[boost] Re: Basic guarantee and synchronized multiple updates.

From: Alexander Terekhov (terekhov_at_[hidden])
Date: 2002-09-27 11:37:01

Fredrik Blomqvist wrote:
[...]
> >         vector1.push_back(e1);
> >         try {
> >             vector2.push_back(e2):
> >         }
> >         catch(...) {
> >             vector1.pop_back();
> >             throw;
> >         }
> 
> Isn't this a good example of something scopeguard ...

would also solve... but >>NOT<< having any nasty portability 
problems [and that's just one reason among a couple of others 
to avoid catch(...)-like-above thing] with respect to "safety" 
of thread cancel and thread exit under currently-quite-common-
threaded-C++-approach known as "forced unwinding"-Y'know-
"catch(...)-doesn't-exist" that, unfortunately, IS used by 
numerous C++ implementations out there... like the following 
one: <for example>

http://publibz.boulder.ibm.com:80/cgi-bin/bookmgr_OS390/BOOKS/EDCLB130/3.522.14
("3.522.14 pthread_cancel() -- Cancel a Thread")

http://publibz.boulder.ibm.com:80/cgi-bin/bookmgr_OS390/BOOKS/EDCLB130/3.522.30
("3.522.30 pthread_exit() -- Exit a Thread")

"....
   Standards / Extensions   C or C++   Dependencies 
   POSIX.4a                 both       POSIX(ON) 
 ....
 Special Behavior for C++ 

 Destructors for automatic objects on the stack will 
 be run when a thread is canceled. The stack is unwound 
 and the destructors are run in reverse order. "

<?>

regards,
alexander.

--
http://groups.google.com/groups?selm=3D89CACB.12BB6122%40web.de
("action_on_propagation_of(...) {")