c++ - 难以理解协议(protocol)定义

Question

我正在阅读 Postgresql 有线协议(protocol)并且在实现一个与简单查询相关的部分时遇到了问题，其中 spec说:

Query (F)
  Byte1('Q')
    Identifies the message as a simple query.

  Int32
    Length of message contents in bytes, including self.

  String
    The query string itself.

我在 C++ 中想出了以下内容:

(std::__1::vector<unsigned char, std::__1::allocator<unsigned char> >) $0 = size=31 {
  [0] = 'Q'
  [1] = '\0'
  [2] = '\0'
  [3] = '\0'
  [4] = '\x1f'
  [5] = 's'
  [6] = 'e'
  [7] = 'l'
  [8] = 'e'
  [9] = 'c'
  [10] = 't'
  [11] = ' '
  [12] = '*'
  [13] = ' '
  [14] = 'f'
  [15] = 'r'
  [16] = 'o'
  [17] = 'm'
  [18] = ' '
  [19] = 'd'
  [20] = 'e'
  [21] = 'p'
  [22] = 'a'
  [23] = 'r'
  [24] = 't'
  [25] = 'm'
  [26] = 'e'
  [27] = 'n'
  [28] = 't'
  [29] = 's'
  [30] = '\0'
}

但我总是从服务器收到错误:

EJSERRORC08P01Minvalid message formatFpqformat.cL652Rpq_getmsgendETSFATALC08P01Minvalid frontend message type 0Fpostgres.cL484RSocketBackend

在这一点上，我厌倦了在服务器上抛出此 vector 的不同版本以始终获得相同的结果，因此我认为我对规范的解释是错误的。你能用代码告诉我这将如何完成吗？不是生成消息的算法，只是一条有效消息。

Answer 1

因为你不分享你的代码，我不能说你错在哪里。

你的消息很好，除了长度应该是 30 ('\x1e') 而不是 31 ('\x1f') – 根据 the documentation ，“消息类型字节”不计入。

我在 Linux 上用 C 尝试过，它对我有用。

我用 send() 通过 PQsocket() 得到的套接字发送消息，我得到的响应包括:

1. 一个 RowDescription 消息(类型 'T')描述列。

2. 每个返回的列一个 DataRow 消息(类型 'D')。

3. 一个 Close 消息(类型 'C')关闭语句。

4. 一条ReadyForQuery(类型'Z')消息。

作为记录，这里是我使用的 C 代码:

#include <libpq-fe.h>
#include <stdio.h>
#include <string.h>
#include <errno.h>
#include <sys/types.h>
#include <sys/socket.h>

#define BUFSIZE 1000

int main(int argc, char **argv) {
    /* PostgreSQL database connection */
    PGconn *conn;
    /* socket for the database connection */
    int s;
    /* simple select statement to send over the socket */
    unsigned char stmt[] = "Q\0\0\0\030select * from amaaa";
    /* buffer for the server response */
    unsigned char buf[BUFSIZE];
    ssize_t len, i;

    /* connect to the database */
    conn = PQconnectdb("");
    if (conn == NULL)
    {
        fprintf(stderr, "Out of memory connecting to PostgreSQL.\n");
        return 1;
    }
    if (PQstatus(conn) != CONNECTION_OK)
    {
        fprintf(stderr, "%s\n", PQerrorMessage(conn));
        PQfinish(conn);
        return 1;
    }

    /* get the TCP socket that corresponds to the connection */
    if ((s = PQsocket(conn)) <= 0)
    {
        fprintf(stderr, "Connection is not open.\n");
        PQfinish(conn);
        return 1;
    }

    /* send the simple select statement */
    if (-1 == send(s, stmt, sizeof(stmt), 0))
    {
        perror("Error in send(2)");
        PQfinish(conn);
        return 1;
    }

    /*
     * The following is oversimplified.
     * In reality, you'd have to call recv(2) until you have
     * the complete response.
     * But I don't want to bother with analyzing the response
     * and hope that it comes in a single packet.
     */

    /* receive the response */
    if (-1 == (len = recv(s, buf, BUFSIZE, MSG_WAITALL)))
    {
        perror("Error in recv(2)");
        PQfinish(conn);
        return 1;
    }

    /* print the server response */
    for (i=0; i<len; ++i) {
        if (buf[i] >= ' ' && buf[i] <= 'z')
            printf("%02X(%c) ", buf[i], buf[i]);
        else
            printf("%02X ", buf[i]);
    }
    printf("\n");

    /* close the database connection */
    PQfinish(conn);

    return 0;
}