Re: [PATCH] Fix to EventQueue 5/7

[Fernando Nasser <fnasser at redhat dot com>]

Thomas Fitzsimmons wrote:
> I have a question though.  The current code allows the following to
> happen:
>
> Thread A calls topEQ.push(newEQ) and topEQ.push executes partially but
> doesn't execute this line:
>
>     newEventQueue.prev = this;
>
> So at this point, newEQ.prev is still null (unless it was already
> initialized before push was called; that shouldn't happen, but if it
> did, the current code could give weird results).  If Thread B calls
> newEQ.pop and executes the lines:
>
>     if (prev == null)
>       throw new EmptyStackException();
>
> then Thread B will get an EmptyStackException, even though the method
> call to push newEQ was issued before newEQ's pop call.  Is that
> acceptable behaviour or should there be more synchronization to ensure
> that push completes on topEQ before pop can be called on newEQ?

In our specific case this will never happen as I made sure that push and 
pop are called from the same thread.  But even in the general case, 
Thread A should not inform anyone about newEQ before push() as this call 
is what activates this EventQueue.  Actually, access to the "current" 
event queue is done through the one returned by getSystemEventQueue() 
and that is an immutable reference (always points to the original 
queue).  This ensures all the proper locks are obtained and in the 
proper order.

Regards,
Fernando

--
Fernando Nasser
Red Hat Canada Ltd.                     E-Mail:  fnasser@redhat.com
2323 Yonge Street, Suite #300
Toronto, Ontario   M4P 2C9