Re: [boost] [thread] condvar wait exception

From: vicente.botet (vicente.botet_at_[hidden])
Date: 2008-06-12 14:19:23

Hello,

----- Original Message ----- 
From: "Emil Dotchevski" <emil_at_[hidden]>
To: <boost_at_[hidden]>
Sent: Thursday, June 12, 2008 12:24 AM
Subject: [boost] [thread] condvar wait exception

> But also, I am puzzled why would wait() throw anything but
> boost::thread_interrupted? Let's say I have two threads call wait() on
> a barrier, and one of them throws an error. What am I supposed to do
> with that exception? How can I recover from such a failure?

How can an exception waiting on a barrier throw an exception?

> I am not an expert but I think that if the exception is used to report
> a bug in user code, it would be better to assert() instead of throwing
> an exception.

I think that you can  get what you want as you can configure what 
BOOST_VERIFY do.
    inline void condition_variable::wait(unique_lock<mutex>& m)
    {
        detail::interruption_checker check_for_interruption(&cond);
        BOOST_VERIFY(!pthread_cond_wait(&cond,m.mutex()->native_handle()));
    }

> One final note, Boost Thread doesn't use boost::throw_exception to
> throw exceptions. This means that the exception_ptr functionality in
> Boost Exception is crippled for exceptions emitted by Boost Thread.

How many boost libraries do not use boost::throw_exception?

> This is unfortunate because transporting exception objects between
> threads is a key motivation for boost::exception_ptr; see
> http://svn.boost.org/svn/boost/trunk/libs/exception/doc/cloning.html.

This is also the case for any piece of code, running in a multi-threaded 
environement. I think Boost.Exception must provided a mechanism to manage 
with 3pp exceptions (why not as I have proposed in other post).

At least it will be great if it ensures that every boost exception can be 
transported between threads.

Regards

Vicente