Re: [Boost-users] Boost serialization : concurrency (was Possibletoappend...)

From: Robert Ramey (ramey_at_[hidden])
Date: 2006-09-17 21:40:34

Hmmm - well that is interesting. I believe I used this idiom in order
to address a problem where global static variables couldn't be guarenteed
to be instantiated in a particular order and this was the fix recommended.

So now what do we do?

Robert Ramey

Daniel F. Savarese wrote:
> In message <eek3v5$hbk$1_at_[hidden]>, "Robert Ramey" writes:
>> So now you know my intention - and I believe that the code reflects
>> that.
>
> Here's what's happening with
>    static oserializer & instantiate(){
>        static oserializer instance;
>        return instance;
>    }
> Thread 1 calls instantiate and is pre-empted in the middle of
> constructing instance the first time control passes through its
> declaration.  Thread 2 then calls instantiate and returns a
> reference to an incompletely constructed instance.  Then thread 2
> bombs in basic_oarchive_impl::save_object the first time it tries to
> dereference the object to invoke a virtual function
> (in this case class_info()) of an incompletely constructed object
> (i.e., no virtual function table).  And now I'm reminded why whenever
> I see a static local variable I have the knee-jerk reaction:
> "not thread-safe."  I had to reeducate myself by exercising the
> debugger and revisiting what the C++ standard has to say about
> the initialization of static storage.
>
> daniel
>
> -#-#-#-#-| Sleep and The Traveller |-#-#-#-#-#-#-#-
> http://www.savarese.org/ In distant lands, I hear the call of my
> home.     #     s a v a r e s e
> Yet my work is not done.  My journey's just begun.-    software
> research -- http://www.sleepandthetraveller.com/          #
> http://www.savarese.com/