CODEWITHSUNDEEP
c++boostboost-asio
Daniel
Daniel

Reputation: 778

Boost asio exits with code 0 for no reason. Setting a breakpoint AFTER the problematic statement solves it

I'm writing a TCP server-client pair with boost asio. It's very simple and synchronous.

The server is supposed to transmit a large amount of binary data through several recursive calls to a function that transmits a packet of data over TCP. The client does the analogue, reading and appending the data through a recursive function that reads incoming packets from the socket.

However, in the middle of receiving this data, most times (around 80%) the client just stops recursion suddenly, always before one of the read calls (shown below). It shouldn't be able to do this, given that there are several other statements and function calls after the recursion.

size_t bytes_transferred = m_socket.read_some(boost::asio::buffer(m_fileReadBuffer, m_fileReadBuffer.size()));

m_fileReadBuffer is a boost::array of char, with size 4096 (although I have tried other buffer formats as well with no success).

There is absolutely no way I can conceive of deducing why this is happening.

Also, it's important to note that the server ALWAYS successfully sends all the data. On top of that, the problem always happens at the very end of transmissions: I can send 8000 bytes and it will exit when around 6000 or 7000 bytes have been transferred, and I can send 8000000 bytes and it will exit when something like 7996000 bytes have been transferred.

I can provide any code necessary, I just have no idea of where the problem could be. Below is the recursive read function on the client:

void TCP_Client::receive_volScan_message()
{
    try
    {
        //If the transfer is complete, exit this loop
        if(m_rollingSum >= (std::streamsize)m_fileSize)
        {
            std::cout << "File transfer complete!\n";
            std::cout << m_fileSize << " "<< m_fileData.size() << "\n\n";               

            return;
        }

        boost::system::error_code error;        

        //Transfer isn't complete, so we read some more        
        size_t bytes_transferred = m_socket.read_some(boost::asio::buffer(m_fileReadBuffer, m_fileReadBuffer.size()));

        std::cout << "Received " << (std::streamsize)bytes_transferred << " bytes\n"; 

        //Copy the bytes_transferred to m_fileData vector. Only copies up to m_fileSize bytes into m_fileData
        if(bytes_transferred+m_rollingSum > m_fileSize) 
        {
            //memcpy(&m_fileData[m_rollingSum], &m_fileReadBuffer, m_fileSize-m_rollingSum);
            m_rollingSum += m_fileSize-m_rollingSum;
        }
        else
        {
           // memcpy(&m_fileData[m_rollingSum], &m_fileReadBuffer, bytes_transferred);
            m_rollingSum += (std::streamsize)bytes_transferred;  
        }                     

        std::cout << "rolling sum: " << m_rollingSum << std::endl;

        this->receive_volScan_message();
    }
    catch(...)
    {
        std::cout << "whoops";
    }                
}

As a suggestion, I have tried changing the recursive loops to for loops on both the client and server. The problem persists, somehow. The only difference is that now instead of exiting 0 before the previously mentioned read_some call, it exits 0 at the end of one of the for loop blocks, just before it starts executing another for loop pass.

EDIT: As it turns out, the error doesn't take place whenever I built the client in debug mode on my IDE.

Upvotes: 0

Views: 490

Answers (1)

Daniel
Daniel

Reputation: 778

I haven't completely understood the problem, however I have managed to fix it entirely.

The root of the issue was that on the client, the boost::asio::read calls made main exit with code 0 if the server messages hadn't arrived yet. That means that a simple

while(m_socket.available() == 0)
{
    ;
}

before all read calls completely prevented the problem. Both in debug and release mode.

This is very strange because as I understand these functions should just block until there is something to read, and even if they encountered errors they should return zero.

I think the debug/release discrepancy happened because the m_readBuffer wasn't initialized to anything whenever the read calls took place. This made the read call return some form of silent error. On debug, uninitialized variables get automatically set to NULL, stealthily fixing my problem.

I have no idea why adding a while loop after the transfer prevented the issue, though. Neither why it normally happened on the end of transfers, after the m_readBuffer had been set and successfully used several times.

On top of that, I have never seen this type of "crash" before, where the program simply exits with code 0 in a random place, with no errors or exceptions thrown.

Upvotes: 0

Related Questions