Hi!

I have a few questions about select function, i wrote this code:

void TCPSerwer::sel()
    {
        struct timeval tv = {1, 0};
        fd_set temp_list = m_RecvList;
        //if(select(m_fdmax + 1, &temp_list, NULL, NULL, &tv) == SOCKET_ERROR)
        if(select(0, &temp_list, NULL, NULL, &tv) == SOCKET_ERROR)
        {
            perror( "select" );
            exit( 1 );
        }
        
        for(int i = 0; i <= m_fdmax; i++ )
        {
            if(FD_ISSET(i, &temp_list))
            {
                // New connection
                if(i == m_ListenSocket)
                {
                    acceptClient();
                }

                // Data from client
                else
                {
                    PacketHeader header;
                    int nbytes = recv(i, (char*)(&header), sizeof(PacketHeader),

                    // Error
                    if(nbytes < 0)
                    {
                        disconnectClient(i);
                    }
                    // success
                    else
                    {
                        std::cout << "type: " << header.type << "   len: " << header.length << std::endl;
                    }
                }
            }
        }
    }

I can give first arg to select function and i can don't do that, but why ? Why a should give first arg to select ? m_fdmax is highest number of socket, but this code working without this arg.

Next question is, why select need timeout ? When i don't give this arg select marks all socket as socket that can be readable but select doing this when socket haven't any data to read. When i give this arg i don't have this problem. But why ?

if m_fdmax is highest number of socket, i have to find next highest number of socket when i close connection, Right ? And i should doing this that:

int size = m_ClientVector.size();
for(int i = 0; i < size; i++)
{
       if(m_ClientVector[i] > m_fdmax)
               m_fdmax = m_ClientVector[i];
}

Basically it checks all of the fd values from zero up to the number you provide. In winsock I think it actually ignores it and just checks everything.

Your iteration method is correct there, but rather than using the argument (m_fdmax + 1), I'd just add 1 to fd_max after finding the higher value. You only need to recalculate it when there's a change or potential change in the maximum value (in other words, opening a new descriptor).

On UNIX, the max-arg matters, because UNIX file descriptors are small integers, and the argument tells the system the number of bits in the fd_set. On Windows, sockets are handles with arbitrary pointer values, so that strategy doesn't work, and thus Windows ignores the argument and defines fd_set differently.

When select times out, it returns an error, and doesn't modify the file descriptor sets you pass in.

Your call to recv() is dangerous, because you're not certain the client has actually sent the full header amount of data. If there is only one byte to return, recv() will return only one byte, not all the bytes you ask for. You need to recv() data into some per-client buffer, and only start to look for packets once that buffer is big enough to contain at least the header size and the minimum packet size. You may still find you don't have enough data, and have to wait until you get more data into the buffer to actually have a complete packet. Only when a full packet has been received and decoded, should it be removed from the buffer, and there may be more data, start of the next packet, after that packet, so don't just clear the buffer!