Reputation: 4274
I am using a conditional variable to stop a thread until another thread has completed processing it's task queue (long story). So, on one thread I lock and wait:
boost::mutex::scoped_lock lock(m_mutex);
m_condition.wait(lock);
Once the other thread has completed it's tasks, it signals the waiting thread as follows:
boost::mutex::scoped_lock lock(m_parent.m_mutex);
m_parent.m_condition.notify_one();
The problem I am seeing is that the waiting thread does not stop waiting unless I set a breakpoint on the instructions following it (I am using xcode, fyi). Yes, this seems strange. Does anyone know why this might be happening? Am I mis-using the condition variable?
Upvotes: 26
Views: 51069
Reputation: 1517
I had implemented an example that illustrates how to use boost condition, based in the discussion.
#include <iostream>
#include <boost/asio.hpp>
#include <boost/thread/mutex.hpp>
#include <boost/thread/thread.hpp>
boost::mutex io_mutex;
bool worker_is_done = false;
boost::condition_variable condition;
void workFunction()
{
std::cout << "Waiting a little..." << std::endl;
boost::this_thread::sleep(boost::posix_time::seconds(1));
worker_is_done = true;
std::cout << "Notifying condition..." << std::endl;
condition.notify_one();
std::cout << "Waiting a little more..." << std::endl;
boost::this_thread::sleep(boost::posix_time::seconds(1));
}
int main()
{
boost::mutex::scoped_lock lock(io_mutex);
boost::thread workThread(&workFunction);
while (!worker_is_done) condition.wait(lock);
std::cout << "Condition notified." << std::endl;
workThread.join();
std::cout << "Thread finished." << std::endl;
return 0;
}
Boost condition variable example
Upvotes: 2
Reputation: 3403
Yes, you are misusing the condition variable. "Condition variables" are really just the signaling mechanism. You also need to be testing a condition. In your case what might be happening is that the thread that is calling notify_one()
actually completes before the thread that calls wait()
even starts. (Or at least, the notify_one()
call is happening before the wait()
call.) This is called a "missed wakeup."
The solution is to actually have a variable which contains the condition you care about:
bool worker_is_done=false;
boost::mutex::scoped_lock lock(m_mutex);
while (!worker_is_done) m_condition.wait(lock);
and
boost::mutex::scoped_lock lock(m_mutex);
worker_is_done = true;
m_condition.notify_one();
If worker_is_done==true
before the other thread even starts waiting then you'll just fall right through the while loop without ever calling wait()
.
This pattern is so common that I'd almost go so far as to say that if you don't have a while
loop wrapping your condition_variable.wait()
then you always have a bug. In fact, when C++11 adopted something similar to the boost::condtion_variable they added a new kind of wait() that takes a predicate lambda expression (essentially it does the while
loop for you):
std::condition_variable cv;
std::mutex m;
bool worker_is_done=false;
std::unique_lock<std::mutex> lk(m);
cv.wait(lk, []{return worker_is_done;});
Upvotes: 44