Managing io_service in another thread, weird problem with exceptions

I have been struggling over this problem for a wee while in my application and so I have tried to reproduce the problem in a simple code snippet.

I am using Boost 1.44 and mingw32-g++ version 4.4.1. on windows XP

My expected result for the following program is:

>messageservice.exe

[debug]BEFORE EXCEPTION ****
[error]EXCEPTION HANDLED ****
[debug]AFTER TRY/CATCH
>

But in reality it sometimes does this, sometimes crashes verbosely "unhandled win32 exception..." (with Just In Time debugger) , and sometimes it crashes silently

So I am at a loss as to why it does this.

If I remove the try...catch block the program runs just fine.

Here is the code:

#include <iostream>
#include <string>

#include <boost/asio.hpp>
#include <boost/thread.hpp>


#define L_(lvl) std::cout<<"开发者_高级运维\n["<<#lvl<<"]"


void nothingfun() { } /* an empty job */

void threadfun() {
    boost::asio::io_service myIoService; /* thread's own io_service object */

    for(int i=0;i<10000;i++) {
        myIoService.post(&nothingfun);
    }
}


int main()
{
    boost::thread t(&threadfun);


    L_(debug)<<"BEFORE EXCEPTION ****";
    try{
         throw "aaah!";
    } catch(...) {
        L_(error)<<"EXCEPTION HANDLED ****";
    }
    L_(debug)<<"AFTER TRY/CATCH";

    t.join();

    return 0;
}

Is it illegal to manage an io_service instance in a non-main thread? or have I done something blindingly obviously wrong!?

Appreciate any help!

Thanks

Thanks to Matt Gruenke for the solution (http://news.gmane.org/gmane.comp.lib.boost.asio.user)

Apparently having thread safe exceptions is an optional extra in mingw. I needed to specify the -mthreads option when linking

I guess im naive in thinking that this should be an always-on core guarantee!