[boost] Re: Boost.threads: interthread exception conveyance

From: Alexander Terekhov (terekhov_at_[hidden])
Date: 2002-08-24 12:00:11

Peter Dimov wrote:
[...]
> (except thread_cancel, which is silently ignored.)

And thread_exit too.

[...]
> A non-void return R implies intent to call "R join()" 

Nah, my intent [including "void R", and accessing thread object via 
"joinable_thread_ptr< R{[, thread]-and/or-[, ...exceptions...]} >" 
smart pointer (in addition to thread_ptr<> and current_thread_ptr<> 
"smarties")]:

A) R* join()

     thread::canceled( jtp->join() ) would yield TRUE if 
     thread was canceled, otherwise result pointer can be 
     safely derefenced ("void R" aside); 

B) R* tryjoin()

     thread::busy( jtp->tryjoin() ) would yield TRUE if 
     thread was still running, 

     thread::canceled( jtp->tryjoin() ) would yield TRUE 
     if thread was canceled, otherwise ("busy" case aside) 
     result pointer can be safely deferenced ("void R" 
     aside); 

C) R* timedjoin(/*timeout*/)

     thread::timedout( jtp->timedjoin(/*timeout*/) ) would 
     yield TRUE if thread was still running by the time of 
     timeout expiration, 

     thread::canceled( jtp->timedjoin(/*timeout*/) ) would 
     yield TRUE if thread was canceled, otherwise ("timedout" 
     case aside) result pointer can be safely deferenced 
     ("void R" aside).

What's wrong with it (the need to communicate exception type 
lists to the "generic" code, aside please)?

> to obtain the thread
> function return value. Threads that are not to be joined should not use
> non-void returns.

Why? Well, I guess, you've NEVER "seen" code like this:

 return_value function();
 .
 .
 .
 function(); // just ignore return value ;-)

regards,
alexander.