SDK sockets: recv=-1 BUT WSAGetLastError=0 (???)

[Xanblax]

Hi everyone!

Something weird is happening with an application of mine:
It works OK on any computer but on one (I have nearly one hundred installations made, up and running)
The computer on which it is not running is: WinXP professional 2002 SP2 on AMD Athlon64

Mainly my app is a TCP server.
Each client has a listening thread with a loop and a call to the platform SDK Windows sockets blocking (synchronous) "recv" function (http://msdn2.microsoft.com/en-us/library/ms740121.aspx) -code below.

According to MS documentation when function "recv" returns SOCKET_ERROR (-1), an error can therefore be retrieved with WSAGetLastError.
As you can see, I check WSAGetLastError immediately after returning recv.
This works perfectly in every computer.
But on this very one computer out of 100, several times a day, "recv" returns SOCKET_ERROR (-1), BUT WSAGetLastError returns 0
Sniffing the TCP traffic via Ethereal software I can see that, in fact, the TCP connection is still up: so it is the "recv" function that is telling me a lie.

Has this happened to anyone out there?
Have you find any Windows or Athlon64 known bug regarding this matter (recv=-1 but WSAGetLastError=0)?
Can you think of a way to make my application work on this problematic computer?

Thank you very much!

Ricardo Vázquez.
Madrid, Spain.

int CClientThread::Run()
{
	char msg[maxLen];

	while (!bExit) {
		int nRet = recv(clientSocket, msg, maxLen, 0);
		if (nRet == SOCKET_ERROR && bExit) {
			// Stop closes clientSocket on exit. WSAETIMEDOUT
			err.Format("Stop closes clientSocket on exit");
			g_logSystem.logWarning(0, err);
			break;
		}
		else if (nRet == SOCKET_ERROR)
		{
			int nErr = WSAGetLastError();
			if (nErr == 0)
				{
				err.Format("recv=-1 but WSAGetLastError=0 from client %s on socket %ld.", clientId, clientSocket);
				g_logSystem.logWarning(0, err);
				continue;
				}
			else
				{
				err.Format("Connection lost with client %s on socket %ld. Error code %ld: recv() = SOCKET_ERROR", clientId, clientSocket, nErr);
				g_logSystem.logWarning(0, err);

				closesocket(clientSocket);
				server->DeleteClient(clientSocket);
				cti_RemoveEventListener(this);
				return SOCKET_ERROR;
				}
		}	
		else if (!nRet) {
			// Connection has been gracefully closed.
 		        err.Format("Client %s with socket %ld has closed the connection: recv() = 0 (apparently)", clientId, clientSocket);
			g_logSystem.logNormal(0, err);

			closesocket(clientSocket);
			server->DeleteClient(clientSocket);
			cti_RemoveEventListener(this);
			return 0;
		}
		else {
			msg[nRet] = 0x0;
			err.Format("[Cli %ld] Received: %s", clientSocket, msg);
			g_logSystem.logNormal(4, err);

			CString sMensaje;
			sMensaje = CStrTok(msg, "%");
			sMensaje.TrimRight();
			sMensaje.TrimLeft();
			while (!sMensaje.IsEmpty())
				{
				parse(sMensaje);
				sMensaje = CStrTok("", "%");
				}
		}
	}

	return 0;
}

[MikeAThon]

Your app is multi-threaded. You might be encountering a race condition on the shared variables (like bExit).

Is the Athlon the only dual-core machine your app is running on?

Try changing the "if-then-else" logic so that the first if tests only bExit, and the second "else if" is changed to a new "if".

Mike