[PEAK] Re: GFs + Events (Was [Python-3000] pep 3124 plans)

[Phillip J. Eby]

At 06:48 AM 7/24/2007 +0100, tav wrote:
>>Not as much as you might think.  If you check out this month's
>>archives of the PEAK mailing list, however, you'll see lots of posts
>>about the new event system I'm doing, called the Trellis, which
>>eliminates the need for explicit publishing and subscribing in order
>>to do event publishing.  It will certainly work with generic
>>functions, but they aren't actually that necessary.  The main place
>>that GF's will be used with the Trellis is in the ORM support layer.
>
>as always, nice work!
>
>perhaps i should have simply said event systems in relation to generic
>functions.
>
>for example, pattern matching (rule sets) limiting event propagation
>(whether through hub-and-spoke or otherwise) is quite similar to what
>RuleDispatch does, no?

Yes, but in the case of the Trellis, such event propagation is 
effectively done by making the dispatch tree part of the trellis 
itself -- by creating cells that reflect a condition by depending on 
more basic conditions.  So it's GF-like, but operating in "parallel", 
able to distribute events to a group simultaneously.

This is a bit more useful for an event-driven system, especially 
something like a message router/bus.  Each client can request an 
event set matching their criteria, and the events propagate 
efficiently to the minimum set of clients needing it.

In fact, such a system is also quite similar to a Linda "tuple space" 
(also implemented by JavaSpaces, PyLinda, and quite a few other 
projects, open source and otherwise).  Of course, to truly implement 
such a thing, you'd need a few other features (like leases and 
transactions) that I haven't thought in depth about.  Yet.  :)