_offset(0),
        // debug(std::cerr.rdbuf()) // or just use `nullptr` to suppress debug output
        debug(nullptr)
    {
        debug.setf(std::ios::unitbuf);
        if (max_blob_size == -1) {
            max_blob_size = sqlite3_blob_bytes(_blob);
            debug << "max_blob_size detected: " << max_blob_size << "\n";
        }
        this->setp(buf, buf + BUFSIZE);
    }

    int overflow (int c = base_type::traits_type::eof())
    {
        auto putpointer = this->pptr();
        if (c!=base_type::traits_type::eof())
        {
            // add the character - even though pptr might be epptr
            *putpointer++ = c;
        }

        if (cur_offset >= size_t(max_blob_size))
            return base_type::traits_type::eof(); // signal failure

        size_t n = std::distance(this->pbase(), putpointer);
        debug << "Overflow " << n << " bytes at " << cur_offset << "\n";
        if (cur_offset+n > size_t(max_blob_size))
        {
            std::cerr << "contents truncated at " << max_blob_size << " bytes\n";
            n = size_t(max_blob_size) - cur_offset;
        }

        if (SQLITE_OK != sqlite3_blob_write(_blob, this->pbase(), n, cur_offset))
        {
            debug << "sqlite3_blob_write reported an error\n";
            return base_type::traits_type::eof(); // signal failure
        }

        cur_offset += n;

        if (this->pptr() > (this->pbase() + n))
        {
            debug << "pending data has not been written";
            return base_type::traits_type::eof(); // signal failure
        }

        // reset buffer
        this->setp(buf, buf + BUFSIZE);

        return base_type::traits_type::not_eof(c);
    }

    int sync()
    {
        return base_type::traits_type::eof() != overflow();
    }

    ~basic_blob_buf() { 
        sqlite3_blob_close(_blob);
    }
};

typedef basic_blob_buf<char> blob_buf;

struct CanBeSerialized
{
    std::string sometext;
    std::vector<double> a_vector;

    template<class Archive>
    void serialize(Archive & ar, const unsigned int version)
    {
        ar & boost::serialization::make_nvp("sometext", sometext);
        ar & boost::serialization::make_nvp("a_vector", a_vector);
    }
};

#define MAX_BLOB_SIZE 256

sqlite3_int64 InsertRecord(sqlite3* db)
{
    sqlite3_stmt *stmt = NULL;
    int rc = sqlite3_prepare_v2(db, "INSERT INTO DEMO(ID, FILE) VALUES(NULL, ?)", -1, &stmt, NULL);

    if (rc != SQLITE_OK) {
        std::cerr << "prepare failed: " << sqlite3_errmsg(db) << "\n";
        exit(255);
    } else {
        rc = sqlite3_bind_zeroblob(stmt, 1, MAX_BLOB_SIZE);
        if (rc != SQLITE_OK) {
            std::cerr << "bind_zeroblob failed: " << sqlite3_errmsg(db) << "\n";
            exit(255);
        }
        rc = sqlite3_step(stmt);
        if (rc != SQLITE_DONE)
        {
            std::cerr << "execution failed: " << sqlite3_errmsg(db) << "\n";
            exit(255);
        }
    }
    rc = sqlite3_finalize(stmt);
    if (rc != SQLITE_OK)
    {
        std::cerr << "finalize stmt failed: " << sqlite3_errmsg(db) << "\n";
        exit(255);
    }

    return sqlite3_last_insert_rowid(db);
}

sqlite3_blob* OpenBlobByRowId(sqlite3* db, sqlite3_int64 rowid)
{
    sqlite3_blob* pBlob = NULL;
    int rc = sqlite3_blob_open(db, "main", "DEMO", "FILE", rowid, 1/*rw*/, &pBlob);

    if (rc != SQLITE_OK) {
        std::cerr << "blob_open failed: " << sqlite3_errmsg(db) << "\n";
        exit(255);
    }
    return pBlob;
}

int main()
{
    sqlite3 *db = NULL;
    int rc = sqlite3_open_v2("test.sqlite3", &db, SQLITE_OPEN_READWRITE, NULL);
    if (rc != SQLITE_OK) {
        std::cerr << "database open failed: " << sqlite3_errmsg(db) << "\n";
        exit(255);
    }

    // 1. insert a record into a table, binding a "zero-blob" of a certain (fixed) size
    sqlite3_int64 inserted = InsertRecord(db);

    {
        // 2. open the blob field in the newly inserted record
        // 3. wrap the blob handle in a custom `blob_buf` object that derives from `std::basic_streambuf<>` and can be used with `std::ostream` to write to that blob
        blob_buf buf(OpenBlobByRowId(db, inserted));
        std::ostream writer(&buf); // this stream now writes to the blob!

        // 4. serialize some data into the `ostream`
        auto payload = CanBeSerialized { "hello world", { 1, 2, 3.4, 1e7, -42.42 } };

        boost::archive::text_oarchive oa(writer);
        oa << payload;

#if 0   // used for testing with larger data
        std::ifstream ifs("test.cpp");
        writer << ifs.rdbuf();
#endif

        // 5. flush
        writer.flush();

        // 6. destruct/cleanup 
    }

    sqlite3_close(db);
}

PS.我一直在处理错误...非常粗暴.您将要引入一个帮助函数来检查sqlite3错误代码并转换为异常.:)