> From: Marc Schier [mailto:[EMAIL PROTECTED]] 
> 
> > From: Berin Loritsch [mailto:[EMAIL PROTECTED]]
> > 
> > However, there are some things that I am curious as to your 
> reasoning. 
> > The version of SandStorm (the SEDA implementation that inspired the 
> > event package) did not have EventHandlerExceptions, and I 
> don't think
> > that they belong.
> 
> I agree. I needed some signalling because all the exceptions 
> in Sandstorm are just print to stdout. I thought I would 
> handle an exceptional condition somewhere else.

I think the signal based system would work best.

> > What should happen is the EventHandler/Stage should enqueue 
> a Signal 
> > that the container can divert to the SignalHandler.  That way we
> > can properly
> > log the event.  As it is now, we would loose information should the
> > EventHandler not properly process or forward the faulty 
> > elements before
> > it throws the exception.  That IMO is a bad thing.
> 
> I agree. Using a signal handler for logging would be a very 
> good way of doing it. 

Not to mention the fact that the signal handler would be able
to do more than *just* log it.  It can do proper notifications
in the event that the signal represents a serious enough issue.


> > Regarding the Stage Queue operations, I think we can clean 
> things up. 
> > If the Stage can get its Sink directly set by the container, we can 
> > set up a multiplexing queue based on the type of element.  However, 
> > that requires
> > the container to have intimate knowledge of the types that 
> come out of
> > the Stage.  I don't think the Stage should want to enqueue things to
> > itself
> > again.  That might cause some other issues such as infinite 
> loops that
> > are
> > hard to detect.  If we keep to the SinkMap format, we should keep it
> > simple.
> > The SinkMap would return the default Sink, and a small collection of
> > named
> > sinks.  That way the Stage can make intelligent decisions 
> as to which
> > Sink
> > it forwards information to.
> 
> Again, I agree, but the sandstorm stages have the notion of 
> enqueuing to their own sink and because I ported the whole 
> thing over, I wanted the stage to get a reference to its own 
> sink.  This is very error prone and probably will lead to 
> really low reuse (But the current stage manager does not very 
> much allow reuse anyway).  It could be handle by designing 
> smaller stages that handle one single event. I would also 
> like to see it go away...

Did you have any talks with Matt regarding his reasoning for
the self queueing stages?


--
To unsubscribe, e-mail:   <mailto:[EMAIL PROTECTED]>
For additional commands, e-mail: <mailto:[EMAIL PROTECTED]>

Reply via email to