Strange crash, analyzing in gdb

Question

I have an application and I am analyzing memory crash dumps of this software.

struct GPS_CONNECTION
{
    int sockfd;
    std::string sendbuf, recvbuf;
    struct sockaddr_in remoteaddr;
};
vector <GPS_CONNECTION> GPSC;

--------------------------------
(cut)
--------------------------------

fd_set master, gps_master, read_fds, gps_read_fds, write_fds, gps_write_fds;

for (;;)
{
    /* Clear */
    FD_ZERO(&gps_read_fds);
    FD_ZERO(&gps_write_fds);

    /* read_fds */
    gps_read_fds = gps_master;

    /* write_fds */
    for (int i=0; i < GPSC.size(); i++)
    {
        if (GPSC[i].sendbuf.empty())
        {
            continue;
        }
        FD_SET(GPSC[i].sockfd, &gps_write_fds);
    }

    /* Timeout struct */
    tv.tv_sec = 0;
    tv.tv_usec = 0;

    /* selectuj write */
    if (select(gps_fdmax+1, &gps_read_fds, &gps_write_fds, NULL, &tv) == -1)
    {
        perror("select");
        return 7;
    }

    --------------------------------
    (cut)
    --------------------------------
}

GDB crash dump says software is crashed in line:

443                 if (GPSC[i].sendbuf.empty())

When I was analyzing the variables I saw this:

(gdb) print i
$1 = -1214807923

I don't understand how this value was overwritten? I don't see any stack overflow issue here, can anyone explain the reason of this crash?

This problem appears recurring - once in 2-days, this is a server working 24/7/365.

---

After g++ expanding this code that results:

 for (int i=0; i < GPSC.size(); i++)
 {
     if (GPSC[i].sendbuf.empty())
     {
         continue;
     }
     __asm__ __volatile__ ("btsl %1,%0" : "=m" (((&gps_write_fds)->fds_bits)[((GPSC[i].sockfd) / (8 * sizeof (__fd_mask)))]) : "r" (((int) (GPSC[i].sockfd)) % (8 * sizeof (__fd_mask))) : "cc","memory");
 }

Answer 1

Code will crashing because of multi-threading that i will not consider before, i think that is working in one thread only, my mistake.