[Boost-users] [Asio] Erratic results with io_service + deadline_timer

Subject: [Boost-users] [Asio] Erratic results with io_service + deadline_timer
From: John Zwinck (jzwinck_at_[hidden])
Date: 2008-09-08 19:47:28

Today I tried using Boost.Asio to make a function get called at a particular
time of day.  My work included some tests, which usually passed but
sometimes failed.  I've boiled it down below.  I'm using gcc 4.0.3 and Boost
1.35 on Linux 2.6.22 (Ubuntu Dapper 32-bit i686, quad-core CPU).  I compiled
the following code with g++ -lboost_thread -lboost_date_time -lboost_system
:

#include <boost/system/error_code.hpp>
#include <boost/thread/thread.hpp>
#include <boost/asio/io_service.hpp>
#include <boost/asio/deadline_timer.hpp>
#include <boost/date_time/posix_time/posix_time.hpp>

using namespace boost;
using namespace boost::asio;
using namespace boost::posix_time;

bool g_called = false;
io_service g_svc;

void func(boost::system::error_code ec)
{
    g_called = true;
}

void work()
{
    for(;;)
    {
        g_svc.run();
    }
}

int main()
{
    thread thr(work);

#ifdef FAIL // if defined, causes program to always assert
    this_thread::sleep(milliseconds(1));
#endif

    deadline_timer tmr(g_svc, second_clock::local_time() + seconds(1));
    tmr.async_wait(&func); // expected to wait 1 second then call func

    this_thread::sleep(milliseconds(1500));

    assert(g_called);
}

I think my first issue was that I didn't recognize the need to call
io_service::run in a loop.  The
docs<http://www.boost.org/doc/libs/1_35_0/doc/html/boost_asio/reference/io_service/run.html>don't
make this clear at all, stating that it will "Run the ... event
processing loop," but I figured from reading the source that the loop in my
own code was needed.  Anyway, that didn't fix it.

What I'm expecting here is that my callback function will be invoked about a
second after registration, and it will set the flag which is then checked
half a second later.  When compiled as-is, this (usually) works.  However,
when compiled with -DFAIL, which just adds a tiny sleep in a place I'd think
it would not matter, it fails always.

Surely I'm doing something wrong.  Or is the combination of this (new)
library and my system just not working correctly (something I rather doubt
given my past experience with Boost)?