I am writing a simple tcp client to send and receive single lines of text. The asynchronous operations are handled by std::future in order to faciliate blocking queries with timeouts. Unfortunately, my test application crashes with an access violation when destructing the server object. Here is my code:
TCPClient.hpp
#ifndef __TCPCLIENT_H__
#define __TCPCLIENT_H__
#include <boost/asio.hpp>
#include <boost/asio/use_future.hpp>
#include <memory>
#include <vector>
#include <future>
#include <thread>
#include <chrono>
#include <iostream>
#include <iterator>
using namespace boost::asio;
class TCPClient {
public:
TCPClient();
~TCPClient();
void connect(const std::string& address, const std::string& port);
void disconnect();
std::string sendMessage(const std::string& msg);
private:
boost::asio::io_service ioservice;
boost::asio::io_service::work work;
std::thread t;
std::unique_ptr<boost::asio::ip::tcp::socket> socket;
};
inline TCPClient::TCPClient() : ioservice(), work(ioservice) {
t = std::thread([&]() {
try {
ioservice.run();
}
catch (const boost::system::system_error& e) {
std::cerr << e.what() << std::endl;
}
});
}
inline TCPClient::~TCPClient() {
disconnect();
ioservice.stop();
if (t.joinable()) t.join();
}
inline void TCPClient::connect(const std::string& address, const std::string& port) {
socket.reset(new ip::tcp::socket(ioservice));
ip::tcp::resolver::query query(address, port);
std::future<ip::tcp::resolver::iterator> conn_result = async_connect(*socket, ip::tcp::resolver(ioservice).resolve(query), use_future);
if (conn_result.wait_for(std::chrono::seconds(6)) != std::future_status::timeout) {
conn_result.get(); // throws boost::system::system_error if the operation fails
}
else {
//socket->close();
// throw timeout_error("Timeout");
throw std::exception("timeout");
}
}
inline void TCPClient::disconnect() {
if (socket) {
try {
socket->shutdown(ip::tcp::socket::shutdown_both);
std::cout << "socket points to " << std::addressof(*socket) << std::endl;
socket->close();
}
catch (const boost::system::system_error& e) {
// ignore
std::cerr << "ignored error " << e.what() << std::endl;
}
}
}
inline std::string TCPClient::sendMessage(const std::string& msg) {
auto time_over = std::chrono::system_clock::now() + std::chrono::seconds(4);
/*
// Doesn't affect the error
std::future<size_t> write_fut = boost::asio::async_write(*socket, boost::asio::buffer(msg), boost::asio::use_future);
try {
write_fut.get();
}
catch (const boost::system::system_error& e) {
std::cerr << e.what() << std::endl;
}
*/
boost::asio::streambuf response;
std::future<std::size_t> read_fut = boost::asio::async_read_until(*socket, response, '\n', boost::asio::use_future);
if (read_fut.wait_until(time_over) != std::future_status::timeout) {
std::cout << "read " << read_fut.get() << " bytes" << std::endl;
return std::string(std::istreambuf_iterator<char>(&response), std::istreambuf_iterator<char>());
}
else {
std::cout << "socket points to " << std::addressof(*socket) << std::endl;
throw std::exception("timeout");
}
}
#endif
main.cpp
#include <iostream>
#include "TCPClient.hpp"
int main(int argc, char* argv[]) {
TCPClient client;
try {
client.connect("localhost", "27015");
std::cout << "Response: " << client.sendMessage("Hello!") << std::endl;
}
catch (const boost::system::system_error& e) {
std::cerr << e.what() << std::endl;
}
catch (const std::exception& e) {
std::cerr << e.what() << std::endl;
}
system("pause");
return 0;
}
The output is "timeout" as expected (test server sends no data on purpose), but ioservice.run()
crashes immediately (access violation) after closing the socket in TCPClient::disconnect()
. Am I doing some memory mismanagment here?
Compiler is MSVC 12.0.31101.00 Update 4 (Visual Studio 2013)
recvmsg
is receiving into a buffer (streambuf
) that was freed after throwing the exception inTCPClient::sendMessage
(line 105, end of scope).You forgot to cancel the asynchronous operation (
async_read_until
) started in line 97. Fix it:Or even, just
Same goes for other timeout paths.
The other answer addresses what went wrong.
On a higher level, though, you're using futures, just to immediately await their return.
It struck me that this is actually not asynchrony at all, and you should be able to do:
.stop()
work
andwork.reset()
unique_ptr<socket>
and the lifetime management that came with itfuture<>
, and the.get()
andfuture_status
checking that come with itAll in all, you can do a lot simpler, e.g. using a simple helper function like this:
E.g.
connect(...)
used to be:It now becomes:
Like wise,
sendMessage
becomes:Note these are significantly simpler. Note, also, that correct exception messages are now thrown, depending on the cause of the failures.
Full Demo
Live On Coliru
BONUS
If you want even more convenience, have a generalized callback handler that just raises the exception:
Now, the bodies become even simpler in absense of lambdas:
See the adapted demo: Live On Coliru too