[zeromq-dev] Mac OS X: test_shutdown_stress sometimes fails

Dhammika Pathirana dhammika at gmail.com
Mon Nov 22 09:02:59 CET 2010

On Sun, Nov 21, 2010 at 7:32 AM, Martin Sustrik <sustrik at 250bpm.com> wrote:
> On 11/21/2010 03:56 PM, Dhammika Pathirana wrote:
>> if (!plugged) is dereferencing this->plugged, but we've deleted this
>> object (engine).
>> So it's going to access deleted memory.
> Right.
>> Different option is to move engine deletion to a proxy object or to a
>> static method.
> Or bubble up the "I am deleted" flag via a return value...
> Actually, there are several places like this in 0MQ codebase. I guess the
> only real solution would be to make the whole thing event-driven.
> Right now individual objects communicate via events
> (send_command/process_command) while the internal state machine of each
> object is implemented using standard function calls.
> So, for example, when engine decides it wants to deallocate itself it can
> trigger a "delete" event on itself and finish what it does decently. Later
> on, when "delete" event is processed, the engine will be physically
> deallocated.
> That way we can decouple individual actions of the state machine instead of
> having them mingled as is the case now.

I'm not sure, does this resolve the race condition?
Engine is not attached to a poller, how do we notify events?


More information about the zeromq-dev mailing list