Re: [Webmacro-devel] More pruning

[Brian G. <br...@qu...>]

>>I think we need some solid use cases first, and then can come up with a 
>>strategy.  Care to describe what your current tools do?
>
>Just because we haven't written X doesn't mean we shouldn't allow X to 
>work well.

Only if we have already
  - agreed what X is AND
  - agreed that X is something we want to work well.

Right now, it all feels very abstract to me.  The fact that I've not yet 
seen a use case that needs it argues that it may well be needless 
abstraction.

>+1 for getting rid of destroy(), but only if it is replaced with a 
>listener mechanism so that CTs can be told when they are no longer needed 
>by WM, without waiting for GC.

Bear in mind that these mechanisms have a cost.  You have to keep track of 
your listeners, and you force users into a mode of explicit 
deallocation.  I don't want users to have to call Context.destroy -- I want 
to let it get garbage collected.

>-1  for getting rid of destroy() without a replacement. finalizers are not 
>good enough. Components must have the opportunity to tap into their lifecycle.

Lifecycles really only make sense for long-lived objects, like 
Brokers.  Contexts should not have a lifecycle -- the overhead of 
maintaining and managing it is too great.

>I think the listener approach gives us the compromise we desire - better 
>performance while retaining useful functionality.

I think you underestimate the cost of the listener approach -- its pretty 
much the whole recycling/pooling crap we want to get rid of.



--
Brian Goetz
Quiotix Corporation
br...@qu...           Tel: 650-843-1300            Fax: 650-324-8032

http://www.quiotix.com